The Concepts section helps you learn about the parts of the Kubernetes system and the abstractions Kubernetes uses to represent your cluster, and helps you obtain a deeper understanding of how Kubernetes works.
This is the multi-page printable view of this section. Click here to print.
Concepts
- 1: Overview
- 1.1: Kubernetes Components
- 1.2: Objects In Kubernetes
- 1.2.1: Kubernetes Object Management
- 1.2.2: Object Names and IDs
- 1.2.3: Labels and Selectors
- 1.2.4: Namespaces
- 1.2.5: Annotations
- 1.2.6: Field Selectors
- 1.2.7: Finalizers
- 1.2.8: Owners and Dependents
- 1.2.9: Recommended Labels
- 1.3: The Kubernetes API
- 2: Cluster Architecture
- 2.1: Nodes
- 2.2: Communication between Nodes and the Control Plane
- 2.3: Controllers
- 2.4: Leases
- 2.5: Cloud Controller Manager
- 2.6: About cgroup v2
- 2.7: Container Runtime Interface (CRI)
- 2.8: Garbage Collection
- 2.9: Mixed Version Proxy
- 3: Containers
- 3.1: Images
- 3.2: Container Environment
- 3.3: Runtime Class
- 3.4: Container Lifecycle Hooks
- 4: Workloads
- 4.1: Pods
- 4.1.1: Pod Lifecycle
- 4.1.2: Init Containers
- 4.1.3: Sidecar Containers
- 4.1.4: Ephemeral Containers
- 4.1.5: Disruptions
- 4.1.6: Pod Quality of Service Classes
- 4.1.7: User Namespaces
- 4.1.8: Downward API
- 4.2: Workload Management
- 4.2.1: Deployments
- 4.2.2: ReplicaSet
- 4.2.3: StatefulSets
- 4.2.4: DaemonSet
- 4.2.5: Jobs
- 4.2.6: Automatic Cleanup for Finished Jobs
- 4.2.7: CronJob
- 4.2.8: ReplicationController
- 4.3: Autoscaling Workloads
- 4.4: Managing Workloads
- 5: Services, Load Balancing, and Networking
- 5.1: Service
- 5.2: Ingress
- 5.3: Ingress Controllers
- 5.4: Gateway API
- 5.5: EndpointSlices
- 5.6: Network Policies
- 5.7: DNS for Services and Pods
- 5.8: IPv4/IPv6 dual-stack
- 5.9: Topology Aware Routing
- 5.10: Networking on Windows
- 5.11: Service ClusterIP allocation
- 5.12: Service Internal Traffic Policy
- 6: Storage
- 6.1: Volumes
- 6.2: Persistent Volumes
- 6.3: Projected Volumes
- 6.4: Ephemeral Volumes
- 6.5: Storage Classes
- 6.6: Volume Attributes Classes
- 6.7: Dynamic Volume Provisioning
- 6.8: Volume Snapshots
- 6.9: Volume Snapshot Classes
- 6.10: CSI Volume Cloning
- 6.11: Storage Capacity
- 6.12: Node-specific Volume Limits
- 6.13: Volume Health Monitoring
- 6.14: Windows Storage
- 7: Configuration
- 7.1: Configuration Best Practices
- 7.2: ConfigMaps
- 7.3: Secrets
- 7.4: Liveness, Readiness, and Startup Probes
- 7.5: Resource Management for Pods and Containers
- 7.6: Organizing Cluster Access Using kubeconfig Files
- 7.7: Resource Management for Windows nodes
- 8: Security
- 8.1: Cloud Native Security and Kubernetes
- 8.2: Pod Security Standards
- 8.3: Pod Security Admission
- 8.4: Service Accounts
- 8.5: Pod Security Policies
- 8.6: Security For Windows Nodes
- 8.7: Controlling Access to the Kubernetes API
- 8.8: Role Based Access Control Good Practices
- 8.9: Good practices for Kubernetes Secrets
- 8.10: Multi-tenancy
- 8.11: Hardening Guide - Authentication Mechanisms
- 8.12: Kubernetes API Server Bypass Risks
- 8.13: Linux kernel security constraints for Pods and containers
- 8.14: Security Checklist
- 8.15: Application Security Checklist
- 9: Policies
- 9.1: Limit Ranges
- 9.2: Resource Quotas
- 9.3: Process ID Limits And Reservations
- 9.4: Node Resource Managers
- 10: Scheduling, Preemption and Eviction
- 10.1: Kubernetes Scheduler
- 10.2: Assigning Pods to Nodes
- 10.3: Pod Overhead
- 10.4: Pod Scheduling Readiness
- 10.5: Pod Topology Spread Constraints
- 10.6: Taints and Tolerations
- 10.7: Scheduling Framework
- 10.8: Dynamic Resource Allocation
- 10.9: Scheduler Performance Tuning
- 10.10: Resource Bin Packing
- 10.11: Pod Priority and Preemption
- 10.12: Node-pressure Eviction
- 10.13: API-initiated Eviction
- 11: Cluster Administration
- 11.1: Node Shutdowns
- 11.2: Certificates
- 11.3: Cluster Networking
- 11.4: Logging Architecture
- 11.5: Metrics For Kubernetes System Components
- 11.6: Metrics for Kubernetes Object States
- 11.7: System Logs
- 11.8: Traces For Kubernetes System Components
- 11.9: Proxies in Kubernetes
- 11.10: API Priority and Fairness
- 11.11: Cluster Autoscaling
- 11.12: Installing Addons
- 11.13: Coordinated Leader Election
- 12: Windows in Kubernetes
- 13: Extending Kubernetes
- 13.1: Compute, Storage, and Networking Extensions
- 13.1.1: Network Plugins
- 13.1.2: Device Plugins
- 13.2: Extending the Kubernetes API
- 13.2.1: Custom Resources
- 13.2.2: Kubernetes API Aggregation Layer
- 13.3: Operator pattern
1 - Overview
This page is an overview of Kubernetes.
The name Kubernetes originates from Greek, meaning helmsman or pilot. K8s as an abbreviation results from counting the eight letters between the "K" and the "s". Google open-sourced the Kubernetes project in 2014. Kubernetes combines over 15 years of Google's experience running production workloads at scale with best-of-breed ideas and practices from the community.
Why you need Kubernetes and what it can do
Containers are a good way to bundle and run your applications. In a production environment, you need to manage the containers that run the applications and ensure that there is no downtime. For example, if a container goes down, another container needs to start. Wouldn't it be easier if this behavior was handled by a system?
That's how Kubernetes comes to the rescue! Kubernetes provides you with a framework to run distributed systems resiliently. It takes care of scaling and failover for your application, provides deployment patterns, and more. For example: Kubernetes can easily manage a canary deployment for your system.
Kubernetes provides you with:
- Service discovery and load balancing Kubernetes can expose a container using the DNS name or using their own IP address. If traffic to a container is high, Kubernetes is able to load balance and distribute the network traffic so that the deployment is stable.
- Storage orchestration Kubernetes allows you to automatically mount a storage system of your choice, such as local storages, public cloud providers, and more.
- Automated rollouts and rollbacks You can describe the desired state for your deployed containers using Kubernetes, and it can change the actual state to the desired state at a controlled rate. For example, you can automate Kubernetes to create new containers for your deployment, remove existing containers and adopt all their resources to the new container.
- Automatic bin packing You provide Kubernetes with a cluster of nodes that it can use to run containerized tasks. You tell Kubernetes how much CPU and memory (RAM) each container needs. Kubernetes can fit containers onto your nodes to make the best use of your resources.
- Self-healing Kubernetes restarts containers that fail, replaces containers, kills containers that don't respond to your user-defined health check, and doesn't advertise them to clients until they are ready to serve.
- Secret and configuration management Kubernetes lets you store and manage sensitive information, such as passwords, OAuth tokens, and SSH keys. You can deploy and update secrets and application configuration without rebuilding your container images, and without exposing secrets in your stack configuration.
- Batch execution In addition to services, Kubernetes can manage your batch and CI workloads, replacing containers that fail, if desired.
- Horizontal scaling Scale your application up and down with a simple command, with a UI, or automatically based on CPU usage.
- IPv4/IPv6 dual-stack Allocation of IPv4 and IPv6 addresses to Pods and Services
- Designed for extensibility Add features to your Kubernetes cluster without changing upstream source code.
What Kubernetes is not
Kubernetes is not a traditional, all-inclusive PaaS (Platform as a Service) system. Since Kubernetes operates at the container level rather than at the hardware level, it provides some generally applicable features common to PaaS offerings, such as deployment, scaling, load balancing, and lets users integrate their logging, monitoring, and alerting solutions. However, Kubernetes is not monolithic, and these default solutions are optional and pluggable. Kubernetes provides the building blocks for building developer platforms, but preserves user choice and flexibility where it is important.
Kubernetes:
- Does not limit the types of applications supported. Kubernetes aims to support an extremely diverse variety of workloads, including stateless, stateful, and data-processing workloads. If an application can run in a container, it should run great on Kubernetes.
- Does not deploy source code and does not build your application. Continuous Integration, Delivery, and Deployment (CI/CD) workflows are determined by organization cultures and preferences as well as technical requirements.
- Does not provide application-level services, such as middleware (for example, message buses), data-processing frameworks (for example, Spark), databases (for example, MySQL), caches, nor cluster storage systems (for example, Ceph) as built-in services. Such components can run on Kubernetes, and/or can be accessed by applications running on Kubernetes through portable mechanisms, such as the Open Service Broker.
- Does not dictate logging, monitoring, or alerting solutions. It provides some integrations as proof of concept, and mechanisms to collect and export metrics.
- Does not provide nor mandate a configuration language/system (for example, Jsonnet). It provides a declarative API that may be targeted by arbitrary forms of declarative specifications.
- Does not provide nor adopt any comprehensive machine configuration, maintenance, management, or self-healing systems.
- Additionally, Kubernetes is not a mere orchestration system. In fact, it eliminates the need for orchestration. The technical definition of orchestration is execution of a defined workflow: first do A, then B, then C. In contrast, Kubernetes comprises a set of independent, composable control processes that continuously drive the current state towards the provided desired state. It shouldn't matter how you get from A to C. Centralized control is also not required. This results in a system that is easier to use and more powerful, robust, resilient, and extensible.
Historical context for Kubernetes
Let's take a look at why Kubernetes is so useful by going back in time.
Traditional deployment era:
Early on, organizations ran applications on physical servers. There was no way to define resource boundaries for applications in a physical server, and this caused resource allocation issues. For example, if multiple applications run on a physical server, there can be instances where one application would take up most of the resources, and as a result, the other applications would underperform. A solution for this would be to run each application on a different physical server. But this did not scale as resources were underutilized, and it was expensive for organizations to maintain many physical servers.
Virtualized deployment era:
As a solution, virtualization was introduced. It allows you to run multiple Virtual Machines (VMs) on a single physical server's CPU. Virtualization allows applications to be isolated between VMs and provides a level of security as the information of one application cannot be freely accessed by another application.
Virtualization allows better utilization of resources in a physical server and allows better scalability because an application can be added or updated easily, reduces hardware costs, and much more. With virtualization you can present a set of physical resources as a cluster of disposable virtual machines.
Each VM is a full machine running all the components, including its own operating system, on top of the virtualized hardware.
Container deployment era:
Containers are similar to VMs, but they have relaxed isolation properties to share the Operating System (OS) among the applications. Therefore, containers are considered lightweight. Similar to a VM, a container has its own filesystem, share of CPU, memory, process space, and more. As they are decoupled from the underlying infrastructure, they are portable across clouds and OS distributions.
Containers have become popular because they provide extra benefits, such as:
- Agile application creation and deployment: increased ease and efficiency of container image creation compared to VM image use.
- Continuous development, integration, and deployment: provides for reliable and frequent container image build and deployment with quick and efficient rollbacks (due to image immutability).
- Dev and Ops separation of concerns: create application container images at build/release time rather than deployment time, thereby decoupling applications from infrastructure.
- Observability: not only surfaces OS-level information and metrics, but also application health and other signals.
- Environmental consistency across development, testing, and production: runs the same on a laptop as it does in the cloud.
- Cloud and OS distribution portability: runs on Ubuntu, RHEL, CoreOS, on-premises, on major public clouds, and anywhere else.
- Application-centric management: raises the level of abstraction from running an OS on virtual hardware to running an application on an OS using logical resources.
- Loosely coupled, distributed, elastic, liberated micro-services: applications are broken into smaller, independent pieces and can be deployed and managed dynamically – not a monolithic stack running on one big single-purpose machine.
- Resource isolation: predictable application performance.
- Resource utilization: high efficiency and density.
What's next
- Take a look at the Kubernetes Components
- Take a look at the The Kubernetes API
- Take a look at the Cluster Architecture
- Ready to Get Started?
1.1 - Kubernetes Components
This page provides a high-level overview of the essential components that make up a Kubernetes cluster.
Core Components
A Kubernetes cluster consists of a control plane and one or more worker nodes. Here's a brief overview of the main components:
Control Plane Components
Manage the overall state of the cluster:
- kube-apiserver
- The core component server that exposes the Kubernetes HTTP API
- etcd
- Consistent and highly-available key value store for all API server data
- kube-scheduler
- Looks for Pods not yet bound to a node, and assigns each Pod to a suitable node.
- kube-controller-manager
- Runs controllers to implement Kubernetes API behavior.
- cloud-controller-manager (optional)
- Integrates with underlying cloud provider(s).
Node Components
Run on every node, maintaining running pods and providing the Kubernetes runtime environment:
- kubelet
- Ensures that Pods are running, including their containers.
- kube-proxy (optional)
- Maintains network rules on nodes to implement Services.
- Container runtime
- Software responsible for running containers. Read Container Runtimes to learn more.
Your cluster may require additional software on each node; for example, you might also run systemd on a Linux node to supervise local components.
Addons
Addons extend the functionality of Kubernetes. A few important examples include:
- DNS
- For cluster-wide DNS resolution
- Web UI (Dashboard)
- For cluster management via a web interface
- Container Resource Monitoring
- For collecting and storing container metrics
- Cluster-level Logging
- For saving container logs to a central log store
Flexibility in Architecture
Kubernetes allows for flexibility in how these components are deployed and managed. The architecture can be adapted to various needs, from small development environments to large-scale production deployments.
For more detailed information about each component and various ways to configure your cluster architecture, see the Cluster Architecture page.
1.2 - Objects In Kubernetes
This page explains how Kubernetes objects are represented in the Kubernetes API, and how you can
express them in .yaml
format.
Understanding Kubernetes objects
Kubernetes objects are persistent entities in the Kubernetes system. Kubernetes uses these entities to represent the state of your cluster. Specifically, they can describe:
- What containerized applications are running (and on which nodes)
- The resources available to those applications
- The policies around how those applications behave, such as restart policies, upgrades, and fault-tolerance
A Kubernetes object is a "record of intent"--once you create the object, the Kubernetes system will constantly work to ensure that the object exists. By creating an object, you're effectively telling the Kubernetes system what you want your cluster's workload to look like; this is your cluster's desired state.
To work with Kubernetes objects—whether to create, modify, or delete them—you'll need to use the
Kubernetes API. When you use the kubectl
command-line
interface, for example, the CLI makes the necessary Kubernetes API calls for you. You can also use
the Kubernetes API directly in your own programs using one of the
Client Libraries.
Object spec and status
Almost every Kubernetes object includes two nested object fields that govern
the object's configuration: the object spec
and the object status
.
For objects that have a spec
, you have to set this when you create the object,
providing a description of the characteristics you want the resource to have:
its desired state.
The status
describes the current state of the object, supplied and updated
by the Kubernetes system and its components. The Kubernetes
control plane continually
and actively manages every object's actual state to match the desired state you
supplied.
For example: in Kubernetes, a Deployment is an object that can represent an
application running on your cluster. When you create the Deployment, you
might set the Deployment spec
to specify that you want three replicas of
the application to be running. The Kubernetes system reads the Deployment
spec and starts three instances of your desired application--updating
the status to match your spec. If any of those instances should fail
(a status change), the Kubernetes system responds to the difference
between spec and status by making a correction--in this case, starting
a replacement instance.
For more information on the object spec, status, and metadata, see the Kubernetes API Conventions.
Describing a Kubernetes object
When you create an object in Kubernetes, you must provide the object spec that describes its
desired state, as well as some basic information about the object (such as a name). When you use
the Kubernetes API to create the object (either directly or via kubectl
), that API request must
include that information as JSON in the request body.
Most often, you provide the information to kubectl
in a file known as a manifest.
By convention, manifests are YAML (you could also use JSON format).
Tools such as kubectl
convert the information from a manifest into JSON or another supported
serialization format when making the API request over HTTP.
Here's an example manifest that shows the required fields and object spec for a Kubernetes Deployment:
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-deployment
spec:
selector:
matchLabels:
app: nginx
replicas: 2 # tells deployment to run 2 pods matching the template
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
One way to create a Deployment using a manifest file like the one above is to use the
kubectl apply
command
in the kubectl
command-line interface, passing the .yaml
file as an argument. Here's an example:
kubectl apply -f https://k8s.io/examples/application/deployment.yaml
The output is similar to this:
deployment.apps/nginx-deployment created
Required fields
In the manifest (YAML or JSON file) for the Kubernetes object you want to create, you'll need to set values for the following fields:
apiVersion
- Which version of the Kubernetes API you're using to create this objectkind
- What kind of object you want to createmetadata
- Data that helps uniquely identify the object, including aname
string,UID
, and optionalnamespace
spec
- What state you desire for the object
The precise format of the object spec
is different for every Kubernetes object, and contains
nested fields specific to that object. The Kubernetes API Reference
can help you find the spec format for all of the objects you can create using Kubernetes.
For example, see the spec
field
for the Pod API reference.
For each Pod, the .spec
field specifies the pod and its desired state (such as the container image name for
each container within that pod).
Another example of an object specification is the
spec
field
for the StatefulSet API. For StatefulSet, the .spec
field specifies the StatefulSet and
its desired state.
Within the .spec
of a StatefulSet is a template
for Pod objects. That template describes Pods that the StatefulSet controller will create in order to
satisfy the StatefulSet specification.
Different kinds of objects can also have different .status
; again, the API reference pages
detail the structure of that .status
field, and its content for each different type of object.
Note:
See Configuration Best Practices for additional information on writing YAML configuration files.Server side field validation
Starting with Kubernetes v1.25, the API server offers server side
field validation
that detects unrecognized or duplicate fields in an object. It provides all the functionality
of kubectl --validate
on the server side.
The kubectl
tool uses the --validate
flag to set the level of field validation. It accepts the
values ignore
, warn
, and strict
while also accepting the values true
(equivalent to strict
)
and false
(equivalent to ignore
). The default validation setting for kubectl
is --validate=true
.
Strict
- Strict field validation, errors on validation failure
Warn
- Field validation is performed, but errors are exposed as warnings rather than failing the request
Ignore
- No server side field validation is performed
When kubectl
cannot connect to an API server that supports field validation it will fall back
to using client-side validation. Kubernetes 1.27 and later versions always offer field validation;
older Kubernetes releases might not. If your cluster is older than v1.27, check the documentation
for your version of Kubernetes.
What's next
If you're new to Kubernetes, read more about the following:
- Pods which are the most important basic Kubernetes objects.
- Deployment objects.
- Controllers in Kubernetes.
- kubectl and kubectl commands.
Kubernetes Object Management
explains how to use kubectl
to manage objects.
You might need to install kubectl if you don't already have it available.
To learn about the Kubernetes API in general, visit:
To learn about objects in Kubernetes in more depth, read other pages in this section:
1.2.1 - Kubernetes Object Management
The kubectl
command-line tool supports several different ways to create and manage
Kubernetes objects. This document provides an overview of the different
approaches. Read the Kubectl book for
details of managing objects by Kubectl.
Management techniques
Warning:
A Kubernetes object should be managed using only one technique. Mixing and matching techniques for the same object results in undefined behavior.Management technique | Operates on | Recommended environment | Supported writers | Learning curve |
---|---|---|---|---|
Imperative commands | Live objects | Development projects | 1+ | Lowest |
Imperative object configuration | Individual files | Production projects | 1 | Moderate |
Declarative object configuration | Directories of files | Production projects | 1+ | Highest |
Imperative commands
When using imperative commands, a user operates directly on live objects
in a cluster. The user provides operations to
the kubectl
command as arguments or flags.
This is the recommended way to get started or to run a one-off task in a cluster. Because this technique operates directly on live objects, it provides no history of previous configurations.
Examples
Run an instance of the nginx container by creating a Deployment object:
kubectl create deployment nginx --image nginx
Trade-offs
Advantages compared to object configuration:
- Commands are expressed as a single action word.
- Commands require only a single step to make changes to the cluster.
Disadvantages compared to object configuration:
- Commands do not integrate with change review processes.
- Commands do not provide an audit trail associated with changes.
- Commands do not provide a source of records except for what is live.
- Commands do not provide a template for creating new objects.
Imperative object configuration
In imperative object configuration, the kubectl command specifies the operation (create, replace, etc.), optional flags and at least one file name. The file specified must contain a full definition of the object in YAML or JSON format.
See the API reference for more details on object definitions.
Warning:
The imperativereplace
command replaces the existing
spec with the newly provided one, dropping all changes to the object missing from
the configuration file. This approach should not be used with resource
types whose specs are updated independently of the configuration file.
Services of type LoadBalancer
, for example, have their externalIPs
field updated
independently from the configuration by the cluster.Examples
Create the objects defined in a configuration file:
kubectl create -f nginx.yaml
Delete the objects defined in two configuration files:
kubectl delete -f nginx.yaml -f redis.yaml
Update the objects defined in a configuration file by overwriting the live configuration:
kubectl replace -f nginx.yaml
Trade-offs
Advantages compared to imperative commands:
- Object configuration can be stored in a source control system such as Git.
- Object configuration can integrate with processes such as reviewing changes before push and audit trails.
- Object configuration provides a template for creating new objects.
Disadvantages compared to imperative commands:
- Object configuration requires basic understanding of the object schema.
- Object configuration requires the additional step of writing a YAML file.
Advantages compared to declarative object configuration:
- Imperative object configuration behavior is simpler and easier to understand.
- As of Kubernetes version 1.5, imperative object configuration is more mature.
Disadvantages compared to declarative object configuration:
- Imperative object configuration works best on files, not directories.
- Updates to live objects must be reflected in configuration files, or they will be lost during the next replacement.
Declarative object configuration
When using declarative object configuration, a user operates on object
configuration files stored locally, however the user does not define the
operations to be taken on the files. Create, update, and delete operations
are automatically detected per-object by kubectl
. This enables working on
directories, where different operations might be needed for different objects.
Note:
Declarative object configuration retains changes made by other writers, even if the changes are not merged back to the object configuration file. This is possible by using thepatch
API operation to write only
observed differences, instead of using the replace
API operation to replace the entire object configuration.Examples
Process all object configuration files in the configs
directory, and create or
patch the live objects. You can first diff
to see what changes are going to be
made, and then apply:
kubectl diff -f configs/
kubectl apply -f configs/
Recursively process directories:
kubectl diff -R -f configs/
kubectl apply -R -f configs/
Trade-offs
Advantages compared to imperative object configuration:
- Changes made directly to live objects are retained, even if they are not merged back into the configuration files.
- Declarative object configuration has better support for operating on directories and automatically detecting operation types (create, patch, delete) per-object.
Disadvantages compared to imperative object configuration:
- Declarative object configuration is harder to debug and understand results when they are unexpected.
- Partial updates using diffs create complex merge and patch operations.
What's next
- Managing Kubernetes Objects Using Imperative Commands
- Imperative Management of Kubernetes Objects Using Configuration Files
- Declarative Management of Kubernetes Objects Using Configuration Files
- Declarative Management of Kubernetes Objects Using Kustomize
- Kubectl Command Reference
- Kubectl Book
- Kubernetes API Reference
1.2.2 - Object Names and IDs
Each object in your cluster has a Name that is unique for that type of resource. Every Kubernetes object also has a UID that is unique across your whole cluster.
For example, you can only have one Pod named myapp-1234
within the same namespace, but you can have one Pod and one Deployment that are each named myapp-1234
.
For non-unique user-provided attributes, Kubernetes provides labels and annotations.
Names
A client-provided string that refers to an object in a resource URL, such as /api/v1/pods/some-name
.
Only one object of a given kind can have a given name at a time. However, if you delete the object, you can make a new object with the same name.
Names must be unique across all API versions of the same resource. API resources are distinguished by their API group, resource type, namespace (for namespaced resources), and name. In other words, API version is irrelevant in this context.
Note:
In cases when objects represent a physical entity, like a Node representing a physical host, when the host is re-created under the same name without deleting and re-creating the Node, Kubernetes treats the new host as the old one, which may lead to inconsistencies.Below are four types of commonly used name constraints for resources.
DNS Subdomain Names
Most resource types require a name that can be used as a DNS subdomain name as defined in RFC 1123. This means the name must:
- contain no more than 253 characters
- contain only lowercase alphanumeric characters, '-' or '.'
- start with an alphanumeric character
- end with an alphanumeric character
RFC 1123 Label Names
Some resource types require their names to follow the DNS label standard as defined in RFC 1123. This means the name must:
- contain at most 63 characters
- contain only lowercase alphanumeric characters or '-'
- start with an alphanumeric character
- end with an alphanumeric character
RFC 1035 Label Names
Some resource types require their names to follow the DNS label standard as defined in RFC 1035. This means the name must:
- contain at most 63 characters
- contain only lowercase alphanumeric characters or '-'
- start with an alphabetic character
- end with an alphanumeric character
Note:
The only difference between the RFC 1035 and RFC 1123 label standards is that RFC 1123 labels are allowed to start with a digit, whereas RFC 1035 labels can start with a lowercase alphabetic character only.Path Segment Names
Some resource types require their names to be able to be safely encoded as a path segment. In other words, the name may not be "." or ".." and the name may not contain "/" or "%".
Here's an example manifest for a Pod named nginx-demo
.
apiVersion: v1
kind: Pod
metadata:
name: nginx-demo
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
Note:
Some resource types have additional restrictions on their names.UIDs
A Kubernetes systems-generated string to uniquely identify objects.
Every object created over the whole lifetime of a Kubernetes cluster has a distinct UID. It is intended to distinguish between historical occurrences of similar entities.
Kubernetes UIDs are universally unique identifiers (also known as UUIDs). UUIDs are standardized as ISO/IEC 9834-8 and as ITU-T X.667.
What's next
- Read about labels and annotations in Kubernetes.
- See the Identifiers and Names in Kubernetes design document.
1.2.3 - Labels and Selectors
Labels are key/value pairs that are attached to objects such as Pods. Labels are intended to be used to specify identifying attributes of objects that are meaningful and relevant to users, but do not directly imply semantics to the core system. Labels can be used to organize and to select subsets of objects. Labels can be attached to objects at creation time and subsequently added and modified at any time. Each object can have a set of key/value labels defined. Each Key must be unique for a given object.
"metadata": {
"labels": {
"key1" : "value1",
"key2" : "value2"
}
}
Labels allow for efficient queries and watches and are ideal for use in UIs and CLIs. Non-identifying information should be recorded using annotations.
Motivation
Labels enable users to map their own organizational structures onto system objects in a loosely coupled fashion, without requiring clients to store these mappings.
Service deployments and batch processing pipelines are often multi-dimensional entities (e.g., multiple partitions or deployments, multiple release tracks, multiple tiers, multiple micro-services per tier). Management often requires cross-cutting operations, which breaks encapsulation of strictly hierarchical representations, especially rigid hierarchies determined by the infrastructure rather than by users.
Example labels:
"release" : "stable"
,"release" : "canary"
"environment" : "dev"
,"environment" : "qa"
,"environment" : "production"
"tier" : "frontend"
,"tier" : "backend"
,"tier" : "cache"
"partition" : "customerA"
,"partition" : "customerB"
"track" : "daily"
,"track" : "weekly"
These are examples of commonly used labels; you are free to develop your own conventions. Keep in mind that label Key must be unique for a given object.
Syntax and character set
Labels are key/value pairs. Valid label keys have two segments: an optional
prefix and name, separated by a slash (/
). The name segment is required and
must be 63 characters or less, beginning and ending with an alphanumeric
character ([a-z0-9A-Z]
) with dashes (-
), underscores (_
), dots (.
),
and alphanumerics between. The prefix is optional. If specified, the prefix
must be a DNS subdomain: a series of DNS labels separated by dots (.
),
not longer than 253 characters in total, followed by a slash (/
).
If the prefix is omitted, the label Key is presumed to be private to the user.
Automated system components (e.g. kube-scheduler
, kube-controller-manager
,
kube-apiserver
, kubectl
, or other third-party automation) which add labels
to end-user objects must specify a prefix.
The kubernetes.io/
and k8s.io/
prefixes are
reserved for Kubernetes core components.
Valid label value:
- must be 63 characters or less (can be empty),
- unless empty, must begin and end with an alphanumeric character (
[a-z0-9A-Z]
), - could contain dashes (
-
), underscores (_
), dots (.
), and alphanumerics between.
For example, here's a manifest for a Pod that has two labels
environment: production
and app: nginx
:
apiVersion: v1
kind: Pod
metadata:
name: label-demo
labels:
environment: production
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
Label selectors
Unlike names and UIDs, labels do not provide uniqueness. In general, we expect many objects to carry the same label(s).
Via a label selector, the client/user can identify a set of objects. The label selector is the core grouping primitive in Kubernetes.
The API currently supports two types of selectors: equality-based and set-based.
A label selector can be made of multiple requirements which are comma-separated.
In the case of multiple requirements, all must be satisfied so the comma separator
acts as a logical AND (&&
) operator.
The semantics of empty or non-specified selectors are dependent on the context, and API types that use selectors should document the validity and meaning of them.
Note:
For some API types, such as ReplicaSets, the label selectors of two instances must not overlap within a namespace, or the controller can see that as conflicting instructions and fail to determine how many replicas should be present.Caution:
For both equality-based and set-based conditions there is no logical OR (||
) operator.
Ensure your filter statements are structured accordingly.Equality-based requirement
Equality- or inequality-based requirements allow filtering by label keys and values.
Matching objects must satisfy all of the specified label constraints, though they may
have additional labels as well. Three kinds of operators are admitted =
,==
,!=
.
The first two represent equality (and are synonyms), while the latter represents inequality.
For example:
environment = production
tier != frontend
The former selects all resources with key equal to environment
and value equal to production
.
The latter selects all resources with key equal to tier
and value distinct from frontend
,
and all resources with no labels with the tier
key. One could filter for resources in production
excluding frontend
using the comma operator: environment=production,tier!=frontend
One usage scenario for equality-based label requirement is for Pods to specify
node selection criteria. For example, the sample Pod below selects nodes where
the accelerator
label exists and is set to nvidia-tesla-p100
.
apiVersion: v1
kind: Pod
metadata:
name: cuda-test
spec:
containers:
- name: cuda-test
image: "registry.k8s.io/cuda-vector-add:v0.1"
resources:
limits:
nvidia.com/gpu: 1
nodeSelector:
accelerator: nvidia-tesla-p100
Set-based requirement
Set-based label requirements allow filtering keys according to a set of values.
Three kinds of operators are supported: in
,notin
and exists
(only the key identifier).
For example:
environment in (production, qa)
tier notin (frontend, backend)
partition
!partition
- The first example selects all resources with key equal to
environment
and value equal toproduction
orqa
. - The second example selects all resources with key equal to
tier
and values other thanfrontend
andbackend
, and all resources with no labels with thetier
key. - The third example selects all resources including a label with key
partition
; no values are checked. - The fourth example selects all resources without a label with key
partition
; no values are checked.
Similarly the comma separator acts as an AND operator. So filtering resources
with a partition
key (no matter the value) and with environment
different
than qa
can be achieved using partition,environment notin (qa)
.
The set-based label selector is a general form of equality since
environment=production
is equivalent to environment in (production)
;
similarly for !=
and notin
.
Set-based requirements can be mixed with equality-based requirements.
For example: partition in (customerA, customerB),environment!=qa
.
API
LIST and WATCH filtering
For list and watch operations, you can specify label selectors to filter the sets of objects returned; you specify the filter using a query parameter. (To learn in detail about watches in Kubernetes, read efficient detection of changes). Both requirements are permitted (presented here as they would appear in a URL query string):
- equality-based requirements:
?labelSelector=environment%3Dproduction,tier%3Dfrontend
- set-based requirements:
?labelSelector=environment+in+%28production%2Cqa%29%2Ctier+in+%28frontend%29
Both label selector styles can be used to list or watch resources via a REST client.
For example, targeting apiserver
with kubectl
and using equality-based one may write:
kubectl get pods -l environment=production,tier=frontend
or using set-based requirements:
kubectl get pods -l 'environment in (production),tier in (frontend)'
As already mentioned set-based requirements are more expressive. For instance, they can implement the OR operator on values:
kubectl get pods -l 'environment in (production, qa)'
or restricting negative matching via notin operator:
kubectl get pods -l 'environment,environment notin (frontend)'
Set references in API objects
Some Kubernetes objects, such as services
and replicationcontrollers
,
also use label selectors to specify sets of other resources, such as
pods.
Service and ReplicationController
The set of pods that a service
targets is defined with a label selector.
Similarly, the population of pods that a replicationcontroller
should
manage is also defined with a label selector.
Label selectors for both objects are defined in json
or yaml
files using maps,
and only equality-based requirement selectors are supported:
"selector": {
"component" : "redis",
}
or
selector:
component: redis
This selector (respectively in json
or yaml
format) is equivalent to
component=redis
or component in (redis)
.
Resources that support set-based requirements
Newer resources, such as Job
,
Deployment
,
ReplicaSet
, and
DaemonSet
,
support set-based requirements as well.
selector:
matchLabels:
component: redis
matchExpressions:
- { key: tier, operator: In, values: [cache] }
- { key: environment, operator: NotIn, values: [dev] }
matchLabels
is a map of {key,value}
pairs. A single {key,value}
in the
matchLabels
map is equivalent to an element of matchExpressions
, whose key
field is "key", the operator
is "In", and the values
array contains only "value".
matchExpressions
is a list of pod selector requirements. Valid operators include
In, NotIn, Exists, and DoesNotExist. The values set must be non-empty in the case of
In and NotIn. All of the requirements, from both matchLabels
and matchExpressions
are ANDed together -- they must all be satisfied in order to match.
Selecting sets of nodes
One use case for selecting over labels is to constrain the set of nodes onto which a pod can schedule. See the documentation on node selection for more information.
Using labels effectively
You can apply a single label to any resources, but this is not always the best practice. There are many scenarios where multiple labels should be used to distinguish resource sets from one another.
For instance, different applications would use different values for the app
label, but a
multi-tier application, such as the guestbook example,
would additionally need to distinguish each tier. The frontend could carry the following labels:
labels:
app: guestbook
tier: frontend
while the Redis master and replica would have different tier
labels, and perhaps even an
additional role
label:
labels:
app: guestbook
tier: backend
role: master
and
labels:
app: guestbook
tier: backend
role: replica
The labels allow for slicing and dicing the resources along any dimension specified by a label:
kubectl apply -f examples/guestbook/all-in-one/guestbook-all-in-one.yaml
kubectl get pods -Lapp -Ltier -Lrole
NAME READY STATUS RESTARTS AGE APP TIER ROLE
guestbook-fe-4nlpb 1/1 Running 0 1m guestbook frontend <none>
guestbook-fe-ght6d 1/1 Running 0 1m guestbook frontend <none>
guestbook-fe-jpy62 1/1 Running 0 1m guestbook frontend <none>
guestbook-redis-master-5pg3b 1/1 Running 0 1m guestbook backend master
guestbook-redis-replica-2q2yf 1/1 Running 0 1m guestbook backend replica
guestbook-redis-replica-qgazl 1/1 Running 0 1m guestbook backend replica
my-nginx-divi2 1/1 Running 0 29m nginx <none> <none>
my-nginx-o0ef1 1/1 Running 0 29m nginx <none> <none>
kubectl get pods -lapp=guestbook,role=replica
NAME READY STATUS RESTARTS AGE
guestbook-redis-replica-2q2yf 1/1 Running 0 3m
guestbook-redis-replica-qgazl 1/1 Running 0 3m
Updating labels
Sometimes you may want to relabel existing pods and other resources before creating
new resources. This can be done with kubectl label
.
For example, if you want to label all your NGINX Pods as frontend tier, run:
kubectl label pods -l app=nginx tier=fe
pod/my-nginx-2035384211-j5fhi labeled
pod/my-nginx-2035384211-u2c7e labeled
pod/my-nginx-2035384211-u3t6x labeled
This first filters all pods with the label "app=nginx", and then labels them with the "tier=fe". To see the pods you labeled, run:
kubectl get pods -l app=nginx -L tier
NAME READY STATUS RESTARTS AGE TIER
my-nginx-2035384211-j5fhi 1/1 Running 0 23m fe
my-nginx-2035384211-u2c7e 1/1 Running 0 23m fe
my-nginx-2035384211-u3t6x 1/1 Running 0 23m fe
This outputs all "app=nginx" pods, with an additional label column of pods' tier
(specified with -L
or --label-columns
).
For more information, please see kubectl label.
What's next
1.2.4 - Namespaces
In Kubernetes, namespaces provide a mechanism for isolating groups of resources within a single cluster. Names of resources need to be unique within a namespace, but not across namespaces. Namespace-based scoping is applicable only for namespaced objects (e.g. Deployments, Services, etc.) and not for cluster-wide objects (e.g. StorageClass, Nodes, PersistentVolumes, etc.).
When to Use Multiple Namespaces
Namespaces are intended for use in environments with many users spread across multiple teams, or projects. For clusters with a few to tens of users, you should not need to create or think about namespaces at all. Start using namespaces when you need the features they provide.
Namespaces provide a scope for names. Names of resources need to be unique within a namespace, but not across namespaces. Namespaces cannot be nested inside one another and each Kubernetes resource can only be in one namespace.
Namespaces are a way to divide cluster resources between multiple users (via resource quota).
It is not necessary to use multiple namespaces to separate slightly different resources, such as different versions of the same software: use labels to distinguish resources within the same namespace.
Note:
For a production cluster, consider not using thedefault
namespace. Instead, make other namespaces and use those.Initial namespaces
Kubernetes starts with four initial namespaces:
default
- Kubernetes includes this namespace so that you can start using your new cluster without first creating a namespace.
kube-node-lease
- This namespace holds Lease objects associated with each node. Node leases allow the kubelet to send heartbeats so that the control plane can detect node failure.
kube-public
- This namespace is readable by all clients (including those not authenticated). This namespace is mostly reserved for cluster usage, in case that some resources should be visible and readable publicly throughout the whole cluster. The public aspect of this namespace is only a convention, not a requirement.
kube-system
- The namespace for objects created by the Kubernetes system.
Working with Namespaces
Creation and deletion of namespaces are described in the Admin Guide documentation for namespaces.
Note:
Avoid creating namespaces with the prefixkube-
, since it is reserved for Kubernetes system namespaces.Viewing namespaces
You can list the current namespaces in a cluster using:
kubectl get namespace
NAME STATUS AGE
default Active 1d
kube-node-lease Active 1d
kube-public Active 1d
kube-system Active 1d
Setting the namespace for a request
To set the namespace for a current request, use the --namespace
flag.
For example:
kubectl run nginx --image=nginx --namespace=<insert-namespace-name-here>
kubectl get pods --namespace=<insert-namespace-name-here>
Setting the namespace preference
You can permanently save the namespace for all subsequent kubectl commands in that context.
kubectl config set-context --current --namespace=<insert-namespace-name-here>
# Validate it
kubectl config view --minify | grep namespace:
Namespaces and DNS
When you create a Service,
it creates a corresponding DNS entry.
This entry is of the form <service-name>.<namespace-name>.svc.cluster.local
, which means
that if a container only uses <service-name>
, it will resolve to the service which
is local to a namespace. This is useful for using the same configuration across
multiple namespaces such as Development, Staging and Production. If you want to reach
across namespaces, you need to use the fully qualified domain name (FQDN).
As a result, all namespace names must be valid RFC 1123 DNS labels.
Warning:
By creating namespaces with the same name as public top-level domains, Services in these namespaces can have short DNS names that overlap with public DNS records. Workloads from any namespace performing a DNS lookup without a trailing dot will be redirected to those services, taking precedence over public DNS.
To mitigate this, limit privileges for creating namespaces to trusted users. If required, you could additionally configure third-party security controls, such as admission webhooks, to block creating any namespace with the name of public TLDs.
Not all objects are in a namespace
Most Kubernetes resources (e.g. pods, services, replication controllers, and others) are in some namespaces. However namespace resources are not themselves in a namespace. And low-level resources, such as nodes and persistentVolumes, are not in any namespace.
To see which Kubernetes resources are and aren't in a namespace:
# In a namespace
kubectl api-resources --namespaced=true
# Not in a namespace
kubectl api-resources --namespaced=false
Automatic labelling
Kubernetes 1.22 [stable]
The Kubernetes control plane sets an immutable label
kubernetes.io/metadata.name
on all namespaces.
The value of the label is the namespace name.
What's next
- Learn more about creating a new namespace.
- Learn more about deleting a namespace.
1.2.5 - Annotations
You can use Kubernetes annotations to attach arbitrary non-identifying metadata to objects. Clients such as tools and libraries can retrieve this metadata.
Attaching metadata to objects
You can use either labels or annotations to attach metadata to Kubernetes objects. Labels can be used to select objects and to find collections of objects that satisfy certain conditions. In contrast, annotations are not used to identify and select objects. The metadata in an annotation can be small or large, structured or unstructured, and can include characters not permitted by labels. It is possible to use labels as well as annotations in the metadata of the same object.
Annotations, like labels, are key/value maps:
"metadata": {
"annotations": {
"key1" : "value1",
"key2" : "value2"
}
}
Note:
The keys and the values in the map must be strings. In other words, you cannot use numeric, boolean, list or other types for either the keys or the values.Here are some examples of information that could be recorded in annotations:
Fields managed by a declarative configuration layer. Attaching these fields as annotations distinguishes them from default values set by clients or servers, and from auto-generated fields and fields set by auto-sizing or auto-scaling systems.
Build, release, or image information like timestamps, release IDs, git branch, PR numbers, image hashes, and registry address.
Pointers to logging, monitoring, analytics, or audit repositories.
Client library or tool information that can be used for debugging purposes: for example, name, version, and build information.
User or tool/system provenance information, such as URLs of related objects from other ecosystem components.
Lightweight rollout tool metadata: for example, config or checkpoints.
Phone or pager numbers of persons responsible, or directory entries that specify where that information can be found, such as a team web site.
Directives from the end-user to the implementations to modify behavior or engage non-standard features.
Instead of using annotations, you could store this type of information in an external database or directory, but that would make it much harder to produce shared client libraries and tools for deployment, management, introspection, and the like.
Syntax and character set
Annotations are key/value pairs. Valid annotation keys have two segments: an optional prefix and name, separated by a slash (/
). The name segment is required and must be 63 characters or less, beginning and ending with an alphanumeric character ([a-z0-9A-Z]
) with dashes (-
), underscores (_
), dots (.
), and alphanumerics between. The prefix is optional. If specified, the prefix must be a DNS subdomain: a series of DNS labels separated by dots (.
), not longer than 253 characters in total, followed by a slash (/
).
If the prefix is omitted, the annotation Key is presumed to be private to the user. Automated system components (e.g. kube-scheduler
, kube-controller-manager
, kube-apiserver
, kubectl
, or other third-party automation) which add annotations to end-user objects must specify a prefix.
The kubernetes.io/
and k8s.io/
prefixes are reserved for Kubernetes core components.
For example, here's a manifest for a Pod that has the annotation imageregistry: https://hub.docker.com/
:
apiVersion: v1
kind: Pod
metadata:
name: annotations-demo
annotations:
imageregistry: "https://hub.docker.com/"
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
What's next
- Learn more about Labels and Selectors.
- Find Well-known labels, Annotations and Taints
1.2.6 - Field Selectors
Field selectors let you select Kubernetes objects based on the value of one or more resource fields. Here are some examples of field selector queries:
metadata.name=my-service
metadata.namespace!=default
status.phase=Pending
This kubectl
command selects all Pods for which the value of the status.phase
field is Running
:
kubectl get pods --field-selector status.phase=Running
Note:
Field selectors are essentially resource filters. By default, no selectors/filters are applied, meaning that all resources of the specified type are selected. This makes thekubectl
queries kubectl get pods
and kubectl get pods --field-selector ""
equivalent.Supported fields
Supported field selectors vary by Kubernetes resource type. All resource types support the metadata.name
and metadata.namespace
fields. Using unsupported field selectors produces an error. For example:
kubectl get ingress --field-selector foo.bar=baz
Error from server (BadRequest): Unable to find "ingresses" that match label selector "", field selector "foo.bar=baz": "foo.bar" is not a known field selector: only "metadata.name", "metadata.namespace"
List of supported fields
Kind | Fields |
---|---|
Pod | spec.nodeName spec.restartPolicy spec.schedulerName spec.serviceAccountName spec.hostNetwork status.phase status.podIP status.nominatedNodeName |
Event | involvedObject.kind involvedObject.namespace involvedObject.name involvedObject.uid involvedObject.apiVersion involvedObject.resourceVersion involvedObject.fieldPath reason reportingComponent source type |
Secret | type |
Namespace | status.phase |
ReplicaSet | status.replicas |
ReplicationController | status.replicas |
Job | status.successful |
Node | spec.unschedulable |
CertificateSigningRequest | spec.signerName |
Supported operators
You can use the =
, ==
, and !=
operators with field selectors (=
and ==
mean the same thing). This kubectl
command, for example, selects all Kubernetes Services that aren't in the default
namespace:
kubectl get services --all-namespaces --field-selector metadata.namespace!=default
Chained selectors
As with label and other selectors, field selectors can be chained together as a comma-separated list. This kubectl
command selects all Pods for which the status.phase
does not equal Running
and the spec.restartPolicy
field equals Always
:
kubectl get pods --field-selector=status.phase!=Running,spec.restartPolicy=Always
Multiple resource types
You can use field selectors across multiple resource types. This kubectl
command selects all Statefulsets and Services that are not in the default
namespace:
kubectl get statefulsets,services --all-namespaces --field-selector metadata.namespace!=default
1.2.7 - Finalizers
Finalizers are namespaced keys that tell Kubernetes to wait until specific conditions are met before it fully deletes resources marked for deletion. Finalizers alert controllers to clean up resources the deleted object owned.
When you tell Kubernetes to delete an object that has finalizers specified for
it, the Kubernetes API marks the object for deletion by populating .metadata.deletionTimestamp
,
and returns a 202
status code (HTTP "Accepted"). The target object remains in a terminating state while the
control plane, or other components, take the actions defined by the finalizers.
After these actions are complete, the controller removes the relevant finalizers
from the target object. When the metadata.finalizers
field is empty,
Kubernetes considers the deletion complete and deletes the object.
You can use finalizers to control garbage collection of resources. For example, you can define a finalizer to clean up related resources or infrastructure before the controller deletes the target resource.
You can use finalizers to control garbage collection of objects by alerting controllers to perform specific cleanup tasks before deleting the target resource.
Finalizers don't usually specify the code to execute. Instead, they are typically lists of keys on a specific resource similar to annotations. Kubernetes specifies some finalizers automatically, but you can also specify your own.
How finalizers work
When you create a resource using a manifest file, you can specify finalizers in
the metadata.finalizers
field. When you attempt to delete the resource, the
API server handling the delete request notices the values in the finalizers
field
and does the following:
- Modifies the object to add a
metadata.deletionTimestamp
field with the time you started the deletion. - Prevents the object from being removed until all items are removed from its
metadata.finalizers
field - Returns a
202
status code (HTTP "Accepted")
The controller managing that finalizer notices the update to the object setting the
metadata.deletionTimestamp
, indicating deletion of the object has been requested.
The controller then attempts to satisfy the requirements of the finalizers
specified for that resource. Each time a finalizer condition is satisfied, the
controller removes that key from the resource's finalizers
field. When the
finalizers
field is emptied, an object with a deletionTimestamp
field set
is automatically deleted. You can also use finalizers to prevent deletion of unmanaged resources.
A common example of a finalizer is kubernetes.io/pv-protection
, which prevents
accidental deletion of PersistentVolume
objects. When a PersistentVolume
object is in use by a Pod, Kubernetes adds the pv-protection
finalizer. If you
try to delete the PersistentVolume
, it enters a Terminating
status, but the
controller can't delete it because the finalizer exists. When the Pod stops
using the PersistentVolume
, Kubernetes clears the pv-protection
finalizer,
and the controller deletes the volume.
Note:
When you
DELETE
an object, Kubernetes adds the deletion timestamp for that object and then immediately starts to restrict changes to the.metadata.finalizers
field for the object that is now pending deletion. You can remove existing finalizers (deleting an entry from thefinalizers
list) but you cannot add a new finalizer. You also cannot modify thedeletionTimestamp
for an object once it is set.After the deletion is requested, you can not resurrect this object. The only way is to delete it and make a new similar object.
Owner references, labels, and finalizers
Like labels, owner references describe the relationships between objects in Kubernetes, but are used for a different purpose. When a controller manages objects like Pods, it uses labels to track changes to groups of related objects. For example, when a Job creates one or more Pods, the Job controller applies labels to those pods and tracks changes to any Pods in the cluster with the same label.
The Job controller also adds owner references to those Pods, pointing at the Job that created the Pods. If you delete the Job while these Pods are running, Kubernetes uses the owner references (not labels) to determine which Pods in the cluster need cleanup.
Kubernetes also processes finalizers when it identifies owner references on a resource targeted for deletion.
In some situations, finalizers can block the deletion of dependent objects, which can cause the targeted owner object to remain for longer than expected without being fully deleted. In these situations, you should check finalizers and owner references on the target owner and dependent objects to troubleshoot the cause.
Note:
In cases where objects are stuck in a deleting state, avoid manually removing finalizers to allow deletion to continue. Finalizers are usually added to resources for a reason, so forcefully removing them can lead to issues in your cluster. This should only be done when the purpose of the finalizer is understood and is accomplished in another way (for example, manually cleaning up some dependent object).What's next
- Read Using Finalizers to Control Deletion on the Kubernetes blog.
1.2.8 - Owners and Dependents
In Kubernetes, some objects are owners of other objects. For example, a ReplicaSet is the owner of a set of Pods. These owned objects are dependents of their owner.
Ownership is different from the labels and selectors
mechanism that some resources also use. For example, consider a Service that
creates EndpointSlice
objects. The Service uses labels to allow the control plane to
determine which EndpointSlice
objects are used for that Service. In addition
to the labels, each EndpointSlice
that is managed on behalf of a Service has
an owner reference. Owner references help different parts of Kubernetes avoid
interfering with objects they don’t control.
Owner references in object specifications
Dependent objects have a metadata.ownerReferences
field that references their
owner object. A valid owner reference consists of the object name and a UID
within the same namespace as the dependent object. Kubernetes sets the value of
this field automatically for objects that are dependents of other objects like
ReplicaSets, DaemonSets, Deployments, Jobs and CronJobs, and ReplicationControllers.
You can also configure these relationships manually by changing the value of
this field. However, you usually don't need to and can allow Kubernetes to
automatically manage the relationships.
Dependent objects also have an ownerReferences.blockOwnerDeletion
field that
takes a boolean value and controls whether specific dependents can block garbage
collection from deleting their owner object. Kubernetes automatically sets this
field to true
if a controller
(for example, the Deployment controller) sets the value of the
metadata.ownerReferences
field. You can also set the value of the
blockOwnerDeletion
field manually to control which dependents block garbage
collection.
A Kubernetes admission controller controls user access to change this field for dependent resources, based on the delete permissions of the owner. This control prevents unauthorized users from delaying owner object deletion.
Note:
Cross-namespace owner references are disallowed by design. Namespaced dependents can specify cluster-scoped or namespaced owners. A namespaced owner must exist in the same namespace as the dependent. If it does not, the owner reference is treated as absent, and the dependent is subject to deletion once all owners are verified absent.
Cluster-scoped dependents can only specify cluster-scoped owners. In v1.20+, if a cluster-scoped dependent specifies a namespaced kind as an owner, it is treated as having an unresolvable owner reference, and is not able to be garbage collected.
In v1.20+, if the garbage collector detects an invalid cross-namespace ownerReference
,
or a cluster-scoped dependent with an ownerReference
referencing a namespaced kind, a warning Event
with a reason of OwnerRefInvalidNamespace
and an involvedObject
of the invalid dependent is reported.
You can check for that kind of Event by running
kubectl get events -A --field-selector=reason=OwnerRefInvalidNamespace
.
Ownership and finalizers
When you tell Kubernetes to delete a resource, the API server allows the
managing controller to process any finalizer rules
for the resource. Finalizers
prevent accidental deletion of resources your cluster may still need to function
correctly. For example, if you try to delete a PersistentVolume that is still
in use by a Pod, the deletion does not happen immediately because the
PersistentVolume
has the kubernetes.io/pv-protection
finalizer on it.
Instead, the volume remains in the Terminating
status until Kubernetes clears
the finalizer, which only happens after the PersistentVolume
is no longer
bound to a Pod.
Kubernetes also adds finalizers to an owner resource when you use either
foreground or orphan cascading deletion.
In foreground deletion, it adds the foreground
finalizer so that the
controller must delete dependent resources that also have
ownerReferences.blockOwnerDeletion=true
before it deletes the owner. If you
specify an orphan deletion policy, Kubernetes adds the orphan
finalizer so
that the controller ignores dependent resources after it deletes the owner
object.
What's next
- Learn more about Kubernetes finalizers.
- Learn about garbage collection.
- Read the API reference for object metadata.
1.2.9 - Recommended Labels
You can visualize and manage Kubernetes objects with more tools than kubectl and the dashboard. A common set of labels allows tools to work interoperably, describing objects in a common manner that all tools can understand.
In addition to supporting tooling, the recommended labels describe applications in a way that can be queried.
The metadata is organized around the concept of an application. Kubernetes is not a platform as a service (PaaS) and doesn't have or enforce a formal notion of an application. Instead, applications are informal and described with metadata. The definition of what an application contains is loose.
Note:
These are recommended labels. They make it easier to manage applications but aren't required for any core tooling.Shared labels and annotations share a common prefix: app.kubernetes.io
. Labels
without a prefix are private to users. The shared prefix ensures that shared labels
do not interfere with custom user labels.
Labels
In order to take full advantage of using these labels, they should be applied on every resource object.
