This is the multi-page printable view of this section. Click here to print.

Return to the regular view of this page.

Concepts

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.

1 - Overview

Kubernetes is a portable, extensible, open source platform for managing containerized workloads and services, that facilitates both declarative configuration and automation. It has a large, rapidly growing ecosystem. Kubernetes services, support, and tools are widely available.

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.

Deployment evolution

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

1.1 - Kubernetes Components

An overview of the key components that make up a Kubernetes cluster.

This page provides a high-level overview of the essential components that make up a Kubernetes cluster.

Components of Kubernetes

The components of 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

Kubernetes objects are persistent entities in the Kubernetes system. Kubernetes uses these entities to represent the state of your cluster. Learn about the Kubernetes object model and how to work with these objects.

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 object
  • kind - What kind of object you want to create
  • metadata - Data that helps uniquely identify the object, including a name string, UID, and optional namespace
  • 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.

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:

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

Management techniqueOperates onRecommended environmentSupported writersLearning curve
Imperative commandsLive objectsDevelopment projects1+Lowest
Imperative object configurationIndividual filesProduction projects1Moderate
Declarative object configurationDirectories of filesProduction projects1+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.

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.

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

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.

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

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

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

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.

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 to production or qa.
  • The second example selects all resources with key equal to tier and values other than frontend and backend, and all resources with no labels with the tier 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.

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.

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.

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

FEATURE STATE: 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

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"
  }
}

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

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

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

KindFields
Podspec.nodeName
spec.restartPolicy
spec.schedulerName
spec.serviceAccountName
spec.hostNetwork
status.phase
status.podIP
status.nominatedNodeName
EventinvolvedObject.kind
involvedObject.namespace
involvedObject.name
involvedObject.uid
involvedObject.apiVersion
involvedObject.resourceVersion
involvedObject.fieldPath
reason
reportingComponent
source
type
Secrettype
Namespacestatus.phase
ReplicaSetstatus.replicas
ReplicationControllerstatus.replicas
Jobstatus.successful
Nodespec.unschedulable
CertificateSigningRequestspec.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.

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.

What's next

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.

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

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.

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.

KeyDescriptionExampleType
app.kubernetes.io/nameThe name of the applicationmysqlstring
app.kubernetes.io/instanceA unique name identifying the instance of an applicationmysql-abcxyzstring
app.kubernetes.io/versionThe current version of the application (e.g., a SemVer 1.0, revision hash, etc.)5.7.21string
app.kubernetes.io/componentThe component within the architecturedatabasestring
app.kubernetes.io/part-ofThe name of a higher level application this one is part ofwordpressstring
app.kubernetes.io/managed-byThe tool being used to manage the operation of an applicationHelmstring

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 Kubernetes API lets you query and manipulate the state of objects in Kubernetes. The core of Kubernetes' control plane is the API server and the HTTP API that it exposes. Users, the different parts of your cluster, and external components all communicate with one another through the API server.

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

FEATURE STATE: 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:

Valid request header values for OpenAPI v2 queries
HeaderPossible valuesNotes
Accept-Encodinggzipnot supplying this header is also acceptable
Acceptapplication/com.github.proto-openapi.spec.v2@v1.0+protobufmainly for intra-cluster use
application/jsondefault
*serves application/json

OpenAPI V3

FEATURE STATE: 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.

Valid request header values for OpenAPI v3 queries
HeaderPossible valuesNotes
Accept-Encodinggzipnot supplying this header is also acceptable
Acceptapplication/com.github.proto-openapi.spec.v3@v1.0+protobufmainly for intra-cluster use
application/jsondefault
*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.

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:

  1. Custom resources let you declaratively define how the API server should provide your chosen resource API.
  2. You can also extend the Kubernetes API by implementing an aggregation layer.

What's next

2 - Cluster Architecture

The architectural concepts behind Kubernetes.

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.

The control plane (kube-apiserver, etcd, kube-controller-manager, kube-scheduler) and several nodes. Each node is running a kubelet and kube-proxy.

Figure 1. Kubernetes cluster components.

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:

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:

  1. The kubelet on a node self-registers to the control plane
  2. 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.

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.

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.

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 the Ready condition to Unknown.
  • 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.

