Horton

Re: SV: Problem with pg_notify / listen

Tom Lane <tgl@sss.pgh.pa.us>

View thread
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Noah Misch <noah@leadboat.com>
Cc: Gustavsson Mikael <mikael.gustavsson@smhi.se>, pgsql-bugs@lists.postgresql.org, Svensson Peter <peter.svensson@smhi.se>, Almen Anders <anders.almen@smhi.se>
Date: 2020-11-28T19:15:11Z
Lists: pgsql-bugs

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix a recently-introduced race condition in LISTEN/NOTIFY handling.

  2. Prevent concurrent SimpleLruTruncate() for any given SLRU.

Noah Misch <noah@leadboat.com> writes:
> On Fri, Nov 27, 2020 at 11:03:40PM -0500, Tom Lane wrote:
>> Hmm, it's not very clear what "stop" means here.  What do you think of
>> QUEUE_OLDEST_PAGE?

> QUEUE_OLDEST_PAGE is fine.  I like it a little less than QUEUE_STOP_PAGE,
> because oldestClogXid is a logical tail, and QUEUE_OLDEST_PAGE would be a
> physical tail.

I went with QUEUE_STOP_PAGE.

In further testing, I noted that the patch as I had it re-introduced the
symptom that 8b7ae5a82 fixed, that running "make installcheck" twice
in a row causes the async-notify isolation test to fail.  That's because
I'd changed asyncQueueUsage() to measure the distance back to the physical
tail, which isn't stable because we only truncate after crossing an
SLRU segment boundary.  So I reverted it to measuring the distance to
QUEUE_TAIL.  You could argue either way about which definition is more
useful to end users, perhaps; but in practice the difference should
usually be too small to matter, for everyone except regression tests
that are looking for distance exactly zero.

Anyhow, pushed with that fix.

Mikael, it appears that you have three options: revert to 11.9 until 11.11
is out, restart your server every time it approaches notify-queue-full, or
apply this patch:

https://git.postgresql.org/gitweb/?p=postgresql.git;a=patch;h=40f2fbe71ad615a2bcaaf5b840ccb9329e4378aa

			regards, tom lane