Key | Description | Example | Type |
---|---|---|---|
app.kubernetes.io/name | The name of the application | mysql | string |
app.kubernetes.io/instance | A unique name identifying the instance of an application | mysql-abcxyz | string |
app.kubernetes.io/version | The current version of the application (e.g., a SemVer 1.0, revision hash, etc.) | 5.7.21 | string |
app.kubernetes.io/component | The component within the architecture | database | string |
app.kubernetes.io/part-of | The name of a higher level application this one is part of | wordpress | string |
app.kubernetes.io/managed-by | The tool being used to manage the operation of an application | Helm | string |
To illustrate these labels in action, consider the following StatefulSet object:
# This is an excerpt
apiVersion: apps/v1
kind: StatefulSet
metadata:
labels:
app.kubernetes.io/name: mysql
app.kubernetes.io/instance: mysql-abcxyz
app.kubernetes.io/version: "5.7.21"
app.kubernetes.io/component: database
app.kubernetes.io/part-of: wordpress
app.kubernetes.io/managed-by: Helm
Applications And Instances Of Applications
An application can be installed one or more times into a Kubernetes cluster and, in some cases, the same namespace. For example, WordPress can be installed more than once where different websites are different installations of WordPress.
The name of an application and the instance name are recorded separately. For
example, WordPress has a app.kubernetes.io/name
of wordpress
while it has
an instance name, represented as app.kubernetes.io/instance
with a value of
wordpress-abcxyz
. This enables the application and instance of the application
to be identifiable. Every instance of an application must have a unique name.
Examples
To illustrate different ways to use these labels the following examples have varying complexity.
A Simple Stateless Service
Consider the case for a simple stateless service deployed using Deployment
and Service
objects. The following two snippets represent how the labels could be used in their simplest form.
The Deployment
is used to oversee the pods running the application itself.
apiVersion: apps/v1
kind: Deployment
metadata:
labels:
app.kubernetes.io/name: myservice
app.kubernetes.io/instance: myservice-abcxyz
...
The Service
is used to expose the application.
apiVersion: v1
kind: Service
metadata:
labels:
app.kubernetes.io/name: myservice
app.kubernetes.io/instance: myservice-abcxyz
...
Web Application With A Database
Consider a slightly more complicated application: a web application (WordPress) using a database (MySQL), installed using Helm. The following snippets illustrate the start of objects used to deploy this application.
The start to the following Deployment
is used for WordPress:
apiVersion: apps/v1
kind: Deployment
metadata:
labels:
app.kubernetes.io/name: wordpress
app.kubernetes.io/instance: wordpress-abcxyz
app.kubernetes.io/version: "4.9.4"
app.kubernetes.io/managed-by: Helm
app.kubernetes.io/component: server
app.kubernetes.io/part-of: wordpress
...
The Service
is used to expose WordPress:
apiVersion: v1
kind: Service
metadata:
labels:
app.kubernetes.io/name: wordpress
app.kubernetes.io/instance: wordpress-abcxyz
app.kubernetes.io/version: "4.9.4"
app.kubernetes.io/managed-by: Helm
app.kubernetes.io/component: server
app.kubernetes.io/part-of: wordpress
...
MySQL is exposed as a StatefulSet
with metadata for both it and the larger application it belongs to:
apiVersion: apps/v1
kind: StatefulSet
metadata:
labels:
app.kubernetes.io/name: mysql
app.kubernetes.io/instance: mysql-abcxyz
app.kubernetes.io/version: "5.7.21"
app.kubernetes.io/managed-by: Helm
app.kubernetes.io/component: database
app.kubernetes.io/part-of: wordpress
...
The Service
is used to expose MySQL as part of WordPress:
apiVersion: v1
kind: Service
metadata:
labels:
app.kubernetes.io/name: mysql
app.kubernetes.io/instance: mysql-abcxyz
app.kubernetes.io/version: "5.7.21"
app.kubernetes.io/managed-by: Helm
app.kubernetes.io/component: database
app.kubernetes.io/part-of: wordpress
...
With the MySQL StatefulSet
and Service
you'll notice information about both MySQL and WordPress, the broader application, are included.
1.3 - The Kubernetes API
The core of Kubernetes' control plane is the API server. The API server exposes an HTTP API that lets end users, different parts of your cluster, and external components communicate with one another.
The Kubernetes API lets you query and manipulate the state of API objects in Kubernetes (for example: Pods, Namespaces, ConfigMaps, and Events).
Most operations can be performed through the kubectl command-line interface or other command-line tools, such as kubeadm, which in turn use the API. However, you can also access the API directly using REST calls. Kubernetes provides a set of client libraries for those looking to write applications using the Kubernetes API.
Each Kubernetes cluster publishes the specification of the APIs that the cluster serves.
There are two mechanisms that Kubernetes uses to publish these API specifications; both are useful
to enable automatic interoperability. For example, the kubectl
tool fetches and caches the API
specification for enabling command-line completion and other features.
The two supported mechanisms are as follows:
The Discovery API provides information about the Kubernetes APIs: API names, resources, versions, and supported operations. This is a Kubernetes specific term as it is a separate API from the Kubernetes OpenAPI. It is intended to be a brief summary of the available resources and it does not detail specific schema for the resources. For reference about resource schemas, please refer to the OpenAPI document.
The Kubernetes OpenAPI Document provides (full) OpenAPI v2.0 and 3.0 schemas for all Kubernetes API endpoints. The OpenAPI v3 is the preferred method for accessing OpenAPI as it provides a more comprehensive and accurate view of the API. It includes all the available API paths, as well as all resources consumed and produced for every operations on every endpoints. It also includes any extensibility components that a cluster supports. The data is a complete specification and is significantly larger than that from the Discovery API.
Discovery API
Kubernetes publishes a list of all group versions and resources supported via the Discovery API. This includes the following for each resource:
- Name
- Cluster or namespaced scope
- Endpoint URL and supported verbs
- Alternative names
- Group, version, kind
The API is available in both aggregated and unaggregated form. The aggregated discovery serves two endpoints, while the unaggregated discovery serves a separate endpoint for each group version.
Aggregated discovery
Kubernetes v1.30 [stable]
(enabled by default: true)Kubernetes offers stable support for aggregated discovery, publishing
all resources supported by a cluster through two endpoints (/api
and
/apis
). Requesting this
endpoint drastically reduces the number of requests sent to fetch the
discovery data from the cluster. You can access the data by
requesting the respective endpoints with an Accept
header indicating
the aggregated discovery resource:
Accept: application/json;v=v2;g=apidiscovery.k8s.io;as=APIGroupDiscoveryList
.
Without indicating the resource type using the Accept
header, the default
response for the /api
and /apis
endpoint is an unaggregated discovery
document.
The discovery document for the built-in resources can be found in the Kubernetes GitHub repository. This Github document can be used as a reference of the base set of the available resources if a Kubernetes cluster is not available to query.
The endpoint also supports ETag and protobuf encoding.
Unaggregated discovery
Without discovery aggregation, discovery is published in levels, with the root endpoints publishing discovery information for downstream documents.
A list of all group versions supported by a cluster is published at
the /api
and /apis
endpoints. Example:
{
"kind": "APIGroupList",
"apiVersion": "v1",
"groups": [
{
"name": "apiregistration.k8s.io",
"versions": [
{
"groupVersion": "apiregistration.k8s.io/v1",
"version": "v1"
}
],
"preferredVersion": {
"groupVersion": "apiregistration.k8s.io/v1",
"version": "v1"
}
},
{
"name": "apps",
"versions": [
{
"groupVersion": "apps/v1",
"version": "v1"
}
],
"preferredVersion": {
"groupVersion": "apps/v1",
"version": "v1"
}
},
...
}
Additional requests are needed to obtain the discovery document for each group version at
/apis/<group>/<version>
(for example:
/apis/rbac.authorization.k8s.io/v1alpha1
), which advertises the list of
resources served under a particular group version. These endpoints are used by
kubectl to fetch the list of resources supported by a cluster.
OpenAPI interface definition
For details about the OpenAPI specifications, see the OpenAPI documentation.
Kubernetes serves both OpenAPI v2.0 and OpenAPI v3.0. OpenAPI v3 is the
preferred method of accessing the OpenAPI because it offers a more comprehensive
(lossless) representation of Kubernetes resources. Due to limitations of OpenAPI
version 2, certain fields are dropped from the published OpenAPI including but not
limited to default
, nullable
, oneOf
.
OpenAPI V2
The Kubernetes API server serves an aggregated OpenAPI v2 spec via the
/openapi/v2
endpoint. You can request the response format using
request headers as follows:
Header | Possible values | Notes |
---|---|---|
Accept-Encoding | gzip | not supplying this header is also acceptable |
Accept | application/com.github.proto-openapi.spec.v2@v1.0+protobuf | mainly for intra-cluster use |
application/json | default | |
* | serves application/json |
OpenAPI V3
Kubernetes v1.27 [stable]
(enabled by default: true)Kubernetes supports publishing a description of its APIs as OpenAPI v3.
A discovery endpoint /openapi/v3
is provided to see a list of all
group/versions available. This endpoint only returns JSON. These
group/versions are provided in the following format:
{
"paths": {
...,
"api/v1": {
"serverRelativeURL": "/openapi/v3/api/v1?hash=CC0E9BFD992D8C59AEC98A1E2336F899E8318D3CF4C68944C3DEC640AF5AB52D864AC50DAA8D145B3494F75FA3CFF939FCBDDA431DAD3CA79738B297795818CF"
},
"apis/admissionregistration.k8s.io/v1": {
"serverRelativeURL": "/openapi/v3/apis/admissionregistration.k8s.io/v1?hash=E19CC93A116982CE5422FC42B590A8AFAD92CDE9AE4D59B5CAAD568F083AD07946E6CB5817531680BCE6E215C16973CD39003B0425F3477CFD854E89A9DB6597"
},
....
}
}
The relative URLs are pointing to immutable OpenAPI descriptions, in
order to improve client-side caching. The proper HTTP caching headers
are also set by the API server for that purpose (Expires
to 1 year in
the future, and Cache-Control
to immutable
). When an obsolete URL is
used, the API server returns a redirect to the newest URL.
The Kubernetes API server publishes an OpenAPI v3 spec per Kubernetes
group version at the /openapi/v3/apis/<group>/<version>?hash=<hash>
endpoint.
Refer to the table below for accepted request headers.
Header | Possible values | Notes |
---|---|---|
Accept-Encoding | gzip | not supplying this header is also acceptable |
Accept | application/com.github.proto-openapi.spec.v3@v1.0+protobuf | mainly for intra-cluster use |
application/json | default | |
* | serves application/json |
A Golang implementation to fetch the OpenAPI V3 is provided in the package
k8s.io/client-go/openapi3
.
Kubernetes 1.31 publishes OpenAPI v2.0 and v3.0; there are no plans to support 3.1 in the near future.
Protobuf serialization
Kubernetes implements an alternative Protobuf based serialization format that is primarily intended for intra-cluster communication. For more information about this format, see the Kubernetes Protobuf serialization design proposal and the Interface Definition Language (IDL) files for each schema located in the Go packages that define the API objects.
Persistence
Kubernetes stores the serialized state of objects by writing them into etcd.
API groups and versioning
To make it easier to eliminate fields or restructure resource representations,
Kubernetes supports multiple API versions, each at a different API path, such
as /api/v1
or /apis/rbac.authorization.k8s.io/v1alpha1
.
Versioning is done at the API level rather than at the resource or field level to ensure that the API presents a clear, consistent view of system resources and behavior, and to enable controlling access to end-of-life and/or experimental APIs.
To make it easier to evolve and to extend its API, Kubernetes implements API groups that can be enabled or disabled.
API resources are distinguished by their API group, resource type, namespace (for namespaced resources), and name. The API server handles the conversion between API versions transparently: all the different versions are actually representations of the same persisted data. The API server may serve the same underlying data through multiple API versions.
For example, suppose there are two API versions, v1
and v1beta1
, for the same
resource. If you originally created an object using the v1beta1
version of its
API, you can later read, update, or delete that object using either the v1beta1
or the v1
API version, until the v1beta1
version is deprecated and removed.
At that point you can continue accessing and modifying the object using the v1
API.
API changes
Any system that is successful needs to grow and change as new use cases emerge or existing ones change. Therefore, Kubernetes has designed the Kubernetes API to continuously change and grow. The Kubernetes project aims to not break compatibility with existing clients, and to maintain that compatibility for a length of time so that other projects have an opportunity to adapt.
In general, new API resources and new resource fields can be added often and frequently. Elimination of resources or fields requires following the API deprecation policy.
Kubernetes makes a strong commitment to maintain compatibility for official Kubernetes APIs
once they reach general availability (GA), typically at API version v1
. Additionally,
Kubernetes maintains compatibility with data persisted via beta API versions of official Kubernetes APIs,
and ensures that data can be converted and accessed via GA API versions when the feature goes stable.
If you adopt a beta API version, you will need to transition to a subsequent beta or stable API version once the API graduates. The best time to do this is while the beta API is in its deprecation period, since objects are simultaneously accessible via both API versions. Once the beta API completes its deprecation period and is no longer served, the replacement API version must be used.
Note:
Although Kubernetes also aims to maintain compatibility for alpha APIs versions, in some circumstances this is not possible. If you use any alpha API versions, check the release notes for Kubernetes when upgrading your cluster, in case the API did change in incompatible ways that require deleting all existing alpha objects prior to upgrade.Refer to API versions reference for more details on the API version level definitions.
API Extension
The Kubernetes API can be extended in one of two ways:
- Custom resources let you declaratively define how the API server should provide your chosen resource API.
- You can also extend the Kubernetes API by implementing an aggregation layer.
What's next
- Learn how to extend the Kubernetes API by adding your own CustomResourceDefinition.
- Controlling Access To The Kubernetes API describes how the cluster manages authentication and authorization for API access.
- Learn about API endpoints, resource types and samples by reading API Reference.
- Learn about what constitutes a compatible change, and how to change the API, from API changes.
2 - Cluster Architecture
A Kubernetes cluster consists of a control plane plus a set of worker machines, called nodes, that run containerized applications. Every cluster needs at least one worker node in order to run Pods.
The worker node(s) host the Pods that are the components of the application workload. The control plane manages the worker nodes and the Pods in the cluster. In production environments, the control plane usually runs across multiple computers and a cluster usually runs multiple nodes, providing fault-tolerance and high availability.
This document outlines the various components you need to have for a complete and working Kubernetes cluster.
About this architecture
The diagram in Figure 1 presents an example reference architecture for a Kubernetes cluster. The actual distribution of components can vary based on specific cluster setups and requirements.
In the diagram, each node runs the kube-proxy
component. You need a
network proxy component on each node to ensure that the
Service API and associated behaviors
are available on your cluster network. However, some network plugins provide their own,
third party implementation of proxying. When you use that kind of network plugin,
the node does not need to run kube-proxy
.
Control plane components
The control plane's components make global decisions about the cluster (for example, scheduling),
as well as detecting and responding to cluster events (for example, starting up a new
pod when a Deployment's
replicas
field is unsatisfied).
Control plane components can be run on any machine in the cluster. However, for simplicity, setup scripts typically start all control plane components on the same machine, and do not run user containers on this machine. See Creating Highly Available clusters with kubeadm for an example control plane setup that runs across multiple machines.
kube-apiserver
The API server is a component of the Kubernetes control plane that exposes the Kubernetes API. The API server is the front end for the Kubernetes control plane.
The main implementation of a Kubernetes API server is kube-apiserver. kube-apiserver is designed to scale horizontally—that is, it scales by deploying more instances. You can run several instances of kube-apiserver and balance traffic between those instances.
etcd
Consistent and highly-available key value store used as Kubernetes' backing store for all cluster data.
If your Kubernetes cluster uses etcd as its backing store, make sure you have a back up plan for the data.
You can find in-depth information about etcd in the official documentation.
kube-scheduler
Control plane component that watches for newly created Pods with no assigned node, and selects a node for them to run on.
Factors taken into account for scheduling decisions include: individual and collective resource requirements, hardware/software/policy constraints, affinity and anti-affinity specifications, data locality, inter-workload interference, and deadlines.
kube-controller-manager
Control plane component that runs controller processes.
Logically, each controller is a separate process, but to reduce complexity, they are all compiled into a single binary and run in a single process.
There are many different types of controllers. Some examples of them are:
- Node controller: Responsible for noticing and responding when nodes go down.
- Job controller: Watches for Job objects that represent one-off tasks, then creates Pods to run those tasks to completion.
- EndpointSlice controller: Populates EndpointSlice objects (to provide a link between Services and Pods).
- ServiceAccount controller: Create default ServiceAccounts for new namespaces.
The above is not an exhaustive list.
cloud-controller-manager
A Kubernetes control plane component that embeds cloud-specific control logic. The cloud controller manager lets you link your cluster into your cloud provider's API, and separates out the components that interact with that cloud platform from components that only interact with your cluster.The cloud-controller-manager only runs controllers that are specific to your cloud provider. If you are running Kubernetes on your own premises, or in a learning environment inside your own PC, the cluster does not have a cloud controller manager.
As with the kube-controller-manager, the cloud-controller-manager combines several logically independent control loops into a single binary that you run as a single process. You can scale horizontally (run more than one copy) to improve performance or to help tolerate failures.
The following controllers can have cloud provider dependencies:
- Node controller: For checking the cloud provider to determine if a node has been deleted in the cloud after it stops responding
- Route controller: For setting up routes in the underlying cloud infrastructure
- Service controller: For creating, updating and deleting cloud provider load balancers
Node components
Node components run on every node, maintaining running pods and providing the Kubernetes runtime environment.
kubelet
An agent that runs on each node in the cluster. It makes sure that containers are running in a Pod.
The kubelet takes a set of PodSpecs that are provided through various mechanisms and ensures that the containers described in those PodSpecs are running and healthy. The kubelet doesn't manage containers which were not created by Kubernetes.
kube-proxy (optional)
kube-proxy is a network proxy that runs on each node in your cluster, implementing part of the Kubernetes Service concept.
kube-proxy maintains network rules on nodes. These network rules allow network communication to your Pods from network sessions inside or outside of your cluster.
kube-proxy uses the operating system packet filtering layer if there is one and it's available. Otherwise, kube-proxy forwards the traffic itself.
If you use a network plugin that implements packet forwarding for Services by itself, and providing equivalent behavior to kube-proxy, then you do not need to run kube-proxy on the nodes in your cluster.Container runtime
A fundamental component that empowers Kubernetes to run containers effectively. It is responsible for managing the execution and lifecycle of containers within the Kubernetes environment.
Kubernetes supports container runtimes such as containerd, CRI-O, and any other implementation of the Kubernetes CRI (Container Runtime Interface).
Addons
Addons use Kubernetes resources (DaemonSet,
Deployment, etc) to implement cluster features.
Because these are providing cluster-level features, namespaced resources for
addons belong within the kube-system
namespace.
Selected addons are described below; for an extended list of available addons, please see Addons.
DNS
While the other addons are not strictly required, all Kubernetes clusters should have cluster DNS, as many examples rely on it.
Cluster DNS is a DNS server, in addition to the other DNS server(s) in your environment, which serves DNS records for Kubernetes services.
Containers started by Kubernetes automatically include this DNS server in their DNS searches.
Web UI (Dashboard)
Dashboard is a general purpose, web-based UI for Kubernetes clusters. It allows users to manage and troubleshoot applications running in the cluster, as well as the cluster itself.
Container resource monitoring
Container Resource Monitoring records generic time-series metrics about containers in a central database, and provides a UI for browsing that data.
Cluster-level Logging
A cluster-level logging mechanism is responsible for saving container logs to a central log store with a search/browsing interface.
Network plugins
Network plugins are software components that implement the container network interface (CNI) specification. They are responsible for allocating IP addresses to pods and enabling them to communicate with each other within the cluster.
Architecture variations
While the core components of Kubernetes remain consistent, the way they are deployed and managed can vary. Understanding these variations is crucial for designing and maintaining Kubernetes clusters that meet specific operational needs.
Control plane deployment options
The control plane components can be deployed in several ways:
- Traditional deployment
- Control plane components run directly on dedicated machines or VMs, often managed as systemd services.
- Static Pods
- Control plane components are deployed as static Pods, managed by the kubelet on specific nodes. This is a common approach used by tools like kubeadm.
- Self-hosted
- The control plane runs as Pods within the Kubernetes cluster itself, managed by Deployments and StatefulSets or other Kubernetes primitives.
- Managed Kubernetes services
- Cloud providers often abstract away the control plane, managing its components as part of their service offering.
Workload placement considerations
The placement of workloads, including the control plane components, can vary based on cluster size, performance requirements, and operational policies:
- In smaller or development clusters, control plane components and user workloads might run on the same nodes.
- Larger production clusters often dedicate specific nodes to control plane components, separating them from user workloads.
- Some organizations run critical add-ons or monitoring tools on control plane nodes.
Cluster management tools
Tools like kubeadm, kops, and Kubespray offer different approaches to deploying and managing clusters, each with its own method of component layout and management.
The flexibility of Kubernetes architecture allows organizations to tailor their clusters to specific needs, balancing factors such as operational complexity, performance, and management overhead.
Customization and extensibility
Kubernetes architecture allows for significant customization:
- Custom schedulers can be deployed to work alongside the default Kubernetes scheduler or to replace it entirely.
- API servers can be extended with CustomResourceDefinitions and API Aggregation.
- Cloud providers can integrate deeply with Kubernetes using the cloud-controller-manager.
The flexibility of Kubernetes architecture allows organizations to tailor their clusters to specific needs, balancing factors such as operational complexity, performance, and management overhead.
What's next
Learn more about the following:
- Nodes and their communication with the control plane.
- Kubernetes controllers.
- kube-scheduler which is the default scheduler for Kubernetes.
- Etcd's official documentation.
- Several container runtimes in Kubernetes.
- Integrating with cloud providers using cloud-controller-manager.
- kubectl commands.
2.1 - Nodes
Kubernetes runs your workload by placing containers into Pods to run on Nodes. A node may be a virtual or physical machine, depending on the cluster. Each node is managed by the control plane and contains the services necessary to run Pods.
Typically you have several nodes in a cluster; in a learning or resource-limited environment, you might have only one node.
The components on a node include the kubelet, a container runtime, and the kube-proxy.
Management
There are two main ways to have Nodes added to the API server:
- The kubelet on a node self-registers to the control plane
- You (or another human user) manually add a Node object
After you create a Node object, or the kubelet on a node self-registers, the control plane checks whether the new Node object is valid. For example, if you try to create a Node from the following JSON manifest:
{
"kind": "Node",
"apiVersion": "v1",
"metadata": {
"name": "10.240.79.157",
"labels": {
"name": "my-first-k8s-node"
}
}
}
Kubernetes creates a Node object internally (the representation). Kubernetes checks
that a kubelet has registered to the API server that matches the metadata.name
field of the Node. If the node is healthy (i.e. all necessary services are running),
then it is eligible to run a Pod. Otherwise, that node is ignored for any cluster activity
until it becomes healthy.
Note:
Kubernetes keeps the object for the invalid Node and continues checking to see whether it becomes healthy.
You, or a controller, must explicitly delete the Node object to stop that health checking.
The name of a Node object must be a valid DNS subdomain name.
Node name uniqueness
The name identifies a Node. Two Nodes cannot have the same name at the same time. Kubernetes also assumes that a resource with the same name is the same object. In case of a Node, it is implicitly assumed that an instance using the same name will have the same state (e.g. network settings, root disk contents) and attributes like node labels. This may lead to inconsistencies if an instance was modified without changing its name. If the Node needs to be replaced or updated significantly, the existing Node object needs to be removed from API server first and re-added after the update.
Self-registration of Nodes
When the kubelet flag --register-node
is true (the default), the kubelet will attempt to
register itself with the API server. This is the preferred pattern, used by most distros.
For self-registration, the kubelet is started with the following options:
--kubeconfig
- Path to credentials to authenticate itself to the API server.--cloud-provider
- How to talk to a cloud provider to read metadata about itself.--register-node
- Automatically register with the API server.--register-with-taints
- Register the node with the given list of taints (comma separated<key>=<value>:<effect>
).No-op if
register-node
is false.--node-ip
- Optional comma-separated list of the IP addresses for the node. You can only specify a single address for each address family. For example, in a single-stack IPv4 cluster, you set this value to be the IPv4 address that the kubelet should use for the node. See configure IPv4/IPv6 dual stack for details of running a dual-stack cluster.If you don't provide this argument, the kubelet uses the node's default IPv4 address, if any; if the node has no IPv4 addresses then the kubelet uses the node's default IPv6 address.
--node-labels
- Labels to add when registering the node in the cluster (see label restrictions enforced by the NodeRestriction admission plugin).--node-status-update-frequency
- Specifies how often kubelet posts its node status to the API server.
When the Node authorization mode and NodeRestriction admission plugin are enabled, kubelets are only authorized to create/modify their own Node resource.
Note:
As mentioned in the Node name uniqueness section,
when Node configuration needs to be updated, it is a good practice to re-register
the node with the API server. For example, if the kubelet is being restarted with
a new set of --node-labels
, but the same Node name is used, the change will
not take effect, as labels are only set (or modified) upon Node registration with the API server.
Pods already scheduled on the Node may misbehave or cause issues if the Node configuration will be changed on kubelet restart. For example, already running Pod may be tainted against the new labels assigned to the Node, while other Pods, that are incompatible with that Pod will be scheduled based on this new label. Node re-registration ensures all Pods will be drained and properly re-scheduled.
Manual Node administration
You can create and modify Node objects using kubectl.
When you want to create Node objects manually, set the kubelet flag --register-node=false
.
You can modify Node objects regardless of the setting of --register-node
.
For example, you can set labels on an existing Node or mark it unschedulable.
You can use labels on Nodes in conjunction with node selectors on Pods to control scheduling. For example, you can constrain a Pod to only be eligible to run on a subset of the available nodes.
Marking a node as unschedulable prevents the scheduler from placing new pods onto that Node but does not affect existing Pods on the Node. This is useful as a preparatory step before a node reboot or other maintenance.
To mark a Node unschedulable, run:
kubectl cordon $NODENAME
See Safely Drain a Node for more details.
Note:
Pods that are part of a DaemonSet tolerate being run on an unschedulable Node. DaemonSets typically provide node-local services that should run on the Node even if it is being drained of workload applications.Node status
A Node's status contains the following information:
You can use kubectl
to view a Node's status and other details:
kubectl describe node <insert-node-name-here>
See Node Status for more details.
Node heartbeats
Heartbeats, sent by Kubernetes nodes, help your cluster determine the availability of each node, and to take action when failures are detected.
For nodes there are two forms of heartbeats:
- Updates to the
.status
of a Node. - Lease objects
within the
kube-node-lease
namespace. Each Node has an associated Lease object.
Node controller
The node controller is a Kubernetes control plane component that manages various aspects of nodes.
The node controller has multiple roles in a node's life. The first is assigning a CIDR block to the node when it is registered (if CIDR assignment is turned on).
The second is keeping the node controller's internal list of nodes up to date with the cloud provider's list of available machines. When running in a cloud environment and whenever a node is unhealthy, the node controller asks the cloud provider if the VM for that node is still available. If not, the node controller deletes the node from its list of nodes.
The third is monitoring the nodes' health. The node controller is responsible for:
- In the case that a node becomes unreachable, updating the
Ready
condition in the Node's.status
field. In this case the node controller sets theReady
condition toUnknown
. - If a node remains unreachable: triggering
API-initiated eviction
for all of the Pods on the unreachable node. By default, the node controller
waits 5 minutes between marking the node as
Unknown
and submitting the first eviction request.
By default, the node controller checks the state of each node every 5 seconds.
This period can be configured using the --node-monitor-period
flag on the
kube-controller-manager
component.
Rate limits on eviction
In most cases, the node controller limits the eviction rate to
--node-eviction-rate
(default 0.1) per second, meaning it won't evict pods
from more than 1 node per 10 seconds.
The node eviction behavior changes when a node in a given availability zone
becomes unhealthy. The node controller checks what percentage of nodes in the zone
are unhealthy (the Ready
condition is Unknown
or False
) at the same time:
- If the fraction of unhealthy nodes is at least
--unhealthy-zone-threshold
(default 0.55), then the eviction rate is reduced. - If the cluster is small (i.e. has less than or equal to
--large-cluster-size-threshold
nodes - default 50), then evictions are stopped. - Otherwise, the eviction rate is reduced to
--secondary-node-eviction-rate
(default 0.01) per second.
The reason these policies are implemented per availability zone is because one availability zone might become partitioned from the control plane while the others remain connected. If your cluster does not span multiple cloud provider availability zones, then the eviction mechanism does not take per-zone unavailability into account.
A key reason for spreading your nodes across availability zones is so that the
workload can be shifted to healthy zones when one entire zone goes down.
Therefore, if all nodes in a zone are unhealthy, then the node controller evicts at
the normal rate of --node-eviction-rate
. The corner case is when all zones are
completely unhealthy (none of the nodes in the cluster are healthy). In such a
case, the node controller assumes that there is some problem with connectivity
between the control plane and the nodes, and doesn't perform any evictions.
(If there has been an outage and some nodes reappear, the node controller does
evict pods from the remaining nodes that are unhealthy or unreachable).
The node controller is also responsible for evicting pods running on nodes with
NoExecute
taints, unless those pods tolerate that taint.
The node controller also adds taints
corresponding to node problems like node unreachable or not ready. This means
that the scheduler won't place Pods onto unhealthy nodes.
Resource capacity tracking
Node objects track information about the Node's resource capacity: for example, the amount of memory available and the number of CPUs. Nodes that self register report their capacity during registration. If you manually add a Node, then you need to set the node's capacity information when you add it.
The Kubernetes scheduler ensures that there are enough resources for all the Pods on a Node. The scheduler checks that the sum of the requests of containers on the node is no greater than the node's capacity. That sum of requests includes all containers managed by the kubelet, but excludes any containers started directly by the container runtime, and also excludes any processes running outside of the kubelet's control.
Note:
If you want to explicitly reserve resources for non-Pod processes, see reserve resources for system daemons.Node topology
Kubernetes v1.27 [stable]
(enabled by default: true)If you have enabled the TopologyManager
feature gate, then
the kubelet can use topology hints when making resource assignment decisions.
See Control Topology Management Policies on a Node
for more information.
Swap memory management
Kubernetes v1.30 [beta]
(enabled by default: true)To enable swap on a node, the NodeSwap
feature gate must be enabled on
the kubelet (default is true), and the --fail-swap-on
command line flag or failSwapOn
configuration setting
must be set to false.
To allow Pods to utilize swap, swapBehavior
should not be set to NoSwap
(which is the default behavior) in the kubelet config.
Warning:
When the memory swap feature is turned on, Kubernetes data such as the content of Secret objects that were written to tmpfs now could be swapped to disk.A user can also optionally configure memorySwap.swapBehavior
in order to
specify how a node will use swap memory. For example,
memorySwap:
swapBehavior: LimitedSwap
NoSwap
(default): Kubernetes workloads will not use swap.LimitedSwap
: The utilization of swap memory by Kubernetes workloads is subject to limitations. Only Pods of Burstable QoS are permitted to employ swap.
If configuration for memorySwap
is not specified and the feature gate is
enabled, by default the kubelet will apply the same behaviour as the
NoSwap
setting.
With LimitedSwap
, Pods that do not fall under the Burstable QoS classification (i.e.
BestEffort
/Guaranteed
Qos Pods) are prohibited from utilizing swap memory.
To maintain the aforementioned security and node health guarantees, these Pods
are not permitted to use swap memory when LimitedSwap
is in effect.
Prior to detailing the calculation of the swap limit, it is necessary to define the following terms:
nodeTotalMemory
: The total amount of physical memory available on the node.totalPodsSwapAvailable
: The total amount of swap memory on the node that is available for use by Pods (some swap memory may be reserved for system use).containerMemoryRequest
: The container's memory request.
Swap limitation is configured as:
(containerMemoryRequest / nodeTotalMemory) * totalPodsSwapAvailable
.
It is important to note that, for containers within Burstable QoS Pods, it is possible to opt-out of swap usage by specifying memory requests that are equal to memory limits. Containers configured in this manner will not have access to swap memory.
Swap is supported only with cgroup v2, cgroup v1 is not supported.
For more information, and to assist with testing and provide feedback, please see the blog-post about Kubernetes 1.28: NodeSwap graduates to Beta1, KEP-2400 and its design proposal.
What's next
Learn more about the following:
- Components that make up a node.
- API definition for Node.
- Node section of the architecture design document.
- Graceful/non-graceful node shutdown.
- Cluster autoscaling to manage the number and size of nodes in your cluster.
- Taints and Tolerations.
- Node Resource Managers.
- Resource Management for Windows nodes.
2.2 - Communication between Nodes and the Control Plane
This document catalogs the communication paths between the API server and the Kubernetes cluster. The intent is to allow users to customize their installation to harden the network configuration such that the cluster can be run on an untrusted network (or on fully public IPs on a cloud provider).
Node to Control Plane
Kubernetes has a "hub-and-spoke" API pattern. All API usage from nodes (or the pods they run) terminates at the API server. None of the other control plane components are designed to expose remote services. The API server is configured to listen for remote connections on a secure HTTPS port (typically 443) with one or more forms of client authentication enabled. One or more forms of authorization should be enabled, especially if anonymous requests or service account tokens are allowed.
Nodes should be provisioned with the public root certificate for the cluster such that they can connect securely to the API server along with valid client credentials. A good approach is that the client credentials provided to the kubelet are in the form of a client certificate. See kubelet TLS bootstrapping for automated provisioning of kubelet client certificates.
Pods that wish to connect to the API server can do so securely by leveraging a service account so
that Kubernetes will automatically inject the public root certificate and a valid bearer token
into the pod when it is instantiated.
The kubernetes
service (in default
namespace) is configured with a virtual IP address that is
redirected (via kube-proxy
) to the HTTPS endpoint on the API server.
The control plane components also communicate with the API server over the secure port.
As a result, the default operating mode for connections from the nodes and pod running on the nodes to the control plane is secured by default and can run over untrusted and/or public networks.
Control plane to node
There are two primary communication paths from the control plane (the API server) to the nodes. The first is from the API server to the kubelet process which runs on each node in the cluster. The second is from the API server to any node, pod, or service through the API server's proxy functionality.
API server to kubelet
The connections from the API server to the kubelet are used for:
- Fetching logs for pods.
- Attaching (usually through
kubectl
) to running pods. - Providing the kubelet's port-forwarding functionality.
These connections terminate at the kubelet's HTTPS endpoint. By default, the API server does not verify the kubelet's serving certificate, which makes the connection subject to man-in-the-middle attacks and unsafe to run over untrusted and/or public networks.
To verify this connection, use the --kubelet-certificate-authority
flag to provide the API
server with a root certificate bundle to use to verify the kubelet's serving certificate.
If that is not possible, use SSH tunneling between the API server and kubelet if required to avoid connecting over an untrusted or public network.
Finally, Kubelet authentication and/or authorization should be enabled to secure the kubelet API.
API server to nodes, pods, and services
The connections from the API server to a node, pod, or service default to plain HTTP connections
and are therefore neither authenticated nor encrypted. They can be run over a secure HTTPS
connection by prefixing https:
to the node, pod, or service name in the API URL, but they will
not validate the certificate provided by the HTTPS endpoint nor provide client credentials. So
while the connection will be encrypted, it will not provide any guarantees of integrity. These
connections are not currently safe to run over untrusted or public networks.
SSH tunnels
Kubernetes supports SSH tunnels to protect the control plane to nodes communication paths. In this configuration, the API server initiates an SSH tunnel to each node in the cluster (connecting to the SSH server listening on port 22) and passes all traffic destined for a kubelet, node, pod, or service through the tunnel. This tunnel ensures that the traffic is not exposed outside of the network in which the nodes are running.
Note:
SSH tunnels are currently deprecated, so you shouldn't opt to use them unless you know what you are doing. The Konnectivity service is a replacement for this communication channel.Konnectivity service
Kubernetes v1.18 [beta]
As a replacement to the SSH tunnels, the Konnectivity service provides TCP level proxy for the control plane to cluster communication. The Konnectivity service consists of two parts: the Konnectivity server in the control plane network and the Konnectivity agents in the nodes network. The Konnectivity agents initiate connections to the Konnectivity server and maintain the network connections. After enabling the Konnectivity service, all control plane to nodes traffic goes through these connections.
Follow the Konnectivity service task to set up the Konnectivity service in your cluster.
What's next
- Read about the Kubernetes control plane components
- Learn more about Hubs and Spoke model
- Learn how to Secure a Cluster
- Learn more about the Kubernetes API
- Set up Konnectivity service
- Use Port Forwarding to Access Applications in a Cluster
- Learn how to Fetch logs for Pods, use kubectl port-forward
2.3 - Controllers
In robotics and automation, a control loop is a non-terminating loop that regulates the state of a system.
Here is one example of a control loop: a thermostat in a room.
When you set the temperature, that's telling the thermostat about your desired state. The actual room temperature is the current state. The thermostat acts to bring the current state closer to the desired state, by turning equipment on or off.
In Kubernetes, controllers are control loops that watch the state of your cluster, then make or request changes where needed. Each controller tries to move the current cluster state closer to the desired state.Controller pattern
A controller tracks at least one Kubernetes resource type. These objects have a spec field that represents the desired state. The controller(s) for that resource are responsible for making the current state come closer to that desired state.
The controller might carry the action out itself; more commonly, in Kubernetes, a controller will send messages to the API server that have useful side effects. You'll see examples of this below.
Control via API server
The Job controller is an example of a Kubernetes built-in controller. Built-in controllers manage state by interacting with the cluster API server.
Job is a Kubernetes resource that runs a Pod, or perhaps several Pods, to carry out a task and then stop.
(Once scheduled, Pod objects become part of the desired state for a kubelet).
When the Job controller sees a new task it makes sure that, somewhere in your cluster, the kubelets on a set of Nodes are running the right number of Pods to get the work done. The Job controller does not run any Pods or containers itself. Instead, the Job controller tells the API server to create or remove Pods. Other components in the control plane act on the new information (there are new Pods to schedule and run), and eventually the work is done.
After you create a new Job, the desired state is for that Job to be completed. The Job controller makes the current state for that Job be nearer to your desired state: creating Pods that do the work you wanted for that Job, so that the Job is closer to completion.
Controllers also update the objects that configure them.
For example: once the work is done for a Job, the Job controller
updates that Job object to mark it Finished
.
(This is a bit like how some thermostats turn a light off to indicate that your room is now at the temperature you set).
Direct control
In contrast with Job, some controllers need to make changes to things outside of your cluster.
For example, if you use a control loop to make sure there are enough Nodes in your cluster, then that controller needs something outside the current cluster to set up new Nodes when needed.
Controllers that interact with external state find their desired state from the API server, then communicate directly with an external system to bring the current state closer in line.
(There actually is a controller that horizontally scales the nodes in your cluster.)
The important point here is that the controller makes some changes to bring about your desired state, and then reports the current state back to your cluster's API server. Other control loops can observe that reported data and take their own actions.
In the thermostat example, if the room is very cold then a different controller might also turn on a frost protection heater. With Kubernetes clusters, the control plane indirectly works with IP address management tools, storage services, cloud provider APIs, and other services by extending Kubernetes to implement that.
Desired versus current state
Kubernetes takes a cloud-native view of systems, and is able to handle constant change.
Your cluster could be changing at any point as work happens and control loops automatically fix failures. This means that, potentially, your cluster never reaches a stable state.
As long as the controllers for your cluster are running and able to make useful changes, it doesn't matter if the overall state is stable or not.
Design
As a tenet of its design, Kubernetes uses lots of controllers that each manage a particular aspect of cluster state. Most commonly, a particular control loop (controller) uses one kind of resource as its desired state, and has a different kind of resource that it manages to make that desired state happen. For example, a controller for Jobs tracks Job objects (to discover new work) and Pod objects (to run the Jobs, and then to see when the work is finished). In this case something else creates the Jobs, whereas the Job controller creates Pods.
It's useful to have simple controllers rather than one, monolithic set of control loops that are interlinked. Controllers can fail, so Kubernetes is designed to allow for that.
Note:
There can be several controllers that create or update the same kind of object. Behind the scenes, Kubernetes controllers make sure that they only pay attention to the resources linked to their controlling resource.
For example, you can have Deployments and Jobs; these both create Pods. The Job controller does not delete the Pods that your Deployment created, because there is information (labels) the controllers can use to tell those Pods apart.
Ways of running controllers
Kubernetes comes with a set of built-in controllers that run inside the kube-controller-manager. These built-in controllers provide important core behaviors.
The Deployment controller and Job controller are examples of controllers that come as part of Kubernetes itself ("built-in" controllers). Kubernetes lets you run a resilient control plane, so that if any of the built-in controllers were to fail, another part of the control plane will take over the work.
You can find controllers that run outside the control plane, to extend Kubernetes. Or, if you want, you can write a new controller yourself. You can run your own controller as a set of Pods, or externally to Kubernetes. What fits best will depend on what that particular controller does.
What's next
- Read about the Kubernetes control plane
- Discover some of the basic Kubernetes objects
- Learn more about the Kubernetes API
- If you want to write your own controller, see Kubernetes extension patterns and the sample-controller repository.
2.4 - Leases
Distributed systems often have a need for leases, which provide a mechanism to lock shared resources
and coordinate activity between members of a set.
In Kubernetes, the lease concept is represented by Lease
objects in the coordination.k8s.io
API Group,
which are used for system-critical capabilities such as node heartbeats and component-level leader election.
Node heartbeats
Kubernetes uses the Lease API to communicate kubelet node heartbeats to the Kubernetes API server.
For every Node
, there is a Lease
object with a matching name in the kube-node-lease
namespace. Under the hood, every kubelet heartbeat is an update request to this Lease
object, updating
the spec.renewTime
field for the Lease. The Kubernetes control plane uses the time stamp of this field
to determine the availability of this Node
.
See Node Lease objects for more details.
Leader election
Kubernetes also uses Leases to ensure only one instance of a component is running at any given time.
This is used by control plane components like kube-controller-manager
and kube-scheduler
in
HA configurations, where only one instance of the component should be actively running while the other
instances are on stand-by.
Read coordinated leader election to learn about how Kubernetes builds on the Lease API to select which component instance acts as leader.
API server identity
Kubernetes v1.26 [beta]
(enabled by default: true)Starting in Kubernetes v1.26, each kube-apiserver
uses the Lease API to publish its identity to the
rest of the system. While not particularly useful on its own, this provides a mechanism for clients to
discover how many instances of kube-apiserver
are operating the Kubernetes control plane.
Existence of kube-apiserver leases enables future capabilities that may require coordination between
each kube-apiserver.
You can inspect Leases owned by each kube-apiserver by checking for lease objects in the kube-system
namespace
with the name kube-apiserver-<sha256-hash>
. Alternatively you can use the label selector apiserver.kubernetes.io/identity=kube-apiserver
:
kubectl -n kube-system get lease -l apiserver.kubernetes.io/identity=kube-apiserver
NAME HOLDER AGE
apiserver-07a5ea9b9b072c4a5f3d1c3702 apiserver-07a5ea9b9b072c4a5f3d1c3702_0c8914f7-0f35-440e-8676-7844977d3a05 5m33s
apiserver-7be9e061c59d368b3ddaf1376e apiserver-7be9e061c59d368b3ddaf1376e_84f2a85d-37c1-4b14-b6b9-603e62e4896f 4m23s
apiserver-1dfef752bcb36637d2763d1868 apiserver-1dfef752bcb36637d2763d1868_c5ffa286-8a9a-45d4-91e7-61118ed58d2e 4m43s
The SHA256 hash used in the lease name is based on the OS hostname as seen by that API server. Each kube-apiserver should be
configured to use a hostname that is unique within the cluster. New instances of kube-apiserver that use the same hostname
will take over existing Leases using a new holder identity, as opposed to instantiating new Lease objects. You can check the
hostname used by kube-apisever by checking the value of the kubernetes.io/hostname
label:
kubectl -n kube-system get lease apiserver-07a5ea9b9b072c4a5f3d1c3702 -o yaml
apiVersion: coordination.k8s.io/v1
kind: Lease
metadata:
creationTimestamp: "2023-07-02T13:16:48Z"
labels:
apiserver.kubernetes.io/identity: kube-apiserver
kubernetes.io/hostname: master-1
name: apiserver-07a5ea9b9b072c4a5f3d1c3702
namespace: kube-system
resourceVersion: "334899"
uid: 90870ab5-1ba9-4523-b215-e4d4e662acb1
spec:
holderIdentity: apiserver-07a5ea9b9b072c4a5f3d1c3702_0c8914f7-0f35-440e-8676-7844977d3a05
leaseDurationSeconds: 3600
renewTime: "2023-07-04T21:58:48.065888Z"
Expired leases from kube-apiservers that no longer exist are garbage collected by new kube-apiservers after 1 hour.
You can disable API server identity leases by disabling the APIServerIdentity
feature gate.
Workloads
Your own workload can define its own use of Leases. For example, you might run a custom
controller where a primary or leader member
performs operations that its peers do not. You define a Lease so that the controller replicas can select
or elect a leader, using the Kubernetes API for coordination.
If you do use a Lease, it's a good practice to define a name for the Lease that is obviously linked to
the product or component. For example, if you have a component named Example Foo, use a Lease named
example-foo
.
If a cluster operator or another end user could deploy multiple instances of a component, select a name prefix and pick a mechanism (such as hash of the name of the Deployment) to avoid name collisions for the Leases.
You can use another approach so long as it achieves the same outcome: different software products do not conflict with one another.
2.5 - Cloud Controller Manager
Kubernetes v1.11 [beta]
Cloud infrastructure technologies let you run Kubernetes on public, private, and hybrid clouds. Kubernetes believes in automated, API-driven infrastructure without tight coupling between components.
The cloud-controller-manager is a Kubernetes control plane component that embeds cloud-specific control logic. The cloud controller manager lets you link your cluster into your cloud provider's API, and separates out the components that interact with that cloud platform from components that only interact with your cluster.
By decoupling the interoperability logic between Kubernetes and the underlying cloud infrastructure, the cloud-controller-manager component enables cloud providers to release features at a different pace compared to the main Kubernetes project.
The cloud-controller-manager is structured using a plugin mechanism that allows different cloud providers to integrate their platforms with Kubernetes.
Design
The cloud controller manager runs in the control plane as a replicated set of processes (usually, these are containers in Pods). Each cloud-controller-manager implements multiple controllers in a single process.
Note:
You can also run the cloud controller manager as a Kubernetes addon rather than as part of the control plane.Cloud controller manager functions
The controllers inside the cloud controller manager include:
Node controller
The node controller is responsible for updating Node objects when new servers are created in your cloud infrastructure. The node controller obtains information about the hosts running inside your tenancy with the cloud provider. The node controller performs the following functions:
- Update a Node object with the corresponding server's unique identifier obtained from the cloud provider API.
- Annotating and labelling the Node object with cloud-specific information, such as the region the node is deployed into and the resources (CPU, memory, etc) that it has available.
- Obtain the node's hostname and network addresses.
- Verifying the node's health. In case a node becomes unresponsive, this controller checks with your cloud provider's API to see if the server has been deactivated / deleted / terminated. If the node has been deleted from the cloud, the controller deletes the Node object from your Kubernetes cluster.
Some cloud provider implementations split this into a node controller and a separate node lifecycle controller.
Route controller
The route controller is responsible for configuring routes in the cloud appropriately so that containers on different nodes in your Kubernetes cluster can communicate with each other.
Depending on the cloud provider, the route controller might also allocate blocks of IP addresses for the Pod network.
Service controller
Services integrate with cloud infrastructure components such as managed load balancers, IP addresses, network packet filtering, and target health checking. The service controller interacts with your cloud provider's APIs to set up load balancers and other infrastructure components when you declare a Service resource that requires them.
Authorization
This section breaks down the access that the cloud controller manager requires on various API objects, in order to perform its operations.
Node controller
The Node controller only works with Node objects. It requires full access to read and modify Node objects.
v1/Node
:
- get
- list
- create
- update
- patch
- watch
- delete
Route controller
The route controller listens to Node object creation and configures routes appropriately. It requires Get access to Node objects.
v1/Node
:
- get
Service controller
The service controller watches for Service object create, update and delete events and then configures Endpoints for those Services appropriately (for EndpointSlices, the kube-controller-manager manages these on demand).
To access Services, it requires list, and watch access. To update Services, it requires patch and update access.
To set up Endpoints resources for the Services, it requires access to create, list, get, watch, and update.
v1/Service
:
- list
- get
- watch
- patch
- update
Others
The implementation of the core of the cloud controller manager requires access to create Event objects, and to ensure secure operation, it requires access to create ServiceAccounts.
v1/Event
:
- create
- patch
- update
v1/ServiceAccount
:
- create
The RBAC ClusterRole for the cloud controller manager looks like:
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
name: cloud-controller-manager
rules:
- apiGroups:
- ""
resources:
- events
verbs:
- create
- patch
- update
- apiGroups:
- ""
resources:
- nodes
verbs:
- '*'
- apiGroups:
- ""
resources:
- nodes/status
verbs:
- patch
- apiGroups:
- ""
resources:
- services
verbs:
- list
- patch
- update
- watch
- apiGroups:
- ""
resources:
- serviceaccounts
verbs:
- create
- apiGroups:
- ""
resources:
- persistentvolumes
verbs:
- get
- list
- update
- watch
- apiGroups:
- ""
resources:
- endpoints
verbs:
- create
- get
- list
- watch
- update
What's next
Cloud Controller Manager Administration has instructions on running and managing the cloud controller manager.
To upgrade a HA control plane to use the cloud controller manager, see Migrate Replicated Control Plane To Use Cloud Controller Manager.
Want to know how to implement your own cloud controller manager, or extend an existing project?
- The cloud controller manager uses Go interfaces, specifically,
CloudProvider
interface defined incloud.go
from kubernetes/cloud-provider to allow implementations from any cloud to be plugged in. - The implementation of the shared controllers highlighted in this document (Node, Route, and Service),
and some scaffolding along with the shared cloudprovider interface, is part of the Kubernetes core.
Implementations specific to cloud providers are outside the core of Kubernetes and implement
the
CloudProvider
interface. - For more information about developing plugins, see Developing Cloud Controller Manager.
- The cloud controller manager uses Go interfaces, specifically,
2.6 - About cgroup v2
On Linux, control groups constrain resources that are allocated to processes.
The kubelet and the underlying container runtime need to interface with cgroups to enforce resource management for pods and containers which includes cpu/memory requests and limits for containerized workloads.
There are two versions of cgroups in Linux: cgroup v1 and cgroup v2. cgroup v2 is
the new generation of the cgroup
API.
What is cgroup v2?
Kubernetes v1.25 [stable]
cgroup v2 is the next version of the Linux cgroup
API. cgroup v2 provides a
unified control system with enhanced resource management
capabilities.
cgroup v2 offers several improvements over cgroup v1, such as the following:
- Single unified hierarchy design in API
- Safer sub-tree delegation to containers
- Newer features like Pressure Stall Information
- Enhanced resource allocation management and isolation across multiple resources
- Unified accounting for different types of memory allocations (network memory, kernel memory, etc)
- Accounting for non-immediate resource changes such as page cache write backs
Some Kubernetes features exclusively use cgroup v2 for enhanced resource management and isolation. For example, the MemoryQoS feature improves memory QoS and relies on cgroup v2 primitives.
Using cgroup v2
The recommended way to use cgroup v2 is to use a Linux distribution that enables and uses cgroup v2 by default.
To check if your distribution uses cgroup v2, refer to Identify cgroup version on Linux nodes.
Requirements
cgroup v2 has the following requirements:
- OS distribution enables cgroup v2
- Linux Kernel version is 5.8 or later
- Container runtime supports cgroup v2. For example:
- containerd v1.4 and later
- cri-o v1.20 and later
- The kubelet and the container runtime are configured to use the systemd cgroup driver
Linux Distribution cgroup v2 support
For a list of Linux distributions that use cgroup v2, refer to the cgroup v2 documentation
- Container Optimized OS (since M97)
- Ubuntu (since 21.10, 22.04+ recommended)
- Debian GNU/Linux (since Debian 11 bullseye)
- Fedora (since 31)
- Arch Linux (since April 2021)
- RHEL and RHEL-like distributions (since 9)
To check if your distribution is using cgroup v2, refer to your distribution's documentation or follow the instructions in Identify the cgroup version on Linux nodes.
You can also enable cgroup v2 manually on your Linux distribution by modifying
the kernel cmdline boot arguments. If your distribution uses GRUB,
systemd.unified_cgroup_hierarchy=1
should be added in GRUB_CMDLINE_LINUX
under /etc/default/grub
, followed by sudo update-grub
. However, the
recommended approach is to use a distribution that already enables cgroup v2 by
default.
Migrating to cgroup v2
To migrate to cgroup v2, ensure that you meet the requirements, then upgrade to a kernel version that enables cgroup v2 by default.
The kubelet automatically detects that the OS is running on cgroup v2 and performs accordingly with no additional configuration required.
There should not be any noticeable difference in the user experience when switching to cgroup v2, unless users are accessing the cgroup file system directly, either on the node or from within the containers.
cgroup v2 uses a different API than cgroup v1, so if there are any applications that directly access the cgroup file system, they need to be updated to newer versions that support cgroup v2. For example:
- Some third-party monitoring and security agents may depend on the cgroup filesystem. Update these agents to versions that support cgroup v2.
- If you run cAdvisor as a stand-alone DaemonSet for monitoring pods and containers, update it to v0.43.0 or later.
- If you deploy Java applications, prefer to use versions which fully support cgroup v2:
- OpenJDK / HotSpot: jdk8u372, 11.0.16, 15 and later
- IBM Semeru Runtimes: 8.0.382.0, 11.0.20.0, 17.0.8.0, and later
- IBM Java: 8.0.8.6 and later
- If you are using the uber-go/automaxprocs package, make sure the version you use is v1.5.1 or higher.
Identify the cgroup version on Linux Nodes
The cgroup version depends on the Linux distribution being used and the
default cgroup version configured on the OS. To check which cgroup version your
distribution uses, run the stat -fc %T /sys/fs/cgroup/
command on
the node:
stat -fc %T /sys/fs/cgroup/
For cgroup v2, the output is cgroup2fs
.
For cgroup v1, the output is tmpfs.
What's next
- Learn more about cgroups
- Learn more about container runtime
- Learn more about cgroup drivers
2.7 - Container Runtime Interface (CRI)
The CRI is a plugin interface which enables the kubelet to use a wide variety of container runtimes, without having a need to recompile the cluster components.
