Services

• 1: Connecting Applications with Services
• 2: Using Source IP
• 3: Explore Termination Behavior for Pods And Their Endpoints

1 - Connecting Applications with Services

The Kubernetes model for connecting containers

Now that you have a continuously running, replicated application you can expose it on a network.

Kubernetes assumes that pods can communicate with other pods, regardless of which host they land on. Kubernetes gives every pod its own cluster-private IP address, so you do not need to explicitly create links between pods or map container ports to host ports. This means that containers within a Pod can all reach each other's ports on localhost, and all pods in a cluster can see each other without NAT. The rest of this document elaborates on how you can run reliable services on such a networking model.

This tutorial uses a simple nginx web server to demonstrate the concept.

Exposing pods to the cluster

We did this in a previous example, but let's do it once again and focus on the networking perspective. Create an nginx Pod, and note that it has a container port specification:

service/networking/run-my-nginx.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-nginx
spec:
  selector:
    matchLabels:
      run: my-nginx
  replicas: 2
  template:
    metadata:
      labels:
        run: my-nginx
    spec:
      containers:
      - name: my-nginx
        image: nginx
        ports:
        - containerPort: 80

This makes it accessible from any node in your cluster. Check the nodes the Pod is running on:

kubectl apply -f ./run-my-nginx.yaml
kubectl get pods -l run=my-nginx -o wide

NAME                        READY     STATUS    RESTARTS   AGE       IP            NODE
my-nginx-3800858182-jr4a2   1/1       Running   0          13s       10.244.3.4    kubernetes-minion-905m
my-nginx-3800858182-kna2y   1/1       Running   0          13s       10.244.2.5    kubernetes-minion-ljyd

Check your pods' IPs:

kubectl get pods -l run=my-nginx -o custom-columns=POD_IP:.status.podIPs
    POD_IP
    [map[ip:10.244.3.4]]
    [map[ip:10.244.2.5]]

You should be able to ssh into any node in your cluster and use a tool such as curl to make queries against both IPs. Note that the containers are not using port 80 on the node, nor are there any special NAT rules to route traffic to the pod. This means you can run multiple nginx pods on the same node all using the same containerPort, and access them from any other pod or node in your cluster using the assigned IP address for the pod. If you want to arrange for a specific port on the host Node to be forwarded to backing Pods, you can - but the networking model should mean that you do not need to do so.

You can read more about the Kubernetes Networking Model if you're curious.

Creating a Service

So we have pods running nginx in a flat, cluster wide, address space. In theory, you could talk to these pods directly, but what happens when a node dies? The pods die with it, and the ReplicaSet inside the Deployment will create new ones, with different IPs. This is the problem a Service solves.

A Kubernetes Service is an abstraction which defines a logical set of Pods running somewhere in your cluster, that all provide the same functionality. When created, each Service is assigned a unique IP address (also called clusterIP). This address is tied to the lifespan of the Service, and will not change while the Service is alive. Pods can be configured to talk to the Service, and know that communication to the Service will be automatically load-balanced out to some pod that is a member of the Service.

You can create a Service for your 2 nginx replicas with kubectl expose:

kubectl expose deployment/my-nginx

service/my-nginx exposed

This is equivalent to kubectl apply -f in the following yaml:

service/networking/nginx-svc.yaml

apiVersion: v1
kind: Service
metadata:
  name: my-nginx
  labels:
    run: my-nginx
spec:
  ports:
  - port: 80
    protocol: TCP
  selector:
    run: my-nginx

This specification will create a Service which targets TCP port 80 on any Pod with the run: my-nginx label, and expose it on an abstracted Service port (targetPort: is the port the container accepts traffic on, port: is the abstracted Service port, which can be any port other pods use to access the Service). View Service API object to see the list of supported fields in service definition. Check your Service:

kubectl get svc my-nginx

NAME       TYPE        CLUSTER-IP     EXTERNAL-IP   PORT(S)   AGE
my-nginx   ClusterIP   10.0.162.149   <none>        80/TCP    21s

