Horton

Re: track_planning causing performance regression

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Fujii Masao <masao.fujii@oss.nttdata.com>
Cc: Julien Rouhaud <rjuju123@gmail.com>, "Tharakan, Robins" <tharar@amazon.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2020-06-29T23:10:15Z
Lists: pgsql-hackers 
Hi,

On 2020-06-29 17:55:28 +0900, Fujii Masao wrote:
> One idea to reduce that lock contention is to separate per-query spinlock
> into two; one is for planning, and the other is for execution. pgss_store()
> determines which lock to use based on the given "kind" argument.
> To make this idea work, also every pgss counters like shared_blks_hit
> need to be separated into two, i.e., for planning and execution.

I suspect that the best thing would be to just turn the spinlock into an
lwlock. Spinlocks deal terribly with contention. I suspect it'd solve
the performance issue entirely. And it might even be possible, further
down the line, to just use a shared lock, and use atomics for the
counters.

Greetings,

Andres Freund

Commits

  1. doc: Fix description about pg_stat_statements.track_planning.

  2. doc: Add note about possible performance overhead by enabling track_planning.

  3. Change default of pg_stat_statements.track_planning to off.