You need a working container runtime on each Node in your cluster, so that the kubelet can launch Pods and their containers.
The Container Runtime Interface (CRI) is the main protocol for the communication between the kubelet and Container Runtime.
The Kubernetes Container Runtime Interface (CRI) defines the main gRPC protocol for the communication between the node components kubelet and container runtime.
The API
Kubernetes v1.23 [stable]
The kubelet acts as a client when connecting to the container runtime via gRPC.
The runtime and image service endpoints have to be available in the container
runtime, which can be configured separately within the kubelet by using the
--image-service-endpoint
command line flags.
For Kubernetes v1.31, the kubelet prefers to use CRI v1
.
If a container runtime does not support v1
of the CRI, then the kubelet tries to
negotiate any older supported version.
The v1.31 kubelet can also negotiate CRI v1alpha2
, but
this version is considered as deprecated.
If the kubelet cannot negotiate a supported CRI version, the kubelet gives up
and doesn't register as a node.
Upgrading
When upgrading Kubernetes, the kubelet tries to automatically select the latest CRI version on restart of the component. If that fails, then the fallback will take place as mentioned above. If a gRPC re-dial was required because the container runtime has been upgraded, then the container runtime must also support the initially selected version or the redial is expected to fail. This requires a restart of the kubelet.
What's next
- Learn more about the CRI protocol definition
2.8 - Garbage Collection
Garbage collection is a collective term for the various mechanisms Kubernetes uses to clean up cluster resources. This allows the clean up of resources like the following:
- Terminated pods
- Completed Jobs
- Objects without owner references
- Unused containers and container images
- Dynamically provisioned PersistentVolumes with a StorageClass reclaim policy of Delete
- Stale or expired CertificateSigningRequests (CSRs)
- Nodes deleted in the following scenarios:
- On a cloud when the cluster uses a cloud controller manager
- On-premises when the cluster uses an addon similar to a cloud controller manager
- Node Lease objects
Owners and dependents
Many objects in Kubernetes link to each other through owner references. Owner references tell the control plane which objects are dependent on others. Kubernetes uses owner references to give the control plane, and other API clients, the opportunity to clean up related resources before deleting an object. In most cases, Kubernetes manages owner references automatically.
Ownership is different from the labels and selectors
mechanism that some resources also use. For example, consider a
Service that creates
EndpointSlice
objects. The Service uses labels to allow the control plane to
determine which EndpointSlice
objects are used for that Service. In addition
to the labels, each EndpointSlice
that is managed on behalf of a Service has
an owner reference. Owner references help different parts of Kubernetes avoid
interfering with objects they don’t control.
Note:
Cross-namespace owner references are disallowed by design. Namespaced dependents can specify cluster-scoped or namespaced owners. A namespaced owner must exist in the same namespace as the dependent. If it does not, the owner reference is treated as absent, and the dependent is subject to deletion once all owners are verified absent.
Cluster-scoped dependents can only specify cluster-scoped owners. In v1.20+, if a cluster-scoped dependent specifies a namespaced kind as an owner, it is treated as having an unresolvable owner reference, and is not able to be garbage collected.
In v1.20+, if the garbage collector detects an invalid cross-namespace ownerReference
,
or a cluster-scoped dependent with an ownerReference
referencing a namespaced kind, a warning Event
with a reason of OwnerRefInvalidNamespace
and an involvedObject
of the invalid dependent is reported.
You can check for that kind of Event by running
kubectl get events -A --field-selector=reason=OwnerRefInvalidNamespace
.
Cascading deletion
Kubernetes checks for and deletes objects that no longer have owner references, like the pods left behind when you delete a ReplicaSet. When you delete an object, you can control whether Kubernetes deletes the object's dependents automatically, in a process called cascading deletion. There are two types of cascading deletion, as follows:
- Foreground cascading deletion
- Background cascading deletion
You can also control how and when garbage collection deletes resources that have owner references using Kubernetes finalizers.
Foreground cascading deletion
In foreground cascading deletion, the owner object you're deleting first enters a deletion in progress state. In this state, the following happens to the owner object:
- The Kubernetes API server sets the object's
metadata.deletionTimestamp
field to the time the object was marked for deletion. - The Kubernetes API server also sets the
metadata.finalizers
field toforegroundDeletion
. - The object remains visible through the Kubernetes API until the deletion process is complete.
After the owner object enters the deletion in progress state, the controller deletes dependents it knows about. After deleting all the dependent objects it knows about, the controller deletes the owner object. At this point, the object is no longer visible in the Kubernetes API.
During foreground cascading deletion, the only dependents that block owner
deletion are those that have the ownerReference.blockOwnerDeletion=true
field
and are in the garbage collection controller cache. The garbage collection controller
cache may not contain objects whose resource type cannot be listed / watched successfully,
or objects that are created concurrent with deletion of an owner object.
See Use foreground cascading deletion
to learn more.
Background cascading deletion
In background cascading deletion, the Kubernetes API server deletes the owner object immediately and the garbage collector controller (custom or default) cleans up the dependent objects in the background. If a finalizer exists, it ensures that objects are not deleted until all necessary clean-up tasks are completed. By default, Kubernetes uses background cascading deletion unless you manually use foreground deletion or choose to orphan the dependent objects.
See Use background cascading deletion to learn more.
Orphaned dependents
When Kubernetes deletes an owner object, the dependents left behind are called orphan objects. By default, Kubernetes deletes dependent objects. To learn how to override this behaviour, see Delete owner objects and orphan dependents.
Garbage collection of unused containers and images
The kubelet performs garbage collection on unused images every two minutes and on unused containers every minute. You should avoid using external garbage collection tools, as these can break the kubelet behavior and remove containers that should exist.
To configure options for unused container and image garbage collection, tune the
kubelet using a configuration file
and change the parameters related to garbage collection using the
KubeletConfiguration
resource type.
Container image lifecycle
Kubernetes manages the lifecycle of all images through its image manager, which is part of the kubelet, with the cooperation of cadvisor. The kubelet considers the following disk usage limits when making garbage collection decisions:
HighThresholdPercent
LowThresholdPercent
Disk usage above the configured HighThresholdPercent
value triggers garbage
collection, which deletes images in order based on the last time they were used,
starting with the oldest first. The kubelet deletes images
until disk usage reaches the LowThresholdPercent
value.
Garbage collection for unused container images
Kubernetes v1.30 [beta]
(enabled by default: true)As a beta feature, you can specify the maximum time a local image can be unused for, regardless of disk usage. This is a kubelet setting that you configure for each node.
To configure the setting, you need to set a value for the imageMaximumGCAge
field in the kubelet configuration file.
The value is specified as a Kubernetes duration. See duration in the glossary for more details.
For example, you can set the configuration field to 12h45m
,
which means 12 hours and 45 minutes.
Note:
This feature does not track image usage across kubelet restarts. If the kubelet is restarted, the tracked image age is reset, causing the kubelet to wait the fullimageMaximumGCAge
duration before qualifying images for garbage collection
based on image age.Container garbage collection
The kubelet garbage collects unused containers based on the following variables, which you can define:
MinAge
: the minimum age at which the kubelet can garbage collect a container. Disable by setting to0
.MaxPerPodContainer
: the maximum number of dead containers each Pod can have. Disable by setting to less than0
.MaxContainers
: the maximum number of dead containers the cluster can have. Disable by setting to less than0
.
In addition to these variables, the kubelet garbage collects unidentified and deleted containers, typically starting with the oldest first.
MaxPerPodContainer
and MaxContainers
may potentially conflict with each other
in situations where retaining the maximum number of containers per Pod
(MaxPerPodContainer
) would go outside the allowable total of global dead
containers (MaxContainers
). In this situation, the kubelet adjusts
MaxPerPodContainer
to address the conflict. A worst-case scenario would be to
downgrade MaxPerPodContainer
to 1
and evict the oldest containers.
Additionally, containers owned by pods that have been deleted are removed once
they are older than MinAge
.
Note:
The kubelet only garbage collects the containers it manages.Configuring garbage collection
You can tune garbage collection of resources by configuring options specific to the controllers managing those resources. The following pages show you how to configure garbage collection:
What's next
- Learn more about ownership of Kubernetes objects.
- Learn more about Kubernetes finalizers.
- Learn about the TTL controller that cleans up finished Jobs.
2.9 - Mixed Version Proxy
Kubernetes v1.28 [alpha]
(enabled by default: false)Kubernetes 1.31 includes an alpha feature that lets an API Server proxy a resource requests to other peer API servers. This is useful when there are multiple API servers running different versions of Kubernetes in one cluster (for example, during a long-lived rollout to a new release of Kubernetes).
This enables cluster administrators to configure highly available clusters that can be upgraded more safely, by directing resource requests (made during the upgrade) to the correct kube-apiserver. That proxying prevents users from seeing unexpected 404 Not Found errors that stem from the upgrade process.
This mechanism is called the Mixed Version Proxy.
Enabling the Mixed Version Proxy
Ensure that UnknownVersionInteroperabilityProxy
feature gate
is enabled when you start the API Server:
kube-apiserver \
--feature-gates=UnknownVersionInteroperabilityProxy=true \
# required command line arguments for this feature
--peer-ca-file=<path to kube-apiserver CA cert>
--proxy-client-cert-file=<path to aggregator proxy cert>,
--proxy-client-key-file=<path to aggregator proxy key>,
--requestheader-client-ca-file=<path to aggregator CA cert>,
# requestheader-allowed-names can be set to blank to allow any Common Name
--requestheader-allowed-names=<valid Common Names to verify proxy client cert against>,
# optional flags for this feature
--peer-advertise-ip=`IP of this kube-apiserver that should be used by peers to proxy requests`
--peer-advertise-port=`port of this kube-apiserver that should be used by peers to proxy requests`
# …and other flags as usual
Proxy transport and authentication between API servers
The source kube-apiserver reuses the existing APIserver client authentication flags
--proxy-client-cert-file
and--proxy-client-key-file
to present its identity that will be verified by its peer (the destination kube-apiserver). The destination API server verifies that peer connection based on the configuration you specify using the--requestheader-client-ca-file
command line argument.To authenticate the destination server's serving certs, you must configure a certificate authority bundle by specifying the
--peer-ca-file
command line argument to the source API server.
Configuration for peer API server connectivity
To set the network location of a kube-apiserver that peers will use to proxy requests, use the
--peer-advertise-ip
and --peer-advertise-port
command line arguments to kube-apiserver or specify
these fields in the API server configuration file.
If these flags are unspecified, peers will use the value from either --advertise-address
or
--bind-address
command line argument to the kube-apiserver.
If those too, are unset, the host's default interface is used.
Mixed version proxying
When you enable mixed version proxying, the aggregation layer loads a special filter that does the following:
- When a resource request reaches an API server that cannot serve that API (either because it is at a version pre-dating the introduction of the API or the API is turned off on the API server) the API server attempts to send the request to a peer API server that can serve the requested API. It does so by identifying API groups / versions / resources that the local server doesn't recognise, and tries to proxy those requests to a peer API server that is capable of handling the request.
- If the peer API server fails to respond, the source API server responds with 503 ("Service Unavailable") error.
How it works under the hood
When an API Server receives a resource request, it first checks which API servers can
serve the requested resource. This check happens using the internal
StorageVersion
API.
If the resource is known to the API server that received the request (for example,
GET /api/v1/pods/some-pod
), the request is handled locally.If there is no internal
StorageVersion
object found for the requested resource (for example,GET /my-api/v1/my-resource
) and the configured APIService specifies proxying to an extension API server, that proxying happens following the usual flow for extension APIs.If a valid internal
StorageVersion
object is found for the requested resource (for example,GET /batch/v1/jobs
) and the API server trying to handle the request (the handling API server) has thebatch
API disabled, then the handling API server fetches the peer API servers that do serve the relevant API group / version / resource (api/v1/batch
in this case) using the information in the fetchedStorageVersion
object. The handling API server then proxies the request to one of the matching peer kube-apiservers that are aware of the requested resource.If there is no peer known for that API group / version / resource, the handling API server passes the request to its own handler chain which should eventually return a 404 ("Not Found") response.
If the handling API server has identified and selected a peer API server, but that peer fails to respond (for reasons such as network connectivity issues, or a data race between the request being received and a controller registering the peer's info into the control plane), then the handling API server responds with a 503 ("Service Unavailable") error.
3 - Containers
This page will discuss containers and container images, as well as their use in operations and solution development.
The word container is an overloaded term. Whenever you use the word, check whether your audience uses the same definition.
Each container that you run is repeatable; the standardization from having dependencies included means that you get the same behavior wherever you run it.
Containers decouple applications from the underlying host infrastructure. This makes deployment easier in different cloud or OS environments.
Each node in a Kubernetes cluster runs the containers that form the Pods assigned to that node. Containers in a Pod are co-located and co-scheduled to run on the same node.
Container images
A container image is a ready-to-run software package containing everything needed to run an application: the code and any runtime it requires, application and system libraries, and default values for any essential settings.
Containers are intended to be stateless and immutable: you should not change the code of a container that is already running. If you have a containerized application and want to make changes, the correct process is to build a new image that includes the change, then recreate the container to start from the updated image.
Container runtimes
A fundamental component that empowers Kubernetes to run containers effectively. It is responsible for managing the execution and lifecycle of containers within the Kubernetes environment.
Kubernetes supports container runtimes such as containerd, CRI-O, and any other implementation of the Kubernetes CRI (Container Runtime Interface).
Usually, you can allow your cluster to pick the default container runtime for a Pod. If you need to use more than one container runtime in your cluster, you can specify the RuntimeClass for a Pod to make sure that Kubernetes runs those containers using a particular container runtime.
You can also use RuntimeClass to run different Pods with the same container runtime but with different settings.
3.1 - Images
A container image represents binary data that encapsulates an application and all its software dependencies. Container images are executable software bundles that can run standalone and that make very well defined assumptions about their runtime environment.
You typically create a container image of your application and push it to a registry before referring to it in a Pod.
This page provides an outline of the container image concept.
Note:
If you are looking for the container images for a Kubernetes release (such as v1.31, the latest minor release), visit Download Kubernetes.Image names
Container images are usually given a name such as pause
, example/mycontainer
, or kube-apiserver
.
Images can also include a registry hostname; for example: fictional.registry.example/imagename
,
and possibly a port number as well; for example: fictional.registry.example:10443/imagename
.
If you don't specify a registry hostname, Kubernetes assumes that you mean the Docker public registry. You can change this behaviour by setting default image registry in container runtime configuration.
After the image name part you can add a tag or digest (in the same way you would when using with commands
like docker
or podman
). Tags let you identify different versions of the same series of images.
Digests are a unique identifier for a specific version of an image. Digests are hashes of the image's content,
and are immutable. Tags can be moved to point to different images, but digests are fixed.
Image tags consist of lowercase and uppercase letters, digits, underscores (_
),
periods (.
), and dashes (-
). It can be up to 128 characters long. And must follow the
next regex pattern: [a-zA-Z0-9_][a-zA-Z0-9._-]{0,127}
You can read more about and find validation regex in the
OCI Distribution Specification.
If you don't specify a tag, Kubernetes assumes you mean the tag latest
.
Image digests consists of a hash algorithm (such as sha256
) and a hash value. For example:
sha256:1ff6c18fbef2045af6b9c16bf034cc421a29027b800e4f9b68ae9b1cb3e9ae07
You can find more information about digests format in the
OCI Image Specification.
Some image name examples that Kubernetes can use are:
busybox
- Image name only, no tag or digest. Kubernetes will use Docker public registry and latest tag. (Same asdocker.io/library/busybox:latest
)busybox:1.32.0
- Image name with tag. Kubernetes will use Docker public registry. (Same asdocker.io/library/busybox:1.32.0
)registry.k8s.io/pause:latest
- Image name with a custom registry and latest tag.registry.k8s.io/pause:3.5
- Image name with a custom registry and non-latest tag.registry.k8s.io/pause@sha256:1ff6c18fbef2045af6b9c16bf034cc421a29027b800e4f9b68ae9b1cb3e9ae07
- Image name with digest.registry.k8s.io/pause:3.5@sha256:1ff6c18fbef2045af6b9c16bf034cc421a29027b800e4f9b68ae9b1cb3e9ae07
- Image name with tag and digest. Only digest will be used for pulling.
Updating images
When you first create a Deployment,
StatefulSet, Pod, or other
object that includes a Pod template, then by default the pull policy of all
containers in that pod will be set to IfNotPresent
if it is not explicitly
specified. This policy causes the
kubelet to skip pulling an
image if it already exists.
Image pull policy
The imagePullPolicy
for a container and the tag of the image affect when the
kubelet attempts to pull (download) the specified image.
Here's a list of the values you can set for imagePullPolicy
and the effects
these values have:
IfNotPresent
- the image is pulled only if it is not already present locally.
Always
- every time the kubelet launches a container, the kubelet queries the container image registry to resolve the name to an image digest. If the kubelet has a container image with that exact digest cached locally, the kubelet uses its cached image; otherwise, the kubelet pulls the image with the resolved digest, and uses that image to launch the container.
Never
- the kubelet does not try fetching the image. If the image is somehow already present locally, the kubelet attempts to start the container; otherwise, startup fails. See pre-pulled images for more details.
The caching semantics of the underlying image provider make even
imagePullPolicy: Always
efficient, as long as the registry is reliably accessible.
Your container runtime can notice that the image layers already exist on the node
so that they don't need to be downloaded again.
Note:
You should avoid using the :latest
tag when deploying containers in production as
it is harder to track which version of the image is running and more difficult to
roll back properly.
Instead, specify a meaningful tag such as v1.42.0
and/or a digest.
To make sure the Pod always uses the same version of a container image, you can specify
the image's digest;
replace <image-name>:<tag>
with <image-name>@<digest>
(for example, image@sha256:45b23dee08af5e43a7fea6c4cf9c25ccf269ee113168c19722f87876677c5cb2
).
When using image tags, if the image registry were to change the code that the tag on that image represents, you might end up with a mix of Pods running the old and new code. An image digest uniquely identifies a specific version of the image, so Kubernetes runs the same code every time it starts a container with that image name and digest specified. Specifying an image by digest fixes the code that you run so that a change at the registry cannot lead to that mix of versions.
There are third-party admission controllers that mutate Pods (and pod templates) when they are created, so that the running workload is defined based on an image digest rather than a tag. That might be useful if you want to make sure that all your workload is running the same code no matter what tag changes happen at the registry.
Default image pull policy
When you (or a controller) submit a new Pod to the API server, your cluster sets the
imagePullPolicy
field when specific conditions are met:
- if you omit the
imagePullPolicy
field, and you specify the digest for the container image, theimagePullPolicy
is automatically set toIfNotPresent
. - if you omit the
imagePullPolicy
field, and the tag for the container image is:latest
,imagePullPolicy
is automatically set toAlways
; - if you omit the
imagePullPolicy
field, and you don't specify the tag for the container image,imagePullPolicy
is automatically set toAlways
; - if you omit the
imagePullPolicy
field, and you specify the tag for the container image that isn't:latest
, theimagePullPolicy
is automatically set toIfNotPresent
.
Note:
The value of imagePullPolicy
of the container is always set when the object is
first created, and is not updated if the image's tag or digest later changes.
For example, if you create a Deployment with an image whose tag is not
:latest
, and later update that Deployment's image to a :latest
tag, the
imagePullPolicy
field will not change to Always
. You must manually change
the pull policy of any object after its initial creation.
Required image pull
If you would like to always force a pull, you can do one of the following:
- Set the
imagePullPolicy
of the container toAlways
. - Omit the
imagePullPolicy
and use:latest
as the tag for the image to use; Kubernetes will set the policy toAlways
when you submit the Pod. - Omit the
imagePullPolicy
and the tag for the image to use; Kubernetes will set the policy toAlways
when you submit the Pod. - Enable the AlwaysPullImages admission controller.
ImagePullBackOff
When a kubelet starts creating containers for a Pod using a container runtime,
it might be possible the container is in Waiting
state because of ImagePullBackOff
.
The status ImagePullBackOff
means that a container could not start because Kubernetes
could not pull a container image (for reasons such as invalid image name, or pulling
from a private registry without imagePullSecret
). The BackOff
part indicates
that Kubernetes will keep trying to pull the image, with an increasing back-off delay.
Kubernetes raises the delay between each attempt until it reaches a compiled-in limit, which is 300 seconds (5 minutes).
Image pull per runtime class
Kubernetes v1.29 [alpha]
(enabled by default: false)If you enable the RuntimeClassInImageCriApi
feature gate,
the kubelet references container images by a tuple of (image name, runtime handler) rather than just the
image name or digest. Your container runtime
may adapt its behavior based on the selected runtime handler.
Pulling images based on runtime class will be helpful for VM based containers like windows hyperV containers.
Serial and parallel image pulls
By default, kubelet pulls images serially. In other words, kubelet sends only one image pull request to the image service at a time. Other image pull requests have to wait until the one being processed is complete.
Nodes make image pull decisions in isolation. Even when you use serialized image pulls, two different nodes can pull the same image in parallel.
If you would like to enable parallel image pulls, you can set the field
serializeImagePulls
to false in the kubelet configuration.
With serializeImagePulls
set to false, image pull requests will be sent to the image service immediately,
and multiple images will be pulled at the same time.
When enabling parallel image pulls, please make sure the image service of your container runtime can handle parallel image pulls.
The kubelet never pulls multiple images in parallel on behalf of one Pod. For example, if you have a Pod that has an init container and an application container, the image pulls for the two containers will not be parallelized. However, if you have two Pods that use different images, the kubelet pulls the images in parallel on behalf of the two different Pods, when parallel image pulls is enabled.
Maximum parallel image pulls
Kubernetes v1.27 [alpha]
When serializeImagePulls
is set to false, the kubelet defaults to no limit on the
maximum number of images being pulled at the same time. If you would like to
limit the number of parallel image pulls, you can set the field maxParallelImagePulls
in kubelet configuration. With maxParallelImagePulls
set to n, only n images
can be pulled at the same time, and any image pull beyond n will have to wait
until at least one ongoing image pull is complete.
Limiting the number parallel image pulls would prevent image pulling from consuming too much network bandwidth or disk I/O, when parallel image pulling is enabled.
You can set maxParallelImagePulls
to a positive number that is greater than or
equal to 1. If you set maxParallelImagePulls
to be greater than or equal to 2, you
must set the serializeImagePulls
to false. The kubelet will fail to start with invalid
maxParallelImagePulls
settings.
Multi-architecture images with image indexes
As well as providing binary images, a container registry can also serve a
container image index.
An image index can point to multiple image manifests
for architecture-specific versions of a container. The idea is that you can have a name for an image
(for example: pause
, example/mycontainer
, kube-apiserver
) and allow different systems to
fetch the right binary image for the machine architecture they are using.
Kubernetes itself typically names container images with a suffix -$(ARCH)
. For backward
compatibility, please generate the older images with suffixes. The idea is to generate say pause
image which has the manifest for all the arch(es) and say pause-amd64
which is backwards
compatible for older configurations or YAML files which may have hard coded the images with
suffixes.
Using a private registry
Private registries may require keys to read images from them.
Credentials can be provided in several ways:
- Configuring Nodes to Authenticate to a Private Registry
- all pods can read any configured private registries
- requires node configuration by cluster administrator
- Kubelet Credential Provider to dynamically fetch credentials for private registries
- kubelet can be configured to use credential provider exec plugin for the respective private registry.
- Pre-pulled Images
- all pods can use any images cached on a node
- requires root access to all nodes to set up
- Specifying ImagePullSecrets on a Pod
- only pods which provide own keys can access the private registry
- Vendor-specific or local extensions
- if you're using a custom node configuration, you (or your cloud provider) can implement your mechanism for authenticating the node to the container registry.
These options are explained in more detail below.
Configuring nodes to authenticate to a private registry
Specific instructions for setting credentials depends on the container runtime and registry you chose to use. You should refer to your solution's documentation for the most accurate information.
For an example of configuring a private container image registry, see the Pull an Image from a Private Registry task. That example uses a private registry in Docker Hub.
Kubelet credential provider for authenticated image pulls
Note:
This approach is especially suitable when kubelet needs to fetch registry credentials dynamically. Most commonly used for registries provided by cloud providers where auth tokens are short-lived.You can configure the kubelet to invoke a plugin binary to dynamically fetch registry credentials for a container image. This is the most robust and versatile way to fetch credentials for private registries, but also requires kubelet-level configuration to enable.
See Configure a kubelet image credential provider for more details.
Interpretation of config.json
The interpretation of config.json
varies between the original Docker
implementation and the Kubernetes interpretation. In Docker, the auths
keys
can only specify root URLs, whereas Kubernetes allows glob URLs as well as
prefix-matched paths. The only limitation is that glob patterns (*
) have to
include the dot (.
) for each subdomain. The amount of matched subdomains has
to be equal to the amount of glob patterns (*.
), for example:
*.kubernetes.io
will not matchkubernetes.io
, butabc.kubernetes.io
*.*.kubernetes.io
will not matchabc.kubernetes.io
, butabc.def.kubernetes.io
prefix.*.io
will matchprefix.kubernetes.io
*-good.kubernetes.io
will matchprefix-good.kubernetes.io
This means that a config.json
like this is valid:
{
"auths": {
"my-registry.io/images": { "auth": "…" },
"*.my-registry.io/images": { "auth": "…" }
}
}
Image pull operations would now pass the credentials to the CRI container runtime for every valid pattern. For example the following container image names would match successfully:
my-registry.io/images
my-registry.io/images/my-image
my-registry.io/images/another-image
sub.my-registry.io/images/my-image
But not:
a.sub.my-registry.io/images/my-image
a.b.sub.my-registry.io/images/my-image
The kubelet performs image pulls sequentially for every found credential. This
means, that multiple entries in config.json
for different paths are possible, too:
{
"auths": {
"my-registry.io/images": {
"auth": "…"
},
"my-registry.io/images/subpath": {
"auth": "…"
}
}
}
If now a container specifies an image my-registry.io/images/subpath/my-image
to be pulled, then the kubelet will try to download them from both
authentication sources if one of them fails.
Pre-pulled images
Note:
This approach is suitable if you can control node configuration. It will not work reliably if your cloud provider manages nodes and replaces them automatically.By default, the kubelet tries to pull each image from the specified registry.
However, if the imagePullPolicy
property of the container is set to IfNotPresent
or Never
,
then a local image is used (preferentially or exclusively, respectively).
If you want to rely on pre-pulled images as a substitute for registry authentication, you must ensure all nodes in the cluster have the same pre-pulled images.
This can be used to preload certain images for speed or as an alternative to authenticating to a private registry.
All pods will have read access to any pre-pulled images.
Specifying imagePullSecrets on a Pod
Note:
This is the recommended approach to run containers based on images in private registries.Kubernetes supports specifying container image registry keys on a Pod.
imagePullSecrets
must all be in the same namespace as the Pod. The referenced
Secrets must be of type kubernetes.io/dockercfg
or kubernetes.io/dockerconfigjson
.
Creating a Secret with a Docker config
You need to know the username, registry password and client email address for authenticating to the registry, as well as its hostname. Run the following command, substituting the appropriate uppercase values:
kubectl create secret docker-registry <name> \
--docker-server=DOCKER_REGISTRY_SERVER \
--docker-username=DOCKER_USER \
--docker-password=DOCKER_PASSWORD \
--docker-email=DOCKER_EMAIL
If you already have a Docker credentials file then, rather than using the above
command, you can import the credentials file as a Kubernetes
Secrets.
Create a Secret based on existing Docker credentials
explains how to set this up.
This is particularly useful if you are using multiple private container
registries, as kubectl create secret docker-registry
creates a Secret that
only works with a single private registry.
Note:
Pods can only reference image pull secrets in their own namespace, so this process needs to be done one time per namespace.Referring to an imagePullSecrets on a Pod
Now, you can create pods which reference that secret by adding an imagePullSecrets
section to a Pod definition. Each item in the imagePullSecrets
array can only
reference a Secret in the same namespace.
For example:
cat <<EOF > pod.yaml
apiVersion: v1
kind: Pod
metadata:
name: foo
namespace: awesomeapps
spec:
containers:
- name: foo
image: janedoe/awesomeapp:v1
imagePullSecrets:
- name: myregistrykey
EOF
cat <<EOF >> ./kustomization.yaml
resources:
- pod.yaml
EOF
This needs to be done for each pod that is using a private registry.
However, setting of this field can be automated by setting the imagePullSecrets in a ServiceAccount resource.
Check Add ImagePullSecrets to a Service Account for detailed instructions.
You can use this in conjunction with a per-node .docker/config.json
. The credentials
will be merged.
Use cases
There are a number of solutions for configuring private registries. Here are some common use cases and suggested solutions.
- Cluster running only non-proprietary (e.g. open-source) images. No need to hide images.
- Use public images from a public registry
- No configuration required.
- Some cloud providers automatically cache or mirror public images, which improves availability and reduces the time to pull images.
- Use public images from a public registry
- Cluster running some proprietary images which should be hidden to those outside the company, but
visible to all cluster users.
- Use a hosted private registry
- Manual configuration may be required on the nodes that need to access to private registry
- Or, run an internal private registry behind your firewall with open read access.
- No Kubernetes configuration is required.
- Use a hosted container image registry service that controls image access
- It will work better with cluster autoscaling than manual node configuration.
- Or, on a cluster where changing the node configuration is inconvenient, use
imagePullSecrets
.
- Use a hosted private registry
- Cluster with proprietary images, a few of which require stricter access control.
- Ensure AlwaysPullImages admission controller is active. Otherwise, all Pods potentially have access to all images.
- Move sensitive data into a "Secret" resource, instead of packaging it in an image.
- A multi-tenant cluster where each tenant needs own private registry.
- Ensure AlwaysPullImages admission controller is active. Otherwise, all Pods of all tenants potentially have access to all images.
- Run a private registry with authorization required.
- Generate registry credential for each tenant, put into secret, and populate secret to each tenant namespace.
- The tenant adds that secret to imagePullSecrets of each namespace.
If you need access to multiple registries, you can create one secret for each registry.
Legacy built-in kubelet credential provider
In older versions of Kubernetes, the kubelet had a direct integration with cloud provider credentials. This gave it the ability to dynamically fetch credentials for image registries.
There were three built-in implementations of the kubelet credential provider integration: ACR (Azure Container Registry), ECR (Elastic Container Registry), and GCR (Google Container Registry).
For more information on the legacy mechanism, read the documentation for the version of Kubernetes that you are using. Kubernetes v1.26 through to v1.31 do not include the legacy mechanism, so you would need to either:
- configure a kubelet image credential provider on each node
- specify image pull credentials using
imagePullSecrets
and at least one Secret
What's next
- Read the OCI Image Manifest Specification.
- Learn about container image garbage collection.
- Learn more about pulling an Image from a Private Registry.
3.2 - Container Environment
This page describes the resources available to Containers in the Container environment.
Container environment
The Kubernetes Container environment provides several important resources to Containers:
- A filesystem, which is a combination of an image and one or more volumes.
- Information about the Container itself.
- Information about other objects in the cluster.
Container information
The hostname of a Container is the name of the Pod in which the Container is running.
It is available through the hostname
command or the
gethostname
function call in libc.
The Pod name and namespace are available as environment variables through the downward API.
User defined environment variables from the Pod definition are also available to the Container, as are any environment variables specified statically in the container image.
Cluster information
A list of all services that were running when a Container was created is available to that Container as environment variables. This list is limited to services within the same namespace as the new Container's Pod and Kubernetes control plane services.
For a service named foo that maps to a Container named bar, the following variables are defined:
FOO_SERVICE_HOST=<the host the service is running on>
FOO_SERVICE_PORT=<the port the service is running on>
Services have dedicated IP addresses and are available to the Container via DNS, if DNS addon is enabled.
What's next
- Learn more about Container lifecycle hooks.
- Get hands-on experience attaching handlers to Container lifecycle events.
3.3 - Runtime Class
Kubernetes v1.20 [stable]
This page describes the RuntimeClass resource and runtime selection mechanism.
RuntimeClass is a feature for selecting the container runtime configuration. The container runtime configuration is used to run a Pod's containers.
Motivation
You can set a different RuntimeClass between different Pods to provide a balance of performance versus security. For example, if part of your workload deserves a high level of information security assurance, you might choose to schedule those Pods so that they run in a container runtime that uses hardware virtualization. You'd then benefit from the extra isolation of the alternative runtime, at the expense of some additional overhead.
You can also use RuntimeClass to run different Pods with the same container runtime but with different settings.
Setup
- Configure the CRI implementation on nodes (runtime dependent)
- Create the corresponding RuntimeClass resources
1. Configure the CRI implementation on nodes
The configurations available through RuntimeClass are Container Runtime Interface (CRI) implementation dependent. See the corresponding documentation (below) for your CRI implementation for how to configure.
Note:
RuntimeClass assumes a homogeneous node configuration across the cluster by default (which means that all nodes are configured the same way with respect to container runtimes). To support heterogeneous node configurations, see Scheduling below.The configurations have a corresponding handler
name, referenced by the RuntimeClass. The
handler must be a valid DNS label name.
2. Create the corresponding RuntimeClass resources
The configurations setup in step 1 should each have an associated handler
name, which identifies
the configuration. For each handler, create a corresponding RuntimeClass object.
The RuntimeClass resource currently only has 2 significant fields: the RuntimeClass name
(metadata.name
) and the handler (handler
). The object definition looks like this:
# RuntimeClass is defined in the node.k8s.io API group
apiVersion: node.k8s.io/v1
kind: RuntimeClass
metadata:
# The name the RuntimeClass will be referenced by.
# RuntimeClass is a non-namespaced resource.
name: myclass
# The name of the corresponding CRI configuration
handler: myconfiguration
The name of a RuntimeClass object must be a valid DNS subdomain name.
Note:
It is recommended that RuntimeClass write operations (create/update/patch/delete) be restricted to the cluster administrator. This is typically the default. See Authorization Overview for more details.Usage
Once RuntimeClasses are configured for the cluster, you can specify a
runtimeClassName
in the Pod spec to use it. For example:
apiVersion: v1
kind: Pod
metadata:
name: mypod
spec:
runtimeClassName: myclass
# ...
This will instruct the kubelet to use the named RuntimeClass to run this pod. If the named
RuntimeClass does not exist, or the CRI cannot run the corresponding handler, the pod will enter the
Failed
terminal phase. Look for a
corresponding event for an
error message.
If no runtimeClassName
is specified, the default RuntimeHandler will be used, which is equivalent
to the behavior when the RuntimeClass feature is disabled.
CRI Configuration
For more details on setting up CRI runtimes, see CRI installation.
containerd
Runtime handlers are configured through containerd's configuration at
/etc/containerd/config.toml
. Valid handlers are configured under the runtimes section:
[plugins."io.containerd.grpc.v1.cri".containerd.runtimes.${HANDLER_NAME}]
See containerd's config documentation for more details:
CRI-O
Runtime handlers are configured through CRI-O's configuration at /etc/crio/crio.conf
. Valid
handlers are configured under the
crio.runtime table:
[crio.runtime.runtimes.${HANDLER_NAME}]
runtime_path = "${PATH_TO_BINARY}"
See CRI-O's config documentation for more details.
Scheduling
Kubernetes v1.16 [beta]
By specifying the scheduling
field for a RuntimeClass, you can set constraints to
ensure that Pods running with this RuntimeClass are scheduled to nodes that support it.
If scheduling
is not set, this RuntimeClass is assumed to be supported by all nodes.
To ensure pods land on nodes supporting a specific RuntimeClass, that set of nodes should have a
common label which is then selected by the runtimeclass.scheduling.nodeSelector
field. The
RuntimeClass's nodeSelector is merged with the pod's nodeSelector in admission, effectively taking
the intersection of the set of nodes selected by each. If there is a conflict, the pod will be
rejected.
If the supported nodes are tainted to prevent other RuntimeClass pods from running on the node, you
can add tolerations
to the RuntimeClass. As with the nodeSelector
, the tolerations are merged
with the pod's tolerations in admission, effectively taking the union of the set of nodes tolerated
by each.
To learn more about configuring the node selector and tolerations, see Assigning Pods to Nodes.
Pod Overhead
Kubernetes v1.24 [stable]
You can specify overhead resources that are associated with running a Pod. Declaring overhead allows the cluster (including the scheduler) to account for it when making decisions about Pods and resources.
Pod overhead is defined in RuntimeClass through the overhead
field. Through the use of this field,
you can specify the overhead of running pods utilizing this RuntimeClass and ensure these overheads
are accounted for in Kubernetes.
What's next
- RuntimeClass Design
- RuntimeClass Scheduling Design
- Read about the Pod Overhead concept
- PodOverhead Feature Design
3.4 - Container Lifecycle Hooks
This page describes how kubelet managed Containers can use the Container lifecycle hook framework to run code triggered by events during their management lifecycle.
Overview
Analogous to many programming language frameworks that have component lifecycle hooks, such as Angular, Kubernetes provides Containers with lifecycle hooks. The hooks enable Containers to be aware of events in their management lifecycle and run code implemented in a handler when the corresponding lifecycle hook is executed.
Container hooks
There are two hooks that are exposed to Containers:
PostStart
This hook is executed immediately after a container is created. However, there is no guarantee that the hook will execute before the container ENTRYPOINT. No parameters are passed to the handler.
PreStop
This hook is called immediately before a container is terminated due to an API request or management
event such as a liveness/startup probe failure, preemption, resource contention and others. A call
to the PreStop
hook fails if the container is already in a terminated or completed state and the
hook must complete before the TERM signal to stop the container can be sent. The Pod's termination
grace period countdown begins before the PreStop
hook is executed, so regardless of the outcome of
the handler, the container will eventually terminate within the Pod's termination grace period. No
parameters are passed to the handler.
A more detailed description of the termination behavior can be found in Termination of Pods.
Hook handler implementations
Containers can access a hook by implementing and registering a handler for that hook. There are three types of hook handlers that can be implemented for Containers:
- Exec - Executes a specific command, such as
pre-stop.sh
, inside the cgroups and namespaces of the Container. Resources consumed by the command are counted against the Container. - HTTP - Executes an HTTP request against a specific endpoint on the Container.
- Sleep - Pauses the container for a specified duration.
This is a beta-level feature default enabled by the
PodLifecycleSleepAction
feature gate.
Hook handler execution
When a Container lifecycle management hook is called,
the Kubernetes management system executes the handler according to the hook action,
httpGet
, tcpSocket
and sleep
are executed by the kubelet process, and exec
is executed in the container.
The PostStart
hook handler call is initiated when a container is created,
meaning the container ENTRYPOINT and the PostStart
hook are triggered simultaneously.
However, if the PostStart
hook takes too long to execute or if it hangs,
it can prevent the container from transitioning to a running
state.
PreStop
hooks are not executed asynchronously from the signal to stop the Container; the hook must
complete its execution before the TERM signal can be sent. If a PreStop
hook hangs during
execution, the Pod's phase will be Terminating
and remain there until the Pod is killed after its
terminationGracePeriodSeconds
expires. This grace period applies to the total time it takes for
both the PreStop
hook to execute and for the Container to stop normally. If, for example,
terminationGracePeriodSeconds
is 60, and the hook takes 55 seconds to complete, and the Container
takes 10 seconds to stop normally after receiving the signal, then the Container will be killed
before it can stop normally, since terminationGracePeriodSeconds
is less than the total time
(55+10) it takes for these two things to happen.
If either a PostStart
or PreStop
hook fails,
it kills the Container.
Users should make their hook handlers as lightweight as possible. There are cases, however, when long running commands make sense, such as when saving state prior to stopping a Container.
Hook delivery guarantees
Hook delivery is intended to be at least once,
which means that a hook may be called multiple times for any given event,
such as for PostStart
or PreStop
.
It is up to the hook implementation to handle this correctly.
Generally, only single deliveries are made. If, for example, an HTTP hook receiver is down and is unable to take traffic, there is no attempt to resend. In some rare cases, however, double delivery may occur. For instance, if a kubelet restarts in the middle of sending a hook, the hook might be resent after the kubelet comes back up.
Debugging Hook handlers
The logs for a Hook handler are not exposed in Pod events.
If a handler fails for some reason, it broadcasts an event.
For PostStart
, this is the FailedPostStartHook
event,
and for PreStop
, this is the FailedPreStopHook
event.
To generate a failed FailedPostStartHook
event yourself, modify the lifecycle-events.yaml file to change the postStart command to "badcommand" and apply it.
Here is some example output of the resulting events you see from running kubectl describe pod lifecycle-demo
:
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal Scheduled 7s default-scheduler Successfully assigned default/lifecycle-demo to ip-XXX-XXX-XX-XX.us-east-2...
Normal Pulled 6s kubelet Successfully pulled image "nginx" in 229.604315ms
Normal Pulling 4s (x2 over 6s) kubelet Pulling image "nginx"
Normal Created 4s (x2 over 5s) kubelet Created container lifecycle-demo-container
Normal Started 4s (x2 over 5s) kubelet Started container lifecycle-demo-container
Warning FailedPostStartHook 4s (x2 over 5s) kubelet Exec lifecycle hook ([badcommand]) for Container "lifecycle-demo-container" in Pod "lifecycle-demo_default(30229739-9651-4e5a-9a32-a8f1688862db)" failed - error: command 'badcommand' exited with 126: , message: "OCI runtime exec failed: exec failed: container_linux.go:380: starting container process caused: exec: \"badcommand\": executable file not found in $PATH: unknown\r\n"
Normal Killing 4s (x2 over 5s) kubelet FailedPostStartHook
Normal Pulled 4s kubelet Successfully pulled image "nginx" in 215.66395ms
Warning BackOff 2s (x2 over 3s) kubelet Back-off restarting failed container
What's next
- Learn more about the Container environment.
- Get hands-on experience attaching handlers to Container lifecycle events.
4 - Workloads
A workload is an application running on Kubernetes. Whether your workload is a single component or several that work together, on Kubernetes you run it inside a set of pods. In Kubernetes, a Pod represents a set of running containers on your cluster.
Kubernetes pods have a defined lifecycle. For example, once a pod is running in your cluster then a critical fault on the node where that pod is running means that all the pods on that node fail. Kubernetes treats that level of failure as final: you would need to create a new Pod to recover, even if the node later becomes healthy.
However, to make life considerably easier, you don't need to manage each Pod directly. Instead, you can use workload resources that manage a set of pods on your behalf. These resources configure controllers that make sure the right number of the right kind of pod are running, to match the state you specified.
Kubernetes provides several built-in workload resources:
- Deployment and ReplicaSet (replacing the legacy resource ReplicationController). Deployment is a good fit for managing a stateless application workload on your cluster, where any Pod in the Deployment is interchangeable and can be replaced if needed.
- StatefulSet lets you run one or more related Pods that do track state somehow. For example, if your workload records data persistently, you can run a StatefulSet that matches each Pod with a PersistentVolume. Your code, running in the Pods for that StatefulSet, can replicate data to other Pods in the same StatefulSet to improve overall resilience.
- DaemonSet defines Pods that provide facilities that are local to nodes. Every time you add a node to your cluster that matches the specification in a DaemonSet, the control plane schedules a Pod for that DaemonSet onto the new node. Each pod in a DaemonSet performs a job similar to a system daemon on a classic Unix / POSIX server. A DaemonSet might be fundamental to the operation of your cluster, such as a plugin to run cluster networking, it might help you to manage the node, or it could provide optional behavior that enhances the container platform you are running.
- Job and CronJob provide different ways to define tasks that run to completion and then stop. You can use a Job to define a task that runs to completion, just once. You can use a CronJob to run the same Job multiple times according a schedule.
In the wider Kubernetes ecosystem, you can find third-party workload resources that provide additional behaviors. Using a custom resource definition, you can add in a third-party workload resource if you want a specific behavior that's not part of Kubernetes' core. For example, if you wanted to run a group of Pods for your application but stop work unless all the Pods are available (perhaps for some high-throughput distributed task), then you can implement or install an extension that does provide that feature.
What's next
As well as reading about each API kind for workload management, you can read how to do specific tasks:
- Run a stateless application using a Deployment
- Run a stateful application either as a single instance or as a replicated set
- Run automated tasks with a CronJob
To learn about Kubernetes' mechanisms for separating code from configuration, visit Configuration.
There are two supporting concepts that provide backgrounds about how Kubernetes manages pods for applications:
- Garbage collection tidies up objects from your cluster after their owning resource has been removed.
- The time-to-live after finished controller removes Jobs once a defined time has passed since they completed.
Once your application is running, you might want to make it available on the internet as a Service or, for web application only, using an Ingress.
4.1 - Pods
Pods are the smallest deployable units of computing that you can create and manage in Kubernetes.
A Pod (as in a pod of whales or pea pod) is a group of one or more containers, with shared storage and network resources, and a specification for how to run the containers. A Pod's contents are always co-located and co-scheduled, and run in a shared context. A Pod models an application-specific "logical host": it contains one or more application containers which are relatively tightly coupled. In non-cloud contexts, applications executed on the same physical or virtual machine are analogous to cloud applications executed on the same logical host.
As well as application containers, a Pod can contain init containers that run during Pod startup. You can also inject ephemeral containers for debugging a running Pod.
What is a Pod?
Note:
You need to install a container runtime into each node in the cluster so that Pods can run there.The shared context of a Pod is a set of Linux namespaces, cgroups, and potentially other facets of isolation - the same things that isolate a container. Within a Pod's context, the individual applications may have further sub-isolations applied.
A Pod is similar to a set of containers with shared namespaces and shared filesystem volumes.
Pods in a Kubernetes cluster are used in two main ways:
Pods that run a single container. The "one-container-per-Pod" model is the most common Kubernetes use case; in this case, you can think of a Pod as a wrapper around a single container; Kubernetes manages Pods rather than managing the containers directly.
Pods that run multiple containers that need to work together. A Pod can encapsulate an application composed of multiple co-located containers that are tightly coupled and need to share resources. These co-located containers form a single cohesive unit.
Grouping multiple co-located and co-managed containers in a single Pod is a relatively advanced use case. You should use this pattern only in specific instances in which your containers are tightly coupled.
You don't need to run multiple containers to provide replication (for resilience or capacity); if you need multiple replicas, see Workload management.
Using Pods
The following is an example of a Pod which consists of a container running the image nginx:1.14.2
.
apiVersion: v1
kind: Pod
metadata:
name: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
To create the Pod shown above, run the following command:
kubectl apply -f https://k8s.io/examples/pods/simple-pod.yaml
Pods are generally not created directly and are created using workload resources. See Working with Pods for more information on how Pods are used with workload resources.
Workload resources for managing pods
Usually you don't need to create Pods directly, even singleton Pods. Instead, create them using workload resources such as Deployment or Job. If your Pods need to track state, consider the StatefulSet resource.
Each Pod is meant to run a single instance of a given application. If you want to scale your application horizontally (to provide more overall resources by running more instances), you should use multiple Pods, one for each instance. In Kubernetes, this is typically referred to as replication. Replicated Pods are usually created and managed as a group by a workload resource and its controller.
See Pods and controllers for more information on how Kubernetes uses workload resources, and their controllers, to implement application scaling and auto-healing.
Pods natively provide two kinds of shared resources for their constituent containers: networking and storage.
Working with Pods
You'll rarely create individual Pods directly in Kubernetes—even singleton Pods. This is because Pods are designed as relatively ephemeral, disposable entities. When a Pod gets created (directly by you, or indirectly by a controller), the new Pod is scheduled to run on a Node in your cluster. The Pod remains on that node until the Pod finishes execution, the Pod object is deleted, the Pod is evicted for lack of resources, or the node fails.
Note:
Restarting a container in a Pod should not be confused with restarting a Pod. A Pod is not a process, but an environment for running container(s). A Pod persists until it is deleted.The name of a Pod must be a valid DNS subdomain value, but this can produce unexpected results for the Pod hostname. For best compatibility, the name should follow the more restrictive rules for a DNS label.
Pod OS
Kubernetes v1.25 [stable]
You should set the .spec.os.name
field to either windows
or linux
to indicate the OS on
which you want the pod to run. These two are the only operating systems supported for now by
Kubernetes. In the future, this list may be expanded.
In Kubernetes v1.31, the value of .spec.os.name
does not affect
how the kube-scheduler
picks a node for the Pod to run on. In any cluster where there is more than one operating system for
running nodes, you should set the
kubernetes.io/os
label correctly on each node, and define pods with a nodeSelector
based on the operating system
label. The kube-scheduler assigns your pod to a node based on other criteria and may or may not
succeed in picking a suitable node placement where the node OS is right for the containers in that Pod.
The Pod security standards also use this
field to avoid enforcing policies that aren't relevant to the operating system.
Pods and controllers
You can use workload resources to create and manage multiple Pods for you. A controller for the resource handles replication and rollout and automatic healing in case of Pod failure. For example, if a Node fails, a controller notices that Pods on that Node have stopped working and creates a replacement Pod. The scheduler places the replacement Pod onto a healthy Node.
Here are some examples of workload resources that manage one or more Pods:
Pod templates
Controllers for workload resources create Pods from a pod template and manage those Pods on your behalf.
PodTemplates are specifications for creating Pods, and are included in workload resources such as Deployments, Jobs, and DaemonSets.
Each controller for a workload resource uses the PodTemplate
inside the workload
object to make actual Pods. The PodTemplate
is part of the desired state of whatever
workload resource you used to run your app.
When you create a Pod, you can include environment variables in the Pod template for the containers that run in the Pod.
The sample below is a manifest for a simple Job with a template
that starts one
container. The container in that Pod prints a message then pauses.
apiVersion: batch/v1
kind: Job
metadata:
name: hello
spec:
template:
# This is the pod template
spec:
containers:
- name: hello
image: busybox:1.28
command: ['sh', '-c', 'echo "Hello, Kubernetes!" && sleep 3600']
restartPolicy: OnFailure
# The pod template ends here
Modifying the pod template or switching to a new pod template has no direct effect on the Pods that already exist. If you change the pod template for a workload resource, that resource needs to create replacement Pods that use the updated template.
For example, the StatefulSet controller ensures that the running Pods match the current pod template for each StatefulSet object. If you edit the StatefulSet to change its pod template, the StatefulSet starts to create new Pods based on the updated template. Eventually, all of the old Pods are replaced with new Pods, and the update is complete.
Each workload resource implements its own rules for handling changes to the Pod template. If you want to read more about StatefulSet specifically, read Update strategy in the StatefulSet Basics tutorial.
On Nodes, the kubelet does not directly observe or manage any of the details around pod templates and updates; those details are abstracted away. That abstraction and separation of concerns simplifies system semantics, and makes it feasible to extend the cluster's behavior without changing existing code.
Pod update and replacement
As mentioned in the previous section, when the Pod template for a workload resource is changed, the controller creates new Pods based on the updated template instead of updating or patching the existing Pods.
Kubernetes doesn't prevent you from managing Pods directly. It is possible to
update some fields of a running Pod, in place. However, Pod update operations
like
patch
, and
replace
have some limitations:
Most of the metadata about a Pod is immutable. For example, you cannot change the
namespace
,name
,uid
, orcreationTimestamp
fields; thegeneration
field is unique. It only accepts updates that increment the field's current value.If the
metadata.deletionTimestamp
is set, no new entry can be added to themetadata.finalizers
list.Pod updates may not change fields other than
spec.containers[*].image
,spec.initContainers[*].image
,spec.activeDeadlineSeconds
orspec.tolerations
. Forspec.tolerations
, you can only add new entries.When updating the
spec.activeDeadlineSeconds
field, two types of updates are allowed:- setting the unassigned field to a positive number;
- updating the field from a positive number to a smaller, non-negative number.
Resource sharing and communication
Pods enable data sharing and communication among their constituent containers.
Storage in Pods
A Pod can specify a set of shared storage volumes. All containers in the Pod can access the shared volumes, allowing those containers to share data. Volumes also allow persistent data in a Pod to survive in case one of the containers within needs to be restarted. See Storage for more information on how Kubernetes implements shared storage and makes it available to Pods.
Pod networking
Each Pod is assigned a unique IP address for each address family. Every
container in a Pod shares the network namespace, including the IP address and
network ports. Inside a Pod (and only then), the containers that belong to the Pod
can communicate with one another using localhost
. When containers in a Pod communicate
with entities outside the Pod,
they must coordinate how they use the shared network resources (such as ports).
Within a Pod, containers share an IP address and port space, and
can find each other via localhost
. The containers in a Pod can also communicate
with each other using standard inter-process communications like SystemV semaphores
or POSIX shared memory. Containers in different Pods have distinct IP addresses
and can not communicate by OS-level IPC without special configuration.
Containers that want to interact with a container running in a different Pod can
use IP networking to communicate.
Containers within the Pod see the system hostname as being the same as the configured
name
for the Pod. There's more about this in the networking
section.
Pod security settings
To set security constraints on Pods and containers, you use the
securityContext
field in the Pod specification. This field gives you
granular control over what a Pod or individual containers can do. For example:
- Drop specific Linux capabilities to avoid the impact of a CVE.
- Force all processes in the Pod to run as a non-root user or as a specific user or group ID.
- Set a specific seccomp profile.
- Set Windows security options, such as whether containers run as HostProcess.
Caution:
You can also use the Pod securityContext to enable privileged mode in Linux containers. Privileged mode overrides many of the other security settings in the securityContext. Avoid using this setting unless you can't grant the equivalent permissions by using other fields in the securityContext. In Kubernetes 1.26 and later, you can run Windows containers in a similarly privileged mode by setting thewindowsOptions.hostProcess
flag on the
security context of the Pod spec. For details and instructions, see
Create a Windows HostProcess Pod.- To learn about kernel-level security constraints that you can use, see Linux kernel security constraints for Pods and containers.
- To learn more about the Pod security context, see Configure a Security Context for a Pod or Container.
Static Pods
Static Pods are managed directly by the kubelet daemon on a specific node, without the API server observing them. Whereas most Pods are managed by the control plane (for example, a Deployment), for static Pods, the kubelet directly supervises each static Pod (and restarts it if it fails).
Static Pods are always bound to one Kubelet on a specific node. The main use for static Pods is to run a self-hosted control plane: in other words, using the kubelet to supervise the individual control plane components.
The kubelet automatically tries to create a mirror Pod on the Kubernetes API server for each static Pod. This means that the Pods running on a node are visible on the API server, but cannot be controlled from there. See the guide Create static Pods for more information.
