Fix the race condition in the test added by 7c99dc587.
authorAmit Kapila <akapila@postgresql.org>
Tue, 13 May 2025 04:24:29 +0000 (09:54 +0530)
committerAmit Kapila <akapila@postgresql.org>
Tue, 13 May 2025 04:24:29 +0000 (09:54 +0530)
commit8ede692de512750ecb6930213e6d78ed90ef06d0
treea08b29d8916b3ce53dfa20efc974b9977a57f250
parentdbf42b84ac7b4e1dfd1ac6e618e247100b6aa08b
Fix the race condition in the test added by 7c99dc587.

After executing ALTER SUBSCRIPTION tap_sub SET PUBLICATION, we did not
wait for the new walsender process to restart. As a result, an INSERT
executed immediately after the ALTER could be decoded and skipped,
considering it is not part of any subscribed publication. And, the old
apply worker could also confirm the LSN of such an INSERT. This could
cause the replication to resume from a point after the INSERT. In such
cases, we miss the expected warning about the missing publication.

To fix this, ensure the walsender has restarted before continuing after
ALTER SUBSCRIPTION.

Reported-by: Tom Lane as per CI
Author: vignesh C <vignesh21@gmail.com>
Reviewed-by: Xuneng Zhou <xunengzhou@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/1230066.1745992333@sss.pgh.pa.us
src/test/subscription/t/024_add_drop_pub.pl