Re: AutoVacuum starvation from sinval messages

Jeff Janes <jeff(dot)janes(at)gmail(dot)com> writes:
> Basically the WaitLatch in autovacuum.c line 602 (in head) never returns.

> This was introduced by a180776f7a1c4554f214b, "Teach unix_latch.c to
> use poll() where available"

> When the poll() gets EINTR by SIGUSR1 for the invalidation, it
> restarts at the full timeout (60 seconds in this case) with no
> accounting for the time already waited.

Hmm. That seems like a problem, all right, but I don't believe the
claim that it applies only to the poll() code path. On many platforms
the select() path would have the same issue.

It's a bit annoying to have to incur an extra gettimeofday call per
WaitLatch operation in case this happens, but there doesn't seem to
be a lot of choice. I guess we can arrange to incur the extra call
only when a timeout is requested --- but IIRC that's going to be
the majority of usages anyway ...

regards, tom lane