VladRodionov commented on PR #8845: URL: https://github.com/apache/hbase/pull/8845#issuecomment-6999831403
@kgeisz If you fix this one, your new feature is going to be bullet-proof (probably the most robust one in HBase :) **Request changes: one remaining P1 issue** in head `048376fc59ad`. The three original findings are fixed at their immediate call sites. **[P1] Preserve repair state before propagating marker-deletion failures.** The new exception from [[TableBackupClient.java:196–197] escapes **after** `cleanupAndRestoreBackupSystem()` restores and deletes the system-table snapshot. That snapshot was taken after acquiring the exclusive lock but before recording the running backup. Consequently, when marker deletion fails: 1. Restoring the snapshot removes the `RUNNING` backup record. 2. The exception prevents persisting `FAILED` and releasing the lock. 3. A subsequent `backup repair` finds no running backup or pending delete/merge operation, so it never retries cleanup or releases the lock. A transient filesystem failure can therefore leave backup operations blocked indefinitely. Move marker invalidation ahead of destructive recovery steps, or preserve a discoverable recovery record until cleanup succeeds. Add an integration test that fails marker deletion during repair, removes the failure, and verifies that another repair succeeds and releases the lock. Validation: the focused filesystem checks pass for the original fixes. I reproduced the remaining ordering problem using extracted production methods with simulated snapshot state; full HBase integration tests were not run. CI diagnostics remain unavailable without a connected GitHub account. -- 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]
