Re: Buffers from parallel workers not accumulated to upper nodes with gather merge

Jehan-Guillaume de Rorthais <jgdr@dalibo.com>

From: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
To: Amit Kapila <amit.kapila16@gmail.com>
Cc: PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>
Date: 2020-07-23T10:29:09Z
Lists: pgsql-bugs
On Thu, 23 Jul 2020 15:44:16 +0530
Amit Kapila <amit.kapila16@gmail.com> wrote:

> On Tue, Jul 21, 2020 at 9:05 PM Jehan-Guillaume de Rorthais
> <jgdr@dalibo.com> wrote:
> >
> > On Mon, 20 Jul 2020 15:29:00 +0530
> > Amit Kapila <amit.kapila16@gmail.com> wrote:
> >
> >  
> > > BTW, can you please test if the problem exist in back-branches, and does
> > > your change fix it there as well?  
> >
> > Gather Merge appears in v10, but I haven't been able to produce a plan
> > using it. 
> 
> I am able to do it for v10.  I think you can try by "set
> enable_hashagg = off;" and following all other steps mentioned in your
> email? 

Indeed. I confirm it produces the same result. I've been able to successfully
test to proposed fix.

> Attached, please find the patch for v10 with a modified commit
> message.  I will do it for other versions as well if you are fine with
> it?

I'm not an english-native writer, but I would rephrase this:

  We can do a similar thing for Gather Merge as well but it seems the better
  to account for stats during nodes shutdown after completing the execution.

Thanks!

Regards,



Commits

  1. Fix buffer usage stats for nodes above Gather Merge.

  2. Fix buffer usage stats for parallel nodes.