Example of Deployment where pods are pinned to a specific node by name

03 Sep, 2023

I was testing a simple deployment that has only one node. Specifically I was testing if the pods, when deploying, could pull the image.

Looking at the pods I see that it was on node4:

$ kubectl get pods -o wide

NAME                                         READY   STATUS             RESTARTS       AGE     IP       NODE    NOMINATED NODE   READINESS GATES
test-pod-866bc57c6c-l46vk                    1/1     Running            0              42s     xxxxx    node4   <none>           <none>

Checking the pod, shows that the image already exists on the machine:

$ kubectl describe pod metabase-6d4c89846f-h5s94 | tail -n 6

Events:
  Type    Reason   Age    From     Message
  ----    ------   ----   ----     -------
  Normal  Pulled   5m14s  kubelet  Container image "test/test-pod:v0.46.8" already present on machine
  Normal  Created  5m14s  kubelet  Created container test-pod
  Normal  Started  5m14s  kubelet  Started container test-pod

So to get it to run on another node, I just added a node name into the deployment description:

$ kubectl edit deployment test-pod

apiVersion: apps/v1
kind: Pod
metadata:
  <... removed some stuff as it is not necessary for the blog post ...>
  labels:
    component: test-pod
    pod-template-hash: 6d4c89846f
  name: metabase-6d4c89846f-h5s94
  namespace: test-namespace
  <... removed some stuff as it is not necessary for the blog post ...>
spec:
  progressDeadlineSeconds: 600
  replicas: 1
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      component: test-pod
  strategy:
    rollingUpdate:
      maxSurge: 25%
      maxUnavailable: 25%
    type: RollingUpdate
  template:
    metadata:
      creationTimestamp: null
      labels:
        component: test-pod
    spec:
      containers:
      - env:
        - name: DB_PASS
          value: <... you wish ...>
        envFrom:
        - configMapRef:
            name: test-config
        image: test/test-pod:v0.46.8
        imagePullPolicy: IfNotPresent
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
      dnsPolicy: ClusterFirst
      nodeName: node4
      restartPolicy: Always
      <... removed other stuff not necessary for blog post ...>

adding in the nodeName worked fine. This allowed me to pick a node that it wasn't used on and then check to ensure the deployment pulled the image.

Another solution would be to change the imagePullPolicy to Always