Horton

Re: Explain buffers wrong counter with parallel plans

Alvaro Herrera <alvherre@2ndquadrant.com>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Amit Kapila <amit.kapila16@gmail.com>, Adrien Nayrat <adrien.nayrat@anayrat.info>, Andres Freund <andres@anarazel.de>, PostgreSQL mailing lists <pgsql-hackers@postgresql.org>, Tom Lane <tgl@sss.pgh.pa.us>
Date: 2018-06-29T18:22:17Z
Lists: pgsql-hackers 
On 2018-Jun-08, Robert Haas wrote:

> On Wed, Jun 6, 2018 at 12:08 PM, Alvaro Herrera
> <alvherre@2ndquadrant.com> wrote:
> > So, apparently this is not a Postgres 11 open item, but rather a bug
> > that goes back to pg10.  However, maybe it would be worth fixing soon
> > anyway?  In particular, if we want to perturb the explain output as
> > suggested upthread, maybe *that* should be considered an open item?
> >
> > Anyway, this has stalled for a month now.
> 
> Yeah.  I'm not willing to do anything here unilaterally.  There is
> neither universal agreement that there is a problem here nor agreement
> on a fix.  Unless that changes, I don't know what to do here.

After going over this whole thing once more, my opinion is that this
reporting is broken.  IMO if we want to display per-process counts, then
let's display them all, or none; if we're going to display aggregated
counts, then we should mark them as such.  Otherwise it's pretty much
impossible to read the output unless you know exactly what the code does
for each value.  (The other option would be to document what the code
does for each value, but that seems suboptimal -- that just requires
people to glue one more cheat sheet to their walls, which is not Pg
style.)

I'm CCing Tom here, as author of the patch that caused (most of) the
issue.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Commits

  1. Adjust comment atop ExecShutdownNode.

  2. Prohibit shutting down resources if there is a possibility of back up.

  3. Fix buffer usage stats for parallel nodes.

  4. Match the buffer usage tracking for leader and worker backends.

  5. Improve division of labor between execParallel.c and nodeGather[Merge].c.