From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
To: Dmitry Shalashov <skaurus@gmail.com>
To: Alex Ignatov <a.ignatov@postgrespro.ru>
Cc: pgsql-performance@postgresql.org <pgsql-performance@postgresql.org>
Subject: Re: Query became very slow after 9.6 -> 10 upgrade
Date: Wed, 22 Nov 2017 16:07:26 +0100
Message-ID: <ab4918a9-bf22-2b80-ec7a-0fa484a8b9a8@2ndquadrant.com> (raw)
In-Reply-To: <CAKPeCUH9S7-Vz9SY4mjVxQ_4pHRvosVV+BZZXCkU3UUAykjOnA@mail.gmail.com>
References: <CAKPeCUGAeHgoh5O=SvcQxREVkoX7UdeJUMj1F5=aBNvoTa+O8w@mail.gmail.com>
<1379201d3639d$8b61bcd0$a2253670$@postgrespro.ru>
<CAKPeCUHuxs8KU=4e9d=9HOjiX5=Y2x7NPNWiQZQ8N5yZpM8DPQ@mail.gmail.com>
<137da01d363a0$603cc970$20b65c50$@postgrespro.ru>
<CAKPeCUH9S7-Vz9SY4mjVxQ_4pHRvosVV+BZZXCkU3UUAykjOnA@mail.gmail.com>
IMHO the problems here are due to poor cardinality estimates.
For example in the first query, the problem is here:
-> Nested Loop (cost=0.42..2.46 rows=1 width=59)
(actual time=2.431..91.330 rows=3173 loops=1)
-> CTE Scan on b (cost=0.00..0.02 rows=1 width=40)
(actual time=2.407..23.115 rows=3173 loops=1)
-> Index Scan using domains_pkey on domains d
(cost=0.42..2.44 rows=1 width=19)
(actual time=0.018..0.018 rows=1 loops=3173)
That is, the database expects the CTE to return 1 row, but it returns
3173 of them, which makes the nested loop very inefficient.
Similarly for the other query, where this happens:
Nested Loop (cost=88.63..25617.31 rows=491 width=16)
(actual time=3.512..733248.271 rows=1442797 loops=1)
-> HashAggregate (cost=88.06..88.07 rows=1 width=4)
(actual time=3.380..13.561 rows=3043 loops=1)
That is, about 1:3000 difference in both cases.
Those estimation errors seem to be caused by a condition that is almost
impossible to estimate, because in both queries it does this:
groups->0->>'provider' ~ '^something'
That is, it's a regexp on an expression. You might try creating an index
on the expression (which is the only way to add expression statistics),
and reformulate the condition as LIKE (which I believe we can estimate
better than regular expressions, but I haven't tried).
So something like
CREATE INDEX ON adroom ((groups->0->>'provider'));
WHERE groups->0->>'provider' LIKE 'something%';
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
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 using the --to and --cc options:
reply via email
To: pgsql-performance@postgresql.org
Cc: tomas.vondra@2ndquadrant.com, skaurus@gmail.com, a.ignatov@postgrespro.ru
Subject: Re: Query became very slow after 9.6 -> 10 upgrade
In-Reply-To: <ab4918a9-bf22-2b80-ec7a-0fa484a8b9a8@2ndquadrant.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