Horton

Re: Explain buffers wrong counter with parallel plans

Amit Kapila <amit.kapila16@gmail.com>

View thread View on postgr.es
From: Amit Kapila <amit.kapila16@gmail.com>
To: "Jonathan S. Katz" <jonathan.katz@excoventures.com>
Cc: Robert Haas <robertmhaas@gmail.com>, Alvaro Herrera <alvherre@2ndquadrant.com>, Adrien Nayrat <adrien.nayrat@anayrat.info>, Andres Freund <andres@anarazel.de>, PostgreSQL mailing lists <pgsql-hackers@postgresql.org>
Date: 2018-07-27T12:31:13Z
Lists: pgsql-hackers 
On Thu, Jul 26, 2018 at 7:31 PM, Jonathan S. Katz
<jonathan.katz@excoventures.com> wrote:
>
>> On Jul 25, 2018, at 10:25 PM, Amit Kapila <amit.kapila16@gmail.com> wrote:
>>
>>
>> You mean to say the number (Buffers: shared read=442478) in HEAD,
>> right?  If so, yeah, I am also wondering why the results of the patch
>> are different in HEAD and 11beta2.  So, if I read correctly, the
>> numbers in 11beta2 appears correct, but on HEAD it is not correct, it
>> should have divided the buffers read by loops.

I want to correct myself here, the numbers on HEAD are correct, but
not on PG11beta2.  Is there any chance that either you forgot to apply
the patch or maybe it is not using correct binaries in case of
11beta2.

>>  I will figure that
>> out, but this is just to clarify that both of us are seeing the
>> results in the same way.
>
> I’ll try it again but patch it against 11beta2 and see what results I get.
>
>>
>>> and there appears to be a performance hit.
>>>
>>
>> Do you mean to say the performance of the same query in 11beta2 and
>> HEAD or something else?
>
> Correct. But per your advice let me try running it against a patched
> version of 11beta2 and see what happens.
>

Yeah, that would be better.  Today, I have tried the patch on both
Head and PG11 and I am getting same and correct results.

-- 
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com

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.