The GitHub Actions job "Flink Kubernetes Operator CI" on flink-kubernetes-operator.git/FLINK-40955 has succeeded. Run started by GitHub user gyfora (triggered by gyfora).
Head commit for run: 4ee70a3a03f18a35bf17295940b4e11379843f60 / Gyula Fora <[email protected]> [FLINK-40955] Give up and remove finalizer after cleanup keeps failing If cleanup keeps failing (e.g. the cluster resources can never be deleted because a pod is stuck Terminating), the failure was retried by JOSDK forever: the finalizer was never removed, the CR stayed pinned in Terminating permanently, and every future redeploy under the same name failed with a 409. Cleanup failures are now propagated so that JOSDK retries them with its regular backoff. On a retry, once the resource has been pending deletion for longer than kubernetes.operator.resource.cleanup.finalizer.timeout (default 120m, measured from metadata.deletionTimestamp), the operator gives up: it emits a Warning/CleanupFailed event and removes the finalizer. The first failed attempt never gives up and a timeout of 0 disables giving up altogether. If the JOSDK retries are exhausted before the timeout the cleanup is rescheduled so that it keeps being retried. This accepts that resources may be orphaned in exchange for keeping the resource name redeployable. The status changes of a failed attempt are patched before the error is rethrown so that they are not lost between retries. Related changes: - Resources that were never deployed are cleaned up right away in the controllers without observing or reconciling, there is nothing to clean up. - A job cancellation that does not complete within resource.cleanup.timeout is considered stuck: session job deletion stops waiting for it (emitting a warning event with the job id, as the job may still be running) and applications fall back to an HA metadata based last-state upgrade when that is allowed. Report URL: https://github.com/apache/flink-kubernetes-operator/actions/runs/41888665334 With regards, GitHub Actions via GitBox