Node topology

FEATURE STATE: 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

FEATURE STATE: 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.

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:

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.

Konnectivity service

FEATURE STATE: 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

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.

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

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

FEATURE STATE: 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

FEATURE STATE: 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

Kubernetes components

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.

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:

  1. Update a Node object with the corresponding server's unique identifier obtained from the cloud provider API.
  2. 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.
  3. Obtain the node's hostname and network addresses.
  4. 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 in cloud.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.

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?

FEATURE STATE: 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:
  • 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:
  • 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

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

FEATURE STATE: 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

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:

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.

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 to foregroundDeletion.
  • 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

FEATURE STATE: 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.

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 to 0.
  • MaxPerPodContainer: the maximum number of dead containers each Pod can have. Disable by setting to less than 0.
  • MaxContainers: the maximum number of dead containers the cluster can have. Disable by setting to less than 0.

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.

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

2.9 - Mixed Version Proxy

FEATURE STATE: 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 the batch 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 fetched StorageVersion 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

Technology for packaging an application along with its runtime dependencies.

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.

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 as docker.io/library/busybox:latest)
  • busybox:1.32.0 - Image name with tag. Kubernetes will use Docker public registry. (Same as docker.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.

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, the imagePullPolicy is automatically set to IfNotPresent.
  • if you omit the imagePullPolicy field, and the tag for the container image is :latest, imagePullPolicy is automatically set to Always;
  • if you omit the imagePullPolicy field, and you don't specify the tag for the container image, imagePullPolicy is automatically set to Always;
  • if you omit the imagePullPolicy field, and you specify the tag for the container image that isn't :latest, the imagePullPolicy is automatically set to IfNotPresent.

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 to Always.
  • Omit the imagePullPolicy and use :latest as the tag for the image to use; Kubernetes will set the policy to Always when you submit the Pod.
  • Omit the imagePullPolicy and the tag for the image to use; Kubernetes will set the policy to Always 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

FEATURE STATE: Kubernetes v1.29 [alpha] (enabled by default: false)
Kubernetes includes alpha support for performing image pulls based on the RuntimeClass of a Pod.

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

FEATURE STATE: 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

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 match kubernetes.io, but abc.kubernetes.io
  • *.*.kubernetes.io will not match abc.kubernetes.io, but abc.def.kubernetes.io
  • prefix.*.io will match prefix.kubernetes.io
  • *-good.kubernetes.io will match prefix-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

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

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.

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.

  1. 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.
  2. 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.
  3. 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.
  4. 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

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

3.3 - Runtime Class

FEATURE STATE: 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

  1. Configure the CRI implementation on nodes (runtime dependent)
  2. 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.

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.

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

FEATURE STATE: 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

FEATURE STATE: 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

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

4 - Workloads

Understand Pods, the smallest deployable compute object in Kubernetes, and the higher-level abstractions that help you to run them.

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:

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:

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?

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.

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

FEATURE STATE: 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, or creationTimestamp fields; the generation 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 the metadata.finalizers list.

  • Pod updates may not change fields other than spec.containers[*].image, spec.initContainers[*].image, spec.activeDeadlineSeconds or spec.tolerations. For spec.tolerations, you can only add new entries.

  • When updating the spec.activeDeadlineSeconds field, two types of updates are allowed:

    1. setting the unassigned field to a positive number;
    2. 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.

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.

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:

Pod creation 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).

FEATURE STATE: 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

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.

A multi-container Pod that contains a file puller sidecar and a web server. The Pod uses an ephemeral emptyDir volume for shared storage between the containers.

Figure 1.

A multi-container Pod that contains a file puller sidecar and a web server. The Pod uses an ephemeral emptyDir volume for shared storage between the containers.

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:

ValueDescription
PendingThe 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.
RunningThe 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.
SucceededAll containers in the Pod have terminated in success, and will not be restarted.
FailedAll 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.
UnknownFor 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.

