improve: log diagnostic for remote deployment junit extension

[csviri]

adds also option to not delete CRDs

Signed-off-by: Attila Mészáros a_meszaros@apple.com

[Copilot]

Pull request overview

Adds richer diagnostics when a cluster-deployed operator times out during deployment, to make remote E2E failures easier to debug.

Changes:

• Catch KubernetesClientTimeoutException during waitUntilReady and emit diagnostics before rethrowing.
• Add logDiagnosticInfo(...) to report deployments, pods, container statuses, related events, and recent pod logs on timeout.

[Copilot]

When pod log retrieval fails, this only logs the exception message and drops the stack trace, which makes diagnosing client/auth/network issues harder. Consider logging the exception itself as the last argument (similar to the diagEx handling below) so the full cause is available when needed.

Suggested change

	" Could not retrieve logs for pod '{}': {}",
	pod.getMetadata().getName(),
	logEx.getMessage());
	" Could not retrieve logs for pod '{}'",
	pod.getMetadata().getName(),
	logEx);

[csviri]

@metacosm @xstefank I restarted a couple times the tomcat E2E test, but it does not want to fail, however this might be useful in general to log issues such a way as in this PR. So we could merge this an see if it fails again.

What do you think?

[xstefank]

OK with me

[csviri]

Now it fails, this seems to be an issue that I also observed with the operations sample PR:
#3291

It seems that CRD is not applied in some cases, will investigate further. This might be actually because of junit extension issue - maybe with newer version of junit? - since we never observed this before.

Or maybe with your recent changes @xstefank ?

[xstefank]

@csviri there is only one test in TomcatOperatorE2ETest so I don't think deleting of the CRD is the issue.

[csviri]

Will mere

@csviri there is only one test in TomcatOperatorE2ETest so I don't think deleting of the CRD is the issue.

No, not the deletion, just thought it might ring some bell, nvm

[csviri]

@csviri there is only one test in TomcatOperatorE2ETest so I don't think deleting of the CRD is the issue.

Yes but there are two test modes local and remote, so that will be the problem. @xstefank

[csviri]

I will do a nicer opt-out from that feature for E2Es

[xstefank]

@csviri all tests passed with the CRD deleting after the test class is executed in my PR. So I don't understand why you think that it caused failures in this PR? Local and Cluster tests run separatedly, so they should both restart the operator and redeploy the CRD. I don't think leaving some CRDs like it is proposed in this PR is good.

[csviri]

@csviri all tests passed with the CRD deleting after the test class is executed in my PR. So I don't understand why you think that it caused failures in this PR? Local and Cluster tests run separatedly, so they should both restart the operator and redeploy the CRD. I don't think leaving some CRDs like it is proposed in this PR is good.

@xstefank they are running against the same cluster, there is probably an issue with cluster test runner regarding the CRD, I will take a look on that later; for this fixes the issue. Also this api to turn off CRD deletion might help for others too. Will continue on this in this PR.

[csviri]

@xstefank btw, currently the deletion of CRD-s would great to handle uniformly also for ClusterDeployedOperatorExtension so implementing it in AbstractOperatorExtension, would you care to create a PR for that?

[csviri]

@xstefank I addressed the CRD issue in separate PR what we discussed. Removed the deletion opt-out, but let there the option for the future / if some might need it.

[csviri]

Delete CRD flag I will move to a separate PR.