Kubernetes：kubectl apply在使用“latest”标签时不会更新pod

Question

I'm using kubectl apply to update my Kubernetes pods:

kubectl apply -f /my-app/service.yaml
kubectl apply -f /my-app/deployment.yaml

Below is my service.yaml:

apiVersion: v1
kind: Service
metadata:
  name: my-app
  labels:
    run: my-app
spec:
  type: NodePort
  selector:
    run: my-app 
  ports:
  - protocol: TCP
    port: 9000
    nodePort: 30769

Below is my deployment.yaml:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-app
spec:  
  selector:
    matchLabels:
      run: my-app
  replicas: 2
  template:
    metadata:
      labels:
        run: my-app
    spec:
      containers:
      - name: my-app
        image: dockerhubaccount/my-app-img:latest
        ports:
        - containerPort: 9000
          protocol: TCP
      imagePullSecrets:
      - name: my-app-img-credentials
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxUnavailable: 25%
      maxSurge: 25%

This works fine the first time, but on subsequent runs, my pods are not getting updated.

I have read the suggested workaround at https://github.com/kubernetes/kubernetes/issues/33664 which is:

kubectl patch deployment my-app -p "{\"spec\":{\"template\":{\"metadata\":{\"labels\":{\"date\":\"`date +'%s'`\"}}}}}"

I was able to run the above command, but it did not resolve the issue for me.

I know that I can trigger pod updates by manually changing the image tag from "latest" to another tag, but I want to make sure I get the latest image without having to check Docker Hub.

Any help would be greatly appreciated.

Answer 1

If nothing changes in the deployment spec, the pods will not be updated for you. This is one of many reasons it is not recommended to use :latest , as the other answer went into more detail on. The Deployment controller is very simple and pretty much just does DeepEquals(old.Spec.Template, new.Spec.Template) , so you need some actual change, such as you have with the PATCH call by setting a label with the current datetime.

Answer 2

You're missing an imagePullPolicy in your deployment. Try this:

containers:
- name: my-app
  image: dockerhubaccount/my-app-img:latest
  imagePullPolicy: Always

The default policy is ifNotPresent which is why yours is not updating.

I will incorporate two notes from the link:

Note: You should avoid using the :latest tag when deploying containers in production as it is harder to track which version of the image is running and more difficult to roll back properly

Note: The caching semantics of the underlying image provider make even imagePullPolicy: Always efficient. With Docker, for example, if the image already exists, the pull attempt is fast because all image layers are cached and no image download is needed

Answer 3

Turns out I misunderstood the workaround command I gave from the link.

I thought it was a one-time command that configured my deployment to treat all future kubectl apply commands as a trigger to update my pods.

I actually just had to run the command every time I wanted to update my pods:

kubectl patch deployment my-app -p "{\"spec\":{\"template\":{\"metadata\":{\"labels\":{\"date\":\"`date +'%s'`\"}}}}}"

Many thanks to everyone who helped!

Answer 4

There are two things here that relates the issue,

1. It is suggested to use kubectl apply for the first time while creating a resource and later recommended to use kubectl replace or kubectl edit or kubectl patch commands which subsequently call the kubectl apply .

2. Once you create a service using either kubectl apply or kubetcl create you cannot replace that service with a yaml file. In other words, service generates a random IP that cannot be patched or replaced . The only option to recreate a service is to delete the service and recreate it with the same name.

NOTE: When I tried replacing a service using kubectl apply command while I was trying to create a backup and restore solution resulted this below error.

kubectl apply -f replace-service.yaml -n restore-proj
The Service "test-q12" is invalid: spec.clusterIP: Invalid value: "10.102.x.x": provided IP is already allocated.