Note:
Thespec
of a static Pod cannot refer to other API objects
(e.g., ServiceAccount,
ConfigMap,
Secret, etc).Pods with multiple containers
Pods are designed to support multiple cooperating processes (as containers) that form a cohesive unit of service. The containers in a Pod are automatically co-located and co-scheduled on the same physical or virtual machine in the cluster. The containers can share resources and dependencies, communicate with one another, and coordinate when and how they are terminated.
Pods in a Kubernetes cluster are used in two main ways:
- Pods that run a single container. The "one-container-per-Pod" model is the most common Kubernetes use case; in this case, you can think of a Pod as a wrapper around a single container; Kubernetes manages Pods rather than managing the containers directly.
- Pods that run multiple containers that need to work together. A Pod can encapsulate an application composed of multiple co-located containers that are tightly coupled and need to share resources. These co-located containers form a single cohesive unit of service—for example, one container serving data stored in a shared volume to the public, while a separate sidecar container refreshes or updates those files. The Pod wraps these containers, storage resources, and an ephemeral network identity together as a single unit.
For example, you might have a container that acts as a web server for files in a shared volume, and a separate sidecar container that updates those files from a remote source, as in the following diagram:
Some Pods have init containers as well as app containers. By default, init containers run and complete before the app containers are started.
You can also have sidecar containers that provide auxiliary services to the main application Pod (for example: a service mesh).
Kubernetes v1.29 [beta]
Enabled by default, the SidecarContainers
feature gate
allows you to specify restartPolicy: Always
for init containers.
Setting the Always
restart policy ensures that the containers where you set it are
treated as sidecars that are kept running during the entire lifetime of the Pod.
Containers that you explicitly define as sidecar containers
start up before the main application Pod and remain running until the Pod is
shut down.
Container probes
A probe is a diagnostic performed periodically by the kubelet on a container. To perform a diagnostic, the kubelet can invoke different actions:
ExecAction
(performed with the help of the container runtime)TCPSocketAction
(checked directly by the kubelet)HTTPGetAction
(checked directly by the kubelet)
You can read more about probes in the Pod Lifecycle documentation.
What's next
- Learn about the lifecycle of a Pod.
- Learn about RuntimeClass and how you can use it to configure different Pods with different container runtime configurations.
- Read about PodDisruptionBudget and how you can use it to manage application availability during disruptions.
- Pod is a top-level resource in the Kubernetes REST API. The Pod object definition describes the object in detail.
- The Distributed System Toolkit: Patterns for Composite Containers explains common layouts for Pods with more than one container.
- Read about Pod topology spread constraints
To understand the context for why Kubernetes wraps a common Pod API in other resources (such as StatefulSets or Deployments), you can read about the prior art, including:
4.1.1 - Pod Lifecycle
This page describes the lifecycle of a Pod. Pods follow a defined lifecycle, starting
in the Pending
phase, moving through Running
if at least one
of its primary containers starts OK, and then through either the Succeeded
or
Failed
phases depending on whether any container in the Pod terminated in failure.
Like individual application containers, Pods are considered to be relatively ephemeral (rather than durable) entities. Pods are created, assigned a unique ID (UID), and scheduled to run on nodes where they remain until termination (according to restart policy) or deletion. If a Node dies, the Pods running on (or scheduled to run on) that node are marked for deletion. The control plane marks the Pods for removal after a timeout period.
Pod lifetime
Whilst a Pod is running, the kubelet is able to restart containers to handle some kind of faults. Within a Pod, Kubernetes tracks different container states and determines what action to take to make the Pod healthy again.
In the Kubernetes API, Pods have both a specification and an actual status. The status for a Pod object consists of a set of Pod conditions. You can also inject custom readiness information into the condition data for a Pod, if that is useful to your application.
Pods are only scheduled once in their lifetime; assigning a Pod to a specific node is called binding, and the process of selecting which node to use is called scheduling. Once a Pod has been scheduled and is bound to a node, Kubernetes tries to run that Pod on the node. The Pod runs on that node until it stops, or until the Pod is terminated; if Kubernetes isn't able to start the Pod on the selected node (for example, if the node crashes before the Pod starts), then that particular Pod never starts.
You can use Pod Scheduling Readiness to delay scheduling for a Pod until all its scheduling gates are removed. For example, you might want to define a set of Pods but only trigger scheduling once all the Pods have been created.
Pods and fault recovery
If one of the containers in the Pod fails, then Kubernetes may try to restart that specific container. Read How Pods handle problems with containers to learn more.
Pods can however fail in a way that the cluster cannot recover from, and in that case Kubernetes does not attempt to heal the Pod further; instead, Kubernetes deletes the Pod and relies on other components to provide automatic healing.
If a Pod is scheduled to a node and that node then fails, the Pod is treated as unhealthy and Kubernetes eventually deletes the Pod. A Pod won't survive an eviction due to a lack of resources or Node maintenance.
Kubernetes uses a higher-level abstraction, called a controller, that handles the work of managing the relatively disposable Pod instances.
A given Pod (as defined by a UID) is never "rescheduled" to a different node; instead,
that Pod can be replaced by a new, near-identical Pod. If you make a replacement Pod, it can
even have same name (as in .metadata.name
) that the old Pod had, but the replacement
would have a different .metadata.uid
from the old Pod.
Kubernetes does not guarantee that a replacement for an existing Pod would be scheduled to the same node as the old Pod that was being replaced.
Associated lifetimes
When something is said to have the same lifetime as a Pod, such as a volume, that means that the thing exists as long as that specific Pod (with that exact UID) exists. If that Pod is deleted for any reason, and even if an identical replacement is created, the related thing (a volume, in this example) is also destroyed and created anew.
Pod phase
A Pod's status
field is a
PodStatus
object, which has a phase
field.
The phase of a Pod is a simple, high-level summary of where the Pod is in its lifecycle. The phase is not intended to be a comprehensive rollup of observations of container or Pod state, nor is it intended to be a comprehensive state machine.
The number and meanings of Pod phase values are tightly guarded.
Other than what is documented here, nothing should be assumed about Pods that
have a given phase
value.
Here are the possible values for phase
:
Value | Description |
---|---|
Pending | The Pod has been accepted by the Kubernetes cluster, but one or more of the containers has not been set up and made ready to run. This includes time a Pod spends waiting to be scheduled as well as the time spent downloading container images over the network. |
Running | The Pod has been bound to a node, and all of the containers have been created. At least one container is still running, or is in the process of starting or restarting. |
Succeeded | All containers in the Pod have terminated in success, and will not be restarted. |
Failed | All containers in the Pod have terminated, and at least one container has terminated in failure. That is, the container either exited with non-zero status or was terminated by the system, and is not set for automatic restarting. |
Unknown | For some reason the state of the Pod could not be obtained. This phase typically occurs due to an error in communicating with the node where the Pod should be running. |
Note:
When a pod is failing to start repeatedly, CrashLoopBackOff
may appear in the Status
field of some kubectl commands. Similarly, when a pod is being deleted, Terminating
may appear in the Status
field of some kubectl commands.
Make sure not to confuse Status, a kubectl display field for user intuition, with the pod's phase
.
Pod phase is an explicit part of the Kubernetes data model and of the
Pod API.
NAMESPACE NAME READY STATUS RESTARTS AGE
alessandras-namespace alessandras-pod 0/1 CrashLoopBackOff 200 2d9h
A Pod is granted a term to terminate gracefully, which defaults to 30 seconds.
You can use the flag --force
to terminate a Pod by force.
Since Kubernetes 1.27, the kubelet transitions deleted Pods, except for
static Pods and
force-deleted Pods
without a finalizer, to a terminal phase (Failed
or Succeeded
depending on
the exit statuses of the pod containers) before their deletion from the API server.
If a node dies or is disconnected from the rest of the cluster, Kubernetes
applies a policy for setting the phase
of all Pods on the lost node to Failed.
Container states
As well as the phase of the Pod overall, Kubernetes tracks the state of each container inside a Pod. You can use container lifecycle hooks to trigger events to run at certain points in a container's lifecycle.
Once the scheduler
assigns a Pod to a Node, the kubelet starts creating containers for that Pod
using a container runtime.
There are three possible container states: Waiting
, Running
, and Terminated
.
To check the state of a Pod's containers, you can use
kubectl describe pod <name-of-pod>
. The output shows the state for each container
within that Pod.
Each state has a specific meaning:
Waiting
If a container is not in either the Running
or Terminated
state, it is Waiting
.
A container in the Waiting
state is still running the operations it requires in
order to complete start up: for example, pulling the container image from a container
image registry, or applying Secret
data.
When you use kubectl
to query a Pod with a container that is Waiting
, you also see
a Reason field to summarize why the container is in that state.
Running
The Running
status indicates that a container is executing without issues. If there
was a postStart
hook configured, it has already executed and finished. When you use
kubectl
to query a Pod with a container that is Running
, you also see information
about when the container entered the Running
state.
Terminated
A container in the Terminated
state began execution and then either ran to
completion or failed for some reason. When you use kubectl
to query a Pod with
a container that is Terminated
, you see a reason, an exit code, and the start and
finish time for that container's period of execution.
If a container has a preStop
hook configured, this hook runs before the container enters
the Terminated
state.
How Pods handle problems with containers
Kubernetes manages container failures within Pods using a restartPolicy
defined in the Pod spec
. This policy determines how Kubernetes reacts to containers exiting due to errors or other reasons, which falls in the following sequence:
- Initial crash: Kubernetes attempts an immediate restart based on the Pod
restartPolicy
. - Repeated crashes: After the initial crash Kubernetes applies an exponential
backoff delay for subsequent restarts, described in
restartPolicy
. This prevents rapid, repeated restart attempts from overloading the system. - CrashLoopBackOff state: This indicates that the backoff delay mechanism is currently in effect for a given container that is in a crash loop, failing and restarting repeatedly.
- Backoff reset: If a container runs successfully for a certain duration (e.g., 10 minutes), Kubernetes resets the backoff delay, treating any new crash as the first one.
In practice, a CrashLoopBackOff
is a condition or event that might be seen as output
from the kubectl
command, while describing or listing Pods, when a container in the Pod
fails to start properly and then continually tries and fails in a loop.
In other words, when a container enters the crash loop, Kubernetes applies the exponential backoff delay mentioned in the Container restart policy. This mechanism prevents a faulty container from overwhelming the system with continuous failed start attempts.
The CrashLoopBackOff
can be caused by issues like the following:
- Application errors that cause the container to exit.
- Configuration errors, such as incorrect environment variables or missing configuration files.
- Resource constraints, where the container might not have enough memory or CPU to start properly.
- Health checks failing if the application doesn't start serving within the expected time.
- Container liveness probes or startup probes returning a
Failure
result as mentioned in the probes section.
To investigate the root cause of a CrashLoopBackOff
issue, a user can:
- Check logs: Use
kubectl logs <name-of-pod>
to check the logs of the container. This is often the most direct way to diagnose the issue causing the crashes. - Inspect events: Use
kubectl describe pod <name-of-pod>
to see events for the Pod, which can provide hints about configuration or resource issues. - Review configuration: Ensure that the Pod configuration, including environment variables and mounted volumes, is correct and that all required external resources are available.
- Check resource limits: Make sure that the container has enough CPU and memory allocated. Sometimes, increasing the resources in the Pod definition can resolve the issue.
- Debug application: There might exist bugs or misconfigurations in the application code. Running this container image locally or in a development environment can help diagnose application specific issues.
Container restart policy
The spec
of a Pod has a restartPolicy
field with possible values Always, OnFailure,
and Never. The default value is Always.
The restartPolicy
for a Pod applies to app containers
in the Pod and to regular init containers.
Sidecar containers
ignore the Pod-level restartPolicy
field: in Kubernetes, a sidecar is defined as an
entry inside initContainers
that has its container-level restartPolicy
set to Always
.
For init containers that exit with an error, the kubelet restarts the init container if
the Pod level restartPolicy
is either OnFailure
or Always
:
Always
: Automatically restarts the container after any termination.OnFailure
: Only restarts the container if it exits with an error (non-zero exit status).Never
: Does not automatically restart the terminated container.
When the kubelet is handling container restarts according to the configured restart
policy, that only applies to restarts that make replacement containers inside the
same Pod and running on the same node. After containers in a Pod exit, the kubelet
restarts them with an exponential backoff delay (10s, 20s, 40s, …), that is capped at
300 seconds (5 minutes). Once a container has executed for 10 minutes without any
problems, the kubelet resets the restart backoff timer for that container.
Sidecar containers and Pod lifecycle
explains the behaviour of init containers
when specify restartpolicy
field on it.
Pod conditions
A Pod has a PodStatus, which has an array of PodConditions through which the Pod has or has not passed. Kubelet manages the following PodConditions:
PodScheduled
: the Pod has been scheduled to a node.PodReadyToStartContainers
: (beta feature; enabled by default) the Pod sandbox has been successfully created and networking configured.ContainersReady
: all containers in the Pod are ready.Initialized
: all init containers have completed successfully.Ready
: the Pod is able to serve requests and should be added to the load balancing pools of all matching Services.
Field name | Description |
---|---|
type | Name of this Pod condition. |
status | Indicates whether that condition is applicable, with possible values "True ", "False ", or "Unknown ". |
lastProbeTime | Timestamp of when the Pod condition was last probed. |
lastTransitionTime | Timestamp for when the Pod last transitioned from one status to another. |
reason | Machine-readable, UpperCamelCase text indicating the reason for the condition's last transition. |
message | Human-readable message indicating details about the last status transition. |
Pod readiness
Kubernetes v1.14 [stable]
Your application can inject extra feedback or signals into PodStatus:
Pod readiness. To use this, set readinessGates
in the Pod's spec
to
specify a list of additional conditions that the kubelet evaluates for Pod readiness.
Readiness gates are determined by the current state of status.condition
fields for the Pod. If Kubernetes cannot find such a condition in the
status.conditions
field of a Pod, the status of the condition
is defaulted to "False
".
Here is an example:
kind: Pod
...
spec:
readinessGates:
- conditionType: "www.example.com/feature-1"
status:
conditions:
- type: Ready # a built in PodCondition
status: "False"
lastProbeTime: null
lastTransitionTime: 2018-01-01T00:00:00Z
- type: "www.example.com/feature-1" # an extra PodCondition
status: "False"
lastProbeTime: null
lastTransitionTime: 2018-01-01T00:00:00Z
containerStatuses:
- containerID: docker://abcd...
ready: true
...
The Pod conditions you add must have names that meet the Kubernetes label key format.
Status for Pod readiness
The kubectl patch
command does not support patching object status.
To set these status.conditions
for the Pod, applications and
operators should use
the PATCH
action.
You can use a Kubernetes client library to
write code that sets custom Pod conditions for Pod readiness.
For a Pod that uses custom conditions, that Pod is evaluated to be ready only when both the following statements apply:
- All containers in the Pod are ready.
- All conditions specified in
readinessGates
areTrue
.
When a Pod's containers are Ready but at least one custom condition is missing or
False
, the kubelet sets the Pod's condition to ContainersReady
.
Pod network readiness
Kubernetes v1.29 [beta]
Note:
During its early development, this condition was namedPodHasNetwork
.After a Pod gets scheduled on a node, it needs to be admitted by the kubelet and
to have any required storage volumes mounted. Once these phases are complete,
the kubelet works with
a container runtime (using Container runtime interface (CRI)) to set up a
runtime sandbox and configure networking for the Pod. If the
PodReadyToStartContainersCondition
feature gate is enabled
(it is enabled by default for Kubernetes 1.31), the
PodReadyToStartContainers
condition will be added to the status.conditions
field of a Pod.
The PodReadyToStartContainers
condition is set to False
by the Kubelet when it detects a
Pod does not have a runtime sandbox with networking configured. This occurs in
the following scenarios:
- Early in the lifecycle of the Pod, when the kubelet has not yet begun to set up a sandbox for the Pod using the container runtime.
- Later in the lifecycle of the Pod, when the Pod sandbox has been destroyed due to either:
- the node rebooting, without the Pod getting evicted
- for container runtimes that use virtual machines for isolation, the Pod sandbox virtual machine rebooting, which then requires creating a new sandbox and fresh container network configuration.
The PodReadyToStartContainers
condition is set to True
by the kubelet after the
successful completion of sandbox creation and network configuration for the Pod
by the runtime plugin. The kubelet can start pulling container images and create
containers after PodReadyToStartContainers
condition has been set to True
.
For a Pod with init containers, the kubelet sets the Initialized
condition to
True
after the init containers have successfully completed (which happens
after successful sandbox creation and network configuration by the runtime
plugin). For a Pod without init containers, the kubelet sets the Initialized
condition to True
before sandbox creation and network configuration starts.
Container probes
A probe is a diagnostic performed periodically by the kubelet on a container. To perform a diagnostic, the kubelet either executes code within the container, or makes a network request.
Check mechanisms
There are four different ways to check a container using a probe. Each probe must define exactly one of these four mechanisms:
exec
- Executes a specified command inside the container. The diagnostic is considered successful if the command exits with a status code of 0.
grpc
- Performs a remote procedure call using gRPC.
The target should implement
gRPC health checks.
The diagnostic is considered successful if the
status
of the response isSERVING
. httpGet
- Performs an HTTP
GET
request against the Pod's IP address on a specified port and path. The diagnostic is considered successful if the response has a status code greater than or equal to 200 and less than 400. tcpSocket
- Performs a TCP check against the Pod's IP address on a specified port. The diagnostic is considered successful if the port is open. If the remote system (the container) closes the connection immediately after it opens, this counts as healthy.
Caution:
Unlike the other mechanisms,exec
probe's implementation involves the creation/forking of multiple processes each time when executed.
As a result, in case of the clusters having higher pod densities, lower intervals of initialDelaySeconds
, periodSeconds
, configuring any probe with exec mechanism might introduce an overhead on the cpu usage of the node.
In such scenarios, consider using the alternative probe mechanisms to avoid the overhead.Probe outcome
Each probe has one of three results:
Success
- The container passed the diagnostic.
Failure
- The container failed the diagnostic.
Unknown
- The diagnostic failed (no action should be taken, and the kubelet will make further checks).
Types of probe
The kubelet can optionally perform and react to three kinds of probes on running containers:
livenessProbe
- Indicates whether the container is running. If
the liveness probe fails, the kubelet kills the container, and the container
is subjected to its restart policy. If a container does not
provide a liveness probe, the default state is
Success
. readinessProbe
- Indicates whether the container is ready to respond to requests.
If the readiness probe fails, the endpoints controller removes the Pod's IP
address from the endpoints of all Services that match the Pod. The default
state of readiness before the initial delay is
Failure
. If a container does not provide a readiness probe, the default state isSuccess
. startupProbe
- Indicates whether the application within the container is started.
All other probes are disabled if a startup probe is provided, until it succeeds.
If the startup probe fails, the kubelet kills the container, and the container
is subjected to its restart policy. If a container does not
provide a startup probe, the default state is
Success
.
For more information about how to set up a liveness, readiness, or startup probe, see Configure Liveness, Readiness and Startup Probes.
When should you use a liveness probe?
If the process in your container is able to crash on its own whenever it
encounters an issue or becomes unhealthy, you do not necessarily need a liveness
probe; the kubelet will automatically perform the correct action in accordance
with the Pod's restartPolicy
.
If you'd like your container to be killed and restarted if a probe fails, then
specify a liveness probe, and specify a restartPolicy
of Always or OnFailure.
When should you use a readiness probe?
If you'd like to start sending traffic to a Pod only when a probe succeeds, specify a readiness probe. In this case, the readiness probe might be the same as the liveness probe, but the existence of the readiness probe in the spec means that the Pod will start without receiving any traffic and only start receiving traffic after the probe starts succeeding.
If you want your container to be able to take itself down for maintenance, you can specify a readiness probe that checks an endpoint specific to readiness that is different from the liveness probe.
If your app has a strict dependency on back-end services, you can implement both a liveness and a readiness probe. The liveness probe passes when the app itself is healthy, but the readiness probe additionally checks that each required back-end service is available. This helps you avoid directing traffic to Pods that can only respond with error messages.
If your container needs to work on loading large data, configuration files, or migrations during startup, you can use a startup probe. However, if you want to detect the difference between an app that has failed and an app that is still processing its startup data, you might prefer a readiness probe.
Note:
If you want to be able to drain requests when the Pod is deleted, you do not necessarily need a readiness probe; on deletion, the Pod automatically puts itself into an unready state regardless of whether the readiness probe exists. The Pod remains in the unready state while it waits for the containers in the Pod to stop.When should you use a startup probe?
Startup probes are useful for Pods that have containers that take a long time to come into service. Rather than set a long liveness interval, you can configure a separate configuration for probing the container as it starts up, allowing a time longer than the liveness interval would allow.
If your container usually starts in more than
initialDelaySeconds + failureThreshold × periodSeconds
, you should specify a
startup probe that checks the same endpoint as the liveness probe. The default for
periodSeconds
is 10s. You should then set its failureThreshold
high enough to
allow the container to start, without changing the default values of the liveness
probe. This helps to protect against deadlocks.
Termination of Pods
Because Pods represent processes running on nodes in the cluster, it is important to
allow those processes to gracefully terminate when they are no longer needed (rather
than being abruptly stopped with a KILL
signal and having no chance to clean up).
The design aim is for you to be able to request deletion and know when processes terminate, but also be able to ensure that deletes eventually complete. When you request deletion of a Pod, the cluster records and tracks the intended grace period before the Pod is allowed to be forcefully killed. With that forceful shutdown tracking in place, the kubelet attempts graceful shutdown.
Typically, with this graceful termination of the pod, kubelet makes requests to the container runtime to attempt to stop the containers in the pod by first sending a TERM (aka. SIGTERM) signal, with a grace period timeout, to the main process in each container. The requests to stop the containers are processed by the container runtime asynchronously. There is no guarantee to the order of processing for these requests. Many container runtimes respect the STOPSIGNAL
value defined in the container image and, if different, send the container image configured STOPSIGNAL instead of TERM.
Once the grace period has expired, the KILL signal is sent to any remaining
processes, and the Pod is then deleted from the
API Server. If the kubelet or the
container runtime's management service is restarted while waiting for processes to terminate, the
cluster retries from the start including the full original grace period.
Pod termination flow, illustrated with an example:
You use the
kubectl
tool to manually delete a specific Pod, with the default grace period (30 seconds).The Pod in the API server is updated with the time beyond which the Pod is considered "dead" along with the grace period. If you use
kubectl describe
to check the Pod you're deleting, that Pod shows up as "Terminating". On the node where the Pod is running: as soon as the kubelet sees that a Pod has been marked as terminating (a graceful shutdown duration has been set), the kubelet begins the local Pod shutdown process.If one of the Pod's containers has defined a
preStop
hook and theterminationGracePeriodSeconds
in the Pod spec is not set to 0, the kubelet runs that hook inside of the container. The defaultterminationGracePeriodSeconds
setting is 30 seconds.If the
preStop
hook is still running after the grace period expires, the kubelet requests a small, one-off grace period extension of 2 seconds.Note:
If thepreStop
hook needs longer to complete than the default grace period allows, you must modifyterminationGracePeriodSeconds
to suit this.The kubelet triggers the container runtime to send a TERM signal to process 1 inside each container.
There is special ordering if the Pod has any sidecar containers defined. Otherwise, the containers in the Pod receive the TERM signal at different times and in an arbitrary order. If the order of shutdowns matters, consider using a
preStop
hook to synchronize (or switch to using sidecar containers).
At the same time as the kubelet is starting graceful shutdown of the Pod, the control plane evaluates whether to remove that shutting-down Pod from EndpointSlice (and Endpoints) objects, where those objects represent a Service with a configured selector. ReplicaSets and other workload resources no longer treat the shutting-down Pod as a valid, in-service replica.
Pods that shut down slowly should not continue to serve regular traffic and should start terminating and finish processing open connections. Some applications need to go beyond finishing open connections and need more graceful termination, for example, session draining and completion.
Any endpoints that represent the terminating Pods are not immediately removed from EndpointSlices, and a status indicating terminating state is exposed from the EndpointSlice API (and the legacy Endpoints API). Terminating endpoints always have their
ready
status asfalse
(for backward compatibility with versions before 1.26), so load balancers will not use it for regular traffic.If traffic draining on terminating Pod is needed, the actual readiness can be checked as a condition
serving
. You can find more details on how to implement connections draining in the tutorial Pods And Endpoints Termination FlowThe kubelet ensures the Pod is shut down and terminated
- When the grace period expires, if there is still any container running in the Pod, the
kubelet triggers forcible shutdown.
The container runtime sends
SIGKILL
to any processes still running in any container in the Pod. The kubelet also cleans up a hiddenpause
container if that container runtime uses one. - The kubelet transitions the Pod into a terminal phase (
Failed
orSucceeded
depending on the end state of its containers). - The kubelet triggers forcible removal of the Pod object from the API server, by setting grace period to 0 (immediate deletion).
- The API server deletes the Pod's API object, which is then no longer visible from any client.
- When the grace period expires, if there is still any container running in the Pod, the
kubelet triggers forcible shutdown.
The container runtime sends
Forced Pod termination
Caution:
Forced deletions can be potentially disruptive for some workloads and their Pods.By default, all deletes are graceful within 30 seconds. The kubectl delete
command supports
the --grace-period=<seconds>
option which allows you to override the default and specify your
own value.
Setting the grace period to 0
forcibly and immediately deletes the Pod from the API
server. If the Pod was still running on a node, that forcible deletion triggers the kubelet to
begin immediate cleanup.
Using kubectl, You must specify an additional flag --force
along with --grace-period=0
in order to perform force deletions.
When a force deletion is performed, the API server does not wait for confirmation from the kubelet that the Pod has been terminated on the node it was running on. It removes the Pod in the API immediately so a new Pod can be created with the same name. On the node, Pods that are set to terminate immediately will still be given a small grace period before being force killed.
Caution:
Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.If you need to force-delete Pods that are part of a StatefulSet, refer to the task documentation for deleting Pods from a StatefulSet.
Pod shutdown and sidecar containers
If your Pod includes one or more sidecar containers (init containers with an Always restart policy), the kubelet will delay sending the TERM signal to these sidecar containers until the last main container has fully terminated. The sidecar containers will be terminated in the reverse order they are defined in the Pod spec. This ensures that sidecar containers continue serving the other containers in the Pod until they are no longer needed.
This means that slow termination of a main container will also delay the termination of the sidecar containers. If the grace period expires before the termination process is complete, the Pod may enter forced termination. In this case, all remaining containers in the Pod will be terminated simultaneously with a short grace period.
Similarly, if the Pod has a preStop
hook that exceeds the termination grace period, emergency termination may occur.
In general, if you have used preStop
hooks to control the termination order without sidecar containers, you can now
remove them and allow the kubelet to manage sidecar termination automatically.
Garbage collection of Pods
For failed Pods, the API objects remain in the cluster's API until a human or controller process explicitly removes them.
The Pod garbage collector (PodGC), which is a controller in the control plane, cleans up
terminated Pods (with a phase of Succeeded
or Failed
), when the number of Pods exceeds the
configured threshold (determined by terminated-pod-gc-threshold
in the kube-controller-manager).
This avoids a resource leak as Pods are created and terminated over time.
Additionally, PodGC cleans up any Pods which satisfy any of the following conditions:
- are orphan Pods - bound to a node which no longer exists,
- are unscheduled terminating Pods,
- are terminating Pods, bound to a non-ready node tainted with
node.kubernetes.io/out-of-service
, when theNodeOutOfServiceVolumeDetach
feature gate is enabled.
Along with cleaning up the Pods, PodGC will also mark them as failed if they are in a non-terminal phase. Also, PodGC adds a Pod disruption condition when cleaning up an orphan Pod. See Pod disruption conditions for more details.
What's next
Get hands-on experience attaching handlers to container lifecycle events.
Get hands-on experience configuring Liveness, Readiness and Startup Probes.
Learn more about container lifecycle hooks.
Learn more about sidecar containers.
For detailed information about Pod and container status in the API, see the API reference documentation covering
status
for Pod.
4.1.2 - Init Containers
This page provides an overview of init containers: specialized containers that run before app containers in a Pod. Init containers can contain utilities or setup scripts not present in an app image.
You can specify init containers in the Pod specification alongside the containers
array (which describes app containers).
In Kubernetes, a sidecar container is a container that starts before the main application container and continues to run. This document is about init containers: containers that run to completion during Pod initialization.
Understanding init containers
A Pod can have multiple containers running apps within it, but it can also have one or more init containers, which are run before the app containers are started.
Init containers are exactly like regular containers, except:
- Init containers always run to completion.
- Each init container must complete successfully before the next one starts.
If a Pod's init container fails, the kubelet repeatedly restarts that init container until it succeeds.
However, if the Pod has a restartPolicy
of Never, and an init container fails during startup of that Pod, Kubernetes treats the overall Pod as failed.
To specify an init container for a Pod, add the initContainers
field into
the Pod specification,
as an array of container
items (similar to the app containers
field and its contents).
See Container in the
API reference for more details.
The status of the init containers is returned in .status.initContainerStatuses
field as an array of the container statuses (similar to the .status.containerStatuses
field).
Differences from regular containers
Init containers support all the fields and features of app containers, including resource limits, volumes, and security settings. However, the resource requests and limits for an init container are handled differently, as documented in Resource sharing within containers.
Regular init containers (in other words: excluding sidecar containers) do not support the
lifecycle
, livenessProbe
, readinessProbe
, or startupProbe
fields. Init containers
must run to completion before the Pod can be ready; sidecar containers continue running
during a Pod's lifetime, and do support some probes. See sidecar container
for further details about sidecar containers.
If you specify multiple init containers for a Pod, kubelet runs each init container sequentially. Each init container must succeed before the next can run. When all of the init containers have run to completion, kubelet initializes the application containers for the Pod and runs them as usual.
Differences from sidecar containers
Init containers run and complete their tasks before the main application container starts. Unlike sidecar containers, init containers are not continuously running alongside the main containers.
Init containers run to completion sequentially, and the main container does not start until all the init containers have successfully completed.
init containers do not support lifecycle
, livenessProbe
, readinessProbe
, or
startupProbe
whereas sidecar containers support all these probes to control their lifecycle.
Init containers share the same resources (CPU, memory, network) with the main application containers but do not interact directly with them. They can, however, use shared volumes for data exchange.
Using init containers
Because init containers have separate images from app containers, they have some advantages for start-up related code:
- Init containers can contain utilities or custom code for setup that are not present in an app
image. For example, there is no need to make an image
FROM
another image just to use a tool likesed
,awk
,python
, ordig
during setup. - The application image builder and deployer roles can work independently without the need to jointly build a single app image.
- Init containers can run with a different view of the filesystem than app containers in the same Pod. Consequently, they can be given access to Secrets that app containers cannot access.
- Because init containers run to completion before any app containers start, init containers offer a mechanism to block or delay app container startup until a set of preconditions are met. Once preconditions are met, all of the app containers in a Pod can start in parallel.
- Init containers can securely run utilities or custom code that would otherwise make an app container image less secure. By keeping unnecessary tools separate you can limit the attack surface of your app container image.
Examples
Here are some ideas for how to use init containers:
Wait for a Service to be created, using a shell one-line command like:
for i in {1..100}; do sleep 1; if nslookup myservice; then exit 0; fi; done; exit 1
Register this Pod with a remote server from the downward API with a command like:
curl -X POST http://$MANAGEMENT_SERVICE_HOST:$MANAGEMENT_SERVICE_PORT/register -d 'instance=$(<POD_NAME>)&ip=$(<POD_IP>)'
Wait for some time before starting the app container with a command like
sleep 60
Clone a Git repository into a Volume
Place values into a configuration file and run a template tool to dynamically generate a configuration file for the main app container. For example, place the
POD_IP
value in a configuration and generate the main app configuration file using Jinja.
Init containers in use
This example defines a simple Pod that has two init containers.
The first waits for myservice
, and the second waits for mydb
. Once both
init containers complete, the Pod runs the app container from its spec
section.
apiVersion: v1
kind: Pod
metadata:
name: myapp-pod
labels:
app.kubernetes.io/name: MyApp
spec:
containers:
- name: myapp-container
image: busybox:1.28
command: ['sh', '-c', 'echo The app is running! && sleep 3600']
initContainers:
- name: init-myservice
image: busybox:1.28
command: ['sh', '-c', "until nslookup myservice.$(cat /var/run/secrets/kubernetes.io/serviceaccount/namespace).svc.cluster.local; do echo waiting for myservice; sleep 2; done"]
- name: init-mydb
image: busybox:1.28
command: ['sh', '-c', "until nslookup mydb.$(cat /var/run/secrets/kubernetes.io/serviceaccount/namespace).svc.cluster.local; do echo waiting for mydb; sleep 2; done"]
You can start this Pod by running:
kubectl apply -f myapp.yaml
The output is similar to this:
pod/myapp-pod created
And check on its status with:
kubectl get -f myapp.yaml
The output is similar to this:
NAME READY STATUS RESTARTS AGE
myapp-pod 0/1 Init:0/2 0 6m
or for more details:
kubectl describe -f myapp.yaml
The output is similar to this:
Name: myapp-pod
Namespace: default
[...]
Labels: app.kubernetes.io/name=MyApp
Status: Pending
[...]
Init Containers:
init-myservice:
[...]
State: Running
[...]
init-mydb:
[...]
State: Waiting
Reason: PodInitializing
Ready: False
[...]
Containers:
myapp-container:
[...]
State: Waiting
Reason: PodInitializing
Ready: False
[...]
Events:
FirstSeen LastSeen Count From SubObjectPath Type Reason Message
--------- -------- ----- ---- ------------- -------- ------ -------
16s 16s 1 {default-scheduler } Normal Scheduled Successfully assigned myapp-pod to 172.17.4.201
16s 16s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Pulling pulling image "busybox"
13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Pulled Successfully pulled image "busybox"
13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Created Created container init-myservice
13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Started Started container init-myservice
To see logs for the init containers in this Pod, run:
kubectl logs myapp-pod -c init-myservice # Inspect the first init container
kubectl logs myapp-pod -c init-mydb # Inspect the second init container
At this point, those init containers will be waiting to discover Services named
mydb
and myservice
.
Here's a configuration you can use to make those Services appear:
---
apiVersion: v1
kind: Service
metadata:
name: myservice
spec:
ports:
- protocol: TCP
port: 80
targetPort: 9376
---
apiVersion: v1
kind: Service
metadata:
name: mydb
spec:
ports:
- protocol: TCP
port: 80
targetPort: 9377
To create the mydb
and myservice
services:
kubectl apply -f services.yaml
The output is similar to this:
service/myservice created
service/mydb created
You'll then see that those init containers complete, and that the myapp-pod
Pod moves into the Running state:
kubectl get -f myapp.yaml
The output is similar to this:
NAME READY STATUS RESTARTS AGE
myapp-pod 1/1 Running 0 9m
This simple example should provide some inspiration for you to create your own init containers. What's next contains a link to a more detailed example.
Detailed behavior
During Pod startup, the kubelet delays running init containers until the networking and storage are ready. Then the kubelet runs the Pod's init containers in the order they appear in the Pod's spec.
Each init container must exit successfully before
the next container starts. If a container fails to start due to the runtime or
exits with failure, it is retried according to the Pod restartPolicy
. However,
if the Pod restartPolicy
is set to Always, the init containers use
restartPolicy
OnFailure.
A Pod cannot be Ready
until all init containers have succeeded. The ports on an
init container are not aggregated under a Service. A Pod that is initializing
is in the Pending
state but should have a condition Initialized
set to false.
If the Pod restarts, or is restarted, all init containers must execute again.
Changes to the init container spec are limited to the container image field.
Directly altering the image
field of an init container does not restart the
Pod or trigger its recreation. If the Pod has yet to start, that change may
have an effect on how the Pod boots up.
For a pod template you can typically change any field for an init container; the impact of making that change depends on where the pod template is used.
Because init containers can be restarted, retried, or re-executed, init container
code should be idempotent. In particular, code that writes into any emptyDir
volume
should be prepared for the possibility that an output file already exists.
Init containers have all of the fields of an app container. However, Kubernetes
prohibits readinessProbe
from being used because init containers cannot
define readiness distinct from completion. This is enforced during validation.
Use activeDeadlineSeconds
on the Pod to prevent init containers from failing forever.
The active deadline includes init containers.
However it is recommended to use activeDeadlineSeconds
only if teams deploy their application
as a Job, because activeDeadlineSeconds
has an effect even after initContainer finished.
The Pod which is already running correctly would be killed by activeDeadlineSeconds
if you set.
The name of each app and init container in a Pod must be unique; a validation error is thrown for any container sharing a name with another.
Resource sharing within containers
Given the order of execution for init, sidecar and app containers, the following rules for resource usage apply:
- The highest of any particular resource request or limit defined on all init containers is the effective init request/limit. If any resource has no resource limit specified this is considered as the highest limit.
- The Pod's effective request/limit for a resource is the higher of:
- the sum of all app containers request/limit for a resource
- the effective init request/limit for a resource
- Scheduling is done based on effective requests/limits, which means init containers can reserve resources for initialization that are not used during the life of the Pod.
- The QoS (quality of service) tier of the Pod's effective QoS tier is the QoS tier for init containers and app containers alike.
Quota and limits are applied based on the effective Pod request and limit.
Init containers and Linux cgroups
On Linux, resource allocations for Pod level control groups (cgroups) are based on the effective Pod request and limit, the same as the scheduler.
Pod restart reasons
A Pod can restart, causing re-execution of init containers, for the following reasons:
- The Pod infrastructure container is restarted. This is uncommon and would have to be done by someone with root access to nodes.
- All containers in a Pod are terminated while
restartPolicy
is set to Always, forcing a restart, and the init container completion record has been lost due to garbage collection.
The Pod will not be restarted when the init container image is changed, or the init container completion record has been lost due to garbage collection. This applies for Kubernetes v1.20 and later. If you are using an earlier version of Kubernetes, consult the documentation for the version you are using.
What's next
Learn more about the following:
- Creating a Pod that has an init container.
- Debug init containers.
- Overview of kubelet and kubectl.
- Types of probes: liveness, readiness, startup probe.
- Sidecar containers.
4.1.3 - Sidecar Containers
Kubernetes v1.29 [beta]
Sidecar containers are the secondary containers that run along with the main application container within the same Pod. These containers are used to enhance or to extend the functionality of the primary app container by providing additional services, or functionality such as logging, monitoring, security, or data synchronization, without directly altering the primary application code.
Typically, you only have one app container in a Pod. For example, if you have a web application that requires a local webserver, the local webserver is a sidecar and the web application itself is the app container.
Sidecar containers in Kubernetes
Kubernetes implements sidecar containers as a special case of init containers; sidecar containers remain running after Pod startup. This document uses the term regular init containers to clearly refer to containers that only run during Pod startup.
Provided that your cluster has the SidecarContainers
feature gate enabled
(the feature is active by default since Kubernetes v1.29), you can specify a restartPolicy
for containers listed in a Pod's initContainers
field.
These restartable sidecar containers are independent from other init containers and from
the main application container(s) within the same pod.
These can be started, stopped, or restarted without effecting the main application container
and other init containers.
You can also run a Pod with multiple containers that are not marked as init or sidecar
containers. This is appropriate if the containers within the Pod are required for the
Pod to work overall, but you don't need to control which containers start or stop first.
You could also do this if you need to support older versions of Kubernetes that don't
support a container-level restartPolicy
field.
Example application
Here's an example of a Deployment with two containers, one of which is a sidecar:
apiVersion: apps/v1
kind: Deployment
metadata:
name: myapp
labels:
app: myapp
spec:
replicas: 1
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
containers:
- name: myapp
image: alpine:latest
command: ['sh', '-c', 'while true; do echo "logging" >> /opt/logs.txt; sleep 1; done']
volumeMounts:
- name: data
mountPath: /opt
initContainers:
- name: logshipper
image: alpine:latest
restartPolicy: Always
command: ['sh', '-c', 'tail -F /opt/logs.txt']
volumeMounts:
- name: data
mountPath: /opt
volumes:
- name: data
emptyDir: {}
Sidecar containers and Pod lifecycle
If an init container is created with its restartPolicy
set to Always
, it will
start and remain running during the entire life of the Pod. This can be helpful for
running supporting services separated from the main application containers.
If a readinessProbe
is specified for this init container, its result will be used
to determine the ready
state of the Pod.
Since these containers are defined as init containers, they benefit from the same ordering and sequential guarantees as regular init containers, allowing you to mix sidecar containers with regular init containers for complex Pod initialization flows.
Compared to regular init containers, sidecars defined within initContainers
continue to
run after they have started. This is important when there is more than one entry inside
.spec.initContainers
for a Pod. After a sidecar-style init container is running (the kubelet
has set the started
status for that init container to true), the kubelet then starts the
next init container from the ordered .spec.initContainers
list.
That status either becomes true because there is a process running in the
container and no startup probe defined, or as a result of its startupProbe
succeeding.
Upon Pod termination, the kubelet postpones terminating sidecar containers until the main application container has fully stopped. The sidecar containers are then shut down in the opposite order of their appearance in the Pod specification. This approach ensures that the sidecars remain operational, supporting other containers within the Pod, until their service is no longer required.
Jobs with sidecar containers
If you define a Job that uses sidecar using Kubernetes-style init containers, the sidecar container in each Pod does not prevent the Job from completing after the main container has finished.
Here's an example of a Job with two containers, one of which is a sidecar:
apiVersion: batch/v1
kind: Job
metadata:
name: myjob
spec:
template:
spec:
containers:
- name: myjob
image: alpine:latest
command: ['sh', '-c', 'echo "logging" > /opt/logs.txt']
volumeMounts:
- name: data
mountPath: /opt
initContainers:
- name: logshipper
image: alpine:latest
restartPolicy: Always
command: ['sh', '-c', 'tail -F /opt/logs.txt']
volumeMounts:
- name: data
mountPath: /opt
restartPolicy: Never
volumes:
- name: data
emptyDir: {}
Differences from application containers
Sidecar containers run alongside app containers in the same pod. However, they do not execute the primary application logic; instead, they provide supporting functionality to the main application.
Sidecar containers have their own independent lifecycles. They can be started, stopped, and restarted independently of app containers. This means you can update, scale, or maintain sidecar containers without affecting the primary application.
Sidecar containers share the same network and storage namespaces with the primary container. This co-location allows them to interact closely and share resources.
From Kubernetes perspective, sidecars graceful termination is less important.
When other containers took all alloted graceful termination time, sidecar containers
will receive the SIGTERM
following with SIGKILL
faster than may be expected.
So exit codes different from 0
(0
indicates successful exit), for sidecar containers are normal
on Pod termination and should be generally ignored by the external tooling.
Differences from init containers
Sidecar containers work alongside the main container, extending its functionality and providing additional services.
Sidecar containers run concurrently with the main application container. They are active throughout the lifecycle of the pod and can be started and stopped independently of the main container. Unlike init containers, sidecar containers support probes to control their lifecycle.
Sidecar containers can interact directly with the main application containers, because like init containers they always share the same network, and can optionally also share volumes (filesystems).
Init containers stop before the main containers start up, so init containers cannot
exchange messages with the app container in a Pod. Any data passing is one-way
(for example, an init container can put information inside an emptyDir
volume).
Resource sharing within containers
Given the order of execution for init, sidecar and app containers, the following rules for resource usage apply:
- The highest of any particular resource request or limit defined on all init containers is the effective init request/limit. If any resource has no resource limit specified this is considered as the highest limit.
- The Pod's effective request/limit for a resource is the sum of
pod overhead and the higher of:
- the sum of all non-init containers(app and sidecar containers) request/limit for a resource
- the effective init request/limit for a resource
- Scheduling is done based on effective requests/limits, which means init containers can reserve resources for initialization that are not used during the life of the Pod.
- The QoS (quality of service) tier of the Pod's effective QoS tier is the QoS tier for all init, sidecar and app containers alike.
Quota and limits are applied based on the effective Pod request and limit.
Sidecar containers and Linux cgroups
On Linux, resource allocations for Pod level control groups (cgroups) are based on the effective Pod request and limit, the same as the scheduler.
What's next
- Learn how to Adopt Sidecar Containers
- Read a blog post on native sidecar containers.
- Read about creating a Pod that has an init container.
- Learn about the types of probes: liveness, readiness, startup probe.
- Learn about pod overhead.
4.1.4 - Ephemeral Containers
Kubernetes v1.25 [stable]
This page provides an overview of ephemeral containers: a special type of container that runs temporarily in an existing Pod to accomplish user-initiated actions such as troubleshooting. You use ephemeral containers to inspect services rather than to build applications.
Understanding ephemeral containers
Pods are the fundamental building block of Kubernetes applications. Since Pods are intended to be disposable and replaceable, you cannot add a container to a Pod once it has been created. Instead, you usually delete and replace Pods in a controlled fashion using deployments.
Sometimes it's necessary to inspect the state of an existing Pod, however, for example to troubleshoot a hard-to-reproduce bug. In these cases you can run an ephemeral container in an existing Pod to inspect its state and run arbitrary commands.
What is an ephemeral container?
Ephemeral containers differ from other containers in that they lack guarantees
for resources or execution, and they will never be automatically restarted, so
they are not appropriate for building applications. Ephemeral containers are
described using the same ContainerSpec
as regular containers, but many fields
are incompatible and disallowed for ephemeral containers.
- Ephemeral containers may not have ports, so fields such as
ports
,livenessProbe
,readinessProbe
are disallowed. - Pod resource allocations are immutable, so setting
resources
is disallowed. - For a complete list of allowed fields, see the EphemeralContainer reference documentation.
Ephemeral containers are created using a special ephemeralcontainers
handler
in the API rather than by adding them directly to pod.spec
, so it's not
possible to add an ephemeral container using kubectl edit
.
Like regular containers, you may not change or remove an ephemeral container after you have added it to a Pod.
Note:
Ephemeral containers are not supported by static pods.Uses for ephemeral containers
Ephemeral containers are useful for interactive troubleshooting when kubectl exec
is insufficient because a container has crashed or a container image
doesn't include debugging utilities.
In particular, distroless images
enable you to deploy minimal container images that reduce attack surface
and exposure to bugs and vulnerabilities. Since distroless images do not include a
shell or any debugging utilities, it's difficult to troubleshoot distroless
images using kubectl exec
alone.
When using ephemeral containers, it's helpful to enable process namespace sharing so you can view processes in other containers.
What's next
- Learn how to debug pods using ephemeral containers.
4.1.5 - Disruptions
This guide is for application owners who want to build highly available applications, and thus need to understand what types of disruptions can happen to Pods.
It is also for cluster administrators who want to perform automated cluster actions, like upgrading and autoscaling clusters.
Voluntary and involuntary disruptions
Pods do not disappear until someone (a person or a controller) destroys them, or there is an unavoidable hardware or system software error.
We call these unavoidable cases involuntary disruptions to an application. Examples are:
- a hardware failure of the physical machine backing the node
- cluster administrator deletes VM (instance) by mistake
- cloud provider or hypervisor failure makes VM disappear
- a kernel panic
- the node disappears from the cluster due to cluster network partition
- eviction of a pod due to the node being out-of-resources.
Except for the out-of-resources condition, all these conditions should be familiar to most users; they are not specific to Kubernetes.
We call other cases voluntary disruptions. These include both actions initiated by the application owner and those initiated by a Cluster Administrator. Typical application owner actions include:
- deleting the deployment or other controller that manages the pod
- updating a deployment's pod template causing a restart
- directly deleting a pod (e.g. by accident)
Cluster administrator actions include:
- Draining a node for repair or upgrade.
- Draining a node from a cluster to scale the cluster down (learn about Cluster Autoscaling).
- Removing a pod from a node to permit something else to fit on that node.
These actions might be taken directly by the cluster administrator, or by automation run by the cluster administrator, or by your cluster hosting provider.
Ask your cluster administrator or consult your cloud provider or distribution documentation to determine if any sources of voluntary disruptions are enabled for your cluster. If none are enabled, you can skip creating Pod Disruption Budgets.
Caution:
Not all voluntary disruptions are constrained by Pod Disruption Budgets. For example, deleting deployments or pods bypasses Pod Disruption Budgets.Dealing with disruptions
Here are some ways to mitigate involuntary disruptions:
- Ensure your pod requests the resources it needs.
- Replicate your application if you need higher availability. (Learn about running replicated stateless and stateful applications.)
- For even higher availability when running replicated applications, spread applications across racks (using anti-affinity) or across zones (if using a multi-zone cluster.)
The frequency of voluntary disruptions varies. On a basic Kubernetes cluster, there are no automated voluntary disruptions (only user-triggered ones). However, your cluster administrator or hosting provider may run some additional services which cause voluntary disruptions. For example, rolling out node software updates can cause voluntary disruptions. Also, some implementations of cluster (node) autoscaling may cause voluntary disruptions to defragment and compact nodes. Your cluster administrator or hosting provider should have documented what level of voluntary disruptions, if any, to expect. Certain configuration options, such as using PriorityClasses in your pod spec can also cause voluntary (and involuntary) disruptions.
Pod disruption budgets
Kubernetes v1.21 [stable]
Kubernetes offers features to help you run highly available applications even when you introduce frequent voluntary disruptions.
As an application owner, you can create a PodDisruptionBudget (PDB) for each application. A PDB limits the number of Pods of a replicated application that are down simultaneously from voluntary disruptions. For example, a quorum-based application would like to ensure that the number of replicas running is never brought below the number needed for a quorum. A web front end might want to ensure that the number of replicas serving load never falls below a certain percentage of the total.
Cluster managers and hosting providers should use tools which respect PodDisruptionBudgets by calling the Eviction API instead of directly deleting pods or deployments.
For example, the kubectl drain
subcommand lets you mark a node as going out of
service. When you run kubectl drain
, the tool tries to evict all of the Pods on
the Node you're taking out of service. The eviction request that kubectl
submits on
your behalf may be temporarily rejected, so the tool periodically retries all failed
requests until all Pods on the target node are terminated, or until a configurable timeout
is reached.
A PDB specifies the number of replicas that an application can tolerate having, relative to how
many it is intended to have. For example, a Deployment which has a .spec.replicas: 5
is
supposed to have 5 pods at any given time. If its PDB allows for there to be 4 at a time,
then the Eviction API will allow voluntary disruption of one (but not two) pods at a time.
The group of pods that comprise the application is specified using a label selector, the same as the one used by the application's controller (deployment, stateful-set, etc).
The "intended" number of pods is computed from the .spec.replicas
of the workload resource
that is managing those pods. The control plane discovers the owning workload resource by
examining the .metadata.ownerReferences
of the Pod.
Involuntary disruptions cannot be prevented by PDBs; however they do count against the budget.
Pods which are deleted or unavailable due to a rolling upgrade to an application do count against the disruption budget, but workload resources (such as Deployment and StatefulSet) are not limited by PDBs when doing rolling upgrades. Instead, the handling of failures during application updates is configured in the spec for the specific workload resource.
It is recommended to set AlwaysAllow
Unhealthy Pod Eviction Policy
to your PodDisruptionBudgets to support eviction of misbehaving applications during a node drain.
The default behavior is to wait for the application pods to become healthy
before the drain can proceed.
When a pod is evicted using the eviction API, it is gracefully
terminated, honoring the
terminationGracePeriodSeconds
setting in its PodSpec.
PodDisruptionBudget example
Consider a cluster with 3 nodes, node-1
through node-3
.
The cluster is running several applications. One of them has 3 replicas initially called
pod-a
, pod-b
, and pod-c
. Another, unrelated pod without a PDB, called pod-x
, is also shown.
Initially, the pods are laid out as follows:
node-1 | node-2 | node-3 |
---|---|---|
pod-a available | pod-b available | pod-c available |
pod-x available |
All 3 pods are part of a deployment, and they collectively have a PDB which requires there be at least 2 of the 3 pods to be available at all times.
For example, assume the cluster administrator wants to reboot into a new kernel version to fix a bug in the kernel.
The cluster administrator first tries to drain node-1
using the kubectl drain
command.
That tool tries to evict pod-a
and pod-x
. This succeeds immediately.
Both pods go into the terminating
state at the same time.
This puts the cluster in this state:
node-1 draining | node-2 | node-3 |
---|---|---|
pod-a terminating | pod-b available | pod-c available |
pod-x terminating |
The deployment notices that one of the pods is terminating, so it creates a replacement
called pod-d
. Since node-1
is cordoned, it lands on another node. Something has
also created pod-y
as a replacement for pod-x
.
(Note: for a StatefulSet, pod-a
, which would be called something like pod-0
, would need
to terminate completely before its replacement, which is also called pod-0
but has a
different UID, could be created. Otherwise, the example applies to a StatefulSet as well.)
Now the cluster is in this state:
node-1 draining | node-2 | node-3 |
---|---|---|
pod-a terminating | pod-b available | pod-c available |
pod-x terminating | pod-d starting | pod-y |
At some point, the pods terminate, and the cluster looks like this:
node-1 drained | node-2 | node-3 |
---|---|---|
pod-b available | pod-c available | |
pod-d starting | pod-y |
At this point, if an impatient cluster administrator tries to drain node-2
or
node-3
, the drain command will block, because there are only 2 available
pods for the deployment, and its PDB requires at least 2. After some time passes, pod-d
becomes available.
The cluster state now looks like this:
node-1 drained | node-2 | node-3 |
---|---|---|
pod-b available | pod-c available | |
pod-d available | pod-y |
Now, the cluster administrator tries to drain node-2
.
The drain command will try to evict the two pods in some order, say
pod-b
first and then pod-d
. It will succeed at evicting pod-b
.
But, when it tries to evict pod-d
, it will be refused because that would leave only
one pod available for the deployment.
The deployment creates a replacement for pod-b
called pod-e
.
Because there are not enough resources in the cluster to schedule
pod-e
the drain will again block. The cluster may end up in this
state:
node-1 drained | node-2 | node-3 | no node |
---|---|---|---|
pod-b terminating | pod-c available | pod-e pending | |
pod-d available | pod-y |
At this point, the cluster administrator needs to add a node back to the cluster to proceed with the upgrade.
You can see how Kubernetes varies the rate at which disruptions can happen, according to:
- how many replicas an application needs
- how long it takes to gracefully shutdown an instance
- how long it takes a new instance to start up
- the type of controller
- the cluster's resource capacity
Pod disruption conditions
Kubernetes v1.31 [stable]
(enabled by default: true)A dedicated Pod DisruptionTarget
condition
is added to indicate
that the Pod is about to be deleted due to a disruption.
The reason
field of the condition additionally
indicates one of the following reasons for the Pod termination:
PreemptionByScheduler
- Pod is due to be preempted by a scheduler in order to accommodate a new Pod with a higher priority. For more information, see Pod priority preemption.
