jojochuang commented on PR #11727: URL: https://github.com/apache/ozone/pull/11727#issuecomment-6480718171
I screened all 324 unresolved issues returned by the Jira filter for `Generated with Specula`. These are the closest related reports; all were Open when checked: | Existing report | Relationship to this PR | | --- | --- | | [HDDS-17267](https://issues.apache.org/jira/browse/HDDS-17267) — Lost CLOSE actions | **Direct reliability dependency:** can lose this PR's volume-failure notification. | | [HDDS-17269](https://issues.apache.org/jira/browse/HDDS-17269) — Queued heartbeats run after re-registration is requested | Another way to drain and lose pending CLOSE actions. | | [HDDS-17064](https://issues.apache.org/jira/browse/HDDS-17064) — Failed heartbeat loses the full container report | SCM can retain the lost replica in its accounting, delaying repair even if closure occurs. | | [HDDS-17062](https://issues.apache.org/jira/browse/HDDS-17062), [HDDS-17063](https://issues.apache.org/jira/browse/HDDS-17063) — Stale ICRs follow the failure FCR | Can re-add the lost replica to SCM's view and delay reconstruction. | | [HDDS-17018](https://issues.apache.org/jira/browse/HDDS-17018) — Container added after the failed-volume sweep | Such a container escapes this PR's sweep and its proactive CLOSE action. | | [HDDS-17019](https://issues.apache.org/jira/browse/HDDS-17019) — DB-volume failure not propagated to data volumes | The data-volume failure hook may never run for affected containers. | | [HDDS-16906](https://issues.apache.org/jira/browse/HDDS-16906) — Dead-member EC pipeline survives SCM leadership changes | Similar allocation/repair symptoms, but a different trigger. A dead datanode cannot send this PR's notification. | These are existing adjacent defects, not regressions introduced by this PR. HDDS-17267 is the most important companion fix for making closure without subsequent writes dependable: the volume-failure hook queues the CLOSE action once and removes the replica locally; a failed heartbeat or SCM re-registration response can discard the action. A later failed EC write re-adds it, but an idle container can remain OPEN. Failure/re-registration coverage would help establish the retry behavior. The earlier recovery-test concern has been addressed: the integration test now waits for CLOSED state and restoration of all five EC indexes. On head `cbb84e256b4`, the 24 focused unit tests and `TestECContainerCloseOnVolumeFailure` passed locally. Generated-by: Codex (GPT-6) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
