In PG 9.2 I’m getting “Index Only Scan Backward” for every partition in the first part of execution plan, when looking for MAX in partitioned table on a similar query:
" -> Index Only Scan Backward using pk_cycle_200610 on gp_cycle_200610 gp_cycle (cost=0.00..8.34 rows=5 width=8) (actual time=0.021..0.021 rows=1 loops=1)"
" Index Cond: (cycle_date_time IS NOT NULL)"
" Heap Fetches: 0"
May be you should upgrade to 9.2.
Regards,
Igor Neyman
From: rudi [mailto:rudolone@gmail.com]
Sent: Tuesday, January 22, 2013 10:08 AM
To: pgsql-performance@postgresql.org
Subject: Re: High CPU usage after partitioning
On Tue, Jan 22, 2013 at 3:46 PM, Andrew Dunstan <andrew@dunslane.net> wrote:
The query is pretty simple and standard, the behaviour (and the plan) is totally different when it comes to a partitioned table.
Partioned table query => explain analyze SELECT "sb_logs".* FROM "sb_logs" WHERE (device_id = 901 AND date_taken = (SELECT MAX(date_taken) FROM sb_logs WHERE device_id = 901));
And there you have it. Constraint exclusion does not work in cases like this. It only works with static expressions (such as a literal date in this case).
Ok, but I would have expected same plant repeated 4 times. When the table is not partitioned, the plan is defintely smarter: it knows that index is reversed and looks for max with an index scan backward). When the table is partitioned, it scan forward and I guess it will always do a full index scan.
--
rd
This is the way the world ends.
Not with a bang, but a whimper.