Re: subscription worker signalling wal writer too much

From: Robert Haas <robertmhaas(at)gmail(dot)com>
To: Jeff Janes <jeff(dot)janes(at)gmail(dot)com>
Cc: Andres Freund <andres(at)anarazel(dot)de>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: subscription worker signalling wal writer too much
Date: 2017-08-25 17:03:50
Message-ID: CA+TgmoaY88+5jjSR2RYEwOpU1jjS0tK5i2XUsMfwz+yLU7OWXQ@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Mon, Jul 3, 2017 at 11:26 PM, Jeff Janes <jeff(dot)janes(at)gmail(dot)com> wrote:
> My understanding is that you can't start on a new file until the old file is
> completely synced, because the book keeping can currently only handle one
> file at a time. So if you signal the wal writer to do the sync, you would
> just end up immediately blocking on it to finish. Getting around that would
> require substantially more changes; we would need to implement a queue of
> full log files not yet synced so that we could move on without waiting.

Yeah. I bet that effort would be reasonably worthwhile. I believe
that the forced end-of-segment syncs are costing us a noticeable
amount of performance. Once or twice I took a very half-hearted run
at doing what you describe here, but gave up pretty quickly; it seems
like a somewhat tricky problem.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Henry M 2017-08-25 17:05:58 Re: [RFC] What would be difficult to make data models pluggable for making PostgreSQL a multi-model database?
Previous Message Robert Haas 2017-08-25 16:52:32 Re: assorted code cleanup