Hello Kuroda-san,
30.07.2026 17:01, Hayato Kuroda (Fujitsu) weote:
Hi hackers,
There was a random buildfarm failure for 030_pg_recvlogical [1].
It tests that pg_recvlogical can re connect to the server if it the connection
is
lost. And we try to ensure by motnitoring the active_pid of the
pg_replication_slots
is changed.
According to the logfile in [1], the PID of the first walsender was 4360, and
pg_terminate_backend() was done toward it. But while waiting a slot was acquired
by the different process, another walsender whose PID was 4360 (!) was started
and it acquired the slot. The test finally failed because the slot was not
acquired by the different process.
Thank you for paying attention to the failure!
FWIW, it reminded me of [1], so reused PIDs on Windows caused us problems
before -- probably if we could somehow force Windows to reuse PIDs more
often, it would help discover other similar anomalies.
[1]
https://www.postgresql.org/message-id/4245b0de-9367-c443-8145-7b62b0ab3c41%40gmail.com
Best regards,
Alexander
