TaskRun stays Running when pod goes to ImagePullBackOff

[dibyom]

Expected Behavior

If the Pod is in ImagePullBackOff state, the TaskRun should fail.

Actual Behavior

The TaskRun's remains "Running" with the message Pending until it eventually times out

Steps to Reproduce the Problem

apiVersion: tekton.dev/v1beta1
kind: TaskRun
metadata:
  generateName: imagepull-fail
spec:
  taskSpec:
    steps:
    - image: whatever
      script: 'true'

[lbernick]

Duplicate of #4890

[imjasonh]

I'm not sure this is a duplicate of #4890, since that one deals with InvalidImageName (e.g., image: "t0T/Al+y//In\/@l1D") vs ImagePullBackoff (image name is valid, but can't be pulled).

In the latter case, I think we've discussed this before, and decided to follow K8s' behavior where it continues to try to pull the image with backoff, until the image exists and is pullable. This can be useful if you accidentally forget to push the image you're trying to run, or don't have auth set up correctly yet.

In K8s's case it will happily sit back and pull forever since there's no concept of a Pod timeout. In Tekton it might make sense to outright fail fast rather than give users an opportunity to fix it.

I suspect this would only work at all today when the image specifies script or command, since otherwise the Pod translation code will try to lookup the image's entrypoint and fail. I don't think it retries, it just gives up. If that's the case, then I think we should aim for consistency and fail fast if the image can't be pulled, even if entrypoint resolution isn't required.

[RafaeLeal]

I'm also suffering from this same issue. There are errors on the pod level such as ImagePullBackoff and FailedMount that make sense for k8s to keep retrying. Still, in a more ephemeral scenario like a CICD Task, most cases are typos and configuration errors that will not recover and I believe it would be nice if Tekton allows us to fail fast those.

[chitrangpatel]

/assign

[chitrangpatel]

cc @dibyom @imjasonh @lbernick : PR #4921 addresses this.

[RafaeLeal]

Should I open a different issue to discuss the FailedMount error?

[dibyom]

@RafaeLeal yes please!

[dibyom]

Closing since the ImagePullBackoff case is now handled