fix: register event sources when dependents are marked for deletion by SamBarker · Pull Request #3250 · operator-framework/java-operator-sdk

…teExecutor

When a BulkDependentResource has an activationCondition and its parent
dependent has a failing reconcilePrecondition, JOSDK's
markDependentsForDelete() cascades to the bulk dependent and fires
NodeDeleteExecutor for it. However, NodeDeleteExecutor does not call
registerOrDeregisterEventSourceBasedOnActivation() before invoking
delete(), so if NodeReconcileExecutor has never run for that node (e.g.
on first reconciliation) the event source is never registered.

The delete() path calls getSecondaryResources() → eventSourceRetriever
.getEventSourceFor() → NoEventSourceForClassException.

This IT demonstrates the bug with a minimal workflow:
  ConfigMapDependentResource  (reconcilePrecondition = ALWAYS_FALSE)
    └── SecretBulkDependentResource  (activationCondition = ALWAYS_TRUE)

The fix is to call registerOrDeregisterEventSourceBasedOnActivation()
in NodeDeleteExecutor.doRun() before calling dependent.delete(),
mirroring what NodeReconcileExecutor already does.

Assisted-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: Sam Barker <sam@quadrocket.co.uk>

- Make lastError and callCount instance fields on the reconciler; hold
  the reconciler instance as a static field in the IT so tests can
  access state without static leakage between tests
- Add callCount so the test can wait for any reconciliation activity
  (reconcile() or updateErrorStatus()) then assert cleanly, rather than
  timing out if the bug is fixed
- Add @disabled linking to issue operator-framework#3249 so this reproducer-only test
  does not break CI
- Add @sample annotation to match the pattern of other workflow ITs

Assisted-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: Sam Barker <sam@quadrocket.co.uk>