DeletionByTaintManager
- Pod is due to be deleted by Taint Manager (which is part of the node lifecycle controller within
kube-controller-manager
) due to aNoExecute
taint that the Pod does not tolerate; see taint-based evictions. EvictionByEvictionAPI
- Pod has been marked for eviction using the Kubernetes API .
DeletionByPodGC
- Pod, that is bound to a no longer existing Node, is due to be deleted by Pod garbage collection.
TerminationByKubelet
- Pod has been terminated by the kubelet, because of either node pressure eviction, the graceful node shutdown, or preemption for system critical pods.
In all other disruption scenarios, like eviction due to exceeding
Pod container limits,
Pods don't receive the DisruptionTarget
condition because the disruptions were
probably caused by the Pod and would reoccur on retry.
Note:
A Pod disruption might be interrupted. The control plane might re-attempt to continue the disruption of the same Pod, but it is not guaranteed. As a result, theDisruptionTarget
condition might be added to a Pod, but that Pod might then not actually be
deleted. In such a situation, after some time, the
Pod disruption condition will be cleared.Along with cleaning up the pods, the Pod garbage collector (PodGC) will also mark them as failed if they are in a non-terminal phase (see also Pod garbage collection).
When using a Job (or CronJob), you may want to use these Pod disruption conditions as part of your Job's Pod failure policy.
Separating Cluster Owner and Application Owner Roles
Often, it is useful to think of the Cluster Manager and Application Owner as separate roles with limited knowledge of each other. This separation of responsibilities may make sense in these scenarios:
- when there are many application teams sharing a Kubernetes cluster, and there is natural specialization of roles
- when third-party tools or services are used to automate cluster management
Pod Disruption Budgets support this separation of roles by providing an interface between the roles.
If you do not have such a separation of responsibilities in your organization, you may not need to use Pod Disruption Budgets.
How to perform Disruptive Actions on your Cluster
If you are a Cluster Administrator, and you need to perform a disruptive action on all the nodes in your cluster, such as a node or system software upgrade, here are some options:
- Accept downtime during the upgrade.
- Failover to another complete replica cluster.
- No downtime, but may be costly both for the duplicated nodes and for human effort to orchestrate the switchover.
- Write disruption tolerant applications and use PDBs.
- No downtime.
- Minimal resource duplication.
- Allows more automation of cluster administration.
- Writing disruption-tolerant applications is tricky, but the work to tolerate voluntary disruptions largely overlaps with work to support autoscaling and tolerating involuntary disruptions.
What's next
Follow steps to protect your application by configuring a Pod Disruption Budget.
Learn more about draining nodes
Learn about updating a deployment including steps to maintain its availability during the rollout.
4.1.6 - Pod Quality of Service Classes
This page introduces Quality of Service (QoS) classes in Kubernetes, and explains how Kubernetes assigns a QoS class to each Pod as a consequence of the resource constraints that you specify for the containers in that Pod. Kubernetes relies on this classification to make decisions about which Pods to evict when there are not enough available resources on a Node.
Quality of Service classes
Kubernetes classifies the Pods that you run and allocates each Pod into a specific
quality of service (QoS) class. Kubernetes uses that classification to influence how different
pods are handled. Kubernetes does this classification based on the
resource requests
of the Containers in that Pod, along with
how those requests relate to resource limits.
This is known as Quality of Service
(QoS) class. Kubernetes assigns every Pod a QoS class based on the resource requests
and limits of its component Containers. QoS classes are used by Kubernetes to decide
which Pods to evict from a Node experiencing
Node Pressure. The possible
QoS classes are Guaranteed
, Burstable
, and BestEffort
. When a Node runs out of resources,
Kubernetes will first evict BestEffort
Pods running on that Node, followed by Burstable
and
finally Guaranteed
Pods. When this eviction is due to resource pressure, only Pods exceeding
resource requests are candidates for eviction.
Guaranteed
Pods that are Guaranteed
have the strictest resource limits and are least likely
to face eviction. They are guaranteed not to be killed until they exceed their limits
or there are no lower-priority Pods that can be preempted from the Node. They may
not acquire resources beyond their specified limits. These Pods can also make
use of exclusive CPUs using the
static
CPU management policy.
Criteria
For a Pod to be given a QoS class of Guaranteed
:
- Every Container in the Pod must have a memory limit and a memory request.
- For every Container in the Pod, the memory limit must equal the memory request.
- Every Container in the Pod must have a CPU limit and a CPU request.
- For every Container in the Pod, the CPU limit must equal the CPU request.
Burstable
Pods that are Burstable
have some lower-bound resource guarantees based on the request, but
do not require a specific limit. If a limit is not specified, it defaults to a
limit equivalent to the capacity of the Node, which allows the Pods to flexibly increase
their resources if resources are available. In the event of Pod eviction due to Node
resource pressure, these Pods are evicted only after all BestEffort
Pods are evicted.
Because a Burstable
Pod can include a Container that has no resource limits or requests, a Pod
that is Burstable
can try to use any amount of node resources.
Criteria
A Pod is given a QoS class of Burstable
if:
- The Pod does not meet the criteria for QoS class
Guaranteed
. - At least one Container in the Pod has a memory or CPU request or limit.
BestEffort
Pods in the BestEffort
QoS class can use node resources that aren't specifically assigned
to Pods in other QoS classes. For example, if you have a node with 16 CPU cores available to the
kubelet, and you assign 4 CPU cores to a Guaranteed
Pod, then a Pod in the BestEffort
QoS class can try to use any amount of the remaining 12 CPU cores.
The kubelet prefers to evict BestEffort
Pods if the node comes under resource pressure.
Criteria
A Pod has a QoS class of BestEffort
if it doesn't meet the criteria for either Guaranteed
or Burstable
. In other words, a Pod is BestEffort
only if none of the Containers in the Pod have a
memory limit or a memory request, and none of the Containers in the Pod have a
CPU limit or a CPU request.
Containers in a Pod can request other resources (not CPU or memory) and still be classified as
BestEffort
.
Memory QoS with cgroup v2
Kubernetes v1.22 [alpha]
(enabled by default: false)Memory QoS uses the memory controller of cgroup v2 to guarantee memory resources in Kubernetes.
Memory requests and limits of containers in pod are used to set specific interfaces memory.min
and memory.high
provided by the memory controller. When memory.min
is set to memory requests,
memory resources are reserved and never reclaimed by the kernel; this is how Memory QoS ensures
memory availability for Kubernetes pods. And if memory limits are set in the container,
this means that the system needs to limit container memory usage; Memory QoS uses memory.high
to throttle workload approaching its memory limit, ensuring that the system is not overwhelmed
by instantaneous memory allocation.
Memory QoS relies on QoS class to determine which settings to apply; however, these are different mechanisms that both provide controls over quality of service.
Some behavior is independent of QoS class
Certain behavior is independent of the QoS class assigned by Kubernetes. For example:
Any Container exceeding a resource limit will be killed and restarted by the kubelet without affecting other Containers in that Pod.
If a Container exceeds its resource request and the node it runs on faces resource pressure, the Pod it is in becomes a candidate for eviction. If this occurs, all Containers in the Pod will be terminated. Kubernetes may create a replacement Pod, usually on a different node.
The resource request of a Pod is equal to the sum of the resource requests of its component Containers, and the resource limit of a Pod is equal to the sum of the resource limits of its component Containers.
The kube-scheduler does not consider QoS class when selecting which Pods to preempt. Preemption can occur when a cluster does not have enough resources to run all the Pods you defined.
What's next
- Learn about resource management for Pods and Containers.
- Learn about Node-pressure eviction.
- Learn about Pod priority and preemption.
- Learn about Pod disruptions.
- Learn how to assign memory resources to containers and pods.
- Learn how to assign CPU resources to containers and pods.
- Learn how to configure Quality of Service for Pods.
4.1.7 - User Namespaces
Kubernetes v1.30 [beta]
This page explains how user namespaces are used in Kubernetes pods. A user namespace isolates the user running inside the container from the one in the host.
A process running as root in a container can run as a different (non-root) user in the host; in other words, the process has full privileges for operations inside the user namespace, but is unprivileged for operations outside the namespace.
You can use this feature to reduce the damage a compromised container can do to the host or other pods in the same node. There are several security vulnerabilities rated either HIGH or CRITICAL that were not exploitable when user namespaces is active. It is expected user namespace will mitigate some future vulnerabilities too.
Before you begin
This is a Linux-only feature and support is needed in Linux for idmap mounts on the filesystems used. This means:
- On the node, the filesystem you use for
/var/lib/kubelet/pods/
, or the custom directory you configure for this, needs idmap mount support. - All the filesystems used in the pod's volumes must support idmap mounts.
In practice this means you need at least Linux 6.3, as tmpfs started supporting idmap mounts in that version. This is usually needed as several Kubernetes features use tmpfs (the service account token that is mounted by default uses a tmpfs, Secrets use a tmpfs, etc.)
Some popular filesystems that support idmap mounts in Linux 6.3 are: btrfs, ext4, xfs, fat, tmpfs, overlayfs.
In addition, the container runtime and its underlying OCI runtime must support user namespaces. The following OCI runtimes offer support:
- crun version 1.9 or greater (it's recommend version 1.13+).
Note:
Many OCI runtimes do not include the support needed for using user namespaces in
Linux pods. If you use a managed Kubernetes, or have downloaded it from packages
and set it up, it's likely that nodes in your cluster use a runtime that doesn't
include this support. For example, the most widely used OCI runtime is runc
,
and version 1.1.z
of runc doesn't support all the features needed by the
Kubernetes implementation of user namespaces.
If there is a newer release of runc than 1.1 available for use, check its documentation and release notes for compatibility (look for idmap mounts support in particular, because that is the missing feature).
To use user namespaces with Kubernetes, you also need to use a CRI container runtime to use this feature with Kubernetes pods:
- CRI-O: version 1.25 (and later) supports user namespaces for containers.
containerd v1.7 is not compatible with the userns support in Kubernetes v1.27 to v1.31. Kubernetes v1.25 and v1.26 used an earlier implementation that is compatible with containerd v1.7, in terms of userns support. If you are using a version of Kubernetes other than 1.31, check the documentation for that version of Kubernetes for the most relevant information. If there is a newer release of containerd than v1.7 available for use, also check the containerd documentation for compatibility information.
You can see the status of user namespaces support in cri-dockerd tracked in an issue on GitHub.
Introduction
User namespaces is a Linux feature that allows to map users in the container to different users in the host. Furthermore, the capabilities granted to a pod in a user namespace are valid only in the namespace and void outside of it.
A pod can opt-in to use user namespaces by setting the pod.spec.hostUsers
field
to false
.
The kubelet will pick host UIDs/GIDs a pod is mapped to, and will do so in a way to guarantee that no two pods on the same node use the same mapping.
The runAsUser
, runAsGroup
, fsGroup
, etc. fields in the pod.spec
always
refer to the user inside the container.
The valid UIDs/GIDs when this feature is enabled is the range 0-65535. This
applies to files and processes (runAsUser
, runAsGroup
, etc.).
Files using a UID/GID outside this range will be seen as belonging to the
overflow ID, usually 65534 (configured in /proc/sys/kernel/overflowuid
and
/proc/sys/kernel/overflowgid
). However, it is not possible to modify those
files, even by running as the 65534 user/group.
Most applications that need to run as root but don't access other host namespaces or resources, should continue to run fine without any changes needed if user namespaces is activated.
Understanding user namespaces for pods
Several container runtimes with their default configuration (like Docker Engine, containerd, CRI-O) use Linux namespaces for isolation. Other technologies exist and can be used with those runtimes too (e.g. Kata Containers uses VMs instead of Linux namespaces). This page is applicable for container runtimes using Linux namespaces for isolation.
When creating a pod, by default, several new namespaces are used for isolation: a network namespace to isolate the network of the container, a PID namespace to isolate the view of processes, etc. If a user namespace is used, this will isolate the users in the container from the users in the node.
This means containers can run as root and be mapped to a non-root user on the
host. Inside the container the process will think it is running as root (and
therefore tools like apt
, yum
, etc. work fine), while in reality the process
doesn't have privileges on the host. You can verify this, for example, if you
check which user the container process is running by executing ps aux
from
the host. The user ps
shows is not the same as the user you see if you
execute inside the container the command id
.
This abstraction limits what can happen, for example, if the container manages to escape to the host. Given that the container is running as a non-privileged user on the host, it is limited what it can do to the host.
Furthermore, as users on each pod will be mapped to different non-overlapping users in the host, it is limited what they can do to other pods too.
Capabilities granted to a pod are also limited to the pod user namespace and mostly invalid out of it, some are even completely void. Here are two examples:
CAP_SYS_MODULE
does not have any effect if granted to a pod using user namespaces, the pod isn't able to load kernel modules.CAP_SYS_ADMIN
is limited to the pod's user namespace and invalid outside of it.
Without using a user namespace a container running as root, in the case of a container breakout, has root privileges on the node. And if some capability were granted to the container, the capabilities are valid on the host too. None of this is true when we use user namespaces.
If you want to know more details about what changes when user namespaces are in
use, see man 7 user_namespaces
.
Set up a node to support user namespaces
By default, the kubelet assigns pods UIDs/GIDs above the range 0-65535, based on the assumption that the host's files and processes use UIDs/GIDs within this range, which is standard for most Linux distributions. This approach prevents any overlap between the UIDs/GIDs of the host and those of the pods.
Avoiding the overlap is important to mitigate the impact of vulnerabilities such as CVE-2021-25741, where a pod can potentially read arbitrary files in the host. If the UIDs/GIDs of the pod and the host don't overlap, it is limited what a pod would be able to do: the pod UID/GID won't match the host's file owner/group.
The kubelet can use a custom range for user IDs and group IDs for pods. To configure a custom range, the node needs to have:
- A user
kubelet
in the system (you cannot use any other username here) - The binary
getsubids
installed (part of shadow-utils) and in thePATH
for the kubelet binary. - A configuration of subordinate UIDs/GIDs for the
kubelet
user (seeman 5 subuid
andman 5 subgid
).
This setting only gathers the UID/GID range configuration and does not change
the user executing the kubelet
.
You must follow some constraints for the subordinate ID range that you assign
to the kubelet
user:
The subordinate user ID, that starts the UID range for Pods, must be a multiple of 65536 and must also be greater than or equal to 65536. In other words, you cannot use any ID from the range 0-65535 for Pods; the kubelet imposes this restriction to make it difficult to create an accidentally insecure configuration.
The subordinate ID count must be a multiple of 65536
The subordinate ID count must be at least
65536 x <maxPods>
where<maxPods>
is the maximum number of pods that can run on the node.You must assign the same range for both user IDs and for group IDs, It doesn't matter if other users have user ID ranges that don't align with the group ID ranges.
None of the assigned ranges should overlap with any other assignment.
The subordinate configuration must be only one line. In other words, you can't have multiple ranges.
For example, you could define /etc/subuid
and /etc/subgid
to both have
these entries for the kubelet
user:
# The format is
# name:firstID:count of IDs
# where
# - firstID is 65536 (the minimum value possible)
# - count of IDs is 110 (default limit for number of) * 65536
kubelet:65536:7208960
Integration with Pod security admission checks
Kubernetes v1.29 [alpha]
For Linux Pods that enable user namespaces, Kubernetes relaxes the application of
Pod Security Standards in a controlled way.
This behavior can be controlled by the feature
gate
UserNamespacesPodSecurityStandards
, which allows an early opt-in for end
users. Admins have to ensure that user namespaces are enabled by all nodes
within the cluster if using the feature gate.
If you enable the associated feature gate and create a Pod that uses user
namespaces, the following fields won't be constrained even in contexts that enforce the
Baseline or Restricted pod security standard. This behavior does not
present a security concern because root
inside a Pod with user namespaces
actually refers to the user inside the container, that is never mapped to a
privileged user on the host. Here's the list of fields that are not checks for Pods in those
circumstances:
spec.securityContext.runAsNonRoot
spec.containers[*].securityContext.runAsNonRoot
spec.initContainers[*].securityContext.runAsNonRoot
spec.ephemeralContainers[*].securityContext.runAsNonRoot
spec.securityContext.runAsUser
spec.containers[*].securityContext.runAsUser
spec.initContainers[*].securityContext.runAsUser
spec.ephemeralContainers[*].securityContext.runAsUser
Limitations
When using a user namespace for the pod, it is disallowed to use other host
namespaces. In particular, if you set hostUsers: false
then you are not
allowed to set any of:
hostNetwork: true
hostIPC: true
hostPID: true
What's next
- Take a look at Use a User Namespace With a Pod
4.1.8 - Downward API
It is sometimes useful for a container to have information about itself, without being overly coupled to Kubernetes. The downward API allows containers to consume information about themselves or the cluster without using the Kubernetes client or API server.
An example is an existing application that assumes a particular well-known environment variable holds a unique identifier. One possibility is to wrap the application, but that is tedious and error-prone, and it violates the goal of low coupling. A better option would be to use the Pod's name as an identifier, and inject the Pod's name into the well-known environment variable.
In Kubernetes, there are two ways to expose Pod and container fields to a running container:
Together, these two ways of exposing Pod and container fields are called the downward API.
Available fields
Only some Kubernetes API fields are available through the downward API. This section lists which fields you can make available.
You can pass information from available Pod-level fields using fieldRef
.
At the API level, the spec
for a Pod always defines at least one
Container.
You can pass information from available Container-level fields using
resourceFieldRef
.
Information available via fieldRef
For some Pod-level fields, you can provide them to a container either as
an environment variable or using a downwardAPI
volume. The fields available
via either mechanism are:
metadata.name
- the pod's name
metadata.namespace
- the pod's namespace
metadata.uid
- the pod's unique ID
metadata.annotations['<KEY>']
- the value of the pod's annotation named
<KEY>
(for example,metadata.annotations['myannotation']
) metadata.labels['<KEY>']
- the text value of the pod's label named
<KEY>
(for example,metadata.labels['mylabel']
)
The following information is available through environment variables but not as a downwardAPI volume fieldRef:
spec.serviceAccountName
- the name of the pod's service account
spec.nodeName
- the name of the node where the Pod is executing
status.hostIP
- the primary IP address of the node to which the Pod is assigned
status.hostIPs
- the IP addresses is a dual-stack version of
status.hostIP
, the first is always the same asstatus.hostIP
. status.podIP
- the pod's primary IP address (usually, its IPv4 address)
status.podIPs
- the IP addresses is a dual-stack version of
status.podIP
, the first is always the same asstatus.podIP
The following information is available through a downwardAPI
volume
fieldRef
, but not as environment variables:
metadata.labels
- all of the pod's labels, formatted as
label-key="escaped-label-value"
with one label per line metadata.annotations
- all of the pod's annotations, formatted as
annotation-key="escaped-annotation-value"
with one annotation per line
Information available via resourceFieldRef
These container-level fields allow you to provide information about requests and limits for resources such as CPU and memory.
resource: limits.cpu
- A container's CPU limit
resource: requests.cpu
- A container's CPU request
resource: limits.memory
- A container's memory limit
resource: requests.memory
- A container's memory request
resource: limits.hugepages-*
- A container's hugepages limit
resource: requests.hugepages-*
- A container's hugepages request
resource: limits.ephemeral-storage
- A container's ephemeral-storage limit
resource: requests.ephemeral-storage
- A container's ephemeral-storage request
Fallback information for resource limits
If CPU and memory limits are not specified for a container, and you use the downward API to try to expose that information, then the kubelet defaults to exposing the maximum allocatable value for CPU and memory based on the node allocatable calculation.
What's next
You can read about downwardAPI
volumes.
You can try using the downward API to expose container- or Pod-level information:
4.2 - Workload Management
Kubernetes provides several built-in APIs for declarative management of your workloads and the components of those workloads.
Ultimately, your applications run as containers inside Pods; however, managing individual Pods would be a lot of effort. For example, if a Pod fails, you probably want to run a new Pod to replace it. Kubernetes can do that for you.
You use the Kubernetes API to create a workload object that represents a higher abstraction level than a Pod, and then the Kubernetes control plane automatically manages Pod objects on your behalf, based on the specification for the workload object you defined.
The built-in APIs for managing workloads are:
Deployment (and, indirectly, ReplicaSet), the most common way to run an application on your cluster. Deployment is a good fit for managing a stateless application workload on your cluster, where any Pod in the Deployment is interchangeable and can be replaced if needed. (Deployments are a replacement for the legacy ReplicationController API).
A StatefulSet lets you manage one or more Pods – all running the same application code – where the Pods rely on having a distinct identity. This is different from a Deployment where the Pods are expected to be interchangeable. The most common use for a StatefulSet is to be able to make a link between its Pods and their persistent storage. For example, you can run a StatefulSet that associates each Pod with a PersistentVolume. If one of the Pods in the StatefulSet fails, Kubernetes makes a replacement Pod that is connected to the same PersistentVolume.
A DaemonSet defines Pods that provide facilities that are local to a specific node; for example, a driver that lets containers on that node access a storage system. You use a DaemonSet when the driver, or other node-level service, has to run on the node where it's useful. Each Pod in a DaemonSet performs a role similar to a system daemon on a classic Unix / POSIX server. A DaemonSet might be fundamental to the operation of your cluster, such as a plugin to let that node access cluster networking, it might help you to manage the node, or it could provide less essential facilities that enhance the container platform you are running. You can run DaemonSets (and their pods) across every node in your cluster, or across just a subset (for example, only install the GPU accelerator driver on nodes that have a GPU installed).
You can use a Job and / or a CronJob to define tasks that run to completion and then stop. A Job represents a one-off task, whereas each CronJob repeats according to a schedule.
Other topics in this section:
4.2.1 - Deployments
A Deployment provides declarative updates for Pods and ReplicaSets.
You describe a desired state in a Deployment, and the Deployment Controller changes the actual state to the desired state at a controlled rate. You can define Deployments to create new ReplicaSets, or to remove existing Deployments and adopt all their resources with new Deployments.
Note:
Do not manage ReplicaSets owned by a Deployment. Consider opening an issue in the main Kubernetes repository if your use case is not covered below.Use Case
The following are typical use cases for Deployments:
- Create a Deployment to rollout a ReplicaSet. The ReplicaSet creates Pods in the background. Check the status of the rollout to see if it succeeds or not.
- Declare the new state of the Pods by updating the PodTemplateSpec of the Deployment. A new ReplicaSet is created and the Deployment manages moving the Pods from the old ReplicaSet to the new one at a controlled rate. Each new ReplicaSet updates the revision of the Deployment.
- Rollback to an earlier Deployment revision if the current state of the Deployment is not stable. Each rollback updates the revision of the Deployment.
- Scale up the Deployment to facilitate more load.
- Pause the rollout of a Deployment to apply multiple fixes to its PodTemplateSpec and then resume it to start a new rollout.
- Use the status of the Deployment as an indicator that a rollout has stuck.
- Clean up older ReplicaSets that you don't need anymore.
Creating a Deployment
The following is an example of a Deployment. It creates a ReplicaSet to bring up three nginx
Pods:
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-deployment
labels:
app: nginx
spec:
replicas: 3
selector:
matchLabels:
app: nginx
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
In this example:
A Deployment named
nginx-deployment
is created, indicated by the.metadata.name
field. This name will become the basis for the ReplicaSets and Pods which are created later. See Writing a Deployment Spec for more details.The Deployment creates a ReplicaSet that creates three replicated Pods, indicated by the
.spec.replicas
field.The
.spec.selector
field defines how the created ReplicaSet finds which Pods to manage. In this case, you select a label that is defined in the Pod template (app: nginx
). However, more sophisticated selection rules are possible, as long as the Pod template itself satisfies the rule.Note:
The.spec.selector.matchLabels
field is a map of {key,value} pairs. A single {key,value} in thematchLabels
map is equivalent to an element ofmatchExpressions
, whosekey
field is "key", theoperator
is "In", and thevalues
array contains only "value". All of the requirements, from bothmatchLabels
andmatchExpressions
, must be satisfied in order to match.The
template
field contains the following sub-fields:- The Pods are labeled
app: nginx
using the.metadata.labels
field. - The Pod template's specification, or
.template.spec
field, indicates that the Pods run one container,nginx
, which runs thenginx
Docker Hub image at version 1.14.2. - Create one container and name it
nginx
using the.spec.template.spec.containers[0].name
field.
- The Pods are labeled
Before you begin, make sure your Kubernetes cluster is up and running. Follow the steps given below to create the above Deployment:
Create the Deployment by running the following command:
kubectl apply -f https://k8s.io/examples/controllers/nginx-deployment.yaml
Run
kubectl get deployments
to check if the Deployment was created.If the Deployment is still being created, the output is similar to the following:
NAME READY UP-TO-DATE AVAILABLE AGE nginx-deployment 0/3 0 0 1s
When you inspect the Deployments in your cluster, the following fields are displayed:
NAME
lists the names of the Deployments in the namespace.READY
displays how many replicas of the application are available to your users. It follows the pattern ready/desired.UP-TO-DATE
displays the number of replicas that have been updated to achieve the desired state.AVAILABLE
displays how many replicas of the application are available to your users.AGE
displays the amount of time that the application has been running.
Notice how the number of desired replicas is 3 according to
.spec.replicas
field.To see the Deployment rollout status, run
kubectl rollout status deployment/nginx-deployment
.The output is similar to:
Waiting for rollout to finish: 2 out of 3 new replicas have been updated... deployment "nginx-deployment" successfully rolled out
Run the
kubectl get deployments
again a few seconds later. The output is similar to this:NAME READY UP-TO-DATE AVAILABLE AGE nginx-deployment 3/3 3 3 18s
Notice that the Deployment has created all three replicas, and all replicas are up-to-date (they contain the latest Pod template) and available.
To see the ReplicaSet (
rs
) created by the Deployment, runkubectl get rs
. The output is similar to this:NAME DESIRED CURRENT READY AGE nginx-deployment-75675f5897 3 3 3 18s
ReplicaSet output shows the following fields:
NAME
lists the names of the ReplicaSets in the namespace.DESIRED
displays the desired number of replicas of the application, which you define when you create the Deployment. This is the desired state.CURRENT
displays how many replicas are currently running.READY
displays how many replicas of the application are available to your users.AGE
displays the amount of time that the application has been running.
Notice that the name of the ReplicaSet is always formatted as
[DEPLOYMENT-NAME]-[HASH]
. This name will become the basis for the Pods which are created.The
HASH
string is the same as thepod-template-hash
label on the ReplicaSet.To see the labels automatically generated for each Pod, run
kubectl get pods --show-labels
. The output is similar to:NAME READY STATUS RESTARTS AGE LABELS nginx-deployment-75675f5897-7ci7o 1/1 Running 0 18s app=nginx,pod-template-hash=75675f5897 nginx-deployment-75675f5897-kzszj 1/1 Running 0 18s app=nginx,pod-template-hash=75675f5897 nginx-deployment-75675f5897-qqcnn 1/1 Running 0 18s app=nginx,pod-template-hash=75675f5897
The created ReplicaSet ensures that there are three
nginx
Pods.
Note:
You must specify an appropriate selector and Pod template labels in a Deployment
(in this case, app: nginx
).
Do not overlap labels or selectors with other controllers (including other Deployments and StatefulSets). Kubernetes doesn't stop you from overlapping, and if multiple controllers have overlapping selectors those controllers might conflict and behave unexpectedly.
Pod-template-hash label
Caution:
Do not change this label.The pod-template-hash
label is added by the Deployment controller to every ReplicaSet that a Deployment creates or adopts.
This label ensures that child ReplicaSets of a Deployment do not overlap. It is generated by hashing the PodTemplate
of the ReplicaSet and using the resulting hash as the label value that is added to the ReplicaSet selector, Pod template labels,
and in any existing Pods that the ReplicaSet might have.
Updating a Deployment
Note:
A Deployment's rollout is triggered if and only if the Deployment's Pod template (that is,.spec.template
)
is changed, for example if the labels or container images of the template are updated. Other updates, such as scaling the Deployment, do not trigger a rollout.Follow the steps given below to update your Deployment:
Let's update the nginx Pods to use the
nginx:1.16.1
image instead of thenginx:1.14.2
image.kubectl set image deployment.v1.apps/nginx-deployment nginx=nginx:1.16.1
or use the following command:
kubectl set image deployment/nginx-deployment nginx=nginx:1.16.1
where
deployment/nginx-deployment
indicates the Deployment,nginx
indicates the Container the update will take place andnginx:1.16.1
indicates the new image and its tag.The output is similar to:
deployment.apps/nginx-deployment image updated
Alternatively, you can
edit
the Deployment and change.spec.template.spec.containers[0].image
fromnginx:1.14.2
tonginx:1.16.1
:kubectl edit deployment/nginx-deployment
The output is similar to:
deployment.apps/nginx-deployment edited
To see the rollout status, run:
kubectl rollout status deployment/nginx-deployment
The output is similar to this:
Waiting for rollout to finish: 2 out of 3 new replicas have been updated...
or
deployment "nginx-deployment" successfully rolled out
Get more details on your updated Deployment:
After the rollout succeeds, you can view the Deployment by running
kubectl get deployments
. The output is similar to this:NAME READY UP-TO-DATE AVAILABLE AGE nginx-deployment 3/3 3 3 36s
Run
kubectl get rs
to see that the Deployment updated the Pods by creating a new ReplicaSet and scaling it up to 3 replicas, as well as scaling down the old ReplicaSet to 0 replicas.kubectl get rs
The output is similar to this:
NAME DESIRED CURRENT READY AGE nginx-deployment-1564180365 3 3 3 6s nginx-deployment-2035384211 0 0 0 36s
Running
get pods
should now show only the new Pods:kubectl get pods
The output is similar to this:
NAME READY STATUS RESTARTS AGE nginx-deployment-1564180365-khku8 1/1 Running 0 14s nginx-deployment-1564180365-nacti 1/1 Running 0 14s nginx-deployment-1564180365-z9gth 1/1 Running 0 14s
Next time you want to update these Pods, you only need to update the Deployment's Pod template again.
Deployment ensures that only a certain number of Pods are down while they are being updated. By default, it ensures that at least 75% of the desired number of Pods are up (25% max unavailable).
Deployment also ensures that only a certain number of Pods are created above the desired number of Pods. By default, it ensures that at most 125% of the desired number of Pods are up (25% max surge).
For example, if you look at the above Deployment closely, you will see that it first creates a new Pod, then deletes an old Pod, and creates another new one. It does not kill old Pods until a sufficient number of new Pods have come up, and does not create new Pods until a sufficient number of old Pods have been killed. It makes sure that at least 3 Pods are available and that at max 4 Pods in total are available. In case of a Deployment with 4 replicas, the number of Pods would be between 3 and 5.
Get details of your Deployment:
kubectl describe deployments
The output is similar to this:
Name: nginx-deployment Namespace: default CreationTimestamp: Thu, 30 Nov 2017 10:56:25 +0000 Labels: app=nginx Annotations: deployment.kubernetes.io/revision=2 Selector: app=nginx Replicas: 3 desired | 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 25% max unavailable, 25% max surge Pod Template: Labels: app=nginx Containers: nginx: Image: nginx:1.16.1 Port: 80/TCP Environment: <none> Mounts: <none> Volumes: <none> Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True NewReplicaSetAvailable OldReplicaSets: <none> NewReplicaSet: nginx-deployment-1564180365 (3/3 replicas created) Events: Type Reason Age From Message ---- ------ ---- ---- ------- Normal ScalingReplicaSet 2m deployment-controller Scaled up replica set nginx-deployment-2035384211 to 3 Normal ScalingReplicaSet 24s deployment-controller Scaled up replica set nginx-deployment-1564180365 to 1 Normal ScalingReplicaSet 22s deployment-controller Scaled down replica set nginx-deployment-2035384211 to 2 Normal ScalingReplicaSet 22s deployment-controller Scaled up replica set nginx-deployment-1564180365 to 2 Normal ScalingReplicaSet 19s deployment-controller Scaled down replica set nginx-deployment-2035384211 to 1 Normal ScalingReplicaSet 19s deployment-controller Scaled up replica set nginx-deployment-1564180365 to 3 Normal ScalingReplicaSet 14s deployment-controller Scaled down replica set nginx-deployment-2035384211 to 0
Here you see that when you first created the Deployment, it created a ReplicaSet (nginx-deployment-2035384211) and scaled it up to 3 replicas directly. When you updated the Deployment, it created a new ReplicaSet (nginx-deployment-1564180365) and scaled it up to 1 and waited for it to come up. Then it scaled down the old ReplicaSet to 2 and scaled up the new ReplicaSet to 2 so that at least 3 Pods were available and at most 4 Pods were created at all times. It then continued scaling up and down the new and the old ReplicaSet, with the same rolling update strategy. Finally, you'll have 3 available replicas in the new ReplicaSet, and the old ReplicaSet is scaled down to 0.
Note:
Kubernetes doesn't count terminating Pods when calculating the number ofavailableReplicas
, which must be between
replicas - maxUnavailable
and replicas + maxSurge
. As a result, you might notice that there are more Pods than
expected during a rollout, and that the total resources consumed by the Deployment is more than replicas + maxSurge
until the terminationGracePeriodSeconds
of the terminating Pods expires.Rollover (aka multiple updates in-flight)
Each time a new Deployment is observed by the Deployment controller, a ReplicaSet is created to bring up
the desired Pods. If the Deployment is updated, the existing ReplicaSet that controls Pods whose labels
match .spec.selector
but whose template does not match .spec.template
are scaled down. Eventually, the new
ReplicaSet is scaled to .spec.replicas
and all old ReplicaSets is scaled to 0.
If you update a Deployment while an existing rollout is in progress, the Deployment creates a new ReplicaSet as per the update and start scaling that up, and rolls over the ReplicaSet that it was scaling up previously -- it will add it to its list of old ReplicaSets and start scaling it down.
For example, suppose you create a Deployment to create 5 replicas of nginx:1.14.2
,
but then update the Deployment to create 5 replicas of nginx:1.16.1
, when only 3
replicas of nginx:1.14.2
had been created. In that case, the Deployment immediately starts
killing the 3 nginx:1.14.2
Pods that it had created, and starts creating
nginx:1.16.1
Pods. It does not wait for the 5 replicas of nginx:1.14.2
to be created
before changing course.
Label selector updates
It is generally discouraged to make label selector updates and it is suggested to plan your selectors up front. In any case, if you need to perform a label selector update, exercise great caution and make sure you have grasped all of the implications.
Note:
In API versionapps/v1
, a Deployment's label selector is immutable after it gets created.- Selector additions require the Pod template labels in the Deployment spec to be updated with the new label too, otherwise a validation error is returned. This change is a non-overlapping one, meaning that the new selector does not select ReplicaSets and Pods created with the old selector, resulting in orphaning all old ReplicaSets and creating a new ReplicaSet.
- Selector updates changes the existing value in a selector key -- result in the same behavior as additions.
- Selector removals removes an existing key from the Deployment selector -- do not require any changes in the Pod template labels. Existing ReplicaSets are not orphaned, and a new ReplicaSet is not created, but note that the removed label still exists in any existing Pods and ReplicaSets.
Rolling Back a Deployment
Sometimes, you may want to rollback a Deployment; for example, when the Deployment is not stable, such as crash looping. By default, all of the Deployment's rollout history is kept in the system so that you can rollback anytime you want (you can change that by modifying revision history limit).
Note:
A Deployment's revision is created when a Deployment's rollout is triggered. This means that the new revision is created if and only if the Deployment's Pod template (.spec.template
) is changed,
for example if you update the labels or container images of the template. Other updates, such as scaling the Deployment,
do not create a Deployment revision, so that you can facilitate simultaneous manual- or auto-scaling.
This means that when you roll back to an earlier revision, only the Deployment's Pod template part is
rolled back.Suppose that you made a typo while updating the Deployment, by putting the image name as
nginx:1.161
instead ofnginx:1.16.1
:kubectl set image deployment/nginx-deployment nginx=nginx:1.161
The output is similar to this:
deployment.apps/nginx-deployment image updated
The rollout gets stuck. You can verify it by checking the rollout status:
kubectl rollout status deployment/nginx-deployment
The output is similar to this:
Waiting for rollout to finish: 1 out of 3 new replicas have been updated...
Press Ctrl-C to stop the above rollout status watch. For more information on stuck rollouts, read more here.
You see that the number of old replicas (adding the replica count from
nginx-deployment-1564180365
andnginx-deployment-2035384211
) is 3, and the number of new replicas (fromnginx-deployment-3066724191
) is 1.kubectl get rs
The output is similar to this:
NAME DESIRED CURRENT READY AGE nginx-deployment-1564180365 3 3 3 25s nginx-deployment-2035384211 0 0 0 36s nginx-deployment-3066724191 1 1 0 6s
Looking at the Pods created, you see that 1 Pod created by new ReplicaSet is stuck in an image pull loop.
kubectl get pods
The output is similar to this:
NAME READY STATUS RESTARTS AGE nginx-deployment-1564180365-70iae 1/1 Running 0 25s nginx-deployment-1564180365-jbqqo 1/1 Running 0 25s nginx-deployment-1564180365-hysrc 1/1 Running 0 25s nginx-deployment-3066724191-08mng 0/1 ImagePullBackOff 0 6s
Note:
The Deployment controller stops the bad rollout automatically, and stops scaling up the new ReplicaSet. This depends on the rollingUpdate parameters (maxUnavailable
specifically) that you have specified. Kubernetes by default sets the value to 25%.Get the description of the Deployment:
kubectl describe deployment
The output is similar to this:
Name: nginx-deployment Namespace: default CreationTimestamp: Tue, 15 Mar 2016 14:48:04 -0700 Labels: app=nginx Selector: app=nginx Replicas: 3 desired | 1 updated | 4 total | 3 available | 1 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 25% max unavailable, 25% max surge Pod Template: Labels: app=nginx Containers: nginx: Image: nginx:1.161 Port: 80/TCP Host Port: 0/TCP Environment: <none> Mounts: <none> Volumes: <none> Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True ReplicaSetUpdated OldReplicaSets: nginx-deployment-1564180365 (3/3 replicas created) NewReplicaSet: nginx-deployment-3066724191 (1/1 replicas created) Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-2035384211 to 3 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 1 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 2 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 2 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 1 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 3 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 0 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 1
To fix this, you need to rollback to a previous revision of Deployment that is stable.
Checking Rollout History of a Deployment
Follow the steps given below to check the rollout history:
First, check the revisions of this Deployment:
kubectl rollout history deployment/nginx-deployment
The output is similar to this:
deployments "nginx-deployment" REVISION CHANGE-CAUSE 1 kubectl apply --filename=https://k8s.io/examples/controllers/nginx-deployment.yaml 2 kubectl set image deployment/nginx-deployment nginx=nginx:1.16.1 3 kubectl set image deployment/nginx-deployment nginx=nginx:1.161
CHANGE-CAUSE
is copied from the Deployment annotationkubernetes.io/change-cause
to its revisions upon creation. You can specify theCHANGE-CAUSE
message by:- Annotating the Deployment with
kubectl annotate deployment/nginx-deployment kubernetes.io/change-cause="image updated to 1.16.1"
- Manually editing the manifest of the resource.
- Annotating the Deployment with
To see the details of each revision, run:
kubectl rollout history deployment/nginx-deployment --revision=2
The output is similar to this:
deployments "nginx-deployment" revision 2 Labels: app=nginx pod-template-hash=1159050644 Annotations: kubernetes.io/change-cause=kubectl set image deployment/nginx-deployment nginx=nginx:1.16.1 Containers: nginx: Image: nginx:1.16.1 Port: 80/TCP QoS Tier: cpu: BestEffort memory: BestEffort Environment Variables: <none> No volumes.
Rolling Back to a Previous Revision
Follow the steps given below to rollback the Deployment from the current version to the previous version, which is version 2.
Now you've decided to undo the current rollout and rollback to the previous revision:
kubectl rollout undo deployment/nginx-deployment
The output is similar to this:
deployment.apps/nginx-deployment rolled back
Alternatively, you can rollback to a specific revision by specifying it with
--to-revision
:kubectl rollout undo deployment/nginx-deployment --to-revision=2
The output is similar to this:
deployment.apps/nginx-deployment rolled back
For more details about rollout related commands, read
kubectl rollout
.The Deployment is now rolled back to a previous stable revision. As you can see, a
DeploymentRollback
event for rolling back to revision 2 is generated from Deployment controller.Check if the rollback was successful and the Deployment is running as expected, run:
kubectl get deployment nginx-deployment
The output is similar to this:
NAME READY UP-TO-DATE AVAILABLE AGE nginx-deployment 3/3 3 3 30m
Get the description of the Deployment:
kubectl describe deployment nginx-deployment
The output is similar to this:
Name: nginx-deployment Namespace: default CreationTimestamp: Sun, 02 Sep 2018 18:17:55 -0500 Labels: app=nginx Annotations: deployment.kubernetes.io/revision=4 kubernetes.io/change-cause=kubectl set image deployment/nginx-deployment nginx=nginx:1.16.1 Selector: app=nginx Replicas: 3 desired | 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 25% max unavailable, 25% max surge Pod Template: Labels: app=nginx Containers: nginx: Image: nginx:1.16.1 Port: 80/TCP Host Port: 0/TCP Environment: <none> Mounts: <none> Volumes: <none> Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True NewReplicaSetAvailable OldReplicaSets: <none> NewReplicaSet: nginx-deployment-c4747d96c (3/3 replicas created) Events: Type Reason Age From Message ---- ------ ---- ---- ------- Normal ScalingReplicaSet 12m deployment-controller Scaled up replica set nginx-deployment-75675f5897 to 3 Normal ScalingReplicaSet 11m deployment-controller Scaled up replica set nginx-deployment-c4747d96c to 1 Normal ScalingReplicaSet 11m deployment-controller Scaled down replica set nginx-deployment-75675f5897 to 2 Normal ScalingReplicaSet 11m deployment-controller Scaled up replica set nginx-deployment-c4747d96c to 2 Normal ScalingReplicaSet 11m deployment-controller Scaled down replica set nginx-deployment-75675f5897 to 1 Normal ScalingReplicaSet 11m deployment-controller Scaled up replica set nginx-deployment-c4747d96c to 3 Normal ScalingReplicaSet 11m deployment-controller Scaled down replica set nginx-deployment-75675f5897 to 0 Normal ScalingReplicaSet 11m deployment-controller Scaled up replica set nginx-deployment-595696685f to 1 Normal DeploymentRollback 15s deployment-controller Rolled back deployment "nginx-deployment" to revision 2 Normal ScalingReplicaSet 15s deployment-controller Scaled down replica set nginx-deployment-595696685f to 0
Scaling a Deployment
You can scale a Deployment by using the following command:
kubectl scale deployment/nginx-deployment --replicas=10
The output is similar to this:
deployment.apps/nginx-deployment scaled
Assuming horizontal Pod autoscaling is enabled in your cluster, you can set up an autoscaler for your Deployment and choose the minimum and maximum number of Pods you want to run based on the CPU utilization of your existing Pods.
kubectl autoscale deployment/nginx-deployment --min=10 --max=15 --cpu-percent=80
The output is similar to this:
deployment.apps/nginx-deployment scaled
Proportional scaling
RollingUpdate Deployments support running multiple versions of an application at the same time. When you or an autoscaler scales a RollingUpdate Deployment that is in the middle of a rollout (either in progress or paused), the Deployment controller balances the additional replicas in the existing active ReplicaSets (ReplicaSets with Pods) in order to mitigate risk. This is called proportional scaling.
For example, you are running a Deployment with 10 replicas, maxSurge=3, and maxUnavailable=2.
Ensure that the 10 replicas in your Deployment are running.
kubectl get deploy
The output is similar to this:
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 10 10 10 10 50s
You update to a new image which happens to be unresolvable from inside the cluster.
kubectl set image deployment/nginx-deployment nginx=nginx:sometag
The output is similar to this:
deployment.apps/nginx-deployment image updated
The image update starts a new rollout with ReplicaSet nginx-deployment-1989198191, but it's blocked due to the
maxUnavailable
requirement that you mentioned above. Check out the rollout status:kubectl get rs
The output is similar to this:
NAME DESIRED CURRENT READY AGE nginx-deployment-1989198191 5 5 0 9s nginx-deployment-618515232 8 8 8 1m
Then a new scaling request for the Deployment comes along. The autoscaler increments the Deployment replicas to 15. The Deployment controller needs to decide where to add these new 5 replicas. If you weren't using proportional scaling, all 5 of them would be added in the new ReplicaSet. With proportional scaling, you spread the additional replicas across all ReplicaSets. Bigger proportions go to the ReplicaSets with the most replicas and lower proportions go to ReplicaSets with less replicas. Any leftovers are added to the ReplicaSet with the most replicas. ReplicaSets with zero replicas are not scaled up.
In our example above, 3 replicas are added to the old ReplicaSet and 2 replicas are added to the new ReplicaSet. The rollout process should eventually move all replicas to the new ReplicaSet, assuming the new replicas become healthy. To confirm this, run:
kubectl get deploy
The output is similar to this:
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
nginx-deployment 15 18 7 8 7m
The rollout status confirms how the replicas were added to each ReplicaSet.
kubectl get rs
The output is similar to this:
NAME DESIRED CURRENT READY AGE
nginx-deployment-1989198191 7 7 0 7m
nginx-deployment-618515232 11 11 11 7m
Pausing and Resuming a rollout of a Deployment
When you update a Deployment, or plan to, you can pause rollouts for that Deployment before you trigger one or more updates. When you're ready to apply those changes, you resume rollouts for the Deployment. This approach allows you to apply multiple fixes in between pausing and resuming without triggering unnecessary rollouts.
For example, with a Deployment that was created:
Get the Deployment details:
kubectl get deploy
The output is similar to this:
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx 3 3 3 3 1m
Get the rollout status:
kubectl get rs
The output is similar to this:
NAME DESIRED CURRENT READY AGE nginx-2142116321 3 3 3 1m
Pause by running the following command:
kubectl rollout pause deployment/nginx-deployment
The output is similar to this:
deployment.apps/nginx-deployment paused
Then update the image of the Deployment:
kubectl set image deployment/nginx-deployment nginx=nginx:1.16.1
The output is similar to this:
deployment.apps/nginx-deployment image updated
Notice that no new rollout started:
kubectl rollout history deployment/nginx-deployment
The output is similar to this:
deployments "nginx" REVISION CHANGE-CAUSE 1 <none>
Get the rollout status to verify that the existing ReplicaSet has not changed:
kubectl get rs
The output is similar to this:
NAME DESIRED CURRENT READY AGE nginx-2142116321 3 3 3 2m
You can make as many updates as you wish, for example, update the resources that will be used:
kubectl set resources deployment/nginx-deployment -c=nginx --limits=cpu=200m,memory=512Mi
The output is similar to this:
deployment.apps/nginx-deployment resource requirements updated
The initial state of the Deployment prior to pausing its rollout will continue its function, but new updates to the Deployment will not have any effect as long as the Deployment rollout is paused.
Eventually, resume the Deployment rollout and observe a new ReplicaSet coming up with all the new updates:
kubectl rollout resume deployment/nginx-deployment
The output is similar to this:
deployment.apps/nginx-deployment resumed
Watch the status of the rollout until it's done.
kubectl get rs --watch
The output is similar to this:
NAME DESIRED CURRENT READY AGE nginx-2142116321 2 2 2 2m nginx-3926361531 2 2 0 6s nginx-3926361531 2 2 1 18s nginx-2142116321 1 2 2 2m nginx-2142116321 1 2 2 2m nginx-3926361531 3 2 1 18s nginx-3926361531 3 2 1 18s nginx-2142116321 1 1 1 2m nginx-3926361531 3 3 1 18s nginx-3926361531 3 3 2 19s nginx-2142116321 0 1 1 2m nginx-2142116321 0 1 1 2m nginx-2142116321 0 0 0 2m nginx-3926361531 3 3 3 20s
Get the status of the latest rollout:
kubectl get rs
The output is similar to this:
NAME DESIRED CURRENT READY AGE nginx-2142116321 0 0 0 2m nginx-3926361531 3 3 3 28s
Note:
You cannot rollback a paused Deployment until you resume it.Deployment status
A Deployment enters various states during its lifecycle. It can be progressing while rolling out a new ReplicaSet, it can be complete, or it can fail to progress.
Progressing Deployment
Kubernetes marks a Deployment as progressing when one of the following tasks is performed:
- The Deployment creates a new ReplicaSet.
- The Deployment is scaling up its newest ReplicaSet.
- The Deployment is scaling down its older ReplicaSet(s).
- New Pods become ready or available (ready for at least MinReadySeconds).
When the rollout becomes “progressing”, the Deployment controller adds a condition with the following
attributes to the Deployment's .status.conditions
:
type: Progressing
status: "True"
reason: NewReplicaSetCreated
|reason: FoundNewReplicaSet
|reason: ReplicaSetUpdated
You can monitor the progress for a Deployment by using kubectl rollout status
.
Complete Deployment
Kubernetes marks a Deployment as complete when it has the following characteristics:
- All of the replicas associated with the Deployment have been updated to the latest version you've specified, meaning any updates you've requested have been completed.
- All of the replicas associated with the Deployment are available.
- No old replicas for the Deployment are running.
When the rollout becomes “complete”, the Deployment controller sets a condition with the following
attributes to the Deployment's .status.conditions
:
type: Progressing
status: "True"
reason: NewReplicaSetAvailable
This Progressing
condition will retain a status value of "True"
until a new rollout
is initiated. The condition holds even when availability of replicas changes (which
does instead affect the Available
condition).
You can check if a Deployment has completed by using kubectl rollout status
. If the rollout completed
successfully, kubectl rollout status
returns a zero exit code.
kubectl rollout status deployment/nginx-deployment
The output is similar to this:
Waiting for rollout to finish: 2 of 3 updated replicas are available...
deployment "nginx-deployment" successfully rolled out
and the exit status from kubectl rollout
is 0 (success):
echo $?
0
Failed Deployment
Your Deployment may get stuck trying to deploy its newest ReplicaSet without ever completing. This can occur due to some of the following factors:
- Insufficient quota
- Readiness probe failures
- Image pull errors
- Insufficient permissions
- Limit ranges
- Application runtime misconfiguration
One way you can detect this condition is to specify a deadline parameter in your Deployment spec:
(.spec.progressDeadlineSeconds
). .spec.progressDeadlineSeconds
denotes the
number of seconds the Deployment controller waits before indicating (in the Deployment status) that the
Deployment progress has stalled.
The following kubectl
command sets the spec with progressDeadlineSeconds
to make the controller report
lack of progress of a rollout for a Deployment after 10 minutes:
kubectl patch deployment/nginx-deployment -p '{"spec":{"progressDeadlineSeconds":600}}'
The output is similar to this:
deployment.apps/nginx-deployment patched
Once the deadline has been exceeded, the Deployment controller adds a DeploymentCondition with the following
attributes to the Deployment's .status.conditions
:
type: Progressing
status: "False"
reason: ProgressDeadlineExceeded
This condition can also fail early and is then set to status value of "False"
due to reasons as ReplicaSetCreateError
.
Also, the deadline is not taken into account anymore once the Deployment rollout completes.
See the Kubernetes API conventions for more information on status conditions.
Note:
Kubernetes takes no action on a stalled Deployment other than to report a status condition withreason: ProgressDeadlineExceeded
. Higher level orchestrators can take advantage of it and act accordingly, for
example, rollback the Deployment to its previous version.Note:
If you pause a Deployment rollout, Kubernetes does not check progress against your specified deadline. You can safely pause a Deployment rollout in the middle of a rollout and resume without triggering the condition for exceeding the deadline.You may experience transient errors with your Deployments, either due to a low timeout that you have set or due to any other kind of error that can be treated as transient. For example, let's suppose you have insufficient quota. If you describe the Deployment you will notice the following section:
kubectl describe deployment nginx-deployment
The output is similar to this:
<...>
Conditions:
Type Status Reason
---- ------ ------
Available True MinimumReplicasAvailable
Progressing True ReplicaSetUpdated
ReplicaFailure True FailedCreate
<...>
If you run kubectl get deployment nginx-deployment -o yaml
, the Deployment status is similar to this:
status:
availableReplicas: 2
conditions:
- lastTransitionTime: 2016-10-04T12:25:39Z
lastUpdateTime: 2016-10-04T12:25:39Z
message: Replica set "nginx-deployment-4262182780" is progressing.
reason: ReplicaSetUpdated
status: "True"
type: Progressing
- lastTransitionTime: 2016-10-04T12:25:42Z
lastUpdateTime: 2016-10-04T12:25:42Z
message: Deployment has minimum availability.
reason: MinimumReplicasAvailable
status: "True"
type: Available
- lastTransitionTime: 2016-10-04T12:25:39Z
lastUpdateTime: 2016-10-04T12:25:39Z
message: 'Error creating: pods "nginx-deployment-4262182780-" is forbidden: exceeded quota:
object-counts, requested: pods=1, used: pods=3, limited: pods=2'
reason: FailedCreate
status: "True"
type: ReplicaFailure
observedGeneration: 3
replicas: 2
unavailableReplicas: 2
Eventually, once the Deployment progress deadline is exceeded, Kubernetes updates the status and the reason for the Progressing condition:
Conditions:
Type Status Reason
---- ------ ------
Available True MinimumReplicasAvailable
Progressing False ProgressDeadlineExceeded
ReplicaFailure True FailedCreate
You can address an issue of insufficient quota by scaling down your Deployment, by scaling down other
controllers you may be running, or by increasing quota in your namespace. If you satisfy the quota
conditions and the Deployment controller then completes the Deployment rollout, you'll see the
Deployment's status update with a successful condition (status: "True"
and reason: NewReplicaSetAvailable
).
Conditions:
Type Status Reason
---- ------ ------
Available True MinimumReplicasAvailable
Progressing True NewReplicaSetAvailable
type: Available
with status: "True"
means that your Deployment has minimum availability. Minimum availability is dictated
by the parameters specified in the deployment strategy. type: Progressing
with status: "True"
means that your Deployment
is either in the middle of a rollout and it is progressing or that it has successfully completed its progress and the minimum
required new replicas are available (see the Reason of the condition for the particulars - in our case
reason: NewReplicaSetAvailable
means that the Deployment is complete).
You can check if a Deployment has failed to progress by using kubectl rollout status
. kubectl rollout status
returns a non-zero exit code if the Deployment has exceeded the progression deadline.
kubectl rollout status deployment/nginx-deployment
The output is similar to this:
Waiting for rollout to finish: 2 out of 3 new replicas have been updated...
error: deployment "nginx" exceeded its progress deadline
and the exit status from kubectl rollout
is 1 (indicating an error):
echo $?