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:

  1. Initial crash: Kubernetes attempts an immediate restart based on the Pod restartPolicy.
  2. 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.
  3. 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.
  4. 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:

  1. 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.
  2. Inspect events: Use kubectl describe pod <name-of-pod> to see events for the Pod, which can provide hints about configuration or resource issues.
  3. Review configuration: Ensure that the Pod configuration, including environment variables and mounted volumes, is correct and that all required external resources are available.
  4. 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.
  5. 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 nameDescription
typeName of this Pod condition.
statusIndicates whether that condition is applicable, with possible values "True", "False", or "Unknown".
lastProbeTimeTimestamp of when the Pod condition was last probed.
lastTransitionTimeTimestamp for when the Pod last transitioned from one status to another.
reasonMachine-readable, UpperCamelCase text indicating the reason for the condition's last transition.
messageHuman-readable message indicating details about the last status transition.

Pod readiness

FEATURE STATE: 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 are True.

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

FEATURE STATE: Kubernetes v1.29 [beta]

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 is SERVING.
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.

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 is Success.
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.

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:

  1. You use the kubectl tool to manually delete a specific Pod, with the default grace period (30 seconds).

  2. 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.

    1. If one of the Pod's containers has defined a preStop hook and the terminationGracePeriodSeconds in the Pod spec is not set to 0, the kubelet runs that hook inside of the container. The default terminationGracePeriodSeconds 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.

    2. 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).

  3. 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 as false (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 Flow

  4. The kubelet ensures the Pod is shut down and terminated

    1. 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 hidden pause container if that container runtime uses one.
    2. The kubelet transitions the Pod into a terminal phase (Failed or Succeeded depending on the end state of its containers).
    3. The kubelet triggers forcible removal of the Pod object from the API server, by setting grace period to 0 (immediate deletion).
    4. The API server deletes the Pod's API object, which is then no longer visible from any client.

Forced Pod termination

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.

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:

  1. are orphan Pods - bound to a node which no longer exists,
  2. are unscheduled terminating Pods,
  3. are terminating Pods, bound to a non-ready node tainted with node.kubernetes.io/out-of-service, when the NodeOutOfServiceVolumeDetach 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

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 like sed, awk, python, or dig 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:

4.1.3 - Sidecar Containers

FEATURE STATE: 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

4.1.4 - Ephemeral Containers

FEATURE STATE: 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.

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

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.

Dealing with disruptions

Here are some ways to mitigate involuntary disruptions:

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

FEATURE STATE: 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-1node-2node-3
pod-a availablepod-b availablepod-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 drainingnode-2node-3
pod-a terminatingpod-b availablepod-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 drainingnode-2node-3
pod-a terminatingpod-b availablepod-c available
pod-x terminatingpod-d startingpod-y

At some point, the pods terminate, and the cluster looks like this:

node-1 drainednode-2node-3
pod-b availablepod-c available
pod-d startingpod-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 drainednode-2node-3
pod-b availablepod-c available
pod-d availablepod-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 drainednode-2node-3no node
pod-b terminatingpod-c availablepod-e pending
pod-d availablepod-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

FEATURE STATE: 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 a NoExecute 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.

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

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

FEATURE STATE: 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

4.1.7 - User Namespaces

FEATURE STATE: 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+).

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 the PATH for the kubelet binary.
  • A configuration of subordinate UIDs/GIDs for the kubelet user (see man 5 subuid and man 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

FEATURE STATE: 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

4.1.8 - Downward API

There are two ways to expose Pod and container fields to a running container: environment variables, and as files that are populated by a special volume type. Together, these two ways of exposing Pod and container fields are called the 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 as status.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 as status.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 manages a set of Pods to run an application workload, usually one that doesn't maintain state.

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.

Use Case

The following are typical use cases for Deployments:

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.

  • The template field contains the following sub-fields:

    • The Pods are labeled app: nginxusing the .metadata.labels field.
    • The Pod template's specification, or .template.spec field, indicates that the Pods run one container, nginx, which runs the nginx Docker Hub image at version 1.14.2.
    • Create one container and name it nginx using the .spec.template.spec.containers[0].name field.

Before you begin, make sure your Kubernetes cluster is up and running. Follow the steps given below to create the above Deployment:

  1. Create the Deployment by running the following command:

    kubectl apply -f https://k8s.io/examples/controllers/nginx-deployment.yaml
    
  2. 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.

  3. 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
    
  4. 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.

  5. To see the ReplicaSet (rs) created by the Deployment, run kubectl 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 the pod-template-hash label on the ReplicaSet.