As mentioned previously, a Service is backed by a group of Pods. These Pods are exposed through EndpointSlices. The Service's selector will be evaluated continuously and the results will be POSTed to an EndpointSlice that is connected to the Service using labels. When a Pod dies, it is automatically removed from the EndpointSlices that contain it as an endpoint. New Pods that match the Service's selector will automatically get added to an EndpointSlice for that Service. Check the endpoints, and note that the IPs are the same as the Pods created in the first step:

kubectl describe svc my-nginx

Name:                my-nginx
Namespace:           default
Labels:              run=my-nginx
Annotations:         <none>
Selector:            run=my-nginx
Type:                ClusterIP
IP Family Policy:    SingleStack
IP Families:         IPv4
IP:                  10.0.162.149
IPs:                 10.0.162.149
Port:                <unset> 80/TCP
TargetPort:          80/TCP
Endpoints:           10.244.2.5:80,10.244.3.4:80
Session Affinity:    None
Events:              <none>

kubectl get endpointslices -l kubernetes.io/service-name=my-nginx

NAME             ADDRESSTYPE   PORTS   ENDPOINTS               AGE
my-nginx-7vzhx   IPv4          80      10.244.2.5,10.244.3.4   21s

You should now be able to curl the nginx Service on <CLUSTER-IP>:<PORT> from any node in your cluster. Note that the Service IP is completely virtual, it never hits the wire. If you're curious about how this works you can read more about the service proxy.

Accessing the Service

Kubernetes supports 2 primary modes of finding a Service - environment variables and DNS. The former works out of the box while the latter requires the CoreDNS cluster addon.

Environment Variables

When a Pod runs on a Node, the kubelet adds a set of environment variables for each active Service. This introduces an ordering problem. To see why, inspect the environment of your running nginx Pods (your Pod name will be different):

kubectl exec my-nginx-3800858182-jr4a2 -- printenv | grep SERVICE

KUBERNETES_SERVICE_HOST=10.0.0.1
KUBERNETES_SERVICE_PORT=443
KUBERNETES_SERVICE_PORT_HTTPS=443

Note there's no mention of your Service. This is because you created the replicas before the Service. Another disadvantage of doing this is that the scheduler might put both Pods on the same machine, which will take your entire Service down if it dies. We can do this the right way by killing the 2 Pods and waiting for the Deployment to recreate them. This time the Service exists before the replicas. This will give you scheduler-level Service spreading of your Pods (provided all your nodes have equal capacity), as well as the right environment variables:

kubectl scale deployment my-nginx --replicas=0; kubectl scale deployment my-nginx --replicas=2;

kubectl get pods -l run=my-nginx -o wide

NAME                        READY     STATUS    RESTARTS   AGE     IP            NODE
my-nginx-3800858182-e9ihh   1/1       Running   0          5s      10.244.2.7    kubernetes-minion-ljyd
my-nginx-3800858182-j4rm4   1/1       Running   0          5s      10.244.3.8    kubernetes-minion-905m

You may notice that the pods have different names, since they are killed and recreated.

kubectl exec my-nginx-3800858182-e9ihh -- printenv | grep SERVICE

KUBERNETES_SERVICE_PORT=443
MY_NGINX_SERVICE_HOST=10.0.162.149
KUBERNETES_SERVICE_HOST=10.0.0.1
MY_NGINX_SERVICE_PORT=80
KUBERNETES_SERVICE_PORT_HTTPS=443

DNS

Kubernetes offers a DNS cluster addon Service that automatically assigns dns names to other Services. You can check if it's running on your cluster:

kubectl get services kube-dns --namespace=kube-system

NAME       TYPE        CLUSTER-IP   EXTERNAL-IP   PORT(S)         AGE
kube-dns   ClusterIP   10.0.0.10    <none>        53/UDP,53/TCP   8m

