Thomas Munro <thomas.munro@gmail.com> writes:
> I made a thing to watch out for low probability BF failures and it
> told me that a similar failure in async-notify might have reappeared
> on brolga:
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=brolga&dt=2020-07-15%2008:30:11
> | REL_10_STABLE
> [ etc ]
Hm, interesting. None of these examples show an actual *failure* to
receive a notification, unlike the example that began this thread.
So it seems unlikely that back-patching 16114f2ea would help. What
we are seeing here, instead, is delayed timing of notify receipt(s).
I suspect that this is a variant of the issue described over here:
https://www.postgresql.org/message-id/flat/2527507.1598237598%40sss.pgh.pa.us
I didn't have a great idea about how to fix it reliably in
insert-conflict-specconflict, and I lack one here too :-(.
It's interesting though that your examples are all in v10 or older.
Could we have done something that indirectly fixes the problem
since then? Or is that just chance?
regards, tom lane