  6. 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.

Pod-template-hash 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

Follow the steps given below to update your Deployment:

  1. Let's update the nginx Pods to use the nginx:1.16.1 image instead of the nginx: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 and nginx: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 from nginx:1.14.2 to nginx:1.16.1:

    kubectl edit deployment/nginx-deployment
    

    The output is similar to:

    deployment.apps/nginx-deployment edited
    
  2. 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.

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.

  • 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).

  • Suppose that you made a typo while updating the Deployment, by putting the image name as nginx:1.161 instead of nginx: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 and nginx-deployment-2035384211) is 3, and the number of new replicas (from nginx-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
    
  • 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:

  1. 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 annotation kubernetes.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.
  2. 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.

  1. 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.

  2. 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
    
  3. 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
    

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.

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.

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.

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.

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

4.2.2 - ReplicaSet

A ReplicaSet's purpose is to maintain a stable set of replica Pods running at any given time. Usually, you define a Deployment and let that Deployment manage ReplicaSets automatically.

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.

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:

  1. Pending (and unschedulable) pods are scaled down first
  2. If controller.kubernetes.io/pod-deletion-cost annotation is set, then the pod with the lower value will come first.
  3. Pods on nodes with more replicas come before pods on nodes with fewer replicas.
  4. 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

FEATURE STATE: 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.

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 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

4.2.3 - StatefulSets

A StatefulSet runs a group of Pods, and maintains a sticky identity for each of those Pods. This is useful for managing applications that need persistent storage or a stable, unique network identity.

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

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

FEATURE STATE: 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

FEATURE STATE: 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 DomainService (ns/name)StatefulSet (ns/name)StatefulSet DomainPod DNSPod Hostname
cluster.localdefault/nginxdefault/webnginx.default.svc.cluster.localweb-{0..N-1}.nginx.default.svc.cluster.localweb-{0..N-1}
cluster.localfoo/nginxfoo/webnginx.foo.svc.cluster.localweb-{0..N-1}.nginx.foo.svc.cluster.localweb-{0..N-1}
kube.localfoo/nginxfoo/webnginx.foo.svc.kube.localweb-{0..N-1}.nginx.foo.svc.kube.localweb-{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

FEATURE STATE: 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 to OnDelete, 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

FEATURE STATE: 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.

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

FEATURE STATE: 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 the whenDeleted policy all PVCs from the volumeClaimTemplate are deleted after their Pods have been deleted. With the whenScaled 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

4.2.4 - DaemonSet

A DaemonSet defines Pods that provide node-local facilities. These might be fundamental to the operation of your cluster, such as a networking helper tool, or be part of an add-on.

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.

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:

Tolerations for DaemonSet pods
Toleration keyEffectDetails
node.kubernetes.io/not-readyNoExecuteDaemonSet 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/unreachableNoExecuteDaemonSet 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-pressureNoScheduleDaemonSet Pods can be scheduled onto nodes with disk pressure issues.
node.kubernetes.io/memory-pressureNoScheduleDaemonSet Pods can be scheduled onto nodes with memory pressure issues.
node.kubernetes.io/pid-pressureNoScheduleDaemonSet Pods can be scheduled onto nodes with process pressure issues.
node.kubernetes.io/unschedulableNoScheduleDaemonSet Pods can be scheduled onto nodes that are unschedulable.
node.kubernetes.io/network-unavailableNoScheduleOnly 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

4.2.5 - Jobs

Jobs represent one-off tasks that run to completion and then stop.

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:

  1. Non-parallel Jobs
    • normally, only one Pod is started, unless the Pod fails.
    • the Job is complete as soon as its Pod terminates successfully.
  2. 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.
  3. 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.

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

FEATURE STATE: 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 implicitly NonIndexed.

  • 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 gate PodIndexLabel 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.

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 to Pending or Running.

If either of the calculations reaches the .spec.backoffLimit, the Job is considered failed.

Backoff limit per index

FEATURE STATE: Kubernetes v1.29 [beta]

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

FEATURE STATE: 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 the backoffLimit is reached the entire Job failed.

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.

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 to Never.
  • 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 inspec.podFailurePolicy.rules[*].onExitCodes.containerName. When not specified the rule applies to all containers. When specified, it should match one the container or initContainer 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.