The rest of this section will assume you have a Service with a long lived IP (my-nginx), and a DNS server that has assigned a name to that IP. Here we use the CoreDNS cluster addon (application name kube-dns), so you can talk to the Service from any pod in your cluster using standard methods (e.g. gethostbyname()). If CoreDNS isn't running, you can enable it referring to the CoreDNS README or Installing CoreDNS. Let's run another curl application to test this:

kubectl run curl --image=radial/busyboxplus:curl -i --tty --rm

Waiting for pod default/curl-131556218-9fnch to be running, status is Pending, pod ready: false
Hit enter for command prompt

Then, hit enter and run nslookup my-nginx:

[ root@curl-131556218-9fnch:/ ]$ nslookup my-nginx
Server:    10.0.0.10
Address 1: 10.0.0.10

Name:      my-nginx
Address 1: 10.0.162.149

Securing the Service

Till now we have only accessed the nginx server from within the cluster. Before exposing the Service to the internet, you want to make sure the communication channel is secure. For this, you will need:

• Self signed certificates for https (unless you already have an identity certificate)
• An nginx server configured to use the certificates
• A secret that makes the certificates accessible to pods

You can acquire all these from the nginx https example. This requires having go and make tools installed. If you don't want to install those, then follow the manual steps later. In short:

make keys KEY=/tmp/nginx.key CERT=/tmp/nginx.crt
kubectl create secret tls nginxsecret --key /tmp/nginx.key --cert /tmp/nginx.crt

secret/nginxsecret created

kubectl get secrets

NAME                  TYPE                                  DATA      AGE
nginxsecret           kubernetes.io/tls                     2         1m

And also the configmap:

kubectl create configmap nginxconfigmap --from-file=default.conf

You can find an example for default.conf in the Kubernetes examples project repo.

configmap/nginxconfigmap created

kubectl get configmaps

NAME             DATA   AGE
nginxconfigmap   1      114s

You can view the details of the nginxconfigmap ConfigMap using the following command:

kubectl describe configmap  nginxconfigmap

The output is similar to:

Name:         nginxconfigmap
Namespace:    default
Labels:       <none>
Annotations:  <none>

Data
====
default.conf:
----
server {
        listen 80 default_server;
        listen [::]:80 default_server ipv6only=on;

        listen 443 ssl;

        root /usr/share/nginx/html;
        index index.html;

        server_name localhost;
        ssl_certificate /etc/nginx/ssl/tls.crt;
        ssl_certificate_key /etc/nginx/ssl/tls.key;

        location / {
                try_files $uri $uri/ =404;
        }
}

BinaryData
====

Events:  <none>

Following are the manual steps to follow in case you run into problems running make (on windows for example):

# Create a public private key pair
openssl req -x509 -nodes -days 365 -newkey rsa:2048 -keyout /d/tmp/nginx.key -out /d/tmp/nginx.crt -subj "/CN=my-nginx/O=my-nginx"
# Convert the keys to base64 encoding
cat /d/tmp/nginx.crt | base64
cat /d/tmp/nginx.key | base64

Use the output from the previous commands to create a yaml file as follows. The base64 encoded value should all be on a single line.

apiVersion: "v1"
kind: "Secret"
metadata:
  name: "nginxsecret"
  namespace: "default"
type: kubernetes.io/tls
data:
  tls.crt: "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"
  tls.key: "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"

Now create the secrets using the file:

kubectl apply -f nginxsecrets.yaml
kubectl get secrets

NAME                  TYPE                                  DATA      AGE
nginxsecret           kubernetes.io/tls                     2         1m