1
Operating on a failed deployment
All actions that apply to a complete Deployment also apply to a failed Deployment. You can scale it up/down, roll back to a previous revision, or even pause it if you need to apply multiple tweaks in the Deployment Pod template.
Clean up Policy
You can set .spec.revisionHistoryLimit
field in a Deployment to specify how many old ReplicaSets for
this Deployment you want to retain. The rest will be garbage-collected in the background. By default,
it is 10.
Note:
Explicitly setting this field to 0, will result in cleaning up all the history of your Deployment thus that Deployment will not be able to roll back.Canary Deployment
If you want to roll out releases to a subset of users or servers using the Deployment, you can create multiple Deployments, one for each release, following the canary pattern described in managing resources.
Writing a Deployment Spec
As with all other Kubernetes configs, a Deployment needs .apiVersion
, .kind
, and .metadata
fields.
For general information about working with config files, see
deploying applications,
configuring containers, and using kubectl to manage resources documents.
When the control plane creates new Pods for a Deployment, the .metadata.name
of the
Deployment is part of the basis for naming those Pods. The name of a Deployment must be a valid
DNS subdomain
value, but this can produce unexpected results for the Pod hostnames. For best compatibility,
the name should follow the more restrictive rules for a
DNS label.
A Deployment also needs a .spec
section.
Pod Template
The .spec.template
and .spec.selector
are the only required fields of the .spec
.
The .spec.template
is a Pod template. It has exactly the same schema as a Pod, except it is nested and does not have an apiVersion
or kind
.
In addition to required fields for a Pod, a Pod template in a Deployment must specify appropriate labels and an appropriate restart policy. For labels, make sure not to overlap with other controllers. See selector.
Only a .spec.template.spec.restartPolicy
equal to Always
is
allowed, which is the default if not specified.
Replicas
.spec.replicas
is an optional field that specifies the number of desired Pods. It defaults to 1.
Should you manually scale a Deployment, example via kubectl scale deployment deployment --replicas=X
, and then you update that Deployment based on a manifest
(for example: by running kubectl apply -f deployment.yaml
),
then applying that manifest overwrites the manual scaling that you previously did.
If a HorizontalPodAutoscaler (or any
similar API for horizontal scaling) is managing scaling for a Deployment, don't set .spec.replicas
.
Instead, allow the Kubernetes
control plane to manage the
.spec.replicas
field automatically.
Selector
.spec.selector
is a required field that specifies a label selector
for the Pods targeted by this Deployment.
.spec.selector
must match .spec.template.metadata.labels
, or it will be rejected by the API.
In API version apps/v1
, .spec.selector
and .metadata.labels
do not default to .spec.template.metadata.labels
if not set. So they must be set explicitly. Also note that .spec.selector
is immutable after creation of the Deployment in apps/v1
.
A Deployment may terminate Pods whose labels match the selector if their template is different
from .spec.template
or if the total number of such Pods exceeds .spec.replicas
. It brings up new
Pods with .spec.template
if the number of Pods is less than the desired number.
Note:
You should not create other Pods whose labels match this selector, either directly, by creating another Deployment, or by creating another controller such as a ReplicaSet or a ReplicationController. If you do so, the first Deployment thinks that it created these other Pods. Kubernetes does not stop you from doing this.If you have multiple controllers that have overlapping selectors, the controllers will fight with each other and won't behave correctly.
Strategy
.spec.strategy
specifies the strategy used to replace old Pods by new ones.
.spec.strategy.type
can be "Recreate" or "RollingUpdate". "RollingUpdate" is
the default value.
Recreate Deployment
All existing Pods are killed before new ones are created when .spec.strategy.type==Recreate
.
Note:
This will only guarantee Pod termination previous to creation for upgrades. If you upgrade a Deployment, all Pods of the old revision will be terminated immediately. Successful removal is awaited before any Pod of the new revision is created. If you manually delete a Pod, the lifecycle is controlled by the ReplicaSet and the replacement will be created immediately (even if the old Pod is still in a Terminating state). If you need an "at most" guarantee for your Pods, you should consider using a StatefulSet.Rolling Update Deployment
The Deployment updates Pods in a rolling update
fashion when .spec.strategy.type==RollingUpdate
. You can specify maxUnavailable
and maxSurge
to control
the rolling update process.
Max Unavailable
.spec.strategy.rollingUpdate.maxUnavailable
is an optional field that specifies the maximum number
of Pods that can be unavailable during the update process. The value can be an absolute number (for example, 5)
or a percentage of desired Pods (for example, 10%). The absolute number is calculated from percentage by
rounding down. The value cannot be 0 if .spec.strategy.rollingUpdate.maxSurge
is 0. The default value is 25%.
For example, when this value is set to 30%, the old ReplicaSet can be scaled down to 70% of desired Pods immediately when the rolling update starts. Once new Pods are ready, old ReplicaSet can be scaled down further, followed by scaling up the new ReplicaSet, ensuring that the total number of Pods available at all times during the update is at least 70% of the desired Pods.
Max Surge
.spec.strategy.rollingUpdate.maxSurge
is an optional field that specifies the maximum number of Pods
that can be created over the desired number of Pods. The value can be an absolute number (for example, 5) or a
percentage of desired Pods (for example, 10%). The value cannot be 0 if MaxUnavailable
is 0. The absolute number
is calculated from the percentage by rounding up. The default value is 25%.
For example, when this value is set to 30%, the new ReplicaSet can be scaled up immediately when the rolling update starts, such that the total number of old and new Pods does not exceed 130% of desired Pods. Once old Pods have been killed, the new ReplicaSet can be scaled up further, ensuring that the total number of Pods running at any time during the update is at most 130% of desired Pods.
Here are some Rolling Update Deployment examples that use the maxUnavailable
and maxSurge
:
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-deployment
labels:
app: nginx
spec:
replicas: 3
selector:
matchLabels:
app: nginx
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
strategy:
type: RollingUpdate
rollingUpdate:
maxUnavailable: 1
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-deployment
labels:
app: nginx
spec:
replicas: 3
selector:
matchLabels:
app: nginx
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
strategy:
type: RollingUpdate
rollingUpdate:
maxSurge: 1
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-deployment
labels:
app: nginx
spec:
replicas: 3
selector:
matchLabels:
app: nginx
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
strategy:
type: RollingUpdate
rollingUpdate:
maxSurge: 1
maxUnavailable: 1
Progress Deadline Seconds
.spec.progressDeadlineSeconds
is an optional field that specifies the number of seconds you want
to wait for your Deployment to progress before the system reports back that the Deployment has
failed progressing - surfaced as a condition with type: Progressing
, status: "False"
.
and reason: ProgressDeadlineExceeded
in the status of the resource. The Deployment controller will keep
retrying the Deployment. This defaults to 600. In the future, once automatic rollback will be implemented, the Deployment
controller will roll back a Deployment as soon as it observes such a condition.
If specified, this field needs to be greater than .spec.minReadySeconds
.
Min Ready Seconds
.spec.minReadySeconds
is an optional field that specifies the minimum number of seconds for which a newly
created Pod should be ready without any of its containers crashing, for it to be considered available.
This defaults to 0 (the Pod will be considered available as soon as it is ready). To learn more about when
a Pod is considered ready, see Container Probes.
Revision History Limit
A Deployment's revision history is stored in the ReplicaSets it controls.
.spec.revisionHistoryLimit
is an optional field that specifies the number of old ReplicaSets to retain
to allow rollback. These old ReplicaSets consume resources in etcd
and crowd the output of kubectl get rs
. The configuration of each Deployment revision is stored in its ReplicaSets; therefore, once an old ReplicaSet is deleted, you lose the ability to rollback to that revision of Deployment. By default, 10 old ReplicaSets will be kept, however its ideal value depends on the frequency and stability of new Deployments.
More specifically, setting this field to zero means that all old ReplicaSets with 0 replicas will be cleaned up. In this case, a new Deployment rollout cannot be undone, since its revision history is cleaned up.
Paused
.spec.paused
is an optional boolean field for pausing and resuming a Deployment. The only difference between
a paused Deployment and one that is not paused, is that any changes into the PodTemplateSpec of the paused
Deployment will not trigger new rollouts as long as it is paused. A Deployment is not paused by default when
it is created.
What's next
- Learn more about Pods.
- Run a stateless application using a Deployment.
- Read the Deployment to understand the Deployment API.
- Read about PodDisruptionBudget and how you can use it to manage application availability during disruptions.
- Use kubectl to create a Deployment.
4.2.2 - ReplicaSet
A ReplicaSet's purpose is to maintain a stable set of replica Pods running at any given time. As such, it is often used to guarantee the availability of a specified number of identical Pods.
How a ReplicaSet works
A ReplicaSet is defined with fields, including a selector that specifies how to identify Pods it can acquire, a number of replicas indicating how many Pods it should be maintaining, and a pod template specifying the data of new Pods it should create to meet the number of replicas criteria. A ReplicaSet then fulfills its purpose by creating and deleting Pods as needed to reach the desired number. When a ReplicaSet needs to create new Pods, it uses its Pod template.
A ReplicaSet is linked to its Pods via the Pods' metadata.ownerReferences field, which specifies what resource the current object is owned by. All Pods acquired by a ReplicaSet have their owning ReplicaSet's identifying information within their ownerReferences field. It's through this link that the ReplicaSet knows of the state of the Pods it is maintaining and plans accordingly.
A ReplicaSet identifies new Pods to acquire by using its selector. If there is a Pod that has no OwnerReference or the OwnerReference is not a Controller and it matches a ReplicaSet's selector, it will be immediately acquired by said ReplicaSet.
When to use a ReplicaSet
A ReplicaSet ensures that a specified number of pod replicas are running at any given time. However, a Deployment is a higher-level concept that manages ReplicaSets and provides declarative updates to Pods along with a lot of other useful features. Therefore, we recommend using Deployments instead of directly using ReplicaSets, unless you require custom update orchestration or don't require updates at all.
This actually means that you may never need to manipulate ReplicaSet objects: use a Deployment instead, and define your application in the spec section.
Example
apiVersion: apps/v1
kind: ReplicaSet
metadata:
name: frontend
labels:
app: guestbook
tier: frontend
spec:
# modify replicas according to your case
replicas: 3
selector:
matchLabels:
tier: frontend
template:
metadata:
labels:
tier: frontend
spec:
containers:
- name: php-redis
image: us-docker.pkg.dev/google-samples/containers/gke/gb-frontend:v5
Saving this manifest into frontend.yaml
and submitting it to a Kubernetes cluster will
create the defined ReplicaSet and the Pods that it manages.
kubectl apply -f https://kubernetes.io/examples/controllers/frontend.yaml
You can then get the current ReplicaSets deployed:
kubectl get rs
And see the frontend one you created:
NAME DESIRED CURRENT READY AGE
frontend 3 3 3 6s
You can also check on the state of the ReplicaSet:
kubectl describe rs/frontend
And you will see output similar to:
Name: frontend
Namespace: default
Selector: tier=frontend
Labels: app=guestbook
tier=frontend
Annotations: <none>
Replicas: 3 current / 3 desired
Pods Status: 3 Running / 0 Waiting / 0 Succeeded / 0 Failed
Pod Template:
Labels: tier=frontend
Containers:
php-redis:
Image: us-docker.pkg.dev/google-samples/containers/gke/gb-frontend:v5
Port: <none>
Host Port: <none>
Environment: <none>
Mounts: <none>
Volumes: <none>
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal SuccessfulCreate 13s replicaset-controller Created pod: frontend-gbgfx
Normal SuccessfulCreate 13s replicaset-controller Created pod: frontend-rwz57
Normal SuccessfulCreate 13s replicaset-controller Created pod: frontend-wkl7w
And lastly you can check for the Pods brought up:
kubectl get pods
You should see Pod information similar to:
NAME READY STATUS RESTARTS AGE
frontend-gbgfx 1/1 Running 0 10m
frontend-rwz57 1/1 Running 0 10m
frontend-wkl7w 1/1 Running 0 10m
You can also verify that the owner reference of these pods is set to the frontend ReplicaSet. To do this, get the yaml of one of the Pods running:
kubectl get pods frontend-gbgfx -o yaml
The output will look similar to this, with the frontend ReplicaSet's info set in the metadata's ownerReferences field:
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: "2024-02-28T22:30:44Z"
generateName: frontend-
labels:
tier: frontend
name: frontend-gbgfx
namespace: default
ownerReferences:
- apiVersion: apps/v1
blockOwnerDeletion: true
controller: true
kind: ReplicaSet
name: frontend
uid: e129deca-f864-481b-bb16-b27abfd92292
...
Non-Template Pod acquisitions
While you can create bare Pods with no problems, it is strongly recommended to make sure that the bare Pods do not have labels which match the selector of one of your ReplicaSets. The reason for this is because a ReplicaSet is not limited to owning Pods specified by its template-- it can acquire other Pods in the manner specified in the previous sections.
Take the previous frontend ReplicaSet example, and the Pods specified in the following manifest:
apiVersion: v1
kind: Pod
metadata:
name: pod1
labels:
tier: frontend
spec:
containers:
- name: hello1
image: gcr.io/google-samples/hello-app:2.0
---
apiVersion: v1
kind: Pod
metadata:
name: pod2
labels:
tier: frontend
spec:
containers:
- name: hello2
image: gcr.io/google-samples/hello-app:1.0
As those Pods do not have a Controller (or any object) as their owner reference and match the selector of the frontend ReplicaSet, they will immediately be acquired by it.
Suppose you create the Pods after the frontend ReplicaSet has been deployed and has set up its initial Pod replicas to fulfill its replica count requirement:
kubectl apply -f https://kubernetes.io/examples/pods/pod-rs.yaml
The new Pods will be acquired by the ReplicaSet, and then immediately terminated as the ReplicaSet would be over its desired count.
Fetching the Pods:
kubectl get pods
The output shows that the new Pods are either already terminated, or in the process of being terminated:
NAME READY STATUS RESTARTS AGE
frontend-b2zdv 1/1 Running 0 10m
frontend-vcmts 1/1 Running 0 10m
frontend-wtsmm 1/1 Running 0 10m
pod1 0/1 Terminating 0 1s
pod2 0/1 Terminating 0 1s
If you create the Pods first:
kubectl apply -f https://kubernetes.io/examples/pods/pod-rs.yaml
And then create the ReplicaSet however:
kubectl apply -f https://kubernetes.io/examples/controllers/frontend.yaml
You shall see that the ReplicaSet has acquired the Pods and has only created new ones according to its spec until the number of its new Pods and the original matches its desired count. As fetching the Pods:
kubectl get pods
Will reveal in its output:
NAME READY STATUS RESTARTS AGE
frontend-hmmj2 1/1 Running 0 9s
pod1 1/1 Running 0 36s
pod2 1/1 Running 0 36s
In this manner, a ReplicaSet can own a non-homogeneous set of Pods
Writing a ReplicaSet manifest
As with all other Kubernetes API objects, a ReplicaSet needs the apiVersion
, kind
, and metadata
fields.
For ReplicaSets, the kind
is always a ReplicaSet.
When the control plane creates new Pods for a ReplicaSet, the .metadata.name
of the
ReplicaSet is part of the basis for naming those Pods. The name of a ReplicaSet must be a valid
DNS subdomain
value, but this can produce unexpected results for the Pod hostnames. For best compatibility,
the name should follow the more restrictive rules for a
DNS label.
A ReplicaSet also needs a .spec
section.
Pod Template
The .spec.template
is a pod template which is also
required to have labels in place. In our frontend.yaml
example we had one label: tier: frontend
.
Be careful not to overlap with the selectors of other controllers, lest they try to adopt this Pod.
For the template's restart policy field,
.spec.template.spec.restartPolicy
, the only allowed value is Always
, which is the default.
Pod Selector
The .spec.selector
field is a label selector. As discussed
earlier these are the labels used to identify potential Pods to acquire. In our
frontend.yaml
example, the selector was:
matchLabels:
tier: frontend
In the ReplicaSet, .spec.template.metadata.labels
must match spec.selector
, or it will
be rejected by the API.
Note:
For 2 ReplicaSets specifying the same.spec.selector
but different
.spec.template.metadata.labels
and .spec.template.spec
fields, each ReplicaSet ignores the
Pods created by the other ReplicaSet.Replicas
You can specify how many Pods should run concurrently by setting .spec.replicas
. The ReplicaSet will create/delete
its Pods to match this number.
If you do not specify .spec.replicas
, then it defaults to 1.
Working with ReplicaSets
Deleting a ReplicaSet and its Pods
To delete a ReplicaSet and all of its Pods, use
kubectl delete
. The
Garbage collector automatically deletes all of
the dependent Pods by default.
When using the REST API or the client-go
library, you must set propagationPolicy
to
Background
or Foreground
in the -d
option. For example:
kubectl proxy --port=8080
curl -X DELETE 'localhost:8080/apis/apps/v1/namespaces/default/replicasets/frontend' \
-d '{"kind":"DeleteOptions","apiVersion":"v1","propagationPolicy":"Foreground"}' \
-H "Content-Type: application/json"
Deleting just a ReplicaSet
You can delete a ReplicaSet without affecting any of its Pods using
kubectl delete
with the --cascade=orphan
option.
When using the REST API or the client-go
library, you must set propagationPolicy
to Orphan
.
For example:
kubectl proxy --port=8080
curl -X DELETE 'localhost:8080/apis/apps/v1/namespaces/default/replicasets/frontend' \
-d '{"kind":"DeleteOptions","apiVersion":"v1","propagationPolicy":"Orphan"}' \
-H "Content-Type: application/json"
Once the original is deleted, you can create a new ReplicaSet to replace it. As long
as the old and new .spec.selector
are the same, then the new one will adopt the old Pods.
However, it will not make any effort to make existing Pods match a new, different pod template.
To update Pods to a new spec in a controlled way, use a
Deployment, as
ReplicaSets do not support a rolling update directly.
Isolating Pods from a ReplicaSet
You can remove Pods from a ReplicaSet by changing their labels. This technique may be used to remove Pods from service for debugging, data recovery, etc. Pods that are removed in this way will be replaced automatically ( assuming that the number of replicas is not also changed).
Scaling a ReplicaSet
A ReplicaSet can be easily scaled up or down by simply updating the .spec.replicas
field. The ReplicaSet controller
ensures that a desired number of Pods with a matching label selector are available and operational.
When scaling down, the ReplicaSet controller chooses which pods to delete by sorting the available pods to prioritize scaling down pods based on the following general algorithm:
- Pending (and unschedulable) pods are scaled down first
- If
controller.kubernetes.io/pod-deletion-cost
annotation is set, then the pod with the lower value will come first. - Pods on nodes with more replicas come before pods on nodes with fewer replicas.
- If the pods' creation times differ, the pod that was created more recently comes before the older pod (the creation times are bucketed on an integer log scale).
If all of the above match, then selection is random.
Pod deletion cost
Kubernetes v1.22 [beta]
Using the controller.kubernetes.io/pod-deletion-cost
annotation, users can set a preference regarding which pods to remove first when downscaling a ReplicaSet.
The annotation should be set on the pod, the range is [-2147483648, 2147483647]. It represents the cost of deleting a pod compared to other pods belonging to the same ReplicaSet. Pods with lower deletion cost are preferred to be deleted before pods with higher deletion cost.
The implicit value for this annotation for pods that don't set it is 0; negative values are permitted. Invalid values will be rejected by the API server.
This feature is beta and enabled by default. You can disable it using the
feature gate
PodDeletionCost
in both kube-apiserver and kube-controller-manager.
Note:
- This is honored on a best-effort basis, so it does not offer any guarantees on pod deletion order.
- Users should avoid updating the annotation frequently, such as updating it based on a metric value, because doing so will generate a significant number of pod updates on the apiserver.
Example Use Case
The different pods of an application could have different utilization levels. On scale down, the application
may prefer to remove the pods with lower utilization. To avoid frequently updating the pods, the application
should update controller.kubernetes.io/pod-deletion-cost
once before issuing a scale down (setting the
annotation to a value proportional to pod utilization level). This works if the application itself controls
the down scaling; for example, the driver pod of a Spark deployment.
ReplicaSet as a Horizontal Pod Autoscaler Target
A ReplicaSet can also be a target for Horizontal Pod Autoscalers (HPA). That is, a ReplicaSet can be auto-scaled by an HPA. Here is an example HPA targeting the ReplicaSet we created in the previous example.
apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
name: frontend-scaler
spec:
scaleTargetRef:
kind: ReplicaSet
name: frontend
minReplicas: 3
maxReplicas: 10
targetCPUUtilizationPercentage: 50
Saving this manifest into hpa-rs.yaml
and submitting it to a Kubernetes cluster should
create the defined HPA that autoscales the target ReplicaSet depending on the CPU usage
of the replicated Pods.
kubectl apply -f https://k8s.io/examples/controllers/hpa-rs.yaml
Alternatively, you can use the kubectl autoscale
command to accomplish the same
(and it's easier!)
kubectl autoscale rs frontend --max=10 --min=3 --cpu-percent=50
Alternatives to ReplicaSet
Deployment (recommended)
Deployment
is an object which can own ReplicaSets and update
them and their Pods via declarative, server-side rolling updates.
While ReplicaSets can be used independently, today they're mainly used by Deployments as a mechanism to orchestrate Pod
creation, deletion and updates. When you use Deployments you don't have to worry about managing the ReplicaSets that
they create. Deployments own and manage their ReplicaSets.
As such, it is recommended to use Deployments when you want ReplicaSets.
Bare Pods
Unlike the case where a user directly created Pods, a ReplicaSet replaces Pods that are deleted or terminated for any reason, such as in the case of node failure or disruptive node maintenance, such as a kernel upgrade. For this reason, we recommend that you use a ReplicaSet even if your application requires only a single Pod. Think of it similarly to a process supervisor, only it supervises multiple Pods across multiple nodes instead of individual processes on a single node. A ReplicaSet delegates local container restarts to some agent on the node such as Kubelet.
Job
Use a Job
instead of a ReplicaSet for Pods that are
expected to terminate on their own (that is, batch jobs).
DaemonSet
Use a DaemonSet
instead of a ReplicaSet for Pods that provide a
machine-level function, such as machine monitoring or machine logging. These Pods have a lifetime that is tied
to a machine lifetime: the Pod needs to be running on the machine before other Pods start, and are
safe to terminate when the machine is otherwise ready to be rebooted/shutdown.
ReplicationController
ReplicaSets are the successors to ReplicationControllers. The two serve the same purpose, and behave similarly, except that a ReplicationController does not support set-based selector requirements as described in the labels user guide. As such, ReplicaSets are preferred over ReplicationControllers
What's next
- Learn about Pods.
- Learn about Deployments.
- Run a Stateless Application Using a Deployment, which relies on ReplicaSets to work.
ReplicaSet
is a top-level resource in the Kubernetes REST API. Read the ReplicaSet object definition to understand the API for replica sets.- Read about PodDisruptionBudget and how you can use it to manage application availability during disruptions.
4.2.3 - StatefulSets
StatefulSet is the workload API object used to manage stateful applications.
Manages the deployment and scaling of a set of Pods, and provides guarantees about the ordering and uniqueness of these Pods.
Like a Deployment, a StatefulSet manages Pods that are based on an identical container spec. Unlike a Deployment, a StatefulSet maintains a sticky identity for each of its Pods. These pods are created from the same spec, but are not interchangeable: each has a persistent identifier that it maintains across any rescheduling.
If you want to use storage volumes to provide persistence for your workload, you can use a StatefulSet as part of the solution. Although individual Pods in a StatefulSet are susceptible to failure, the persistent Pod identifiers make it easier to match existing volumes to the new Pods that replace any that have failed.
Using StatefulSets
StatefulSets are valuable for applications that require one or more of the following.
- Stable, unique network identifiers.
- Stable, persistent storage.
- Ordered, graceful deployment and scaling.
- Ordered, automated rolling updates.
In the above, stable is synonymous with persistence across Pod (re)scheduling. If an application doesn't require any stable identifiers or ordered deployment, deletion, or scaling, you should deploy your application using a workload object that provides a set of stateless replicas. Deployment or ReplicaSet may be better suited to your stateless needs.
Limitations
- The storage for a given Pod must either be provisioned by a PersistentVolume Provisioner (examples here) based on the requested storage class, or pre-provisioned by an admin.
- Deleting and/or scaling a StatefulSet down will not delete the volumes associated with the StatefulSet. This is done to ensure data safety, which is generally more valuable than an automatic purge of all related StatefulSet resources.
- StatefulSets currently require a Headless Service to be responsible for the network identity of the Pods. You are responsible for creating this Service.
- StatefulSets do not provide any guarantees on the termination of pods when a StatefulSet is deleted. To achieve ordered and graceful termination of the pods in the StatefulSet, it is possible to scale the StatefulSet down to 0 prior to deletion.
- When using Rolling Updates with the default
Pod Management Policy (
OrderedReady
), it's possible to get into a broken state that requires manual intervention to repair.
Components
The example below demonstrates the components of a StatefulSet.
apiVersion: v1
kind: Service
metadata:
name: nginx
labels:
app: nginx
spec:
ports:
- port: 80
name: web
clusterIP: None
selector:
app: nginx
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: web
spec:
selector:
matchLabels:
app: nginx # has to match .spec.template.metadata.labels
serviceName: "nginx"
replicas: 3 # by default is 1
minReadySeconds: 10 # by default is 0
template:
metadata:
labels:
app: nginx # has to match .spec.selector.matchLabels
spec:
terminationGracePeriodSeconds: 10
containers:
- name: nginx
image: registry.k8s.io/nginx-slim:0.24
ports:
- containerPort: 80
name: web
volumeMounts:
- name: www
mountPath: /usr/share/nginx/html
volumeClaimTemplates:
- metadata:
name: www
spec:
accessModes: [ "ReadWriteOnce" ]
storageClassName: "my-storage-class"
resources:
requests:
storage: 1Gi
Note:
This example uses theReadWriteOnce
access mode, for simplicity. For
production use, the Kubernetes project recommends using the ReadWriteOncePod
access mode instead.In the above example:
- A Headless Service, named
nginx
, is used to control the network domain. - The StatefulSet, named
web
, has a Spec that indicates that 3 replicas of the nginx container will be launched in unique Pods. - The
volumeClaimTemplates
will provide stable storage using PersistentVolumes provisioned by a PersistentVolume Provisioner.
The name of a StatefulSet object must be a valid DNS label.
Pod Selector
You must set the .spec.selector
field of a StatefulSet to match the labels of its
.spec.template.metadata.labels
. Failing to specify a matching Pod Selector will result in a
validation error during StatefulSet creation.
Volume Claim Templates
You can set the .spec.volumeClaimTemplates
field to create a
PersistentVolumeClaim.
This will provide stable storage to the StatefulSet if either
- The StorageClass specified for the volume claim is set up to use dynamic provisioning, or
- The cluster already contains a PersistentVolume with the correct StorageClass and sufficient available storage space.
Minimum ready seconds
Kubernetes v1.25 [stable]
.spec.minReadySeconds
is an optional field that specifies the minimum number of seconds for which a newly
created Pod should be running and ready without any of its containers crashing, for it to be considered available.
This is used to check progression of a rollout when using a Rolling Update strategy.
This field defaults to 0 (the Pod will be considered available as soon as it is ready). To learn more about when
a Pod is considered ready, see Container Probes.
Pod Identity
StatefulSet Pods have a unique identity that consists of an ordinal, a stable network identity, and stable storage. The identity sticks to the Pod, regardless of which node it's (re)scheduled on.
Ordinal Index
For a StatefulSet with N replicas, each Pod in the StatefulSet
will be assigned an integer ordinal, that is unique over the Set. By default,
pods will be assigned ordinals from 0 up through N-1. The StatefulSet controller
will also add a pod label with this index: apps.kubernetes.io/pod-index
.
Start ordinal
Kubernetes v1.31 [stable]
(enabled by default: true).spec.ordinals
is an optional field that allows you to configure the integer
ordinals assigned to each Pod. It defaults to nil. Within the field, you can
configure the following options:
.spec.ordinals.start
: If the.spec.ordinals.start
field is set, Pods will be assigned ordinals from.spec.ordinals.start
up through.spec.ordinals.start + .spec.replicas - 1
.
Stable Network ID
Each Pod in a StatefulSet derives its hostname from the name of the StatefulSet
and the ordinal of the Pod. The pattern for the constructed hostname
is $(statefulset name)-$(ordinal)
. The example above will create three Pods
named web-0,web-1,web-2
.
A StatefulSet can use a Headless Service
to control the domain of its Pods. The domain managed by this Service takes the form:
$(service name).$(namespace).svc.cluster.local
, where "cluster.local" is the
cluster domain.
As each Pod is created, it gets a matching DNS subdomain, taking the form:
$(podname).$(governing service domain)
, where the governing service is defined
by the serviceName
field on the StatefulSet.
Depending on how DNS is configured in your cluster, you may not be able to look up the DNS name for a newly-run Pod immediately. This behavior can occur when other clients in the cluster have already sent queries for the hostname of the Pod before it was created. Negative caching (normal in DNS) means that the results of previous failed lookups are remembered and reused, even after the Pod is running, for at least a few seconds.
If you need to discover Pods promptly after they are created, you have a few options:
- Query the Kubernetes API directly (for example, using a watch) rather than relying on DNS lookups.
- Decrease the time of caching in your Kubernetes DNS provider (typically this means editing the config map for CoreDNS, which currently caches for 30 seconds).
As mentioned in the limitations section, you are responsible for creating the Headless Service responsible for the network identity of the pods.
Here are some examples of choices for Cluster Domain, Service name, StatefulSet name, and how that affects the DNS names for the StatefulSet's Pods.
Cluster Domain | Service (ns/name) | StatefulSet (ns/name) | StatefulSet Domain | Pod DNS | Pod Hostname |
---|---|---|---|---|---|
cluster.local | default/nginx | default/web | nginx.default.svc.cluster.local | web-{0..N-1}.nginx.default.svc.cluster.local | web-{0..N-1} |
cluster.local | foo/nginx | foo/web | nginx.foo.svc.cluster.local | web-{0..N-1}.nginx.foo.svc.cluster.local | web-{0..N-1} |
kube.local | foo/nginx | foo/web | nginx.foo.svc.kube.local | web-{0..N-1}.nginx.foo.svc.kube.local | web-{0..N-1} |
Stable Storage
For each VolumeClaimTemplate entry defined in a StatefulSet, each Pod receives one
PersistentVolumeClaim. In the nginx example above, each Pod receives a single PersistentVolume
with a StorageClass of my-storage-class
and 1 GiB of provisioned storage. If no StorageClass
is specified, then the default StorageClass will be used. When a Pod is (re)scheduled
onto a node, its volumeMounts
mount the PersistentVolumes associated with its
PersistentVolume Claims. Note that, the PersistentVolumes associated with the
Pods' PersistentVolume Claims are not deleted when the Pods, or StatefulSet are deleted.
This must be done manually.
Pod Name Label
When the StatefulSet controller creates a Pod,
it adds a label, statefulset.kubernetes.io/pod-name
, that is set to the name of
the Pod. This label allows you to attach a Service to a specific Pod in
the StatefulSet.
Pod index label
Kubernetes v1.28 [beta]
When the StatefulSet controller creates a Pod,
the new Pod is labelled with apps.kubernetes.io/pod-index
. The value of this label is the ordinal index of
the Pod. This label allows you to route traffic to a particular pod index, filter logs/metrics
using the pod index label, and more. Note the feature gate PodIndexLabel
must be enabled for this
feature, and it is enabled by default.
Deployment and Scaling Guarantees
- For a StatefulSet with N replicas, when Pods are being deployed, they are created sequentially, in order from {0..N-1}.
- When Pods are being deleted, they are terminated in reverse order, from {N-1..0}.
- Before a scaling operation is applied to a Pod, all of its predecessors must be Running and Ready.
- Before a Pod is terminated, all of its successors must be completely shutdown.
The StatefulSet should not specify a pod.Spec.TerminationGracePeriodSeconds
of 0. This practice
is unsafe and strongly discouraged. For further explanation, please refer to
force deleting StatefulSet Pods.
When the nginx example above is created, three Pods will be deployed in the order web-0, web-1, web-2. web-1 will not be deployed before web-0 is Running and Ready, and web-2 will not be deployed until web-1 is Running and Ready. If web-0 should fail, after web-1 is Running and Ready, but before web-2 is launched, web-2 will not be launched until web-0 is successfully relaunched and becomes Running and Ready.
If a user were to scale the deployed example by patching the StatefulSet such that
replicas=1
, web-2 would be terminated first. web-1 would not be terminated until web-2
is fully shutdown and deleted. If web-0 were to fail after web-2 has been terminated and
is completely shutdown, but prior to web-1's termination, web-1 would not be terminated
until web-0 is Running and Ready.
Pod Management Policies
StatefulSet allows you to relax its ordering guarantees while
preserving its uniqueness and identity guarantees via its .spec.podManagementPolicy
field.
OrderedReady Pod Management
OrderedReady
pod management is the default for StatefulSets. It implements the behavior
described above.
Parallel Pod Management
Parallel
pod management tells the StatefulSet controller to launch or
terminate all Pods in parallel, and to not wait for Pods to become Running
and Ready or completely terminated prior to launching or terminating another
Pod. This option only affects the behavior for scaling operations. Updates are not
affected.
Update strategies
A StatefulSet's .spec.updateStrategy
field allows you to configure
and disable automated rolling updates for containers, labels, resource request/limits, and
annotations for the Pods in a StatefulSet. There are two possible values:
OnDelete
- When a StatefulSet's
.spec.updateStrategy.type
is set toOnDelete
, the StatefulSet controller will not automatically update the Pods in a StatefulSet. Users must manually delete Pods to cause the controller to create new Pods that reflect modifications made to a StatefulSet's.spec.template
. RollingUpdate
- The
RollingUpdate
update strategy implements automated, rolling updates for the Pods in a StatefulSet. This is the default update strategy.
Rolling Updates
When a StatefulSet's .spec.updateStrategy.type
is set to RollingUpdate
, the
StatefulSet controller will delete and recreate each Pod in the StatefulSet. It will proceed
in the same order as Pod termination (from the largest ordinal to the smallest), updating
each Pod one at a time.
The Kubernetes control plane waits until an updated Pod is Running and Ready prior
to updating its predecessor. If you have set .spec.minReadySeconds
(see
Minimum Ready Seconds), the control plane additionally waits that
amount of time after the Pod turns ready, before moving on.
Partitioned rolling updates
The RollingUpdate
update strategy can be partitioned, by specifying a
.spec.updateStrategy.rollingUpdate.partition
. If a partition is specified, all Pods with an
ordinal that is greater than or equal to the partition will be updated when the StatefulSet's
.spec.template
is updated. All Pods with an ordinal that is less than the partition will not
be updated, and, even if they are deleted, they will be recreated at the previous version. If a
StatefulSet's .spec.updateStrategy.rollingUpdate.partition
is greater than its .spec.replicas
,
updates to its .spec.template
will not be propagated to its Pods.
In most cases you will not need to use a partition, but they are useful if you want to stage an
update, roll out a canary, or perform a phased roll out.
Maximum unavailable Pods
Kubernetes v1.24 [alpha]
You can control the maximum number of Pods that can be unavailable during an update
by specifying the .spec.updateStrategy.rollingUpdate.maxUnavailable
field.
The value can be an absolute number (for example, 5
) or a percentage of desired
Pods (for example, 10%
). Absolute number is calculated from the percentage value
by rounding it up. This field cannot be 0. The default setting is 1.
This field applies to all Pods in the range 0
to replicas - 1
. If there is any
unavailable Pod in the range 0
to replicas - 1
, it will be counted towards
maxUnavailable
.
Note:
ThemaxUnavailable
field is in Alpha stage and it is honored only by API servers
that are running with the MaxUnavailableStatefulSet
feature gate
enabled.Forced rollback
When using Rolling Updates with the default
Pod Management Policy (OrderedReady
),
it's possible to get into a broken state that requires manual intervention to repair.
If you update the Pod template to a configuration that never becomes Running and Ready (for example, due to a bad binary or application-level configuration error), StatefulSet will stop the rollout and wait.
In this state, it's not enough to revert the Pod template to a good configuration. Due to a known issue, StatefulSet will continue to wait for the broken Pod to become Ready (which never happens) before it will attempt to revert it back to the working configuration.
After reverting the template, you must also delete any Pods that StatefulSet had already attempted to run with the bad configuration. StatefulSet will then begin to recreate the Pods using the reverted template.
PersistentVolumeClaim retention
Kubernetes v1.27 [beta]
The optional .spec.persistentVolumeClaimRetentionPolicy
field controls if
and how PVCs are deleted during the lifecycle of a StatefulSet. You must enable the
StatefulSetAutoDeletePVC
feature gate
on the API server and the controller manager to use this field.
Once enabled, there are two policies you can configure for each StatefulSet:
whenDeleted
- configures the volume retention behavior that applies when the StatefulSet is deleted
whenScaled
- configures the volume retention behavior that applies when the replica count of the StatefulSet is reduced; for example, when scaling down the set.
For each policy that you can configure, you can set the value to either Delete
or Retain
.
Delete
- The PVCs created from the StatefulSet
volumeClaimTemplate
are deleted for each Pod affected by the policy. With thewhenDeleted
policy all PVCs from thevolumeClaimTemplate
are deleted after their Pods have been deleted. With thewhenScaled
policy, only PVCs corresponding to Pod replicas being scaled down are deleted, after their Pods have been deleted. Retain
(default)- PVCs from the
volumeClaimTemplate
are not affected when their Pod is deleted. This is the behavior before this new feature.
Bear in mind that these policies only apply when Pods are being removed due to the StatefulSet being deleted or scaled down. For example, if a Pod associated with a StatefulSet fails due to node failure, and the control plane creates a replacement Pod, the StatefulSet retains the existing PVC. The existing volume is unaffected, and the cluster will attach it to the node where the new Pod is about to launch.
The default for policies is Retain
, matching the StatefulSet behavior before this new feature.
Here is an example policy.
apiVersion: apps/v1
kind: StatefulSet
...
spec:
persistentVolumeClaimRetentionPolicy:
whenDeleted: Retain
whenScaled: Delete
...
The StatefulSet controller adds
owner references
to its PVCs, which are then deleted by the garbage collector after the Pod is terminated. This enables the Pod to
cleanly unmount all volumes before the PVCs are deleted (and before the backing PV and
volume are deleted, depending on the retain policy). When you set the whenDeleted
policy to Delete
, an owner reference to the StatefulSet instance is placed on all PVCs
associated with that StatefulSet.
The whenScaled
policy must delete PVCs only when a Pod is scaled down, and not when a
Pod is deleted for another reason. When reconciling, the StatefulSet controller compares
its desired replica count to the actual Pods present on the cluster. Any StatefulSet Pod
whose id greater than the replica count is condemned and marked for deletion. If the
whenScaled
policy is Delete
, the condemned Pods are first set as owners to the
associated StatefulSet template PVCs, before the Pod is deleted. This causes the PVCs
to be garbage collected after only the condemned Pods have terminated.
This means that if the controller crashes and restarts, no Pod will be deleted before its owner reference has been updated appropriate to the policy. If a condemned Pod is force-deleted while the controller is down, the owner reference may or may not have been set up, depending on when the controller crashed. It may take several reconcile loops to update the owner references, so some condemned Pods may have set up owner references and others may not. For this reason we recommend waiting for the controller to come back up, which will verify owner references before terminating Pods. If that is not possible, the operator should verify the owner references on PVCs to ensure the expected objects are deleted when Pods are force-deleted.
Replicas
.spec.replicas
is an optional field that specifies the number of desired Pods. It defaults to 1.
Should you manually scale a deployment, example via kubectl scale statefulset statefulset --replicas=X
, and then you update that StatefulSet
based on a manifest (for example: by running kubectl apply -f statefulset.yaml
), then applying that manifest overwrites the manual scaling
that you previously did.
If a HorizontalPodAutoscaler
(or any similar API for horizontal scaling) is managing scaling for a
Statefulset, don't set .spec.replicas
. Instead, allow the Kubernetes
control plane to manage
the .spec.replicas
field automatically.
What's next
- Learn about Pods.
- Find out how to use StatefulSets
- Follow an example of deploying a stateful application.
- Follow an example of deploying Cassandra with Stateful Sets.
- Follow an example of running a replicated stateful application.
- Learn how to scale a StatefulSet.
- Learn what's involved when you delete a StatefulSet.
- Learn how to configure a Pod to use a volume for storage.
- Learn how to configure a Pod to use a PersistentVolume for storage.
StatefulSet
is a top-level resource in the Kubernetes REST API. Read the StatefulSet object definition to understand the API for stateful sets.- Read about PodDisruptionBudget and how you can use it to manage application availability during disruptions.
4.2.4 - DaemonSet
A DaemonSet ensures that all (or some) Nodes run a copy of a Pod. As nodes are added to the cluster, Pods are added to them. As nodes are removed from the cluster, those Pods are garbage collected. Deleting a DaemonSet will clean up the Pods it created.
Some typical uses of a DaemonSet are:
- running a cluster storage daemon on every node
- running a logs collection daemon on every node
- running a node monitoring daemon on every node
In a simple case, one DaemonSet, covering all nodes, would be used for each type of daemon. A more complex setup might use multiple DaemonSets for a single type of daemon, but with different flags and/or different memory and cpu requests for different hardware types.
Writing a DaemonSet Spec
Create a DaemonSet
You can describe a DaemonSet in a YAML file. For example, the daemonset.yaml
file below
describes a DaemonSet that runs the fluentd-elasticsearch Docker image:
apiVersion: apps/v1
kind: DaemonSet
metadata:
name: fluentd-elasticsearch
namespace: kube-system
labels:
k8s-app: fluentd-logging
spec:
selector:
matchLabels:
name: fluentd-elasticsearch
template:
metadata:
labels:
name: fluentd-elasticsearch
spec:
tolerations:
# these tolerations are to have the daemonset runnable on control plane nodes
# remove them if your control plane nodes should not run pods
- key: node-role.kubernetes.io/control-plane
operator: Exists
effect: NoSchedule
- key: node-role.kubernetes.io/master
operator: Exists
effect: NoSchedule
containers:
- name: fluentd-elasticsearch
image: quay.io/fluentd_elasticsearch/fluentd:v2.5.2
resources:
limits:
memory: 200Mi
requests:
cpu: 100m
memory: 200Mi
volumeMounts:
- name: varlog
mountPath: /var/log
# it may be desirable to set a high priority class to ensure that a DaemonSet Pod
# preempts running Pods
# priorityClassName: important
terminationGracePeriodSeconds: 30
volumes:
- name: varlog
hostPath:
path: /var/log
Create a DaemonSet based on the YAML file:
kubectl apply -f https://k8s.io/examples/controllers/daemonset.yaml
Required Fields
As with all other Kubernetes config, a DaemonSet needs apiVersion
, kind
, and metadata
fields. For
general information about working with config files, see
running stateless applications
and object management using kubectl.
The name of a DaemonSet object must be a valid DNS subdomain name.
A DaemonSet also needs a
.spec
section.
Pod Template
The .spec.template
is one of the required fields in .spec
.
The .spec.template
is a pod template.
It has exactly the same schema as a Pod,
except it is nested and does not have an apiVersion
or kind
.
In addition to required fields for a Pod, a Pod template in a DaemonSet has to specify appropriate labels (see pod selector).
A Pod Template in a DaemonSet must have a RestartPolicy
equal to Always
, or be unspecified, which defaults to Always
.
Pod Selector
The .spec.selector
field is a pod selector. It works the same as the .spec.selector
of
a Job.
You must specify a pod selector that matches the labels of the
.spec.template
.
Also, once a DaemonSet is created,
its .spec.selector
can not be mutated. Mutating the pod selector can lead to the
unintentional orphaning of Pods, and it was found to be confusing to users.
The .spec.selector
is an object consisting of two fields:
matchLabels
- works the same as the.spec.selector
of a ReplicationController.matchExpressions
- allows to build more sophisticated selectors by specifying key, list of values and an operator that relates the key and values.
When the two are specified the result is ANDed.
The .spec.selector
must match the .spec.template.metadata.labels
.
Config with these two not matching will be rejected by the API.
Running Pods on select Nodes
If you specify a .spec.template.spec.nodeSelector
, then the DaemonSet controller will
create Pods on nodes which match that node selector.
Likewise if you specify a .spec.template.spec.affinity
,
then DaemonSet controller will create Pods on nodes which match that
node affinity.
If you do not specify either, then the DaemonSet controller will create Pods on all nodes.
How Daemon Pods are scheduled
A DaemonSet can be used to ensure that all eligible nodes run a copy of a Pod.
The DaemonSet controller creates a Pod for each eligible node and adds the
spec.affinity.nodeAffinity
field of the Pod to match the target host. After
the Pod is created, the default scheduler typically takes over and then binds
the Pod to the target host by setting the .spec.nodeName
field. If the new
Pod cannot fit on the node, the default scheduler may preempt (evict) some of
the existing Pods based on the
priority
of the new Pod.
Note:
If it's important that the DaemonSet pod run on each node, it's often desirable to set the.spec.template.spec.priorityClassName
of the DaemonSet to a
PriorityClass
with a higher priority to ensure that this eviction occurs.The user can specify a different scheduler for the Pods of the DaemonSet, by
setting the .spec.template.spec.schedulerName
field of the DaemonSet.
The original node affinity specified at the
.spec.template.spec.affinity.nodeAffinity
field (if specified) is taken into
consideration by the DaemonSet controller when evaluating the eligible nodes,
but is replaced on the created Pod with the node affinity that matches the name
of the eligible node.
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchFields:
- key: metadata.name
operator: In
values:
- target-host-name
Taints and tolerations
The DaemonSet controller automatically adds a set of tolerations to DaemonSet Pods:
Toleration key | Effect | Details |
---|---|---|
node.kubernetes.io/not-ready | NoExecute | DaemonSet Pods can be scheduled onto nodes that are not healthy or ready to accept Pods. Any DaemonSet Pods running on such nodes will not be evicted. |
node.kubernetes.io/unreachable | NoExecute | DaemonSet Pods can be scheduled onto nodes that are unreachable from the node controller. Any DaemonSet Pods running on such nodes will not be evicted. |
node.kubernetes.io/disk-pressure | NoSchedule | DaemonSet Pods can be scheduled onto nodes with disk pressure issues. |
node.kubernetes.io/memory-pressure | NoSchedule | DaemonSet Pods can be scheduled onto nodes with memory pressure issues. |
node.kubernetes.io/pid-pressure | NoSchedule | DaemonSet Pods can be scheduled onto nodes with process pressure issues. |
node.kubernetes.io/unschedulable | NoSchedule | DaemonSet Pods can be scheduled onto nodes that are unschedulable. |
node.kubernetes.io/network-unavailable | NoSchedule | Only added for DaemonSet Pods that request host networking, i.e., Pods having spec.hostNetwork: true . Such DaemonSet Pods can be scheduled onto nodes with unavailable network. |
You can add your own tolerations to the Pods of a DaemonSet as well, by defining these in the Pod template of the DaemonSet.
Because the DaemonSet controller sets the
node.kubernetes.io/unschedulable:NoSchedule
toleration automatically,
Kubernetes can run DaemonSet Pods on nodes that are marked as unschedulable.
If you use a DaemonSet to provide an important node-level function, such as cluster networking, it is helpful that Kubernetes places DaemonSet Pods on nodes before they are ready. For example, without that special toleration, you could end up in a deadlock situation where the node is not marked as ready because the network plugin is not running there, and at the same time the network plugin is not running on that node because the node is not yet ready.
Communicating with Daemon Pods
Some possible patterns for communicating with Pods in a DaemonSet are:
- Push: Pods in the DaemonSet are configured to send updates to another service, such as a stats database. They do not have clients.
- NodeIP and Known Port: Pods in the DaemonSet can use a
hostPort
, so that the pods are reachable via the node IPs. Clients know the list of node IPs somehow, and know the port by convention. - DNS: Create a headless service
with the same pod selector, and then discover DaemonSets using the
endpoints
resource or retrieve multiple A records from DNS. - Service: Create a service with the same Pod selector, and use the service to reach a daemon on a random node. (No way to reach specific node.)
Updating a DaemonSet
If node labels are changed, the DaemonSet will promptly add Pods to newly matching nodes and delete Pods from newly not-matching nodes.
You can modify the Pods that a DaemonSet creates. However, Pods do not allow all fields to be updated. Also, the DaemonSet controller will use the original template the next time a node (even with the same name) is created.
You can delete a DaemonSet. If you specify --cascade=orphan
with kubectl
, then the Pods
will be left on the nodes. If you subsequently create a new DaemonSet with the same selector,
the new DaemonSet adopts the existing Pods. If any Pods need replacing the DaemonSet replaces
them according to its updateStrategy
.
You can perform a rolling update on a DaemonSet.
Alternatives to DaemonSet
Init scripts
It is certainly possible to run daemon processes by directly starting them on a node (e.g. using
init
, upstartd
, or systemd
). This is perfectly fine. However, there are several advantages to
running such processes via a DaemonSet:
- Ability to monitor and manage logs for daemons in the same way as applications.
- Same config language and tools (e.g. Pod templates,
kubectl
) for daemons and applications. - Running daemons in containers with resource limits increases isolation between daemons from app containers. However, this can also be accomplished by running the daemons in a container but not in a Pod.
Bare Pods
It is possible to create Pods directly which specify a particular node to run on. However, a DaemonSet replaces Pods that are deleted or terminated for any reason, such as in the case of node failure or disruptive node maintenance, such as a kernel upgrade. For this reason, you should use a DaemonSet rather than creating individual Pods.
Static Pods
It is possible to create Pods by writing a file to a certain directory watched by Kubelet. These are called static pods. Unlike DaemonSet, static Pods cannot be managed with kubectl or other Kubernetes API clients. Static Pods do not depend on the apiserver, making them useful in cluster bootstrapping cases. Also, static Pods may be deprecated in the future.
Deployments
DaemonSets are similar to Deployments in that they both create Pods, and those Pods have processes which are not expected to terminate (e.g. web servers, storage servers).
Use a Deployment for stateless services, like frontends, where scaling up and down the number of replicas and rolling out updates are more important than controlling exactly which host the Pod runs on. Use a DaemonSet when it is important that a copy of a Pod always run on all or certain hosts, if the DaemonSet provides node-level functionality that allows other Pods to run correctly on that particular node.
For example, network plugins often include a component that runs as a DaemonSet. The DaemonSet component makes sure that the node where it's running has working cluster networking.
What's next
- Learn about Pods.
- Learn about static Pods, which are useful for running Kubernetes control plane components.
- Find out how to use DaemonSets
- Perform a rolling update on a DaemonSet
- Perform a rollback on a DaemonSet (for example, if a roll out didn't work how you expected).
- Understand how Kubernetes assigns Pods to Nodes.
- Learn about device plugins and add ons, which often run as DaemonSets.
DaemonSet
is a top-level resource in the Kubernetes REST API. Read the DaemonSet object definition to understand the API for daemon sets.
4.2.5 - Jobs
A Job creates one or more Pods and will continue to retry execution of the Pods until a specified number of them successfully terminate. As pods successfully complete, the Job tracks the successful completions. When a specified number of successful completions is reached, the task (ie, Job) is complete. Deleting a Job will clean up the Pods it created. Suspending a Job will delete its active Pods until the Job is resumed again.
A simple case is to create one Job object in order to reliably run one Pod to completion. The Job object will start a new Pod if the first Pod fails or is deleted (for example due to a node hardware failure or a node reboot).
You can also use a Job to run multiple Pods in parallel.
If you want to run a Job (either a single task, or several in parallel) on a schedule, see CronJob.
Running an example Job
Here is an example Job config. It computes π to 2000 places and prints it out. It takes around 10s to complete.
apiVersion: batch/v1
kind: Job
metadata:
name: pi
spec:
template:
spec:
containers:
- name: pi
image: perl:5.34.0
command: ["perl", "-Mbignum=bpi", "-wle", "print bpi(2000)"]
restartPolicy: Never
backoffLimit: 4
You can run the example with this command:
kubectl apply -f https://kubernetes.io/examples/controllers/job.yaml
The output is similar to this:
job.batch/pi created
Check on the status of the Job with kubectl
:
Name: pi
Namespace: default
Selector: batch.kubernetes.io/controller-uid=c9948307-e56d-4b5d-8302-ae2d7b7da67c
Labels: batch.kubernetes.io/controller-uid=c9948307-e56d-4b5d-8302-ae2d7b7da67c
batch.kubernetes.io/job-name=pi
...
Annotations: batch.kubernetes.io/job-tracking: ""
Parallelism: 1
Completions: 1
Start Time: Mon, 02 Dec 2019 15:20:11 +0200
Completed At: Mon, 02 Dec 2019 15:21:16 +0200
Duration: 65s
Pods Statuses: 0 Running / 1 Succeeded / 0 Failed
Pod Template:
Labels: batch.kubernetes.io/controller-uid=c9948307-e56d-4b5d-8302-ae2d7b7da67c
batch.kubernetes.io/job-name=pi
Containers:
pi:
Image: perl:5.34.0
Port: <none>
Host Port: <none>
Command:
perl
-Mbignum=bpi
-wle
print bpi(2000)
Environment: <none>
Mounts: <none>
Volumes: <none>
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal SuccessfulCreate 21s job-controller Created pod: pi-xf9p4
Normal Completed 18s job-controller Job completed
apiVersion: batch/v1
kind: Job
metadata:
annotations: batch.kubernetes.io/job-tracking: ""
...
creationTimestamp: "2022-11-10T17:53:53Z"
generation: 1
labels:
batch.kubernetes.io/controller-uid: 863452e6-270d-420e-9b94-53a54146c223
batch.kubernetes.io/job-name: pi
name: pi
namespace: default
resourceVersion: "4751"
uid: 204fb678-040b-497f-9266-35ffa8716d14
spec:
backoffLimit: 4
completionMode: NonIndexed
completions: 1
parallelism: 1
selector:
matchLabels:
batch.kubernetes.io/controller-uid: 863452e6-270d-420e-9b94-53a54146c223
suspend: false
template:
metadata:
creationTimestamp: null
labels:
batch.kubernetes.io/controller-uid: 863452e6-270d-420e-9b94-53a54146c223
batch.kubernetes.io/job-name: pi
spec:
containers:
- command:
- perl
- -Mbignum=bpi
- -wle
- print bpi(2000)
image: perl:5.34.0
imagePullPolicy: IfNotPresent
name: pi
resources: {}
terminationMessagePath: /dev/termination-log
terminationMessagePolicy: File
dnsPolicy: ClusterFirst
restartPolicy: Never
schedulerName: default-scheduler
securityContext: {}
terminationGracePeriodSeconds: 30
status:
active: 1
ready: 0
startTime: "2022-11-10T17:53:57Z"
uncountedTerminatedPods: {}
To view completed Pods of a Job, use kubectl get pods
.