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

FEATURE STATE: Kubernetes v1.31 [beta] (enabled by default: true)

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 the succeededIndexes succeed, the job controller marks the Job as succeeded. The succeededIndexes 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 the succeededCount, the job controller marks the Job as succeeded.
  • When you specify both succeededIndexes and succeededCount, once the number of succeeded indexes from the subset of indexes specified in the succeededIndexes reaches the succeededCount, 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.

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 the FailJob 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

FEATURE STATE: 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.

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.

PatternSingle Job objectFewer pods than work items?Use app unmodified?
Queue with Pod Per Work Itemsometimes
Queue with Variable Pod Count
Indexed Job with Static Work Assignment
Job with Pod-to-Pod Communicationsometimessometimes
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 ItemWany
Queue with Variable Pod Countnullany
Indexed Job with Static Work AssignmentWany
Job with Pod-to-Pod CommunicationWW
Job Template Expansion1should be 1

Advanced usage

Suspending a Job

FEATURE STATE: 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

FEATURE STATE: 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

FEATURE STATE: 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.

Elastic Indexed Jobs

FEATURE STATE: 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

FEATURE STATE: Kubernetes v1.29 [beta]

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

FEATURE STATE: Kubernetes v1.30 [alpha] (enabled by default: false)

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.

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

4.2.6 - Automatic Cleanup for Finished Jobs

A time-to-live mechanism to clean up old Jobs that have finished execution.
FEATURE STATE: 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

4.2.7 - CronJob

A CronJob starts one-time Jobs on a repeating schedule.
FEATURE STATE: 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 is 0,2,4,6,8,10,12,14,16,18,20,22). Steps are also permitted after an asterisk, so if you want to say "every two hours", just use */2.

Other than the standard syntax, some macros like @monthly can also be used:

EntryDescriptionEquivalent to
@yearly (or @annually)Run once a year at midnight of 1 January0 0 1 1 *
@monthlyRun once a month at midnight of the first day of the month0 0 1 * *
@weeklyRun once a week at midnight on Sunday morning0 0 * * 0
@daily (or @midnight)Run once a day at midnight0 0 * * *
@hourlyRun once an hour at the beginning of the hour0 * * * *

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 Jobs
  • Forbid: 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.

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 is 3. Setting this field to 0 will not keep any successful jobs.

  • .spec.failedJobsHistoryLimit: This field specifies the number of failed finished jobs to keep. The default value is 1. Setting this field to 0 will not keep any failed jobs.

For another way to clean up Jobs automatically, see Clean up finished Jobs automatically.

Time zones

FEATURE STATE: 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.

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

Legacy API for managing workloads that can scale horizontally. Superseded by the Deployment and ReplicaSet APIs.

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 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

With autoscaling, you can automatically update your workloads in one way or another. This allows your cluster to react to changes in resource demand more elastically and efficiently.

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.

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

FEATURE STATE: 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.

At the moment, the VPA can operate in four different modes:

Different modes of the VPA
ModeDescription
AutoCurrently, Recreate might change to in-place updates in the future
RecreateThe 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
InitialThe VPA only assigns resource requests on pod creation and never changes them later.
OffThe 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

FEATURE STATE: 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

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

Concepts and resources behind networking in Kubernetes.

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.
  • 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.

  • 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

Expose an application running in your cluster behind a single outward-facing endpoint, even when the workload is split across multiple backends.

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.

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.

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.

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

FEATURE STATE: 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

FEATURE STATE: 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:

  • IANA standard service names.

  • Implementation-defined prefixed names such as mycompany.com/my-custom-protocol.

  • Kubernetes-defined prefixed names:

ProtocolDescription
kubernetes.io/h2cHTTP/2 over cleartext as described in RFC 7540
kubernetes.io/wsWebSocket over cleartext as described in RFC 6455
kubernetes.io/wssWebSocket 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

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 of type: 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 hostname api.foo.bar.example). The mapping configures your cluster's DNS server to return a CNAME 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.)

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.

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

FEATURE STATE: 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.

Disabling load balancer NodePort allocation

FEATURE STATE: 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

FEATURE STATE: 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

FEATURE STATE: 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"