Now modify your nginx replicas to start an https server using the certificate in the secret, and the Service, to expose both ports (80 and 443):

service/networking/nginx-secure-app.yaml

apiVersion: v1
kind: Service
metadata:
  name: my-nginx
  labels:
    run: my-nginx
spec:
  type: NodePort
  ports:
  - port: 8080
    targetPort: 80
    protocol: TCP
    name: http
  - port: 443
    protocol: TCP
    name: https
  selector:
    run: my-nginx
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-nginx
spec:
  selector:
    matchLabels:
      run: my-nginx
  replicas: 1
  template:
    metadata:
      labels:
        run: my-nginx
    spec:
      volumes:
      - name: secret-volume
        secret:
          secretName: nginxsecret
      - name: configmap-volume
        configMap:
          name: nginxconfigmap
      containers:
      - name: nginxhttps
        image: bprashanth/nginxhttps:1.0
        ports:
        - containerPort: 443
        - containerPort: 80
        volumeMounts:
        - mountPath: /etc/nginx/ssl
          name: secret-volume
        - mountPath: /etc/nginx/conf.d
          name: configmap-volume

Noteworthy points about the nginx-secure-app manifest:

• It contains both Deployment and Service specification in the same file.
• The nginx server serves HTTP traffic on port 80 and HTTPS traffic on 443, and nginx Service exposes both ports.
• Each container has access to the keys through a volume mounted at /etc/nginx/ssl. This is set up before the nginx server is started.

kubectl delete deployments,svc my-nginx; kubectl create -f ./nginx-secure-app.yaml

At this point you can reach the nginx server from any node.

kubectl get pods -l run=my-nginx -o custom-columns=POD_IP:.status.podIPs
    POD_IP
    [map[ip:10.244.3.5]]

node $ curl -k https://10.244.3.5
...
<h1>Welcome to nginx!</h1>

Note how we supplied the -k parameter to curl in the last step, this is because we don't know anything about the pods running nginx at certificate generation time, so we have to tell curl to ignore the CName mismatch. By creating a Service we linked the CName used in the certificate with the actual DNS name used by pods during Service lookup. Let's test this from a pod (the same secret is being reused for simplicity, the pod only needs nginx.crt to access the Service):

service/networking/curlpod.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: curl-deployment
spec:
  selector:
    matchLabels:
      app: curlpod
  replicas: 1
  template:
    metadata:
      labels:
        app: curlpod
    spec:
      volumes:
      - name: secret-volume
        secret:
          secretName: nginxsecret
      containers:
      - name: curlpod
        command:
        - sh
        - -c
        - while true; do sleep 1; done
        image: radial/busyboxplus:curl
        volumeMounts:
        - mountPath: /etc/nginx/ssl
          name: secret-volume

kubectl apply -f ./curlpod.yaml
kubectl get pods -l app=curlpod

NAME                               READY     STATUS    RESTARTS   AGE
curl-deployment-1515033274-1410r   1/1       Running   0          1m

kubectl exec curl-deployment-1515033274-1410r -- curl https://my-nginx --cacert /etc/nginx/ssl/tls.crt
...
<title>Welcome to nginx!</title>
...

Exposing the Service

For some parts of your applications you may want to expose a Service onto an external IP address. Kubernetes supports two ways of doing this: NodePorts and LoadBalancers. The Service created in the last section already used NodePort, so your nginx HTTPS replica is ready to serve traffic on the internet if your node has a public IP.

kubectl get svc my-nginx -o yaml | grep nodePort -C 5
  uid: 07191fb3-f61a-11e5-8ae5-42010af00002
spec:
  clusterIP: 10.0.162.149
  ports:
  - name: http
    nodePort: 31704
    port: 8080
    protocol: TCP
    targetPort: 80
  - name: https
    nodePort: 32453
    port: 443
    protocol: TCP
    targetPort: 443
  selector:
    run: my-nginx

kubectl get nodes -o yaml | grep ExternalIP -C 1
    - address: 104.197.41.11
      type: ExternalIP
    allocatable:
--
    - address: 23.251.152.56
      type: ExternalIP
    allocatable:
...