To list all the Pods that belong to a Job in a machine readable form, you can use a command like this:
pods=$(kubectl get pods --selector=batch.kubernetes.io/job-name=pi --output=jsonpath='{.items[*].metadata.name}')
echo $pods
The output is similar to this:
pi-5rwd7
Here, the selector is the same as the selector for the Job. The --output=jsonpath
option specifies an expression
with the name from each Pod in the returned list.
View the standard output of one of the pods:
kubectl logs $pods
Another way to view the logs of a Job:
kubectl logs jobs/pi
The output is similar to this:
3.1415926535897932384626433832795028841971693993751058209749445923078164062862089986280348253421170679821480865132823066470938446095505822317253594081284811174502841027019385211055596446229489549303819644288109756659334461284756482337867831652712019091456485669234603486104543266482133936072602491412737245870066063155881748815209209628292540917153643678925903600113305305488204665213841469519415116094330572703657595919530921861173819326117931051185480744623799627495673518857527248912279381830119491298336733624406566430860213949463952247371907021798609437027705392171762931767523846748184676694051320005681271452635608277857713427577896091736371787214684409012249534301465495853710507922796892589235420199561121290219608640344181598136297747713099605187072113499999983729780499510597317328160963185950244594553469083026425223082533446850352619311881710100031378387528865875332083814206171776691473035982534904287554687311595628638823537875937519577818577805321712268066130019278766111959092164201989380952572010654858632788659361533818279682303019520353018529689957736225994138912497217752834791315155748572424541506959508295331168617278558890750983817546374649393192550604009277016711390098488240128583616035637076601047101819429555961989467678374494482553797747268471040475346462080466842590694912933136770289891521047521620569660240580381501935112533824300355876402474964732639141992726042699227967823547816360093417216412199245863150302861829745557067498385054945885869269956909272107975093029553211653449872027559602364806654991198818347977535663698074265425278625518184175746728909777727938000816470600161452491921732172147723501414419735685481613611573525521334757418494684385233239073941433345477624168625189835694855620992192221842725502542568876717904946016534668049886272327917860857843838279679766814541009538837863609506800642251252051173929848960841284886269456042419652850222106611863067442786220391949450471237137869609563643719172874677646575739624138908658326459958133904780275901
Writing a Job spec
As with all other Kubernetes config, a Job needs apiVersion
, kind
, and metadata
fields.
When the control plane creates new Pods for a Job, the .metadata.name
of the
Job is part of the basis for naming those Pods. The name of a Job must be a valid
DNS subdomain
value, but this can produce unexpected results for the Pod hostnames. For best compatibility,
the name should follow the more restrictive rules for a
DNS label.
Even when the name is a DNS subdomain, the name must be no longer than 63
characters.
A Job also needs a .spec
section.
Job Labels
Job labels will have batch.kubernetes.io/
prefix for job-name
and controller-uid
.
Pod Template
The .spec.template
is the only required field of the .spec
.
The .spec.template
is a pod template.
It has exactly the same schema as a Pod,
except it is nested and does not have an apiVersion
or kind
.
In addition to required fields for a Pod, a pod template in a Job must specify appropriate labels (see pod selector) and an appropriate restart policy.
Only a RestartPolicy
equal to Never
or OnFailure
is allowed.
Pod selector
The .spec.selector
field is optional. In almost all cases you should not specify it.
See section specifying your own pod selector.
Parallel execution for Jobs
There are three main types of task suitable to run as a Job:
- Non-parallel Jobs
- normally, only one Pod is started, unless the Pod fails.
- the Job is complete as soon as its Pod terminates successfully.
- Parallel Jobs with a fixed completion count:
- specify a non-zero positive value for
.spec.completions
. - the Job represents the overall task, and is complete when there are
.spec.completions
successful Pods. - when using
.spec.completionMode="Indexed"
, each Pod gets a different index in the range 0 to.spec.completions-1
.
- specify a non-zero positive value for
- Parallel Jobs with a work queue:
- do not specify
.spec.completions
, default to.spec.parallelism
. - the Pods must coordinate amongst themselves or an external service to determine what each should work on. For example, a Pod might fetch a batch of up to N items from the work queue.
- each Pod is independently capable of determining whether or not all its peers are done, and thus that the entire Job is done.
- when any Pod from the Job terminates with success, no new Pods are created.
- once at least one Pod has terminated with success and all Pods are terminated, then the Job is completed with success.
- once any Pod has exited with success, no other Pod should still be doing any work for this task or writing any output. They should all be in the process of exiting.
- do not specify
For a non-parallel Job, you can leave both .spec.completions
and .spec.parallelism
unset.
When both are unset, both are defaulted to 1.
For a fixed completion count Job, you should set .spec.completions
to the number of completions needed.
You can set .spec.parallelism
, or leave it unset and it will default to 1.
For a work queue Job, you must leave .spec.completions
unset, and set .spec.parallelism
to
a non-negative integer.
For more information about how to make use of the different types of job, see the job patterns section.
Controlling parallelism
The requested parallelism (.spec.parallelism
) can be set to any non-negative value.
If it is unspecified, it defaults to 1.
If it is specified as 0, then the Job is effectively paused until it is increased.
Actual parallelism (number of pods running at any instant) may be more or less than requested parallelism, for a variety of reasons:
- For fixed completion count Jobs, the actual number of pods running in parallel will not exceed the number of
remaining completions. Higher values of
.spec.parallelism
are effectively ignored. - For work queue Jobs, no new Pods are started after any Pod has succeeded -- remaining Pods are allowed to complete, however.
- If the Job Controller has not had time to react.
- If the Job controller failed to create Pods for any reason (lack of
ResourceQuota
, lack of permission, etc.), then there may be fewer pods than requested. - The Job controller may throttle new Pod creation due to excessive previous pod failures in the same Job.
- When a Pod is gracefully shut down, it takes time to stop.
Completion mode
Kubernetes v1.24 [stable]
Jobs with fixed completion count - that is, jobs that have non null
.spec.completions
- can have a completion mode that is specified in .spec.completionMode
:
NonIndexed
(default): the Job is considered complete when there have been.spec.completions
successfully completed Pods. In other words, each Pod completion is homologous to each other. Note that Jobs that have null.spec.completions
are implicitlyNonIndexed
.Indexed
: the Pods of a Job get an associated completion index from 0 to.spec.completions-1
. The index is available through four mechanisms:- The Pod annotation
batch.kubernetes.io/job-completion-index
. - The Pod label
batch.kubernetes.io/job-completion-index
(for v1.28 and later). Note the feature gatePodIndexLabel
must be enabled to use this label, and it is enabled by default. - As part of the Pod hostname, following the pattern
$(job-name)-$(index)
. When you use an Indexed Job in combination with a Service, Pods within the Job can use the deterministic hostnames to address each other via DNS. For more information about how to configure this, see Job with Pod-to-Pod Communication. - From the containerized task, in the environment variable
JOB_COMPLETION_INDEX
.
The Job is considered complete when there is one successfully completed Pod for each index. For more information about how to use this mode, see Indexed Job for Parallel Processing with Static Work Assignment.
- The Pod annotation
Note:
Although rare, more than one Pod could be started for the same index (due to various reasons such as node failures, kubelet restarts, or Pod evictions). In this case, only the first Pod that completes successfully will count towards the completion count and update the status of the Job. The other Pods that are running or completed for the same index will be deleted by the Job controller once they are detected.Handling Pod and container failures
A container in a Pod may fail for a number of reasons, such as because the process in it exited with
a non-zero exit code, or the container was killed for exceeding a memory limit, etc. If this
happens, and the .spec.template.spec.restartPolicy = "OnFailure"
, then the Pod stays
on the node, but the container is re-run. Therefore, your program needs to handle the case when it is
restarted locally, or else specify .spec.template.spec.restartPolicy = "Never"
.
See pod lifecycle for more information on restartPolicy
.
An entire Pod can also fail, for a number of reasons, such as when the pod is kicked off the node
(node is upgraded, rebooted, deleted, etc.), or if a container of the Pod fails and the
.spec.template.spec.restartPolicy = "Never"
. When a Pod fails, then the Job controller
starts a new Pod. This means that your application needs to handle the case when it is restarted in a new
pod. In particular, it needs to handle temporary files, locks, incomplete output and the like
caused by previous runs.
By default, each pod failure is counted towards the .spec.backoffLimit
limit,
see pod backoff failure policy. However, you can
customize handling of pod failures by setting the Job's pod failure policy.
Additionally, you can choose to count the pod failures independently for each
index of an Indexed Job by setting the .spec.backoffLimitPerIndex
field
(for more information, see backoff limit per index).
Note that even if you specify .spec.parallelism = 1
and .spec.completions = 1
and
.spec.template.spec.restartPolicy = "Never"
, the same program may
sometimes be started twice.
If you do specify .spec.parallelism
and .spec.completions
both greater than 1, then there may be
multiple pods running at once. Therefore, your pods must also be tolerant of concurrency.
If you specify the .spec.podFailurePolicy
field, the Job controller does not consider a terminating
Pod (a pod that has a .metadata.deletionTimestamp
field set) as a failure until that Pod is
terminal (its .status.phase
is Failed
or Succeeded
). However, the Job controller
creates a replacement Pod as soon as the termination becomes apparent. Once the
pod terminates, the Job controller evaluates .backoffLimit
and .podFailurePolicy
for the relevant Job, taking this now-terminated Pod into consideration.
If either of these requirements is not satisfied, the Job controller counts
a terminating Pod as an immediate failure, even if that Pod later terminates
with phase: "Succeeded"
.
Pod backoff failure policy
There are situations where you want to fail a Job after some amount of retries
due to a logical error in configuration etc.
To do so, set .spec.backoffLimit
to specify the number of retries before
considering a Job as failed. The back-off limit is set by default to 6. Failed
Pods associated with the Job are recreated by the Job controller with an
exponential back-off delay (10s, 20s, 40s ...) capped at six minutes.
The number of retries is calculated in two ways:
- The number of Pods with
.status.phase = "Failed"
. - When using
restartPolicy = "OnFailure"
, the number of retries in all the containers of Pods with.status.phase
equal toPending
orRunning
.
If either of the calculations reaches the .spec.backoffLimit
, the Job is
considered failed.
Note:
If your job hasrestartPolicy = "OnFailure"
, keep in mind that your Pod running the Job
will be terminated once the job backoff limit has been reached. This can make debugging
the Job's executable more difficult. We suggest setting
restartPolicy = "Never"
when debugging the Job or using a logging system to ensure output
from failed Jobs is not lost inadvertently.Backoff limit per index
Kubernetes v1.29 [beta]
Note:
You can only configure the backoff limit per index for an Indexed Job, if you have theJobBackoffLimitPerIndex
feature gate
enabled in your cluster.When you run an indexed Job, you can choose to handle retries
for pod failures independently for each index. To do so, set the
.spec.backoffLimitPerIndex
to specify the maximal number of pod failures
per index.
When the per-index backoff limit is exceeded for an index, Kubernetes considers the index as failed and adds it to the
.status.failedIndexes
field. The succeeded indexes, those with a successfully
executed pods, are recorded in the .status.completedIndexes
field, regardless of whether you set
the backoffLimitPerIndex
field.
Note that a failing index does not interrupt execution of other indexes. Once all indexes finish for a Job where you specified a backoff limit per index, if at least one of those indexes did fail, the Job controller marks the overall Job as failed, by setting the Failed condition in the status. The Job gets marked as failed even if some, potentially nearly all, of the indexes were processed successfully.
You can additionally limit the maximal number of indexes marked failed by
setting the .spec.maxFailedIndexes
field.
When the number of failed indexes exceeds the maxFailedIndexes
field, the
Job controller triggers termination of all remaining running Pods for that Job.
Once all pods are terminated, the entire Job is marked failed by the Job
controller, by setting the Failed condition in the Job status.
Here is an example manifest for a Job that defines a backoffLimitPerIndex
:
apiVersion: batch/v1
kind: Job
metadata:
name: job-backoff-limit-per-index-example
spec:
completions: 10
parallelism: 3
completionMode: Indexed # required for the feature
backoffLimitPerIndex: 1 # maximal number of failures per index
maxFailedIndexes: 5 # maximal number of failed indexes before terminating the Job execution
template:
spec:
restartPolicy: Never # required for the feature
containers:
- name: example
image: python
command: # The jobs fails as there is at least one failed index
# (all even indexes fail in here), yet all indexes
# are executed as maxFailedIndexes is not exceeded.
- python3
- -c
- |
import os, sys
print("Hello world")
if int(os.environ.get("JOB_COMPLETION_INDEX")) % 2 == 0:
sys.exit(1)
In the example above, the Job controller allows for one restart for each of the indexes. When the total number of failed indexes exceeds 5, then the entire Job is terminated.
Once the job is finished, the Job status looks as follows:
kubectl get -o yaml job job-backoff-limit-per-index-example
status:
completedIndexes: 1,3,5,7,9
failedIndexes: 0,2,4,6,8
succeeded: 5 # 1 succeeded pod for each of 5 succeeded indexes
failed: 10 # 2 failed pods (1 retry) for each of 5 failed indexes
conditions:
- message: Job has failed indexes
reason: FailedIndexes
status: "True"
type: FailureTarget
- message: Job has failed indexes
reason: FailedIndexes
status: "True"
type: Failed
The Job controller adds the FailureTarget
Job condition to trigger
Job termination and cleanup. When all of the
Job Pods are terminated, the Job controller adds the Failed
condition
with the same values for reason
and message
as the FailureTarget
Job
condition. For details, see Termination of Job Pods.
Additionally, you may want to use the per-index backoff along with a
pod failure policy. When using
per-index backoff, there is a new FailIndex
action available which allows you to
avoid unnecessary retries within an index.
Pod failure policy
Kubernetes v1.31 [stable]
(enabled by default: true)A Pod failure policy, defined with the .spec.podFailurePolicy
field, enables
your cluster to handle Pod failures based on the container exit codes and the
Pod conditions.
In some situations, you may want to have a better control when handling Pod
failures than the control provided by the Pod backoff failure policy,
which is based on the Job's .spec.backoffLimit
. These are some examples of use cases:
- To optimize costs of running workloads by avoiding unnecessary Pod restarts, you can terminate a Job as soon as one of its Pods fails with an exit code indicating a software bug.
- To guarantee that your Job finishes even if there are disruptions, you can
ignore Pod failures caused by disruptions (such as preemption,
API-initiated eviction
or taint-based eviction) so
that they don't count towards the
.spec.backoffLimit
limit of retries.
You can configure a Pod failure policy, in the .spec.podFailurePolicy
field,
to meet the above use cases. This policy can handle Pod failures based on the
container exit codes and the Pod conditions.
Here is a manifest for a Job that defines a podFailurePolicy
:
apiVersion: batch/v1
kind: Job
metadata:
name: job-pod-failure-policy-example
spec:
completions: 12
parallelism: 3
template:
spec:
restartPolicy: Never
containers:
- name: main
image: docker.io/library/bash:5
command: ["bash"] # example command simulating a bug which triggers the FailJob action
args:
- -c
- echo "Hello world!" && sleep 5 && exit 42
backoffLimit: 6
podFailurePolicy:
rules:
- action: FailJob
onExitCodes:
containerName: main # optional
operator: In # one of: In, NotIn
values: [42]
- action: Ignore # one of: Ignore, FailJob, Count
onPodConditions:
- type: DisruptionTarget # indicates Pod disruption
In the example above, the first rule of the Pod failure policy specifies that
the Job should be marked failed if the main
container fails with the 42 exit
code. The following are the rules for the main
container specifically:
- an exit code of 0 means that the container succeeded
- an exit code of 42 means that the entire Job failed
- any other exit code represents that the container failed, and hence the entire
Pod. The Pod will be re-created if the total number of restarts is
below
backoffLimit
. If thebackoffLimit
is reached the entire Job failed.
Note:
Because the Pod template specifies arestartPolicy: Never
,
the kubelet does not restart the main
container in that particular Pod.The second rule of the Pod failure policy, specifying the Ignore
action for
failed Pods with condition DisruptionTarget
excludes Pod disruptions from
being counted towards the .spec.backoffLimit
limit of retries.
Note:
If the Job failed, either by the Pod failure policy or Pod backoff failure policy, and the Job is running multiple Pods, Kubernetes terminates all the Pods in that Job that are still Pending or Running.These are some requirements and semantics of the API:
- if you want to use a
.spec.podFailurePolicy
field for a Job, you must also define that Job's pod template with.spec.restartPolicy
set toNever
. - the Pod failure policy rules you specify under
spec.podFailurePolicy.rules
are evaluated in order. Once a rule matches a Pod failure, the remaining rules are ignored. When no rule matches the Pod failure, the default handling applies. - you may want to restrict a rule to a specific container by specifying its name
in
spec.podFailurePolicy.rules[*].onExitCodes.containerName
. When not specified the rule applies to all containers. When specified, it should match one the container orinitContainer
names in the Pod template. - you may specify the action taken when a Pod failure policy is matched by
spec.podFailurePolicy.rules[*].action
. Possible values are:FailJob
: use to indicate that the Pod's job should be marked as Failed and all running Pods should be terminated.Ignore
: use to indicate that the counter towards the.spec.backoffLimit
should not be incremented and a replacement Pod should be created.Count
: use to indicate that the Pod should be handled in the default way. The counter towards the.spec.backoffLimit
should be incremented.FailIndex
: use this action along with backoff limit per index to avoid unnecessary retries within the index of a failed pod.
Note:
When you use apodFailurePolicy
, the job controller only matches Pods in the
Failed
phase. Pods with a deletion timestamp that are not in a terminal phase
(Failed
or Succeeded
) are considered still terminating. This implies that
terminating pods retain a tracking finalizer
until they reach a terminal phase.
Since Kubernetes 1.27, Kubelet transitions deleted pods to a terminal phase
(see: Pod Phase). This
ensures that deleted pods have their finalizers removed by the Job controller.Note:
Starting with Kubernetes v1.28, when Pod failure policy is used, the Job controller recreates terminating Pods only once these Pods reach the terminalFailed
phase. This behavior is similar
to podReplacementPolicy: Failed
. For more information, see Pod replacement policy.When you use the podFailurePolicy
, and the Job fails due to the pod
matching the rule with the FailJob
action, then the Job controller triggers
the Job termination process by adding the FailureTarget
condition.
For more details, see Job termination and cleanup.
Success policy
Kubernetes v1.31 [beta]
(enabled by default: true)Note:
You can only configure a success policy for an Indexed Job if you have theJobSuccessPolicy
feature gate
enabled in your cluster.When creating an Indexed Job, you can define when a Job can be declared as succeeded using a .spec.successPolicy
,
based on the pods that succeeded.
By default, a Job succeeds when the number of succeeded Pods equals .spec.completions
.
These are some situations where you might want additional control for declaring a Job succeeded:
- When running simulations with different parameters, you might not need all the simulations to succeed for the overall Job to be successful.
- When following a leader-worker pattern, only the success of the leader determines the success or failure of a Job. Examples of this are frameworks like MPI and PyTorch etc.
You can configure a success policy, in the .spec.successPolicy
field,
to meet the above use cases. This policy can handle Job success based on the
succeeded pods. After the Job meets the success policy, the job controller terminates the lingering Pods.
A success policy is defined by rules. Each rule can take one of the following forms:
- When you specify the
succeededIndexes
only, once all indexes specified in thesucceededIndexes
succeed, the job controller marks the Job as succeeded. ThesucceededIndexes
must be a list of intervals between 0 and.spec.completions-1
. - When you specify the
succeededCount
only, once the number of succeeded indexes reaches thesucceededCount
, the job controller marks the Job as succeeded. - When you specify both
succeededIndexes
andsucceededCount
, once the number of succeeded indexes from the subset of indexes specified in thesucceededIndexes
reaches thesucceededCount
, the job controller marks the Job as succeeded.
Note that when you specify multiple rules in the .spec.successPolicy.rules
,
the job controller evaluates the rules in order. Once the Job meets a rule, the job controller ignores remaining rules.
Here is a manifest for a Job with successPolicy
:
apiVersion: batch/v1
kind: Job
metadata:
name: job-success
spec:
parallelism: 10
completions: 10
completionMode: Indexed # Required for the success policy
successPolicy:
rules:
- succeededIndexes: 0,2-3
succeededCount: 1
template:
spec:
containers:
- name: main
image: python
command: # Provided that at least one of the Pods with 0, 2, and 3 indexes has succeeded,
# the overall Job is a success.
- python3
- -c
- |
import os, sys
if os.environ.get("JOB_COMPLETION_INDEX") == "2":
sys.exit(0)
else:
sys.exit(1)
restartPolicy: Never
In the example above, both succeededIndexes
and succeededCount
have been specified.
Therefore, the job controller will mark the Job as succeeded and terminate the lingering Pods
when either of the specified indexes, 0, 2, or 3, succeed.
The Job that meets the success policy gets the SuccessCriteriaMet
condition with a SuccessPolicy
reason.
After the removal of the lingering Pods is issued, the Job gets the Complete
condition.
Note that the succeededIndexes
is represented as intervals separated by a hyphen.
The number are listed in represented by the first and last element of the series, separated by a hyphen.
Note:
When you specify both a success policy and some terminating policies such as.spec.backoffLimit
and .spec.podFailurePolicy
,
once the Job meets either policy, the job controller respects the terminating policy and ignores the success policy.Job termination and cleanup
When a Job completes, no more Pods are created, but the Pods are usually not deleted either.
Keeping them around allows you to still view the logs of completed pods to check for errors, warnings, or other diagnostic output.
The job object also remains after it is completed so that you can view its status. It is up to the user to delete
old jobs after noting their status. Delete the job with kubectl
(e.g. kubectl delete jobs/pi
or kubectl delete -f ./job.yaml
).
When you delete the job using kubectl
, all the pods it created are deleted too.
By default, a Job will run uninterrupted unless a Pod fails (restartPolicy=Never
)
or a Container exits in error (restartPolicy=OnFailure
), at which point the Job defers to the
.spec.backoffLimit
described above. Once .spec.backoffLimit
has been reached the Job will
be marked as failed and any running Pods will be terminated.
Another way to terminate a Job is by setting an active deadline.
Do this by setting the .spec.activeDeadlineSeconds
field of the Job to a number of seconds.
The activeDeadlineSeconds
applies to the duration of the job, no matter how many Pods are created.
Once a Job reaches activeDeadlineSeconds
, all of its running Pods are terminated and the Job status
will become type: Failed
with reason: DeadlineExceeded
.
Note that a Job's .spec.activeDeadlineSeconds
takes precedence over its .spec.backoffLimit
.
Therefore, a Job that is retrying one or more failed Pods will not deploy additional Pods once
it reaches the time limit specified by activeDeadlineSeconds
, even if the backoffLimit
is not yet reached.
Example:
apiVersion: batch/v1
kind: Job
metadata:
name: pi-with-timeout
spec:
backoffLimit: 5
activeDeadlineSeconds: 100
template:
spec:
containers:
- name: pi
image: perl:5.34.0
command: ["perl", "-Mbignum=bpi", "-wle", "print bpi(2000)"]
restartPolicy: Never
Note that both the Job spec and the Pod template spec
within the Job have an activeDeadlineSeconds
field. Ensure that you set this field at the proper level.
Keep in mind that the restartPolicy
applies to the Pod, and not to the Job itself:
there is no automatic Job restart once the Job status is type: Failed
.
That is, the Job termination mechanisms activated with .spec.activeDeadlineSeconds
and .spec.backoffLimit
result in a permanent Job failure that requires manual intervention to resolve.
Terminal Job conditions
A Job has two possible terminal states, each of which has a corresponding Job condition:
- Succeeded: Job condition
Complete
- Failed: Job condition
Failed
Jobs fail for the following reasons:
- The number of Pod failures exceeded the specified
.spec.backoffLimit
in the Job specification. For details, see Pod backoff failure policy. - The Job runtime exceeded the specified
.spec.activeDeadlineSeconds
- An indexed Job that used
.spec.backoffLimitPerIndex
has failed indexes. For details, see Backoff limit per index. - The number of failed indexes in the Job exceeded the specified
spec.maxFailedIndexes
. For details, see Backoff limit per index - A failed Pod matches a rule in
.spec.podFailurePolicy
that has theFailJob
action. For details about how Pod failure policy rules might affect failure evaluation, see Pod failure policy.
Jobs succeed for the following reasons:
- The number of succeeded Pods reached the specified
.spec.completions
- The criteria specified in
.spec.successPolicy
are met. For details, see Success policy.
In Kubernetes v1.31 and later the Job controller delays the addition of the
terminal conditions,Failed
or Complete
, until all of the Job Pods are terminated.
In Kubernetes v1.30 and earlier, the Job controller added the Complete
or the
Failed
Job terminal conditions as soon as the Job termination process was
triggered and all Pod finalizers were removed. However, some Pods would still
be running or terminating at the moment that the terminal condition was added.
In Kubernetes v1.31 and later, the controller only adds the Job terminal conditions
after all of the Pods are terminated. You can enable this behavior by using the
JobManagedBy
or the JobPodReplacementPolicy
(enabled by default)
feature gates.
Termination of Job pods
The Job controller adds the FailureTarget
condition or the SuccessCriteriaMet
condition to the Job to trigger Pod termination after a Job meets either the
success or failure criteria.
Factors like terminationGracePeriodSeconds
might increase the amount of time
from the moment that the Job controller adds the FailureTarget
condition or the
SuccessCriteriaMet
condition to the moment that all of the Job Pods terminate
and the Job controller adds a terminal condition
(Failed
or Complete
).
You can use the FailureTarget
or the SuccessCriteriaMet
condition to evaluate
whether the Job has failed or succeeded without having to wait for the controller
to add a terminal condition.
For example, you might want to decide when to create a replacement Job
that replaces a failed Job. If you replace the failed Job when the FailureTarget
condition appears, your replacement Job runs sooner, but could result in Pods
from the failed and the replacement Job running at the same time, using
extra compute resources.
Alternatively, if your cluster has limited resource capacity, you could choose to
wait until the Failed
condition appears on the Job, which would delay your
replacement Job but would ensure that you conserve resources by waiting
until all of the failed Pods are removed.
Clean up finished jobs automatically
Finished Jobs are usually no longer needed in the system. Keeping them around in the system will put pressure on the API server. If the Jobs are managed directly by a higher level controller, such as CronJobs, the Jobs can be cleaned up by CronJobs based on the specified capacity-based cleanup policy.
TTL mechanism for finished Jobs
Kubernetes v1.23 [stable]
Another way to clean up finished Jobs (either Complete
or Failed
)
automatically is to use a TTL mechanism provided by a
TTL controller for
finished resources, by specifying the .spec.ttlSecondsAfterFinished
field of
the Job.
When the TTL controller cleans up the Job, it will delete the Job cascadingly, i.e. delete its dependent objects, such as Pods, together with the Job. Note that when the Job is deleted, its lifecycle guarantees, such as finalizers, will be honored.
For example:
apiVersion: batch/v1
kind: Job
metadata:
name: pi-with-ttl
spec:
ttlSecondsAfterFinished: 100
template:
spec:
containers:
- name: pi
image: perl:5.34.0
command: ["perl", "-Mbignum=bpi", "-wle", "print bpi(2000)"]
restartPolicy: Never
The Job pi-with-ttl
will be eligible to be automatically deleted, 100
seconds after it finishes.
If the field is set to 0
, the Job will be eligible to be automatically deleted
immediately after it finishes. If the field is unset, this Job won't be cleaned
up by the TTL controller after it finishes.
Note:
It is recommended to set ttlSecondsAfterFinished
field because unmanaged jobs
(Jobs that you created directly, and not indirectly through other workload APIs
such as CronJob) have a default deletion
policy of orphanDependents
causing Pods created by an unmanaged Job to be left around
after that Job is fully deleted.
Even though the control plane eventually
garbage collects
the Pods from a deleted Job after they either fail or complete, sometimes those
lingering pods may cause cluster performance degradation or in worst case cause the
cluster to go offline due to this degradation.
You can use LimitRanges and ResourceQuotas to place a cap on the amount of resources that a particular namespace can consume.
Job patterns
The Job object can be used to process a set of independent but related work items. These might be emails to be sent, frames to be rendered, files to be transcoded, ranges of keys in a NoSQL database to scan, and so on.
In a complex system, there may be multiple different sets of work items. Here we are just considering one set of work items that the user wants to manage together — a batch job.
There are several different patterns for parallel computation, each with strengths and weaknesses. The tradeoffs are:
- One Job object for each work item, versus a single Job object for all work items. One Job per work item creates some overhead for the user and for the system to manage large numbers of Job objects. A single Job for all work items is better for large numbers of items.
- Number of Pods created equals number of work items, versus each Pod can process multiple work items. When the number of Pods equals the number of work items, the Pods typically requires less modification to existing code and containers. Having each Pod process multiple work items is better for large numbers of items.
- Several approaches use a work queue. This requires running a queue service, and modifications to the existing program or container to make it use the work queue. Other approaches are easier to adapt to an existing containerised application.
- When the Job is associated with a headless Service, you can enable the Pods within a Job to communicate with each other to collaborate in a computation.
The tradeoffs are summarized here, with columns 2 to 4 corresponding to the above tradeoffs. The pattern names are also links to examples and more detailed description.
Pattern | Single Job object | Fewer pods than work items? | Use app unmodified? |
---|---|---|---|
Queue with Pod Per Work Item | ✓ | sometimes | |
Queue with Variable Pod Count | ✓ | ✓ | |
Indexed Job with Static Work Assignment | ✓ | ✓ | |
Job with Pod-to-Pod Communication | ✓ | sometimes | sometimes |
Job Template Expansion | ✓ |
When you specify completions with .spec.completions
, each Pod created by the Job controller
has an identical spec
.
This means that all pods for a task will have the same command line and the same
image, the same volumes, and (almost) the same environment variables. These patterns
are different ways to arrange for pods to work on different things.
This table shows the required settings for .spec.parallelism
and .spec.completions
for each of the patterns.
Here, W
is the number of work items.
Pattern | .spec.completions | .spec.parallelism |
---|---|---|
Queue with Pod Per Work Item | W | any |
Queue with Variable Pod Count | null | any |
Indexed Job with Static Work Assignment | W | any |
Job with Pod-to-Pod Communication | W | W |
Job Template Expansion | 1 | should be 1 |
Advanced usage
Suspending a Job
Kubernetes v1.24 [stable]
When a Job is created, the Job controller will immediately begin creating Pods to satisfy the Job's requirements and will continue to do so until the Job is complete. However, you may want to temporarily suspend a Job's execution and resume it later, or start Jobs in suspended state and have a custom controller decide later when to start them.
To suspend a Job, you can update the .spec.suspend
field of
the Job to true; later, when you want to resume it again, update it to false.
Creating a Job with .spec.suspend
set to true will create it in the suspended
state.
When a Job is resumed from suspension, its .status.startTime
field will be
reset to the current time. This means that the .spec.activeDeadlineSeconds
timer will be stopped and reset when a Job is suspended and resumed.
When you suspend a Job, any running Pods that don't have a status of Completed
will be terminated
with a SIGTERM signal. The Pod's graceful termination period will be honored and
your Pod must handle this signal in this period. This may involve saving
progress for later or undoing changes. Pods terminated this way will not count
towards the Job's completions
count.
An example Job definition in the suspended state can be like so:
kubectl get job myjob -o yaml
apiVersion: batch/v1
kind: Job
metadata:
name: myjob
spec:
suspend: true
parallelism: 1
completions: 5
template:
spec:
...
You can also toggle Job suspension by patching the Job using the command line.
Suspend an active Job:
kubectl patch job/myjob --type=strategic --patch '{"spec":{"suspend":true}}'
Resume a suspended Job:
kubectl patch job/myjob --type=strategic --patch '{"spec":{"suspend":false}}'
The Job's status can be used to determine if a Job is suspended or has been suspended in the past:
kubectl get jobs/myjob -o yaml
apiVersion: batch/v1
kind: Job
# .metadata and .spec omitted
status:
conditions:
- lastProbeTime: "2021-02-05T13:14:33Z"
lastTransitionTime: "2021-02-05T13:14:33Z"
status: "True"
type: Suspended
startTime: "2021-02-05T13:13:48Z"
The Job condition of type "Suspended" with status "True" means the Job is
suspended; the lastTransitionTime
field can be used to determine how long the
Job has been suspended for. If the status of that condition is "False", then the
Job was previously suspended and is now running. If such a condition does not
exist in the Job's status, the Job has never been stopped.
Events are also created when the Job is suspended and resumed:
kubectl describe jobs/myjob
Name: myjob
...
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal SuccessfulCreate 12m job-controller Created pod: myjob-hlrpl
Normal SuccessfulDelete 11m job-controller Deleted pod: myjob-hlrpl
Normal Suspended 11m job-controller Job suspended
Normal SuccessfulCreate 3s job-controller Created pod: myjob-jvb44
Normal Resumed 3s job-controller Job resumed
The last four events, particularly the "Suspended" and "Resumed" events, are
directly a result of toggling the .spec.suspend
field. In the time between
these two events, we see that no Pods were created, but Pod creation restarted
as soon as the Job was resumed.
Mutable Scheduling Directives
Kubernetes v1.27 [stable]
In most cases, a parallel job will want the pods to run with constraints, like all in the same zone, or all either on GPU model x or y but not a mix of both.
The suspend field is the first step towards achieving those semantics. Suspend allows a custom queue controller to decide when a job should start; However, once a job is unsuspended, a custom queue controller has no influence on where the pods of a job will actually land.
This feature allows updating a Job's scheduling directives before it starts, which gives custom queue controllers the ability to influence pod placement while at the same time offloading actual pod-to-node assignment to kube-scheduler. This is allowed only for suspended Jobs that have never been unsuspended before.
The fields in a Job's pod template that can be updated are node affinity, node selector, tolerations, labels, annotations and scheduling gates.
Specifying your own Pod selector
Normally, when you create a Job object, you do not specify .spec.selector
.
The system defaulting logic adds this field when the Job is created.
It picks a selector value that will not overlap with any other jobs.
However, in some cases, you might need to override this automatically set selector.
To do this, you can specify the .spec.selector
of the Job.
Be very careful when doing this. If you specify a label selector which is not
unique to the pods of that Job, and which matches unrelated Pods, then pods of the unrelated
job may be deleted, or this Job may count other Pods as completing it, or one or both
Jobs may refuse to create Pods or run to completion. If a non-unique selector is
chosen, then other controllers (e.g. ReplicationController) and their Pods may behave
in unpredictable ways too. Kubernetes will not stop you from making a mistake when
specifying .spec.selector
.
Here is an example of a case when you might want to use this feature.
Say Job old
is already running. You want existing Pods
to keep running, but you want the rest of the Pods it creates
to use a different pod template and for the Job to have a new name.
You cannot update the Job because these fields are not updatable.
Therefore, you delete Job old
but leave its pods
running, using kubectl delete jobs/old --cascade=orphan
.
Before deleting it, you make a note of what selector it uses:
kubectl get job old -o yaml
The output is similar to this:
kind: Job
metadata:
name: old
...
spec:
selector:
matchLabels:
batch.kubernetes.io/controller-uid: a8f3d00d-c6d2-11e5-9f87-42010af00002
...
Then you create a new Job with name new
and you explicitly specify the same selector.
Since the existing Pods have label batch.kubernetes.io/controller-uid=a8f3d00d-c6d2-11e5-9f87-42010af00002
,
they are controlled by Job new
as well.
You need to specify manualSelector: true
in the new Job since you are not using
the selector that the system normally generates for you automatically.
kind: Job
metadata:
name: new
...
spec:
manualSelector: true
selector:
matchLabels:
batch.kubernetes.io/controller-uid: a8f3d00d-c6d2-11e5-9f87-42010af00002
...
The new Job itself will have a different uid from a8f3d00d-c6d2-11e5-9f87-42010af00002
. Setting
manualSelector: true
tells the system that you know what you are doing and to allow this
mismatch.
Job tracking with finalizers
Kubernetes v1.26 [stable]
The control plane keeps track of the Pods that belong to any Job and notices if
any such Pod is removed from the API server. To do that, the Job controller
creates Pods with the finalizer batch.kubernetes.io/job-tracking
. The
controller removes the finalizer only after the Pod has been accounted for in
the Job status, allowing the Pod to be removed by other controllers or users.
Note:
See My pod stays terminating if you observe that pods from a Job are stuck with the tracking finalizer.Elastic Indexed Jobs
Kubernetes v1.31 [stable]
(enabled by default: true)You can scale Indexed Jobs up or down by mutating both .spec.parallelism
and .spec.completions
together such that .spec.parallelism == .spec.completions
.
When scaling down, Kubernetes removes the Pods with higher indexes.
Use cases for elastic Indexed Jobs include batch workloads which require scaling an indexed Job, such as MPI, Horovod, Ray, and PyTorch training jobs.
Delayed creation of replacement pods
Kubernetes v1.29 [beta]
Note:
You can only setpodReplacementPolicy
on Jobs if you enable the JobPodReplacementPolicy
feature gate
(enabled by default).By default, the Job controller recreates Pods as soon they either fail or are terminating (have a deletion timestamp).
This means that, at a given time, when some of the Pods are terminating, the number of running Pods for a Job
can be greater than parallelism
or greater than one Pod per index (if you are using an Indexed Job).
You may choose to create replacement Pods only when the terminating Pod is fully terminal (has status.phase: Failed
).
To do this, set the .spec.podReplacementPolicy: Failed
.
The default replacement policy depends on whether the Job has a podFailurePolicy
set.
With no Pod failure policy defined for a Job, omitting the podReplacementPolicy
field selects the
TerminatingOrFailed
replacement policy:
the control plane creates replacement Pods immediately upon Pod deletion
(as soon as the control plane sees that a Pod for this Job has deletionTimestamp
set).
For Jobs with a Pod failure policy set, the default podReplacementPolicy
is Failed
, and no other
value is permitted.
See Pod failure policy to learn more about Pod failure policies for Jobs.
kind: Job
metadata:
name: new
...
spec:
podReplacementPolicy: Failed
...
Provided your cluster has the feature gate enabled, you can inspect the .status.terminating
field of a Job.
The value of the field is the number of Pods owned by the Job that are currently terminating.
kubectl get jobs/myjob -o yaml
apiVersion: batch/v1
kind: Job
# .metadata and .spec omitted
status:
terminating: 3 # three Pods are terminating and have not yet reached the Failed phase
Delegation of managing a Job object to external controller
Kubernetes v1.30 [alpha]
(enabled by default: false)Note:
You can only set themanagedBy
field on Jobs if you enable the JobManagedBy
feature gate
(disabled by default).This feature allows you to disable the built-in Job controller, for a specific Job, and delegate reconciliation of the Job to an external controller.
You indicate the controller that reconciles the Job by setting a custom value
for the spec.managedBy
field - any value
other than kubernetes.io/job-controller
. The value of the field is immutable.
Note:
When using this feature, make sure the controller indicated by the field is installed, otherwise the Job may not be reconciled at all.Note:
When developing an external Job controller be aware that your controller needs to operate in a fashion conformant with the definitions of the API spec and status fields of the Job object.
Please review these in detail in the Job API. We also recommend that you run the e2e conformance tests for the Job object to verify your implementation.
Finally, when developing an external Job controller make sure it does not use the
batch.kubernetes.io/job-tracking
finalizer, reserved for the built-in controller.
Warning:
If you are considering to disable theJobManagedBy
feature gate, or to
downgrade the cluster to a version without the feature gate enabled, check if
there are jobs with a custom value of the spec.managedBy
field. If there
are such jobs, there is a risk that they might be reconciled by two controllers
after the operation: the built-in Job controller and the external controller
indicated by the field value.Alternatives
Bare Pods
When the node that a Pod is running on reboots or fails, the pod is terminated and will not be restarted. However, a Job will create new Pods to replace terminated ones. For this reason, we recommend that you use a Job rather than a bare Pod, even if your application requires only a single Pod.
Replication Controller
Jobs are complementary to Replication Controllers. A Replication Controller manages Pods which are not expected to terminate (e.g. web servers), and a Job manages Pods that are expected to terminate (e.g. batch tasks).
As discussed in Pod Lifecycle, Job
is only appropriate
for pods with RestartPolicy
equal to OnFailure
or Never
.
(Note: If RestartPolicy
is not set, the default value is Always
.)
Single Job starts controller Pod
Another pattern is for a single Job to create a Pod which then creates other Pods, acting as a sort of custom controller for those Pods. This allows the most flexibility, but may be somewhat complicated to get started with and offers less integration with Kubernetes.
One example of this pattern would be a Job which starts a Pod which runs a script that in turn starts a Spark master controller (see spark example), runs a spark driver, and then cleans up.
An advantage of this approach is that the overall process gets the completion guarantee of a Job object, but maintains complete control over what Pods are created and how work is assigned to them.
What's next
- Learn about Pods.
- Read about different ways of running Jobs:
- Coarse Parallel Processing Using a Work Queue
- Fine Parallel Processing Using a Work Queue
- Use an indexed Job for parallel processing with static work assignment
- Create multiple Jobs based on a template: Parallel Processing using Expansions
- Follow the links within Clean up finished jobs automatically to learn more about how your cluster can clean up completed and / or failed tasks.
Job
is part of the Kubernetes REST API. Read the Job object definition to understand the API for jobs.- Read about
CronJob
, which you can use to define a series of Jobs that will run based on a schedule, similar to the UNIX toolcron
. - Practice how to configure handling of retriable and non-retriable pod failures
using
podFailurePolicy
, based on the step-by-step examples.
4.2.6 - Automatic Cleanup for Finished Jobs
Kubernetes v1.23 [stable]
When your Job has finished, it's useful to keep that Job in the API (and not immediately delete the Job) so that you can tell whether the Job succeeded or failed.
Kubernetes' TTL-after-finished controller provides a TTL (time to live) mechanism to limit the lifetime of Job objects that have finished execution.
Cleanup for finished Jobs
The TTL-after-finished controller is only supported for Jobs. You can use this mechanism to clean
up finished Jobs (either Complete
or Failed
) automatically by specifying the
.spec.ttlSecondsAfterFinished
field of a Job, as in this
example.
The TTL-after-finished controller assumes that a Job is eligible to be cleaned up
TTL seconds after the Job has finished. The timer starts once the
status condition of the Job changes to show that the Job is either Complete
or Failed
; once the TTL has
expired, that Job becomes eligible for
cascading removal. When the
TTL-after-finished controller cleans up a job, it will delete it cascadingly, that is to say it will delete
its dependent objects together with it.
Kubernetes honors object lifecycle guarantees on the Job, such as waiting for finalizers.
You can set the TTL seconds at any time. Here are some examples for setting the
.spec.ttlSecondsAfterFinished
field of a Job:
- Specify this field in the Job manifest, so that a Job can be cleaned up automatically some time after it finishes.
- Manually set this field of existing, already finished Jobs, so that they become eligible for cleanup.
- Use a mutating admission webhook to set this field dynamically at Job creation time. Cluster administrators can use this to enforce a TTL policy for finished jobs.
- Use a
mutating admission webhook
to set this field dynamically after the Job has finished, and choose
different TTL values based on job status, labels. For this case, the webhook needs
to detect changes to the
.status
of the Job and only set a TTL when the Job is being marked as completed. - Write your own controller to manage the cleanup TTL for Jobs that match a particular selector.
Caveats
Updating TTL for finished Jobs
You can modify the TTL period, e.g. .spec.ttlSecondsAfterFinished
field of Jobs,
after the job is created or has finished. If you extend the TTL period after the
existing ttlSecondsAfterFinished
period has expired, Kubernetes doesn't guarantee
to retain that Job, even if an update to extend the TTL returns a successful API
response.
Time skew
Because the TTL-after-finished controller uses timestamps stored in the Kubernetes jobs to determine whether the TTL has expired or not, this feature is sensitive to time skew in your cluster, which may cause the control plane to clean up Job objects at the wrong time.
Clocks aren't always correct, but the difference should be very small. Please be aware of this risk when setting a non-zero TTL.
What's next
Refer to the Kubernetes Enhancement Proposal (KEP) for adding this mechanism.
4.2.7 - CronJob
Kubernetes v1.21 [stable]
A CronJob creates Jobs on a repeating schedule.
CronJob is meant for performing regular scheduled actions such as backups, report generation, and so on. One CronJob object is like one line of a crontab (cron table) file on a Unix system. It runs a Job periodically on a given schedule, written in Cron format.
CronJobs have limitations and idiosyncrasies. For example, in certain circumstances, a single CronJob can create multiple concurrent Jobs. See the limitations below.
When the control plane creates new Jobs and (indirectly) Pods for a CronJob, the .metadata.name
of the CronJob is part of the basis for naming those Pods. The name of a CronJob must be a valid
DNS subdomain
value, but this can produce unexpected results for the Pod hostnames. For best compatibility,
the name should follow the more restrictive rules for a
DNS label.
Even when the name is a DNS subdomain, the name must be no longer than 52
characters. This is because the CronJob controller will automatically append
11 characters to the name you provide and there is a constraint that the
length of a Job name is no more than 63 characters.
Example
This example CronJob manifest prints the current time and a hello message every minute:
apiVersion: batch/v1
kind: CronJob
metadata:
name: hello
spec:
schedule: "* * * * *"
jobTemplate:
spec:
template:
spec:
containers:
- name: hello
image: busybox:1.28
imagePullPolicy: IfNotPresent
command:
- /bin/sh
- -c
- date; echo Hello from the Kubernetes cluster
restartPolicy: OnFailure
(Running Automated Tasks with a CronJob takes you through this example in more detail).
Writing a CronJob spec
Schedule syntax
The .spec.schedule
field is required. The value of that field follows the Cron syntax:
# ┌───────────── minute (0 - 59)
# │ ┌───────────── hour (0 - 23)
# │ │ ┌───────────── day of the month (1 - 31)
# │ │ │ ┌───────────── month (1 - 12)
# │ │ │ │ ┌───────────── day of the week (0 - 6) (Sunday to Saturday)
# │ │ │ │ │ OR sun, mon, tue, wed, thu, fri, sat
# │ │ │ │ │
# │ │ │ │ │
# * * * * *
For example, 0 3 * * 1
means this task is scheduled to run weekly on a Monday at 3 AM.
The format also includes extended "Vixie cron" step values. As explained in the FreeBSD manual:
Step values can be used in conjunction with ranges. Following a range with
/<number>
specifies skips of the number's value through the range. For example,0-23/2
can be used in the hours field to specify command execution every other hour (the alternative in the V7 standard is0,2,4,6,8,10,12,14,16,18,20,22
). Steps are also permitted after an asterisk, so if you want to say "every two hours", just use*/2
.
Note:
A question mark (?
) in the schedule has the same meaning as an asterisk *
, that is,
it stands for any of available value for a given field.Other than the standard syntax, some macros like @monthly
can also be used:
Entry | Description | Equivalent to |
---|---|---|
@yearly (or @annually) | Run once a year at midnight of 1 January | 0 0 1 1 * |
@monthly | Run once a month at midnight of the first day of the month | 0 0 1 * * |
@weekly | Run once a week at midnight on Sunday morning | 0 0 * * 0 |
@daily (or @midnight) | Run once a day at midnight | 0 0 * * * |
@hourly | Run once an hour at the beginning of the hour | 0 * * * * |
To generate CronJob schedule expressions, you can also use web tools like crontab.guru.
Job template
The .spec.jobTemplate
defines a template for the Jobs that the CronJob creates, and it is required.
It has exactly the same schema as a Job, except that
it is nested and does not have an apiVersion
or kind
.
You can specify common metadata for the templated Jobs, such as
labels or
annotations.
For information about writing a Job .spec
, see Writing a Job Spec.
Deadline for delayed Job start
The .spec.startingDeadlineSeconds
field is optional.
This field defines a deadline (in whole seconds) for starting the Job, if that Job misses its scheduled time
for any reason.
After missing the deadline, the CronJob skips that instance of the Job (future occurrences are still scheduled). For example, if you have a backup Job that runs twice a day, you might allow it to start up to 8 hours late, but no later, because a backup taken any later wouldn't be useful: you would instead prefer to wait for the next scheduled run.
For Jobs that miss their configured deadline, Kubernetes treats them as failed Jobs.
If you don't specify startingDeadlineSeconds
for a CronJob, the Job occurrences have no deadline.
If the .spec.startingDeadlineSeconds
field is set (not null), the CronJob
controller measures the time between when a Job is expected to be created and
now. If the difference is higher than that limit, it will skip this execution.
For example, if it is set to 200
, it allows a Job to be created for up to 200
seconds after the actual schedule.
Concurrency policy
The .spec.concurrencyPolicy
field is also optional.
It specifies how to treat concurrent executions of a Job that is created by this CronJob.
The spec may specify only one of the following concurrency policies:
Allow
(default): The CronJob allows concurrently running JobsForbid
: The CronJob does not allow concurrent runs; if it is time for a new Job run and the previous Job run hasn't finished yet, the CronJob skips the new Job run. Also note that when the previous Job run finishes,.spec.startingDeadlineSeconds
is still taken into account and may result in a new Job run.Replace
: If it is time for a new Job run and the previous Job run hasn't finished yet, the CronJob replaces the currently running Job run with a new Job run
Note that concurrency policy only applies to the Jobs created by the same CronJob. If there are multiple CronJobs, their respective Jobs are always allowed to run concurrently.
Schedule suspension
You can suspend execution of Jobs for a CronJob, by setting the optional .spec.suspend
field
to true. The field defaults to false.
This setting does not affect Jobs that the CronJob has already started.
If you do set that field to true, all subsequent executions are suspended (they remain scheduled, but the CronJob controller does not start the Jobs to run the tasks) until you unsuspend the CronJob.
Caution:
Executions that are suspended during their scheduled time count as missed Jobs. When.spec.suspend
changes from true
to false
on an existing CronJob without a
starting deadline, the missed Jobs are scheduled immediately.Jobs history limits
The .spec.successfulJobsHistoryLimit
and .spec.failedJobsHistoryLimit
fields specify
how many completed and failed Jobs should be kept. Both fields are optional.
.spec.successfulJobsHistoryLimit
: This field specifies the number of successful finished jobs to keep. The default value is3
. Setting this field to0
will not keep any successful jobs..spec.failedJobsHistoryLimit
: This field specifies the number of failed finished jobs to keep. The default value is1
. Setting this field to0
will not keep any failed jobs.
For another way to clean up Jobs automatically, see Clean up finished Jobs automatically.
Time zones
Kubernetes v1.27 [stable]
For CronJobs with no time zone specified, the kube-controller-manager interprets schedules relative to its local time zone.
You can specify a time zone for a CronJob by setting .spec.timeZone
to the name
of a valid time zone.
For example, setting .spec.timeZone: "Etc/UTC"
instructs Kubernetes to interpret
the schedule relative to Coordinated Universal Time.
A time zone database from the Go standard library is included in the binaries and used as a fallback in case an external database is not available on the system.
CronJob limitations
Unsupported TimeZone specification
Specifying a timezone using CRON_TZ
or TZ
variables inside .spec.schedule
is not officially supported (and never has been).
Starting with Kubernetes 1.29 if you try to set a schedule that includes TZ
or CRON_TZ
timezone specification, Kubernetes will fail to create the resource with a validation
error.
Updates to CronJobs already using TZ
or CRON_TZ
will continue to report a
warning to the client.
Modifying a CronJob
By design, a CronJob contains a template for new Jobs. If you modify an existing CronJob, the changes you make will apply to new Jobs that start to run after your modification is complete. Jobs (and their Pods) that have already started continue to run without changes. That is, the CronJob does not update existing Jobs, even if those remain running.
Job creation
A CronJob creates a Job object approximately once per execution time of its schedule. The scheduling is approximate because there are certain circumstances where two Jobs might be created, or no Job might be created. Kubernetes tries to avoid those situations, but does not completely prevent them. Therefore, the Jobs that you define should be idempotent.
If startingDeadlineSeconds
is set to a large value or left unset (the default)
and if concurrencyPolicy
is set to Allow
, the Jobs will always run
at least once.
Caution:
IfstartingDeadlineSeconds
is set to a value less than 10 seconds, the CronJob may not be scheduled. This is because the CronJob controller checks things every 10 seconds.For every CronJob, the CronJob Controller checks how many schedules it missed in the duration from its last scheduled time until now. If there are more than 100 missed schedules, then it does not start the Job and logs the error.
Cannot determine if job needs to be started. Too many missed start time (> 100). Set or decrease .spec.startingDeadlineSeconds or check clock skew.
It is important to note that if the startingDeadlineSeconds
field is set (not nil
), the controller counts how many missed Jobs occurred from the value of startingDeadlineSeconds
until now rather than from the last scheduled time until now. For example, if startingDeadlineSeconds
is 200
, the controller counts how many missed Jobs occurred in the last 200 seconds.
A CronJob is counted as missed if it has failed to be created at its scheduled time. For example, if concurrencyPolicy
is set to Forbid
and a CronJob was attempted to be scheduled when there was a previous schedule still running, then it would count as missed.
For example, suppose a CronJob is set to schedule a new Job every one minute beginning at 08:30:00
, and its
startingDeadlineSeconds
field is not set. If the CronJob controller happens to
be down from 08:29:00
to 10:21:00
, the Job will not start as the number of missed Jobs which missed their schedule is greater than 100.
To illustrate this concept further, suppose a CronJob is set to schedule a new Job every one minute beginning at 08:30:00
, and its
startingDeadlineSeconds
is set to 200 seconds. If the CronJob controller happens to
be down for the same period as the previous example (08:29:00
to 10:21:00
,) the Job will still start at 10:22:00. This happens as the controller now checks how many missed schedules happened in the last 200 seconds (i.e., 3 missed schedules), rather than from the last scheduled time until now.
The CronJob is only responsible for creating Jobs that match its schedule, and the Job in turn is responsible for the management of the Pods it represents.
What's next
- Learn about Pods and Jobs, two concepts that CronJobs rely upon.
- Read about the detailed format
of CronJob
.spec.schedule
fields. - For instructions on creating and working with CronJobs, and for an example of a CronJob manifest, see Running automated tasks with CronJobs.
CronJob
is part of the Kubernetes REST API. Read the CronJob API reference for more details.
