A couple of state-dumping issues in OpenShift E2E test workflow

[MikeSpreitzer]

It Deletes a Pod Before Showing Its Logs

llm-d-fast-model-actuation/.github/workflows/ci-e2e-openshift.yaml

Line 913 in c47c25b

kubectl scale deployment "$FMA_RELEASE_NAME" -n "$FMA_NAMESPACE" --replicas=0 || true

deletes the one and only controller Pod. Later,

llm-d-fast-model-actuation/.github/workflows/ci-e2e-openshift.yaml

Lines 927 to 928 in c47c25b

	kubectl logs deployment/"$FMA_RELEASE_NAME" -n "$FMA_NAMESPACE" --previous 2>/dev/null || true
	kubectl logs deployment/"$FMA_RELEASE_NAME" -n "$FMA_NAMESPACE" 2>/dev/null || true

tries to display log from the latest and previous container run in the now-non-existent Pod.

It Only Shows State If There Was a Failure

It would be better to dump the state unconditionally, so that we have a good baseline to compare with when diagnosing a failure.

State is Not Deleted on Failure

But normal developers are not authorized to see the state in the failure case. So the state left around is just junk, serving no purpose.

[MikeSpreitzer]

@Copilot: In more prescriptive terms: this issue would be resolved by restructuring the workflow so that it does state dumping unconditionally and then does an unconditional deletion of the test objects (including the namespace).

I recommend that each kind of state be dumped in its own job step, so that each kind of state appears under a separate twistie in GitHub's web UI for workflow runs.