$ curl https://<EXTERNAL-IP>:<NODE-PORT> -k
...
<h1>Welcome to nginx!</h1>

Let's now recreate the Service to use a cloud load balancer. Change the Type of my-nginx Service from NodePort to LoadBalancer:

kubectl edit svc my-nginx
kubectl get svc my-nginx

NAME       TYPE           CLUSTER-IP     EXTERNAL-IP        PORT(S)               AGE
my-nginx   LoadBalancer   10.0.162.149     xx.xxx.xxx.xxx     8080:30163/TCP        21s

curl https://<EXTERNAL-IP> -k
...
<title>Welcome to nginx!</title>

The IP address in the EXTERNAL-IP column is the one that is available on the public internet. The CLUSTER-IP is only available inside your cluster/private cloud network.

Note that on AWS, type LoadBalancer creates an ELB, which uses a (long) hostname, not an IP. It's too long to fit in the standard kubectl get svc output, in fact, so you'll need to do kubectl describe service my-nginx to see it. You'll see something like this:

kubectl describe service my-nginx
...
LoadBalancer Ingress:   a320587ffd19711e5a37606cf4a74574-1142138393.us-east-1.elb.amazonaws.com
...

What's next

• Learn more about Using a Service to Access an Application in a Cluster
• Learn more about Connecting a Front End to a Back End Using a Service
• Learn more about Creating an External Load Balancer

2 - Using Source IP

Applications running in a Kubernetes cluster find and communicate with each other, and the outside world, through the Service abstraction. This document explains what happens to the source IP of packets sent to different types of Services, and how you can toggle this behavior according to your needs.

Before you begin

Terminology

This document makes use of the following terms:

NAT

Network address translation

Source NAT

Replacing the source IP on a packet; in this page, that usually means replacing with the IP address of a node.

Destination NAT

Replacing the destination IP on a packet; in this page, that usually means replacing with the IP address of a Pod

VIP

A virtual IP address, such as the one assigned to every Service in Kubernetes

kube-proxy

A network daemon that orchestrates Service VIP management on every node

Prerequisites

You need to have a Kubernetes cluster, and the kubectl command-line tool must be configured to communicate with your cluster. It is recommended to run this tutorial on a cluster with at least two nodes that are not acting as control plane hosts. If you do not already have a cluster, you can create one by using minikube or you can use one of these Kubernetes playgrounds:

• Killercoda
• Play with Kubernetes

The examples use a small nginx webserver that echoes back the source IP of requests it receives through an HTTP header. You can create it as follows:

kubectl create deployment source-ip-app --image=registry.k8s.io/echoserver:1.10

The output is:

deployment.apps/source-ip-app created

Objectives

• Expose a simple application through various types of Services
• Understand how each Service type handles source IP NAT
• Understand the tradeoffs involved in preserving source IP

Source IP for Services with Type=ClusterIP

Packets sent to ClusterIP from within the cluster are never source NAT'd if you're running kube-proxy in iptables mode, (the default). You can query the kube-proxy mode by fetching http://localhost:10249/proxyMode on the node where kube-proxy is running.

kubectl get nodes

The output is similar to this:

NAME                           STATUS     ROLES    AGE     VERSION
kubernetes-node-6jst   Ready      <none>   2h      v1.13.0
kubernetes-node-cx31   Ready      <none>   2h      v1.13.0
kubernetes-node-jj1t   Ready      <none>   2h      v1.13.0

Get the proxy mode on one of the nodes (kube-proxy listens on port 10249):

# Run this in a shell on the node you want to query.
curl http://localhost:10249/proxyMode

The output is:

iptables

You can test source IP preservation by creating a Service over the source IP app:

kubectl expose deployment source-ip-app --name=clusterip --port=80 --target-port=8080

The output is:

service/clusterip exposed

kubectl get svc clusterip

The output is similar to:

NAME         TYPE        CLUSTER-IP    EXTERNAL-IP   PORT(S)   AGE
clusterip    ClusterIP   10.0.170.92   <none>        80/TCP    51s

And hitting the ClusterIP from a pod in the same cluster:

kubectl run busybox -it --image=busybox:1.28 --restart=Never --rm

The output is similar to this:

Waiting for pod default/busybox to be running, status is Pending, pod ready: false
If you don't see a command prompt, try pressing enter.

You can then run a command inside that Pod:

# Run this inside the terminal from "kubectl run"
ip addr

1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
3: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1460 qdisc noqueue
    link/ether 0a:58:0a:f4:03:08 brd ff:ff:ff:ff:ff:ff
    inet 10.244.3.8/24 scope global eth0
       valid_lft forever preferred_lft forever
    inet6 fe80::188a:84ff:feb0:26a5/64 scope link
       valid_lft forever preferred_lft forever

…then use wget to query the local webserver

# Replace "10.0.170.92" with the IPv4 address of the Service named "clusterip"
wget -qO - 10.0.170.92

CLIENT VALUES:
client_address=10.244.3.8
command=GET
...

The client_address is always the client pod's IP address, whether the client pod and server pod are in the same node or in different nodes.

Source IP for Services with Type=NodePort

Packets sent to Services with Type=NodePort are source NAT'd by default. You can test this by creating a NodePort Service:

kubectl expose deployment source-ip-app --name=nodeport --port=80 --target-port=8080 --type=NodePort

The output is:

service/nodeport exposed

NODEPORT=$(kubectl get -o jsonpath="{.spec.ports[0].nodePort}" services nodeport)
NODES=$(kubectl get nodes -o jsonpath='{ $.items[*].status.addresses[?(@.type=="InternalIP")].address }')

If you're running on a cloud provider, you may need to open up a firewall-rule for the nodes:nodeport reported above. Now you can try reaching the Service from outside the cluster through the node port allocated above.

for node in $NODES; do curl -s $node:$NODEPORT | grep -i client_address; done

The output is similar to:

client_address=10.180.1.1
client_address=10.240.0.5
client_address=10.240.0.3

Note that these are not the correct client IPs, they're cluster internal IPs. This is what happens:

• Client sends packet to node2:nodePort
• node2 replaces the source IP address (SNAT) in the packet with its own IP address
• node2 replaces the destination IP on the packet with the pod IP
• packet is routed to node 1, and then to the endpoint
• the pod's reply is routed back to node2
• the pod's reply is sent back to the client

Visually:

To avoid this, Kubernetes has a feature to preserve the client source IP. If you set service.spec.externalTrafficPolicy to the value Local, kube-proxy only proxies proxy requests to local endpoints, and does not forward traffic to other nodes. This approach preserves the original source IP address. If there are no local endpoints, packets sent to the node are dropped, so you can rely on the correct source-ip in any packet processing rules you might apply a packet that make it through to the endpoint.

Set the service.spec.externalTrafficPolicy field as follows:

kubectl patch svc nodeport -p '{"spec":{"externalTrafficPolicy":"Local"}}'

The output is:

service/nodeport patched

Now, re-run the test:

for node in $NODES; do curl --connect-timeout 1 -s $node:$NODEPORT | grep -i client_address; done

The output is similar to:

client_address=198.51.100.79

Note that you only got one reply, with the right client IP, from the one node on which the endpoint pod is running.

This is what happens:

• client sends packet to node2:nodePort, which doesn't have any endpoints
• packet is dropped
• client sends packet to node1:nodePort, which does have endpoints
• node1 routes packet to endpoint with the correct source IP

Visually:

Source IP for Services with Type=LoadBalancer

Packets sent to Services with Type=LoadBalancer are source NAT'd by default, because all schedulable Kubernetes nodes in the Ready state are eligible for load-balanced traffic. So if packets arrive at a node without an endpoint, the system proxies it to a node with an endpoint, replacing the source IP on the packet with the IP of the node (as described in the previous section).

You can test this by exposing the source-ip-app through a load balancer:

kubectl expose deployment source-ip-app --name=loadbalancer --port=80 --target-port=8080 --type=LoadBalancer

The output is:

service/loadbalancer exposed

Print out the IP addresses of the Service:

kubectl get svc loadbalancer

The output is similar to this:

NAME           TYPE           CLUSTER-IP    EXTERNAL-IP       PORT(S)   AGE
loadbalancer   LoadBalancer   10.0.65.118   203.0.113.140     80/TCP    5m

Next, send a request to this Service's external-ip:

curl 203.0.113.140

The output is similar to this:

CLIENT VALUES:
client_address=10.240.0.5
...

