pg.ddx.io  pgsql-admin@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: MichaelDBA <MichaelDBA@sqlexec.com>
To: Laurenz Albe <laurenz.albe@cybertec.at>
Cc: Mark Steben <mark.steben@drivedominion.com>
Cc: pgsql-admin <pgsql-admin@postgresql.org>
Subject: Re: Long running query in new production, not so long in old
Date: Thu, 28 Mar 2019 09:23:55 -0400
Message-ID: <44995f85-a65c-6ac9-cba8-b12c1ffd0602@sqlexec.com> (raw)
In-Reply-To: <55e1328ed76390092c1ea5f1af4b59ce593e732f.camel@cybertec.at>
References: <CADyzmyy9afP526per2wRO6PBn=uqZe6jtpsRLyiFE5CG4PoQiA@mail.gmail.com>
	<55e1328ed76390092c1ea5f1af4b59ce593e732f.camel@cybertec.at>

My question is about the new VM environment...

 1. What vendor for the VMs?
 2. Make sure memory and CPU are not overallocated in the hypervisor.
 3. Evaluate the resource needs of the VM group/cluster to which it belongs.
 4. Regarding VMWare, I dealt with a problem where the "cpu ready
    percentage" was too high (above 5-10%), which caused cpu cycle delays.


Regards,
Michael Vitale



Laurenz Albe wrote on 3/28/2019 8:56 AM:
> On Wed, 2019-03-27 at 19:27 -0400, Mark Steben wrote:
>> We are moving to a new VM environment (expedient) and have one query  that typically runs in 22 - 25
>> seconds in our old environment, but is running in about 1 hour, 20 minutes in our new.
>> I'd like some insight as to why the explain is showing shared buffer hits numbering over 113 milliion
>> in the new environment and only 445 thousand in the old.  I have sent the explains along with the
>> table descriptions, row counts, the one function that I know causes the bottleneck,  the query,
>> some relevant configuration settings in postgresql conf (identical in both environments)
>> and a listing from top in both environments, showing memory, shared memory, and cpu.
>>
>> Everything seems to be identical or close, except for the shared buffer count in the explain.
>> Any insight would be appreciated.
> Slow plan:
>
>   ->  Index Scan using emailrcpts_4columns on emailrcpts  (cost=0.56..119078.48 rows=5 width=29) (actual time=4873080.765..4873080.765 rows=0 loops=1)
>         Index Cond: ((cid = 1784539) AND (removed = false) AND (active = true) AND (bounce < 3))
>         Filter: ((NOT removed) AND active AND (email IS NOT NULL) AND (((fname IS NOT NULL) AND (length((fname)::text) <> 0)) OR ((lname IS NOT NULL) AND (length((lname)::text) <> 0))) AND
> (number_of_sends(id, 30, 1) < 3) AND (lower(get_make(cid, (vin)::text, (make_purchased)::text)) = 'ohgoshnonotthebees!!!'::text) AND (NOT (SubPlan 1)))
>         Rows Removed by Filter: 19952
>         Buffers: shared hit=113768530 read=6244
>         SubPlan 1
>           ->  Limit  (cost=0.28..15.58 rows=1 width=0) (never executed)
>
> Fast plan:
>
>   ->  Index Scan using emailrcpts_4columns on emailrcpts  (cost=0.56..113162.26 rows=5 width=29) (actual time=21086.555..21086.555 rows=0 loops=1)
>         Index Cond: ((cid = 1784539) AND (removed = false) AND (active = true) AND (bounce < 3))
>         Filter: ((NOT removed) AND active AND (email IS NOT NULL) AND (((fname IS NOT NULL) AND (length((fname)::text) <> 0)) OR ((lname IS NOT NULL) AND (length((lname)::text) <> 0))) AND
> (number_of_sends(id, 30, 1) < 3) AND (lower(get_make(cid, (vin)::text, (make_purchased)::text)) = 'ohgoshnonotthebees!!!'::text) AND (NOT (SubPlan 1)))
>         Rows Removed by Filter: 19952
>         Buffers: shared hit=445188 read=61756
>         SubPlan 1
>           ->  Limit  (cost=0.28..15.58 rows=1 width=0) (never executed)
>
> Hmm. These are the ideas I can come up with:
>
> 1. There are many index tuples belonging to dead heap tuples.
>     Then re-running the query should produce way fewer buffer hits.
>     VACUUM would fix that issue.
>
> 2. The index is terribly fragmented.
>     REINDEX INDEX emailrcpts_4columns
>     would improve that one.
>
> Yours,
> Laurenz Albe

view thread (6+ messages)  latest in thread

Message-ID: <44995f85-a65c-6ac9-cba8-b12c1ffd0602@sqlexec.com>
Permalink:  ../44995f85-a65c-6ac9-cba8-b12c1ffd0602@sqlexec.com/
Also on:    postgresql.org/message-id/44995f85-a65c-6ac9-cba8-b12c1ffd0602@sqlexec.com

 · 

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients from your mail client; the link fills in
  the headers below: reply via email
  Some clients, webmail especially, drop In-Reply-To and References from
  a link, and the reply then starts a new thread: use the mbox below.

  To: pgsql-admin@postgresql.org
  Cc: MichaelDBA@sqlexec.com, laurenz.albe@cybertec.at, mark.steben@drivedominion.com
  Subject: Re: Long running query in new production, not so long in old
  In-Reply-To: <44995f85-a65c-6ac9-cba8-b12c1ffd0602@sqlexec.com>
  References: <CADyzmyy9afP526per2wRO6PBn=uqZe6jtpsRLyiFE5CG4PoQiA@mail.gmail.com> <55e1328ed76390092c1ea5f1af4b59ce593e732f.camel@cybertec.at> <44995f85-a65c-6ac9-cba8-b12c1ffd0602@sqlexec.com>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox