Re: [PATCH] Fix pg_rewind false positives caused by shutdown-only WAL

1. Control file changes:
You are correct that the control file always changes on shutdown, and that pg_rewind cannot simply ignore those updates. My earlier patch proposal did not address that, and I now understand why the server itself would reject a mismatch here.

2. Other WAL records (RUNNING_XACTS):
I see now that a clean shutdown generates both RUNNING_XACTS and CHECKPOINT_SHUTDOWN. My patch only skipped over the latter, so in practice rewind would still be triggered incorrectly. I will extend the logic to also consider this sequence properly.

3. Server-side consistency:
As noted, even if pg_rewind skips shutdown-only WAL records, the restarted old primary can still fail due to control file divergence (infinite loop issue). That means it needs a more holistic fix that considers both pg_rewind and server startup behavior.

4. RMID verification:
I did not guard the filtering with an XLogRecGetRmid() check. I’ll fix this to avoid misclassification, following the walsummarizer.c example as you suggested.

Plan forward:

• Revise the patch so that pg_rewind correctly checks RMIDs and handles both RUNNING_XACTS + CHECKPOINT_SHUTDOWN sequences, not just shutdown checkpoints.

• Investigate whether control file normalization is required (or whether server-side startup logic also needs adjustments) so that an old primary can rejoin cleanly without looping.

• Ensure consistent patch authorship (my name + email will match the commit and submission).

• Add regression coverage under src/bin/pg_rewind/t/ to reproduce this clean-shutdown failover scenario automatically.

Best regards,
Soumya