pg_back should check the "pause" state after the request to pause recovery.

[yhuelf]

From the documentation, when pg_wal_replay_pause() returns, it doesn't mean that recovery stops right away. So I think that pg_back should loop around pg_get_wal_replay_pause_state() until it returns pause.

[orgrim]

Hi. Did you run into issues with that in production?

Regards

[yhuelf]

I think so. See the following (anonymised) logs:

2025/12/18 01:12:30 INFO: dumping globals
2025/12/18 01:12:30 INFO: dumping instance configuration
2025/12/18 01:12:30 INFO: pausing replication
2025/12/18 01:12:30 INFO: replication paused
2025/12/18 01:12:30 INFO: dumping database foo
2025/12/18 01:15:07 ERROR: [foo] pg_dump: Dumping the contents of table "bar" failed: PQgetResult() failed.
2025/12/18 01:15:07 ERROR: [foo] pg_dump: Error message from server: ERROR: canceling statement due to conflict with recovery
2025/12/18 01:15:07 ERROR: [foo] DETAIL: User query might have needed to see row versions that must be removed.
2025/12/18 01:15:07 ERROR: [foo] pg_dump: The command was: COPY foo.bar (id, tag, payload) TO stdout;

[orgrim]

Hi, looks reasonable. Could you send a patch to ensure recovery is paused? It should not be too hard to write.

Thanks

[yhuelf]

@OrioQuezac might be interested in doing the patch.