Horton

Re: Problem with synchronous replication

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: masao.fujii@gmail.com, lingce.ldm@alibaba-inc.com, pgsql-hackers@postgresql.org
Date: 2019-10-31T02:11:03Z
Lists: pgsql-hackers

Attachments

On Wed, Oct 30, 2019 at 05:43:04PM +0900, Kyotaro Horiguchi wrote:
> At Wed, 30 Oct 2019 17:21:17 +0900, Fujii Masao <masao.fujii@gmail.com> wrote in 
>> This change causes every ending backends to always take the exclusive lock
>> even when it's not in SyncRep queue. This may be problematic, for example,
>> when terminating multiple backends at the same time? If yes,
>> it might be better to check SHMQueueIsDetached() again after taking the lock.
>> That is,
> 
> I'm not sure how much that harms but double-checked locking
> (releasing) is simple enough for reducing possible congestion here, I
> think.

FWIW, I could not measure any actual difference with pgbench -C, up to
500 sessions and an empty input file (just have one meta-command) and
-c 20.

I have added some comments in SyncRepCleanupAtProcExit(), and adjusted
the patch with the suggestion from Fujii-san.  Any comments?
--
Michael

Commits

  1. Add some assertions in syncrep.c

  2. Fix race condition at backend exit when deleting element in syncrep queue

  3. Fix typo in comment of syncrep.c