However, if you're running on Google Kubernetes Engine/GCE, setting the same service.spec.externalTrafficPolicy field to Local forces nodes without Service endpoints to remove themselves from the list of nodes eligible for loadbalanced traffic by deliberately failing health checks.

Visually:

You can test this by setting the annotation:

kubectl patch svc loadbalancer -p '{"spec":{"externalTrafficPolicy":"Local"}}'

You should immediately see the service.spec.healthCheckNodePort field allocated by Kubernetes:

kubectl get svc loadbalancer -o yaml | grep -i healthCheckNodePort

The output is similar to this:

  healthCheckNodePort: 32122

The service.spec.healthCheckNodePort field points to a port on every node serving the health check at /healthz. You can test this:

kubectl get pod -o wide -l app=source-ip-app

The output is similar to this:

NAME                            READY     STATUS    RESTARTS   AGE       IP             NODE
source-ip-app-826191075-qehz4   1/1       Running   0          20h       10.180.1.136   kubernetes-node-6jst

Use curl to fetch the /healthz endpoint on various nodes:

# Run this locally on a node you choose
curl localhost:32122/healthz

1 Service Endpoints found

On a different node you might get a different result:

# Run this locally on a node you choose
curl localhost:32122/healthz

No Service Endpoints Found

A controller running on the control plane is responsible for allocating the cloud load balancer. The same controller also allocates HTTP health checks pointing to this port/path on each node. Wait about 10 seconds for the 2 nodes without endpoints to fail health checks, then use curl to query the IPv4 address of the load balancer:

curl 203.0.113.140

The output is similar to this:

CLIENT VALUES:
client_address=198.51.100.79
...

Cross-platform support

Only some cloud providers offer support for source IP preservation through Services with Type=LoadBalancer. The cloud provider you're running on might fulfill the request for a loadbalancer in a few different ways:

1. With a proxy that terminates the client connection and opens a new connection to your nodes/endpoints. In such cases the source IP will always be that of the cloud LB, not that of the client.

2. With a packet forwarder, such that requests from the client sent to the loadbalancer VIP end up at the node with the source IP of the client, not an intermediate proxy.

Load balancers in the first category must use an agreed upon protocol between the loadbalancer and backend to communicate the true client IP such as the HTTP Forwarded or X-FORWARDED-FOR headers, or the proxy protocol. Load balancers in the second category can leverage the feature described above by creating an HTTP health check pointing at the port stored in the service.spec.healthCheckNodePort field on the Service.

Cleaning up

Delete the Services:

kubectl delete svc -l app=source-ip-app

Delete the Deployment, ReplicaSet and Pod:

kubectl delete deployment source-ip-app

What's next

• Learn more about connecting applications via services
• Read how to Create an External Load Balancer

3 - Explore Termination Behavior for Pods And Their Endpoints

Once you connected your Application with Service following steps like those outlined in Connecting Applications with Services, you have a continuously running, replicated application, that is exposed on a network. This tutorial helps you look at the termination flow for Pods and to explore ways to implement graceful connection draining.

Termination process for Pods and their endpoints

There are often cases when you need to terminate a Pod - be it to upgrade or scale down. In order to improve application availability, it may be important to implement a proper active connections draining.

This tutorial explains the flow of Pod termination in connection with the corresponding endpoint state and removal by using a simple nginx web server to demonstrate the concept.

Example flow with endpoint termination

The following is the example flow described in the Termination of Pods document.

