OLM doesn't revert modified deployment of operator

[zkyle95]

Bug Report

What did you do?

1. Subscribe an operator on operatorhub console

2. Scale down operator deployment

What did you expect to see?

Deployment be scaled up or rolled back soon.

What did you see instead? Under which circumstances?

The replicas of deployment keep 0 and CSV phase remains succeed

Environment

• operator-lifecycle-manager version:

  0e65642

• Kubernetes version information:

  v1.19.1

• Kubernetes cluster kind:

  Kubernetes created with kind

Possible Solution

Additional context

Remove the deployment and it will be rebuild soon.

[zkyle95]

/kind bug

[zkyle95]

I found that OLM compares the hash set in label of existing deployment with calculated from CSV.spec.install.spec.deployments, but both of these hash values won't be updated when deployment is modified directly, which made them equled always.

operator-lifecycle-manager/pkg/controller/install/deployment.go

Line 281 in 0e65642

_, calculatedDeploymentHash, err := i.deploymentForSpec(spec.Name, spec.Spec)

I'd like to solve this bug if my analysis is right 😃 @awgreene

[awgreene]

Thank you for submitting this bug @zkyle95 - after reading your most recent comment, I believe that OLM is working as intended.

OLM uses the deploymentHash to ensure that the resources created by OLM match those defined in the CSV. This approach allows a user to edit the resources once they are installed and prevents OLM from entering an endless loop when a webhook is changing the resource before it is saved to the object store.

I am planning to close this issue as things are working as intended.