Kubernetes - Old replica sets remains after deployment

In this post i will write about annoying Kubernetes problem I managed to solve (at least I think it solved)
after long struggle.

So, here is the problem:

After running update-deployment-image command:

kubectl set image deployment/my-app my-app=gcr.io/my-cloud/my-app:new-image-tag 

The kubernetes showed that although the new image successfully updated, the old replicaset still remains:

Old replicates are all with 0 pods, but not removed.

Solution

The solution I found is to change the "selector" node in deployment configuration -
From:

  selector:
    matchLabels:
      app: my-app
      version: $SHORT_SHA

To

  selector:
    matchLabels:
      app: my-app

(Remove the "version" label)
And to add revisionHistoryLimit entry

spec:
  replicas: 2
  revisionHistoryLimit: 0
  selector:
    matchLabels:
      app: my-app