Let's say you have a Deployment containing a single nginx replica (say just for the sake of demonstration purposes) and a Service:

service/pod-with-graceful-termination.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 1
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      terminationGracePeriodSeconds: 120 # extra long grace period
      containers:
      - name: nginx
        image: nginx:latest
        ports:
        - containerPort: 80
        lifecycle:
          preStop:
            exec:
              # Real life termination may take any time up to terminationGracePeriodSeconds.
              # In this example - just hang around for at least the duration of terminationGracePeriodSeconds,
              # at 120 seconds container will be forcibly terminated.
              # Note, all this time nginx will keep processing requests.
              command: [
                "/bin/sh", "-c", "sleep 180"
              ]

service/explore-graceful-termination-nginx.yaml

apiVersion: v1
kind: Service
metadata:
  name: nginx-service
spec:
  selector:
    app: nginx
  ports:
    - protocol: TCP
      port: 80
      targetPort: 80

Now create the Deployment Pod and Service using the above files:

kubectl apply -f pod-with-graceful-termination.yaml
kubectl apply -f explore-graceful-termination-nginx.yaml

Once the Pod and Service are running, you can get the name of any associated EndpointSlices:

kubectl get endpointslice

The output is similar to this:

NAME                  ADDRESSTYPE   PORTS   ENDPOINTS                 AGE
nginx-service-6tjbr   IPv4          80      10.12.1.199,10.12.1.201   22m

You can see its status, and validate that there is one endpoint registered:

kubectl get endpointslices -o json -l kubernetes.io/service-name=nginx-service

The output is similar to this:

{
    "addressType": "IPv4",
    "apiVersion": "discovery.k8s.io/v1",
    "endpoints": [
        {
            "addresses": [
                "10.12.1.201"
            ],
            "conditions": {
                "ready": true,
                "serving": true,
                "terminating": false

Now let's terminate the Pod and validate that the Pod is being terminated respecting the graceful termination period configuration:

kubectl delete pod nginx-deployment-7768647bf9-b4b9s

All pods:

kubectl get pods

The output is similar to this:

NAME                                READY   STATUS        RESTARTS      AGE
nginx-deployment-7768647bf9-b4b9s   1/1     Terminating   0             4m1s
nginx-deployment-7768647bf9-rkxlw   1/1     Running       0             8s

You can see that the new pod got scheduled.

While the new endpoint is being created for the new Pod, the old endpoint is still around in the terminating state:

kubectl get endpointslice -o json nginx-service-6tjbr

The output is similar to this:

{
    "addressType": "IPv4",
    "apiVersion": "discovery.k8s.io/v1",
    "endpoints": [
        {
            "addresses": [
                "10.12.1.201"
            ],
            "conditions": {
                "ready": false,
                "serving": true,
                "terminating": true
            },
            "nodeName": "gke-main-default-pool-dca1511c-d17b",
            "targetRef": {
                "kind": "Pod",
                "name": "nginx-deployment-7768647bf9-b4b9s",
                "namespace": "default",
                "uid": "66fa831c-7eb2-407f-bd2c-f96dfe841478"
            },
            "zone": "us-central1-c"
        },
        {
            "addresses": [
                "10.12.1.202"
            ],
            "conditions": {
                "ready": true,
                "serving": true,
                "terminating": false
            },
            "nodeName": "gke-main-default-pool-dca1511c-d17b",
            "targetRef": {
                "kind": "Pod",
                "name": "nginx-deployment-7768647bf9-rkxlw",
                "namespace": "default",
                "uid": "722b1cbe-dcd7-4ed4-8928-4a4d0e2bbe35"
            },
            "zone": "us-central1-c"

This allows applications to communicate their state during termination and clients (such as load balancers) to implement connection draining functionality. These clients may detect terminating endpoints and implement a special logic for them.

In Kubernetes, endpoints that are terminating always have their ready status set as false. This needs to happen for backward compatibility, so existing 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.

When Pod is deleted, the old endpoint will also be deleted.

What's next

• Learn how to Connect Applications with Services
• Learn more about Using a Service to Access an Application in a Cluster
• Learn more about Connecting a Front End to a Back End Using a Service
• Learn more about Creating an External Load Balancer