4.2.8 - ReplicationController
A ReplicationController ensures that a specified number of pod replicas are running at any one time. In other words, a ReplicationController makes sure that a pod or a homogeneous set of pods is always up and available.
How a ReplicationController works
If there are too many pods, the ReplicationController terminates the extra pods. If there are too few, the ReplicationController starts more pods. Unlike manually created pods, the pods maintained by a ReplicationController are automatically replaced if they fail, are deleted, or are terminated. For example, your pods are re-created on a node after disruptive maintenance such as a kernel upgrade. For this reason, you should use a ReplicationController even if your application requires only a single pod. A ReplicationController is similar to a process supervisor, but instead of supervising individual processes on a single node, the ReplicationController supervises multiple pods across multiple nodes.
ReplicationController is often abbreviated to "rc" in discussion, and as a shortcut in kubectl commands.
A simple case is to create one ReplicationController object to reliably run one instance of a Pod indefinitely. A more complex use case is to run several identical replicas of a replicated service, such as web servers.
Running an example ReplicationController
This example ReplicationController config runs three copies of the nginx web server.
apiVersion: v1
kind: ReplicationController
metadata:
name: nginx
spec:
replicas: 3
selector:
app: nginx
template:
metadata:
name: nginx
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx
ports:
- containerPort: 80
Run the example job by downloading the example file and then running this command:
kubectl apply -f https://k8s.io/examples/controllers/replication.yaml
The output is similar to this:
replicationcontroller/nginx created
Check on the status of the ReplicationController using this command:
kubectl describe replicationcontrollers/nginx
The output is similar to this:
Name: nginx
Namespace: default
Selector: app=nginx
Labels: app=nginx
Annotations: <none>
Replicas: 3 current / 3 desired
Pods Status: 0 Running / 3 Waiting / 0 Succeeded / 0 Failed
Pod Template:
Labels: app=nginx
Containers:
nginx:
Image: nginx
Port: 80/TCP
Environment: <none>
Mounts: <none>
Volumes: <none>
Events:
FirstSeen LastSeen Count From SubobjectPath Type Reason Message
--------- -------- ----- ---- ------------- ---- ------ -------
20s 20s 1 {replication-controller } Normal SuccessfulCreate Created pod: nginx-qrm3m
20s 20s 1 {replication-controller } Normal SuccessfulCreate Created pod: nginx-3ntk0
20s 20s 1 {replication-controller } Normal SuccessfulCreate Created pod: nginx-4ok8v
Here, three pods are created, but none is running yet, perhaps because the image is being pulled. A little later, the same command may show:
Pods Status: 3 Running / 0 Waiting / 0 Succeeded / 0 Failed
To list all the pods that belong to the ReplicationController in a machine readable form, you can use a command like this:
pods=$(kubectl get pods --selector=app=nginx --output=jsonpath={.items..metadata.name})
echo $pods
The output is similar to this:
nginx-3ntk0 nginx-4ok8v nginx-qrm3m
Here, the selector is the same as the selector for the ReplicationController (seen in the
kubectl describe
output), and in a different form in replication.yaml
. The --output=jsonpath
option
specifies an expression with the name from each pod in the returned list.
Writing a ReplicationController Manifest
As with all other Kubernetes config, a ReplicationController needs apiVersion
, kind
, and metadata
fields.
When the control plane creates new Pods for a ReplicationController, the .metadata.name
of the
ReplicationController is part of the basis for naming those Pods. The name of a ReplicationController must be a valid
DNS subdomain
value, but this can produce unexpected results for the Pod hostnames. For best compatibility,
the name should follow the more restrictive rules for a
DNS label.
For general information about working with configuration files, see object management.
A ReplicationController also needs a .spec
section.
Pod Template
The .spec.template
is the only required field of the .spec
.
The .spec.template
is a pod template. It has exactly the same schema as a Pod, except it is nested and does not have an apiVersion
or kind
.
In addition to required fields for a Pod, a pod template in a ReplicationController must specify appropriate labels and an appropriate restart policy. For labels, make sure not to overlap with other controllers. See pod selector.
Only a .spec.template.spec.restartPolicy
equal to Always
is allowed, which is the default if not specified.
For local container restarts, ReplicationControllers delegate to an agent on the node, for example the Kubelet.
Labels on the ReplicationController
The ReplicationController can itself have labels (.metadata.labels
). Typically, you
would set these the same as the .spec.template.metadata.labels
; if .metadata.labels
is not specified
then it defaults to .spec.template.metadata.labels
. However, they are allowed to be
different, and the .metadata.labels
do not affect the behavior of the ReplicationController.
Pod Selector
The .spec.selector
field is a label selector. A ReplicationController
manages all the pods with labels that match the selector. It does not distinguish
between pods that it created or deleted and pods that another person or process created or
deleted. This allows the ReplicationController to be replaced without affecting the running pods.
If specified, the .spec.template.metadata.labels
must be equal to the .spec.selector
, or it will
be rejected by the API. If .spec.selector
is unspecified, it will be defaulted to
.spec.template.metadata.labels
.
Also you should not normally create any pods whose labels match this selector, either directly, with another ReplicationController, or with another controller such as Job. If you do so, the ReplicationController thinks that it created the other pods. Kubernetes does not stop you from doing this.
If you do end up with multiple controllers that have overlapping selectors, you will have to manage the deletion yourself (see below).
Multiple Replicas
You can specify how many pods should run concurrently by setting .spec.replicas
to the number
of pods you would like to have running concurrently. The number running at any time may be higher
or lower, such as if the replicas were just increased or decreased, or if a pod is gracefully
shutdown, and a replacement starts early.
If you do not specify .spec.replicas
, then it defaults to 1.
Working with ReplicationControllers
Deleting a ReplicationController and its Pods
To delete a ReplicationController and all its pods, use kubectl delete
. Kubectl will scale the ReplicationController to zero and wait
for it to delete each pod before deleting the ReplicationController itself. If this kubectl
command is interrupted, it can be restarted.
When using the REST API or client library, you need to do the steps explicitly (scale replicas to 0, wait for pod deletions, then delete the ReplicationController).
Deleting only a ReplicationController
You can delete a ReplicationController without affecting any of its pods.
Using kubectl, specify the --cascade=orphan
option to kubectl delete
.
When using the REST API or client library, you can delete the ReplicationController object.
Once the original is deleted, you can create a new ReplicationController to replace it. As long
as the old and new .spec.selector
are the same, then the new one will adopt the old pods.
However, it will not make any effort to make existing pods match a new, different pod template.
To update pods to a new spec in a controlled way, use a rolling update.
Isolating pods from a ReplicationController
Pods may be removed from a ReplicationController's target set by changing their labels. This technique may be used to remove pods from service for debugging and data recovery. Pods that are removed in this way will be replaced automatically (assuming that the number of replicas is not also changed).
Common usage patterns
Rescheduling
As mentioned above, whether you have 1 pod you want to keep running, or 1000, a ReplicationController will ensure that the specified number of pods exists, even in the event of node failure or pod termination (for example, due to an action by another control agent).
Scaling
The ReplicationController enables scaling the number of replicas up or down, either manually or by an auto-scaling control agent, by updating the replicas
field.
Rolling updates
The ReplicationController is designed to facilitate rolling updates to a service by replacing pods one-by-one.
As explained in #1353, the recommended approach is to create a new ReplicationController with 1 replica, scale the new (+1) and old (-1) controllers one by one, and then delete the old controller after it reaches 0 replicas. This predictably updates the set of pods regardless of unexpected failures.
Ideally, the rolling update controller would take application readiness into account, and would ensure that a sufficient number of pods were productively serving at any given time.
The two ReplicationControllers would need to create pods with at least one differentiating label, such as the image tag of the primary container of the pod, since it is typically image updates that motivate rolling updates.
Multiple release tracks
In addition to running multiple releases of an application while a rolling update is in progress, it's common to run multiple releases for an extended period of time, or even continuously, using multiple release tracks. The tracks would be differentiated by labels.
For instance, a service might target all pods with tier in (frontend), environment in (prod)
. Now say you have 10 replicated pods that make up this tier. But you want to be able to 'canary' a new version of this component. You could set up a ReplicationController with replicas
set to 9 for the bulk of the replicas, with labels tier=frontend, environment=prod, track=stable
, and another ReplicationController with replicas
set to 1 for the canary, with labels tier=frontend, environment=prod, track=canary
. Now the service is covering both the canary and non-canary pods. But you can mess with the ReplicationControllers separately to test things out, monitor the results, etc.
Using ReplicationControllers with Services
Multiple ReplicationControllers can sit behind a single service, so that, for example, some traffic goes to the old version, and some goes to the new version.
A ReplicationController will never terminate on its own, but it isn't expected to be as long-lived as services. Services may be composed of pods controlled by multiple ReplicationControllers, and it is expected that many ReplicationControllers may be created and destroyed over the lifetime of a service (for instance, to perform an update of pods that run the service). Both services themselves and their clients should remain oblivious to the ReplicationControllers that maintain the pods of the services.
Writing programs for Replication
Pods created by a ReplicationController are intended to be fungible and semantically identical, though their configurations may become heterogeneous over time. This is an obvious fit for replicated stateless servers, but ReplicationControllers can also be used to maintain availability of master-elected, sharded, and worker-pool applications. Such applications should use dynamic work assignment mechanisms, such as the RabbitMQ work queues, as opposed to static/one-time customization of the configuration of each pod, which is considered an anti-pattern. Any pod customization performed, such as vertical auto-sizing of resources (for example, cpu or memory), should be performed by another online controller process, not unlike the ReplicationController itself.
Responsibilities of the ReplicationController
The ReplicationController ensures that the desired number of pods matches its label selector and are operational. Currently, only terminated pods are excluded from its count. In the future, readiness and other information available from the system may be taken into account, we may add more controls over the replacement policy, and we plan to emit events that could be used by external clients to implement arbitrarily sophisticated replacement and/or scale-down policies.
The ReplicationController is forever constrained to this narrow responsibility. It itself will not perform readiness nor liveness probes. Rather than performing auto-scaling, it is intended to be controlled by an external auto-scaler (as discussed in #492), which would change its replicas
field. We will not add scheduling policies (for example, spreading) to the ReplicationController. Nor should it verify that the pods controlled match the currently specified template, as that would obstruct auto-sizing and other automated processes. Similarly, completion deadlines, ordering dependencies, configuration expansion, and other features belong elsewhere. We even plan to factor out the mechanism for bulk pod creation (#170).
The ReplicationController is intended to be a composable building-block primitive. We expect higher-level APIs and/or tools to be built on top of it and other complementary primitives for user convenience in the future. The "macro" operations currently supported by kubectl (run, scale) are proof-of-concept examples of this. For instance, we could imagine something like Asgard managing ReplicationControllers, auto-scalers, services, scheduling policies, canaries, etc.
API Object
Replication controller is a top-level resource in the Kubernetes REST API. More details about the API object can be found at: ReplicationController API object.
Alternatives to ReplicationController
ReplicaSet
ReplicaSet
is the next-generation ReplicationController that supports the new set-based label selector.
It's mainly used by Deployment as a mechanism to orchestrate pod creation, deletion and updates.
Note that we recommend using Deployments instead of directly using Replica Sets, unless you require custom update orchestration or don't require updates at all.
Deployment (Recommended)
Deployment
is a higher-level API object that updates its underlying Replica Sets and their Pods. Deployments are recommended if you want the rolling update functionality, because they are declarative, server-side, and have additional features.
Bare Pods
Unlike in the case where a user directly created pods, a ReplicationController replaces pods that are deleted or terminated for any reason, such as in the case of node failure or disruptive node maintenance, such as a kernel upgrade. For this reason, we recommend that you use a ReplicationController even if your application requires only a single pod. Think of it similarly to a process supervisor, only it supervises multiple pods across multiple nodes instead of individual processes on a single node. A ReplicationController delegates local container restarts to some agent on the node, such as the kubelet.
Job
Use a Job
instead of a ReplicationController for pods that are expected to terminate on their own
(that is, batch jobs).
DaemonSet
Use a DaemonSet
instead of a ReplicationController for pods that provide a
machine-level function, such as machine monitoring or machine logging. These pods have a lifetime that is tied
to a machine lifetime: the pod needs to be running on the machine before other pods start, and are
safe to terminate when the machine is otherwise ready to be rebooted/shutdown.
What's next
- Learn about Pods.
- Learn about Deployment, the replacement for ReplicationController.
ReplicationController
is part of the Kubernetes REST API. Read the ReplicationController object definition to understand the API for replication controllers.
4.3 - Autoscaling Workloads
In Kubernetes, you can scale a workload depending on the current demand of resources. This allows your cluster to react to changes in resource demand more elastically and efficiently.
When you scale a workload, you can either increase or decrease the number of replicas managed by the workload, or adjust the resources available to the replicas in-place.
The first approach is referred to as horizontal scaling, while the second is referred to as vertical scaling.
There are manual and automatic ways to scale your workloads, depending on your use case.
Scaling workloads manually
Kubernetes supports manual scaling of workloads. Horizontal scaling can be done
using the kubectl
CLI.
For vertical scaling, you need to patch the resource definition of your workload.
See below for examples of both strategies.
- Horizontal scaling: Running multiple instances of your app
- Vertical scaling: Resizing CPU and memory resources assigned to containers
Scaling workloads automatically
Kubernetes also supports automatic scaling of workloads, which is the focus of this page.
The concept of Autoscaling in Kubernetes refers to the ability to automatically update an object that manages a set of Pods (for example a Deployment).
Scaling workloads horizontally
In Kubernetes, you can automatically scale a workload horizontally using a HorizontalPodAutoscaler (HPA).
It is implemented as a Kubernetes API resource and a controller and periodically adjusts the number of replicas in a workload to match observed resource utilization such as CPU or memory usage.
There is a walkthrough tutorial of configuring a HorizontalPodAutoscaler for a Deployment.
Scaling workloads vertically
Kubernetes v1.25 [stable]
You can automatically scale a workload vertically using a VerticalPodAutoscaler (VPA). Unlike the HPA, the VPA doesn't come with Kubernetes by default, but is a separate project that can be found on GitHub.
Once installed, it allows you to create CustomResourceDefinitions (CRDs) for your workloads which define how and when to scale the resources of the managed replicas.
Note:
You will need to have the Metrics Server installed to your cluster for the HPA to work.At the moment, the VPA can operate in four different modes:
Mode | Description |
---|---|
Auto | Currently, Recreate might change to in-place updates in the future |
Recreate | The VPA assigns resource requests on pod creation as well as updates them on existing pods by evicting them when the requested resources differ significantly from the new recommendation |
Initial | The VPA only assigns resource requests on pod creation and never changes them later. |
Off | The VPA does not automatically change the resource requirements of the pods. The recommendations are calculated and can be inspected in the VPA object. |
Requirements for in-place resizing
Kubernetes v1.27 [alpha]
Resizing a workload in-place without restarting the Pods
or its Containers requires Kubernetes version 1.27 or later.
Additionally, the InPlaceVerticalScaling
feature gate needs to be enabled.
InPlacePodVerticalScaling
: Enables in-place Pod vertical scaling.
Autoscaling based on cluster size
For workloads that need to be scaled based on the size of the cluster (for example
cluster-dns
or other system components), you can use the
Cluster Proportional Autoscaler.
Just like the VPA, it is not part of the Kubernetes core, but hosted as its
own project on GitHub.
The Cluster Proportional Autoscaler watches the number of schedulable nodes and cores and scales the number of replicas of the target workload accordingly.
If the number of replicas should stay the same, you can scale your workloads vertically according to the cluster size using the Cluster Proportional Vertical Autoscaler. The project is currently in beta and can be found on GitHub.
While the Cluster Proportional Autoscaler scales the number of replicas of a workload, the Cluster Proportional Vertical Autoscaler adjusts the resource requests for a workload (for example a Deployment or DaemonSet) based on the number of nodes and/or cores in the cluster.
Event driven Autoscaling
It is also possible to scale workloads based on events, for example using the Kubernetes Event Driven Autoscaler (KEDA).
KEDA is a CNCF graduated enabling you to scale your workloads based on the number of events to be processed, for example the amount of messages in a queue. There exists a wide range of adapters for different event sources to choose from.
Autoscaling based on schedules
Another strategy for scaling your workloads is to schedule the scaling operations, for example in order to reduce resource consumption during off-peak hours.
Similar to event driven autoscaling, such behavior can be achieved using KEDA in conjunction with
its Cron
scaler. The Cron
scaler allows you to define schedules
(and time zones) for scaling your workloads in or out.
Scaling cluster infrastructure
If scaling workloads isn't enough to meet your needs, you can also scale your cluster infrastructure itself.
Scaling the cluster infrastructure normally means adding or removing nodes. Read cluster autoscaling for more information.
What's next
- Learn more about scaling horizontally
- Resize Container Resources In-Place
- Autoscale the DNS Service in a Cluster
- Learn about cluster autoscaling
4.4 - Managing Workloads
You've deployed your application and exposed it via a Service. Now what? Kubernetes provides a number of tools to help you manage your application deployment, including scaling and updating.
Organizing resource configurations
Many applications require multiple resources to be created, such as a Deployment along with a Service.
Management of multiple resources can be simplified by grouping them together in the same file
(separated by ---
in YAML). For example:
apiVersion: v1
kind: Service
metadata:
name: my-nginx-svc
labels:
app: nginx
spec:
type: LoadBalancer
ports:
- port: 80
selector:
app: nginx
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: my-nginx
labels:
app: nginx
spec:
replicas: 3
selector:
matchLabels:
app: nginx
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
ports:
- containerPort: 80
Multiple resources can be created the same way as a single resource:
kubectl apply -f https://k8s.io/examples/application/nginx-app.yaml
service/my-nginx-svc created
deployment.apps/my-nginx created
The resources will be created in the order they appear in the manifest. Therefore, it's best to specify the Service first, since that will ensure the scheduler can spread the pods associated with the Service as they are created by the controller(s), such as Deployment.
kubectl apply
also accepts multiple -f
arguments:
kubectl apply -f https://k8s.io/examples/application/nginx/nginx-svc.yaml \
-f https://k8s.io/examples/application/nginx/nginx-deployment.yaml
It is a recommended practice to put resources related to the same microservice or application tier into the same file, and to group all of the files associated with your application in the same directory. If the tiers of your application bind to each other using DNS, you can deploy all of the components of your stack together.
A URL can also be specified as a configuration source, which is handy for deploying directly from manifests in your source control system:
kubectl apply -f https://k8s.io/examples/application/nginx/nginx-deployment.yaml
deployment.apps/my-nginx created
If you need to define more manifests, such as adding a ConfigMap, you can do that too.
External tools
This section lists only the most common tools used for managing workloads on Kubernetes. To see a larger list, view Application definition and image build in the CNCF Landscape.
Helm
Helm is a tool for managing packages of pre-configured Kubernetes resources. These packages are known as Helm charts.
Kustomize
Kustomize traverses a Kubernetes manifest to add, remove or update configuration options. It is available both as a standalone binary and as a native feature of kubectl.
Bulk operations in kubectl
Resource creation isn't the only operation that kubectl
can perform in bulk. It can also extract
resource names from configuration files in order to perform other operations, in particular to
delete the same resources you created:
kubectl delete -f https://k8s.io/examples/application/nginx-app.yaml
deployment.apps "my-nginx" deleted
service "my-nginx-svc" deleted
In the case of two resources, you can specify both resources on the command line using the resource/name syntax:
kubectl delete deployments/my-nginx services/my-nginx-svc
For larger numbers of resources, you'll find it easier to specify the selector (label query)
specified using -l
or --selector
, to filter resources by their labels:
kubectl delete deployment,services -l app=nginx
deployment.apps "my-nginx" deleted
service "my-nginx-svc" deleted
Chaining and filtering
Because kubectl
outputs resource names in the same syntax it accepts, you can chain operations
using $()
or xargs
:
kubectl get $(kubectl create -f docs/concepts/cluster-administration/nginx/ -o name | grep service/ )
kubectl create -f docs/concepts/cluster-administration/nginx/ -o name | grep service/ | xargs -i kubectl get '{}'
The output might be similar to:
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
my-nginx-svc LoadBalancer 10.0.0.208 <pending> 80/TCP 0s
With the above commands, first you create resources under examples/application/nginx/
and print
the resources created with -o name
output format (print each resource as resource/name).
Then you grep
only the Service, and then print it with kubectl get
.
Recursive operations on local files
If you happen to organize your resources across several subdirectories within a particular
directory, you can recursively perform the operations on the subdirectories also, by specifying
--recursive
or -R
alongside the --filename
/-f
argument.
For instance, assume there is a directory project/k8s/development
that holds all of the
manifests needed for the development environment,
organized by resource type:
project/k8s/development
├── configmap
│ └── my-configmap.yaml
├── deployment
│ └── my-deployment.yaml
└── pvc
└── my-pvc.yaml
By default, performing a bulk operation on project/k8s/development
will stop at the first level
of the directory, not processing any subdirectories. If you had tried to create the resources in
this directory using the following command, we would have encountered an error:
kubectl apply -f project/k8s/development
error: you must provide one or more resources by argument or filename (.json|.yaml|.yml|stdin)
Instead, specify the --recursive
or -R
command line argument along with the --filename
/-f
argument:
kubectl apply -f project/k8s/development --recursive
configmap/my-config created
deployment.apps/my-deployment created
persistentvolumeclaim/my-pvc created
The --recursive
argument works with any operation that accepts the --filename
/-f
argument such as:
kubectl create
, kubectl get
, kubectl delete
, kubectl describe
, or even kubectl rollout
.
The --recursive
argument also works when multiple -f
arguments are provided:
kubectl apply -f project/k8s/namespaces -f project/k8s/development --recursive
namespace/development created
namespace/staging created
configmap/my-config created
deployment.apps/my-deployment created
persistentvolumeclaim/my-pvc created
If you're interested in learning more about kubectl
, go ahead and read
Command line tool (kubectl).
Updating your application without an outage
At some point, you'll eventually need to update your deployed application, typically by specifying
a new image or image tag. kubectl
supports several update operations, each of which is applicable
to different scenarios.
You can run multiple copies of your app, and use a rollout to gradually shift the traffic to new healthy Pods. Eventually, all the running Pods would have the new software.
This section of the page guides you through how to create and update applications with Deployments.
Let's say you were running version 1.14.2 of nginx:
kubectl create deployment my-nginx --image=nginx:1.14.2
deployment.apps/my-nginx created
Ensure that there is 1 replica:
kubectl scale --replicas 1 deployments/my-nginx --subresource='scale' --type='merge' -p '{"spec":{"replicas": 1}}'
deployment.apps/my-nginx scaled
and allow Kubernetes to add more temporary replicas during a rollout, by setting a surge maximum of 100%:
kubectl patch --type='merge' -p '{"spec":{"strategy":{"rollingUpdate":{"maxSurge": "100%" }}}}'
deployment.apps/my-nginx patched
To update to version 1.16.1, change .spec.template.spec.containers[0].image
from nginx:1.14.2
to nginx:1.16.1
using kubectl edit
:
kubectl edit deployment/my-nginx
# Change the manifest to use the newer container image, then save your changes
That's it! The Deployment will declaratively update the deployed nginx application progressively behind the scene. It ensures that only a certain number of old replicas may be down while they are being updated, and only a certain number of new replicas may be created above the desired number of pods. To learn more details about how this happens, visit Deployment.
You can use rollouts with DaemonSets, Deployments, or StatefulSets.
Managing rollouts
You can use kubectl rollout
to manage a
progressive update of an existing application.
For example:
kubectl apply -f my-deployment.yaml
# wait for rollout to finish
kubectl rollout status deployment/my-deployment --timeout 10m # 10 minute timeout
or
kubectl apply -f backing-stateful-component.yaml
# don't wait for rollout to finish, just check the status
kubectl rollout status statefulsets/backing-stateful-component --watch=false
You can also pause, resume or cancel a rollout.
Visit kubectl rollout
to learn more.
Canary deployments
Another scenario where multiple labels are needed is to distinguish deployments of different releases or configurations of the same component. It is common practice to deploy a canary of a new application release (specified via image tag in the pod template) side by side with the previous release so that the new release can receive live production traffic before fully rolling it out.
For instance, you can use a track
label to differentiate different releases.
The primary, stable release would have a track
label with value as stable
:
name: frontend
replicas: 3
...
labels:
app: guestbook
tier: frontend
track: stable
...
image: gb-frontend:v3
and then you can create a new release of the guestbook frontend that carries the track
label
with different value (i.e. canary
), so that two sets of pods would not overlap:
name: frontend-canary
replicas: 1
...
labels:
app: guestbook
tier: frontend
track: canary
...
image: gb-frontend:v4
The frontend service would span both sets of replicas by selecting the common subset of their
labels (i.e. omitting the track
label), so that the traffic will be redirected to both
applications:
selector:
app: guestbook
tier: frontend
You can tweak the number of replicas of the stable and canary releases to determine the ratio of each release that will receive live production traffic (in this case, 3:1). Once you're confident, you can update the stable track to the new application release and remove the canary one.
Updating annotations
Sometimes you would want to attach annotations to resources. Annotations are arbitrary
non-identifying metadata for retrieval by API clients such as tools or libraries.
This can be done with kubectl annotate
. For example:
kubectl annotate pods my-nginx-v4-9gw19 description='my frontend running nginx'
kubectl get pods my-nginx-v4-9gw19 -o yaml
apiVersion: v1
kind: pod
metadata:
annotations:
description: my frontend running nginx
...
For more information, see annotations and kubectl annotate.
Scaling your application
When load on your application grows or shrinks, use kubectl
to scale your application.
For instance, to decrease the number of nginx replicas from 3 to 1, do:
kubectl scale deployment/my-nginx --replicas=1
deployment.apps/my-nginx scaled
Now you only have one pod managed by the deployment.
kubectl get pods -l app=nginx
NAME READY STATUS RESTARTS AGE
my-nginx-2035384211-j5fhi 1/1 Running 0 30m
To have the system automatically choose the number of nginx replicas as needed, ranging from 1 to 3, do:
# This requires an existing source of container and Pod metrics
kubectl autoscale deployment/my-nginx --min=1 --max=3
horizontalpodautoscaler.autoscaling/my-nginx autoscaled
Now your nginx replicas will be scaled up and down as needed, automatically.
For more information, please see kubectl scale, kubectl autoscale and horizontal pod autoscaler document.
In-place updates of resources
Sometimes it's necessary to make narrow, non-disruptive updates to resources you've created.
kubectl apply
It is suggested to maintain a set of configuration files in source control
(see configuration as code),
so that they can be maintained and versioned along with the code for the resources they configure.
Then, you can use kubectl apply
to push your configuration changes to the cluster.
This command will compare the version of the configuration that you're pushing with the previous version and apply the changes you've made, without overwriting any automated changes to properties you haven't specified.
kubectl apply -f https://k8s.io/examples/application/nginx/nginx-deployment.yaml
deployment.apps/my-nginx configured
To learn more about the underlying mechanism, read server-side apply.
kubectl edit
Alternatively, you may also update resources with kubectl edit
:
kubectl edit deployment/my-nginx
This is equivalent to first get
the resource, edit it in text editor, and then apply
the
resource with the updated version:
kubectl get deployment my-nginx -o yaml > /tmp/nginx.yaml
vi /tmp/nginx.yaml
# do some edit, and then save the file
kubectl apply -f /tmp/nginx.yaml
deployment.apps/my-nginx configured
rm /tmp/nginx.yaml
This allows you to do more significant changes more easily. Note that you can specify the editor
with your EDITOR
or KUBE_EDITOR
environment variables.
For more information, please see kubectl edit.
kubectl patch
You can use kubectl patch
to update API objects in place.
This subcommand supports JSON patch,
JSON merge patch, and strategic merge patch.
See Update API Objects in Place Using kubectl patch for more details.
Disruptive updates
In some cases, you may need to update resource fields that cannot be updated once initialized, or
you may want to make a recursive change immediately, such as to fix broken pods created by a
Deployment. To change such fields, use replace --force
, which deletes and re-creates the
resource. In this case, you can modify your original configuration file:
kubectl replace -f https://k8s.io/examples/application/nginx/nginx-deployment.yaml --force
deployment.apps/my-nginx deleted
deployment.apps/my-nginx replaced
What's next
5 - Services, Load Balancing, and Networking
The Kubernetes network model
The Kubernetes network model is built out of several pieces:
Each pod in a cluster gets its own unique cluster-wide IP address.
- A pod has its own private network namespace which is shared by
all of the containers within the pod. Processes running in
different containers in the same pod can communicate with each
other over
localhost
.
- A pod has its own private network namespace which is shared by
all of the containers within the pod. Processes running in
different containers in the same pod can communicate with each
other over
The pod network (also called a cluster network) handles communication between pods. It ensures that (barring intentional network segmentation):
All pods can communicate with all other pods, whether they are on the same node or on different nodes. Pods can communicate with each other directly, without the use of proxies or address translation (NAT).
On Windows, this rule does not apply to host-network pods.
Agents on a node (such as system daemons, or kubelet) can communicate with all pods on that node.
The Service API lets you provide a stable (long lived) IP address or hostname for a service implemented by one or more backend pods, where the individual pods making up the service can change over time.
Kubernetes automatically manages EndpointSlice objects to provide information about the pods currently backing a Service.
A service proxy implementation monitors the set of Service and EndpointSlice objects, and programs the data plane to route service traffic to its backends, by using operating system or cloud provider APIs to intercept or rewrite packets.
The Gateway API (or its predecessor, Ingress) allows you to make Services accessible to clients that are outside the cluster.
- A simpler, but less-configurable, mechanism for cluster
ingress is available via the Service API's
type: LoadBalancer
, when using a supported Cloud Provider.
- A simpler, but less-configurable, mechanism for cluster
ingress is available via the Service API's
NetworkPolicy is a built-in Kubernetes API that allows you to control traffic between pods, or between pods and the outside world.
In older container systems, there was no automatic connectivity between containers on different hosts, and so it was often necessary to explicitly create links between containers, or to map container ports to host ports to make them reachable by containers on other hosts. This is not needed in Kubernetes; Kubernetes's model is that pods can be treated much like VMs or physical hosts from the perspectives of port allocation, naming, service discovery, load balancing, application configuration, and migration.
Only a few parts of this model are implemented by Kubernetes itself. For the other parts, Kubernetes defines the APIs, but the corresponding functionality is provided by external components, some of which are optional:
Pod network namespace setup is handled by system-level software implementing the Container Runtime Interface.
The pod network itself is managed by a pod network implementation. On Linux, most container runtimes use the Container Networking Interface (CNI) to interact with the pod network implementation, so these implementations are often called CNI plugins.
Kubernetes provides a default implementation of service proxying, called kube-proxy, but some pod network implementations instead use their own service proxy that is more tightly integrated with the rest of the implementation.
NetworkPolicy is generally also implemented by the pod network implementation. (Some simpler pod network implementations don't implement NetworkPolicy, or an administrator may choose to configure the pod network without NetworkPolicy support. In these cases, the API will still be present, but it will have no effect.)
There are many implementations of the Gateway API, some of which are specific to particular cloud environments, some more focused on "bare metal" environments, and others more generic.
What's next
The Connecting Applications with Services tutorial lets you learn about Services and Kubernetes networking with a hands-on example.
Cluster Networking explains how to set up networking for your cluster, and also provides an overview of the technologies involved.
5.1 - Service
In Kubernetes, a Service is a method for exposing a network application that is running as one or more Pods in your cluster.
A key aim of Services in Kubernetes is that you don't need to modify your existing application to use an unfamiliar service discovery mechanism. You can run code in Pods, whether this is a code designed for a cloud-native world, or an older app you've containerized. You use a Service to make that set of Pods available on the network so that clients can interact with it.
If you use a Deployment to run your app, that Deployment can create and destroy Pods dynamically. From one moment to the next, you don't know how many of those Pods are working and healthy; you might not even know what those healthy Pods are named. Kubernetes Pods are created and destroyed to match the desired state of your cluster. Pods are ephemeral resources (you should not expect that an individual Pod is reliable and durable).
Each Pod gets its own IP address (Kubernetes expects network plugins to ensure this). For a given Deployment in your cluster, the set of Pods running in one moment in time could be different from the set of Pods running that application a moment later.
This leads to a problem: if some set of Pods (call them "backends") provides functionality to other Pods (call them "frontends") inside your cluster, how do the frontends find out and keep track of which IP address to connect to, so that the frontend can use the backend part of the workload?
Enter Services.
Services in Kubernetes
The Service API, part of Kubernetes, is an abstraction to help you expose groups of Pods over a network. Each Service object defines a logical set of endpoints (usually these endpoints are Pods) along with a policy about how to make those pods accessible.
For example, consider a stateless image-processing backend which is running with 3 replicas. Those replicas are fungible—frontends do not care which backend they use. While the actual Pods that compose the backend set may change, the frontend clients should not need to be aware of that, nor should they need to keep track of the set of backends themselves.
The Service abstraction enables this decoupling.
The set of Pods targeted by a Service is usually determined by a selector that you define. To learn about other ways to define Service endpoints, see Services without selectors.
If your workload speaks HTTP, you might choose to use an Ingress to control how web traffic reaches that workload. Ingress is not a Service type, but it acts as the entry point for your cluster. An Ingress lets you consolidate your routing rules into a single resource, so that you can expose multiple components of your workload, running separately in your cluster, behind a single listener.
The Gateway API for Kubernetes provides extra capabilities beyond Ingress and Service. You can add Gateway to your cluster - it is a family of extension APIs, implemented using CustomResourceDefinitions - and then use these to configure access to network services that are running in your cluster.
Cloud-native service discovery
If you're able to use Kubernetes APIs for service discovery in your application, you can query the API server for matching EndpointSlices. Kubernetes updates the EndpointSlices for a Service whenever the set of Pods in a Service changes.
For non-native applications, Kubernetes offers ways to place a network port or load balancer in between your application and the backend Pods.
Either way, your workload can use these service discovery mechanisms to find the target it wants to connect to.
Defining a Service
A Service is an object
(the same way that a Pod or a ConfigMap is an object). You can create,
view or modify Service definitions using the Kubernetes API. Usually
you use a tool such as kubectl
to make those API calls for you.
For example, suppose you have a set of Pods that each listen on TCP port 9376
and are labelled as app.kubernetes.io/name=MyApp
. You can define a Service to
publish that TCP listener:
apiVersion: v1
kind: Service
metadata:
name: my-service
spec:
selector:
app.kubernetes.io/name: MyApp
ports:
- protocol: TCP
port: 80
targetPort: 9376
Applying this manifest creates a new Service named "my-service" with the default
ClusterIP service type. The Service
targets TCP port 9376 on any Pod with the app.kubernetes.io/name: MyApp
label.
Kubernetes assigns this Service an IP address (the cluster IP), that is used by the virtual IP address mechanism. For more details on that mechanism, read Virtual IPs and Service Proxies.
The controller for that Service continuously scans for Pods that match its selector, and then makes any necessary updates to the set of EndpointSlices for the Service.
The name of a Service object must be a valid RFC 1035 label name.
Note:
A Service can map any incomingport
to a targetPort
. By default and
for convenience, the targetPort
is set to the same value as the port
field.Port definitions
Port definitions in Pods have names, and you can reference these names in the
targetPort
attribute of a Service. For example, we can bind the targetPort
of the Service to the Pod port in the following way:
apiVersion: v1
kind: Pod
metadata:
name: nginx
labels:
app.kubernetes.io/name: proxy
spec:
containers:
- name: nginx
image: nginx:stable
ports:
- containerPort: 80
name: http-web-svc
---
apiVersion: v1
kind: Service
metadata:
name: nginx-service
spec:
selector:
app.kubernetes.io/name: proxy
ports:
- name: name-of-service-port
protocol: TCP
port: 80
targetPort: http-web-svc
This works even if there is a mixture of Pods in the Service using a single configured name, with the same network protocol available via different port numbers. This offers a lot of flexibility for deploying and evolving your Services. For example, you can change the port numbers that Pods expose in the next version of your backend software, without breaking clients.
The default protocol for Services is TCP; you can also use any other supported protocol.
Because many Services need to expose more than one port, Kubernetes supports
multiple port definitions for a single Service.
Each port definition can have the same protocol
, or a different one.
Services without selectors
Services most commonly abstract access to Kubernetes Pods thanks to the selector, but when used with a corresponding set of EndpointSlices objects and without a selector, the Service can abstract other kinds of backends, including ones that run outside the cluster.
For example:
- You want to have an external database cluster in production, but in your test environment you use your own databases.
- You want to point your Service to a Service in a different Namespace or on another cluster.
- You are migrating a workload to Kubernetes. While evaluating the approach, you run only a portion of your backends in Kubernetes.
In any of these scenarios you can define a Service without specifying a selector to match Pods. For example:
apiVersion: v1
kind: Service
metadata:
name: my-service
spec:
ports:
- name: http
protocol: TCP
port: 80
targetPort: 9376
Because this Service has no selector, the corresponding EndpointSlice (and legacy Endpoints) objects are not created automatically. You can map the Service to the network address and port where it's running, by adding an EndpointSlice object manually. For example:
apiVersion: discovery.k8s.io/v1
kind: EndpointSlice
metadata:
name: my-service-1 # by convention, use the name of the Service
# as a prefix for the name of the EndpointSlice
labels:
# You should set the "kubernetes.io/service-name" label.
# Set its value to match the name of the Service
kubernetes.io/service-name: my-service
addressType: IPv4
ports:
- name: http # should match with the name of the service port defined above
appProtocol: http
protocol: TCP
port: 9376
endpoints:
- addresses:
- "10.4.5.6"
- addresses:
- "10.1.2.3"
Custom EndpointSlices
When you create an EndpointSlice object for a Service, you can
use any name for the EndpointSlice. Each EndpointSlice in a namespace must have a
unique name. You link an EndpointSlice to a Service by setting the
kubernetes.io/service-name
label
on that EndpointSlice.
Note:
The endpoint IPs must not be: loopback (127.0.0.0/8 for IPv4, ::1/128 for IPv6), or link-local (169.254.0.0/16 and 224.0.0.0/24 for IPv4, fe80::/64 for IPv6).
The endpoint IP addresses cannot be the cluster IPs of other Kubernetes Services, because kube-proxy doesn't support virtual IPs as a destination.
For an EndpointSlice that you create yourself, or in your own code,
you should also pick a value to use for the label
endpointslice.kubernetes.io/managed-by
.
If you create your own controller code to manage EndpointSlices, consider using a
value similar to "my-domain.example/name-of-controller"
. If you are using a third
party tool, use the name of the tool in all-lowercase and change spaces and other
punctuation to dashes (-
).
If people are directly using a tool such as kubectl
to manage EndpointSlices,
use a name that describes this manual management, such as "staff"
or
"cluster-admins"
. You should
avoid using the reserved value "controller"
, which identifies EndpointSlices
managed by Kubernetes' own control plane.
Accessing a Service without a selector
Accessing a Service without a selector works the same as if it had a selector. In the example for a Service without a selector, traffic is routed to one of the two endpoints defined in the EndpointSlice manifest: a TCP connection to 10.1.2.3 or 10.4.5.6, on port 9376.
Note:
The Kubernetes API server does not allow proxying to endpoints that are not mapped to pods. Actions such askubectl port-forward service/<service-name> forwardedPort:servicePort
where the service has no
selector will fail due to this constraint. This prevents the Kubernetes API server
from being used as a proxy to endpoints the caller may not be authorized to access.An ExternalName
Service is a special case of Service that does not have
selectors and uses DNS names instead. For more information, see the
ExternalName section.
EndpointSlices
Kubernetes v1.21 [stable]
EndpointSlices are objects that represent a subset (a slice) of the backing network endpoints for a Service.
Your Kubernetes cluster tracks how many endpoints each EndpointSlice represents. If there are so many endpoints for a Service that a threshold is reached, then Kubernetes adds another empty EndpointSlice and stores new endpoint information there. By default, Kubernetes makes a new EndpointSlice once the existing EndpointSlices all contain at least 100 endpoints. Kubernetes does not make the new EndpointSlice until an extra endpoint needs to be added.
See EndpointSlices for more information about this API.
Endpoints
In the Kubernetes API, an Endpoints (the resource kind is plural) defines a list of network endpoints, typically referenced by a Service to define which Pods the traffic can be sent to.
The EndpointSlice API is the recommended replacement for Endpoints.
Over-capacity endpoints
Kubernetes limits the number of endpoints that can fit in a single Endpoints object. When there are over 1000 backing endpoints for a Service, Kubernetes truncates the data in the Endpoints object. Because a Service can be linked with more than one EndpointSlice, the 1000 backing endpoint limit only affects the legacy Endpoints API.
In that case, Kubernetes selects at most 1000 possible backend endpoints to store
into the Endpoints object, and sets an
annotation on the Endpoints:
endpoints.kubernetes.io/over-capacity: truncated
.
The control plane also removes that annotation if the number of backend Pods drops below 1000.
Traffic is still sent to backends, but any load balancing mechanism that relies on the legacy Endpoints API only sends traffic to at most 1000 of the available backing endpoints.
The same API limit means that you cannot manually update an Endpoints to have more than 1000 endpoints.
Application protocol
Kubernetes v1.20 [stable]
The appProtocol
field provides a way to specify an application protocol for
each Service port. This is used as a hint for implementations to offer
richer behavior for protocols that they understand.
The value of this field is mirrored by the corresponding
Endpoints and EndpointSlice objects.
This field follows standard Kubernetes label syntax. Valid values are one of:
Implementation-defined prefixed names such as
mycompany.com/my-custom-protocol
.Kubernetes-defined prefixed names:
Protocol | Description |
---|---|
kubernetes.io/h2c | HTTP/2 over cleartext as described in RFC 7540 |
kubernetes.io/ws | WebSocket over cleartext as described in RFC 6455 |
kubernetes.io/wss | WebSocket over TLS as described in RFC 6455 |
Multi-port Services
For some Services, you need to expose more than one port. Kubernetes lets you configure multiple port definitions on a Service object. When using multiple ports for a Service, you must give all of your ports names so that these are unambiguous. For example:
apiVersion: v1
kind: Service
metadata:
name: my-service
spec:
selector:
app.kubernetes.io/name: MyApp
ports:
- name: http
protocol: TCP
port: 80
targetPort: 9376
- name: https
protocol: TCP
port: 443
targetPort: 9377
Note:
As with Kubernetes names in general, names for ports
must only contain lowercase alphanumeric characters and -
. Port names must
also start and end with an alphanumeric character.
For example, the names 123-abc
and web
are valid, but 123_abc
and -web
are not.
Service type
For some parts of your application (for example, frontends) you may want to expose a Service onto an external IP address, one that's accessible from outside of your cluster.
Kubernetes Service types allow you to specify what kind of Service you want.
The available type
values and their behaviors are:
ClusterIP
- Exposes the Service on a cluster-internal IP. Choosing this value
makes the Service only reachable from within the cluster. This is the
default that is used if you don't explicitly specify a
type
for a Service. You can expose the Service to the public internet using an Ingress or a Gateway. NodePort
- Exposes the Service on each Node's IP at a static port (the
NodePort
). To make the node port available, Kubernetes sets up a cluster IP address, the same as if you had requested a Service oftype: ClusterIP
. LoadBalancer
- Exposes the Service externally using an external load balancer. Kubernetes does not directly offer a load balancing component; you must provide one, or you can integrate your Kubernetes cluster with a cloud provider.
ExternalName
- Maps the Service to the contents of the
externalName
field (for example, to the hostnameapi.foo.bar.example
). The mapping configures your cluster's DNS server to return aCNAME
record with that external hostname value. No proxying of any kind is set up.
The type
field in the Service API is designed as nested functionality - each level
adds to the previous. However there is an exception to this nested design. You can
define a LoadBalancer
Service by
disabling the load balancer NodePort
allocation.
type: ClusterIP
This default Service type assigns an IP address from a pool of IP addresses that your cluster has reserved for that purpose.
Several of the other types for Service build on the ClusterIP
type as a
foundation.
If you define a Service that has the .spec.clusterIP
set to "None"
then
Kubernetes does not assign an IP address. See headless Services
for more information.
Choosing your own IP address
You can specify your own cluster IP address as part of a Service
creation
request. To do this, set the .spec.clusterIP
field. For example, if you
already have an existing DNS entry that you wish to reuse, or legacy systems
that are configured for a specific IP address and difficult to re-configure.
The IP address that you choose must be a valid IPv4 or IPv6 address from within the
service-cluster-ip-range
CIDR range that is configured for the API server.
If you try to create a Service with an invalid clusterIP
address value, the API
server will return a 422 HTTP status code to indicate that there's a problem.
Read avoiding collisions to learn how Kubernetes helps reduce the risk and impact of two different Services both trying to use the same IP address.
type: NodePort
If you set the type
field to NodePort
, the Kubernetes control plane
allocates a port from a range specified by --service-node-port-range
flag (default: 30000-32767).
Each node proxies that port (the same port number on every Node) into your Service.
Your Service reports the allocated port in its .spec.ports[*].nodePort
field.
Using a NodePort gives you the freedom to set up your own load balancing solution, to configure environments that are not fully supported by Kubernetes, or even to expose one or more nodes' IP addresses directly.
For a node port Service, Kubernetes additionally allocates a port (TCP, UDP or
SCTP to match the protocol of the Service). Every node in the cluster configures
itself to listen on that assigned port and to forward traffic to one of the ready
endpoints associated with that Service. You'll be able to contact the type: NodePort
Service, from outside the cluster, by connecting to any node using the appropriate
protocol (for example: TCP), and the appropriate port (as assigned to that Service).
Choosing your own port
If you want a specific port number, you can specify a value in the nodePort
field. The control plane will either allocate you that port or report that
the API transaction failed.
This means that you need to take care of possible port collisions yourself.
You also have to use a valid port number, one that's inside the range configured
for NodePort use.
Here is an example manifest for a Service of type: NodePort
that specifies
a NodePort value (30007, in this example):
apiVersion: v1
kind: Service
metadata:
name: my-service
spec:
type: NodePort
selector:
app.kubernetes.io/name: MyApp
ports:
- port: 80
# By default and for convenience, the `targetPort` is set to
# the same value as the `port` field.
targetPort: 80
# Optional field
# By default and for convenience, the Kubernetes control plane
# will allocate a port from a range (default: 30000-32767)
nodePort: 30007
Reserve Nodeport ranges to avoid collisions
The policy for assigning ports to NodePort services applies to both the auto-assignment and the manual assignment scenarios. When a user wants to create a NodePort service that uses a specific port, the target port may conflict with another port that has already been assigned.
To avoid this problem, the port range for NodePort services is divided into two bands. Dynamic port assignment uses the upper band by default, and it may use the lower band once the upper band has been exhausted. Users can then allocate from the lower band with a lower risk of port collision.
Custom IP address configuration for type: NodePort
Services
You can set up nodes in your cluster to use a particular IP address for serving node port services. You might want to do this if each node is connected to multiple networks (for example: one network for application traffic, and another network for traffic between nodes and the control plane).
If you want to specify particular IP address(es) to proxy the port, you can set the
--nodeport-addresses
flag for kube-proxy or the equivalent nodePortAddresses
field of the kube-proxy configuration file
to particular IP block(s).
This flag takes a comma-delimited list of IP blocks (e.g. 10.0.0.0/8
, 192.0.2.0/25
)
to specify IP address ranges that kube-proxy should consider as local to this node.
For example, if you start kube-proxy with the --nodeport-addresses=127.0.0.0/8
flag,
kube-proxy only selects the loopback interface for NodePort Services.
The default for --nodeport-addresses
is an empty list.
This means that kube-proxy should consider all available network interfaces for NodePort.
(That's also compatible with earlier Kubernetes releases.)
Note:
This Service is visible as<NodeIP>:spec.ports[*].nodePort
and .spec.clusterIP:spec.ports[*].port
.
If the --nodeport-addresses
flag for kube-proxy or the equivalent field
in the kube-proxy configuration file is set, <NodeIP>
would be a filtered
node IP address (or possibly IP addresses).type: LoadBalancer
On cloud providers which support external load balancers, setting the type
field to LoadBalancer
provisions a load balancer for your Service.
The actual creation of the load balancer happens asynchronously, and
information about the provisioned balancer is published in the Service's
.status.loadBalancer
field.
For example:
apiVersion: v1
kind: Service
metadata:
name: my-service
spec:
selector:
app.kubernetes.io/name: MyApp
ports:
- protocol: TCP
port: 80
targetPort: 9376
clusterIP: 10.0.171.239
type: LoadBalancer
status:
loadBalancer:
ingress:
- ip: 192.0.2.127
Traffic from the external load balancer is directed at the backend Pods. The cloud provider decides how it is load balanced.
To implement a Service of type: LoadBalancer
, Kubernetes typically starts off
by making the changes that are equivalent to you requesting a Service of
type: NodePort
. The cloud-controller-manager component then configures the external
load balancer to forward traffic to that assigned node port.
You can configure a load balanced Service to omit assigning a node port, provided that the cloud provider implementation supports this.
Some cloud providers allow you to specify the loadBalancerIP
. In those cases, the load-balancer is created
with the user-specified loadBalancerIP
. If the loadBalancerIP
field is not specified,
the load balancer is set up with an ephemeral IP address. If you specify a loadBalancerIP
but your cloud provider does not support the feature, the loadbalancerIP
field that you
set is ignored.
Note:
The.spec.loadBalancerIP
field for a Service was deprecated in Kubernetes v1.24.
This field was under-specified and its meaning varies across implementations. It also cannot support dual-stack networking. This field may be removed in a future API version.
If you're integrating with a provider that supports specifying the load balancer IP address(es) for a Service via a (provider specific) annotation, you should switch to doing that.
If you are writing code for a load balancer integration with Kubernetes, avoid using this field. You can integrate with Gateway rather than Service, or you can define your own (provider specific) annotations on the Service that specify the equivalent detail.
Node liveness impact on load balancer traffic
Load balancer health checks are critical to modern applications. They are used to
determine which server (virtual machine, or IP address) the load balancer should
dispatch traffic to. The Kubernetes APIs do not define how health checks have to be
implemented for Kubernetes managed load balancers, instead it's the cloud providers
(and the people implementing integration code) who decide on the behavior. Load
balancer health checks are extensively used within the context of supporting the
externalTrafficPolicy
field for Services.
Load balancers with mixed protocol types
Kubernetes v1.26 [stable]
(enabled by default: true)By default, for LoadBalancer type of Services, when there is more than one port defined, all ports must have the same protocol, and the protocol must be one which is supported by the cloud provider.
The feature gate MixedProtocolLBService
(enabled by default for the kube-apiserver as of v1.24) allows the use of
different protocols for LoadBalancer type of Services, when there is more than one port defined.
Note:
The set of protocols that can be used for load balanced Services is defined by your cloud provider; they may impose restrictions beyond what the Kubernetes API enforces.Disabling load balancer NodePort allocation
Kubernetes v1.24 [stable]
You can optionally disable node port allocation for a Service of type: LoadBalancer
, by setting
the field spec.allocateLoadBalancerNodePorts
to false
. This should only be used for load balancer implementations
that route traffic directly to pods as opposed to using node ports. By default, spec.allocateLoadBalancerNodePorts
is true
and type LoadBalancer Services will continue to allocate node ports. If spec.allocateLoadBalancerNodePorts
is set to false
on an existing Service with allocated node ports, those node ports will not be de-allocated automatically.
You must explicitly remove the nodePorts
entry in every Service port to de-allocate those node ports.
Specifying class of load balancer implementation
Kubernetes v1.24 [stable]
For a Service with type
set to LoadBalancer
, the .spec.loadBalancerClass
field
enables you to use a load balancer implementation other than the cloud provider default.
By default, .spec.loadBalancerClass
is not set and a LoadBalancer
type of Service uses the cloud provider's default load balancer implementation if the
cluster is configured with a cloud provider using the --cloud-provider
component
flag.
If you specify .spec.loadBalancerClass
, it is assumed that a load balancer
implementation that matches the specified class is watching for Services.
Any default load balancer implementation (for example, the one provided by
the cloud provider) will ignore Services that have this field set.
spec.loadBalancerClass
can be set on a Service of type LoadBalancer
only.
Once set, it cannot be changed.
The value of spec.loadBalancerClass
must be a label-style identifier,
with an optional prefix such as "internal-vip
" or "example.com/internal-vip
".
Unprefixed names are reserved for end-users.
Specifying IPMode of load balancer status
Kubernetes v1.30 [beta]
(enabled by default: true)As a Beta feature in Kubernetes 1.30,
a feature gate
named LoadBalancerIPMode
allows you to set the .status.loadBalancer.ingress.ipMode
for a Service with type
set to LoadBalancer
.
The .status.loadBalancer.ingress.ipMode
specifies how the load-balancer IP behaves.
It may be specified only when the .status.loadBalancer.ingress.ip
field is also specified.
There are two possible values for .status.loadBalancer.ingress.ipMode
: "VIP" and "Proxy".
The default value is "VIP" meaning that traffic is delivered to the node
with the destination set to the load-balancer's IP and port.
There are two cases when setting this to "Proxy", depending on how the load-balancer
from the cloud provider delivers the traffics:
- If the traffic is delivered to the node then DNATed to the pod, the destination would be set to the node's IP and node port;
- If the traffic is delivered directly to the pod, the destination would be set to the pod's IP and port.
Service implementations may use this information to adjust traffic routing.
Internal load balancer
In a mixed environment it is sometimes necessary to route traffic from Services inside the same (virtual) network address block.
In a split-horizon DNS environment you would need two Services to be able to route both external and internal traffic to your endpoints.
To set an internal load balancer, add one of the following annotations to your Service depending on the cloud service provider you're using:
Select one of the tabs.
metadata:
name: my-service
annotations:
networking.gke.io/load-balancer-type: "Internal"
metadata:
name: my-service
annotations:
service.beta.kubernetes.io/aws-load-balancer-internal: "true"
metadata:
name: my-service
annotations:
service.beta.kubernetes.io/azure-load-balancer-internal: "true"
metadata:
name: my-service
annotations:
service.kubernetes.io/ibm-load-balancer-cloud-provider-ip-type: "private"
metadata:
name: my-service
annotations:
service.beta.kubernetes.io/openstack-internal-load-balancer: "true"
metadata:
name: my-service
annotations:
service.beta.kubernetes.io/cce-load-balancer-internal-vpc: "true"
metadata:
annotations:
service.kubernetes.io/qcloud-loadbalancer-internal-subnetid: subnet-xxxxx
metadata:
annotations:
service.beta.kubernetes.io/alibaba-cloud-loadbalancer-address-type: "intranet"