Message-ID: <4C114381.3030900@collab.net>
Date: Thu, 10 Jun 2010 12:56:49 -0700
From: Anne Rosset <arosset@collab.net>
User-Agent: Thunderbird 2.0.0.24 (Windows/20100228)
MIME-Version: 1.0
To: Craig James <craig_james@emolecules.com>
CC: pgsql-performance@postgresql.org
Subject: Re: Need to increase performance of a query
References: <4C1125F0.8090101@collab.net> <4C112B19.6080908@krogh.cc>
	<4C113098.5040105@collab.net> <98414706.20100610212211@erwied.eu>
	<4C113E2F.6000305@collab.net> <4C11413B.8080107@emolecules.com>
In-Reply-To: <4C11413B.8080107@emolecules.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Craig James wrote:
> On 6/10/10 12:34 PM, Anne Rosset wrote:
>> Jochen Erwied wrote:
>>> Thursday, June 10, 2010, 8:36:08 PM you wrote:
>>>
>>>> psrdb=# (SELECT
>>>> psrdb(# MAX(item_rank.rank) AS maxRank
>>>> psrdb(# FROM
>>>> psrdb(# item_rank item_rank
>>>> psrdb(# WHERE
>>>> psrdb(# item_rank.project_id='proj2783'
>>>> psrdb(# AND item_rank.pf_id IS NULL
>>>> psrdb(#
>>>> psrdb(# )
>>>> psrdb-# ORDER BY
>>>> psrdb-# maxRank DESC;
>>>
>>> Don't think it does really matter, but why do you sort a resultset
>>> consisting of only one row?
>>>
>> Sorry, I should have removed the ORDER by (the full query has a union).
>> So without the ORDER by, here are the results:
>> psrdb=# SELECT
>> psrdb-# MAX(item_rank.rank) AS maxRank
>> psrdb-# FROM
>> psrdb-# item_rank item_rank
>> psrdb-# WHERE
>> psrdb-# item_rank.pf_id='plan1408';
>> maxrank
>> -------------
>> 20504000000
>> (1 row)
>>
>> Time: 1.516 ms
>> psrdb=# SELECT
>> psrdb-# MAX(item_rank.rank) AS maxRank
>> psrdb-# FROM
>> psrdb-# item_rank item_rank
>> psrdb-# WHERE
>> psrdb-# item_rank.project_id='proj2783'
>> psrdb-# AND item_rank.pf_id IS NULL;
>> maxrank
>> -------------
>> 20200000000
>> (1 row)
>>
>> Time: 13.177 ms
>>
>> Is there anything that can be done for the second one?
>
> Postgres normally doesn't index NULL values even if the column is 
> indexed, so it has to do a table scan when your query includes an IS 
> NULL condition.  You need to create an index that includes the "IS 
> NULL" condition.
>
>   create index item_rank_null_idx on item_rank(pf_id)
>    where item_rank.pf_id is null;
>
> Craig
>
Hi Craig,
I tried again after adding your suggested index but I didn't see any 
improvements: (seems that the index is not used)
psrdb=# explain analyze SELECT
psrdb-#        MAX(item_rank.rank) AS maxRank
psrdb-# FROM
psrdb-#        item_rank item_rank
psrdb-# WHERE
psrdb-#        item_rank.project_id='proj2783'
psrdb-#         AND item_rank.pf_id IS NULL;
                                                                        
QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------------------
 Result  (cost=0.17..0.18 rows=1 width=0) (actual time=11.942..11.943 
rows=1 loops=1)
   InitPlan
     ->  Limit  (cost=0.00..0.17 rows=1 width=8) (actual 
time=11.931..11.932 rows=1 loops=1)
           ->  Index Scan Backward using item_rank_rank on item_rank  
(cost=0.00..2933.84 rows=17558 width=8) (actual time=11.926..11.926 
rows=1 loops=1)
                 Filter: ((rank IS NOT NULL) AND (pf_id IS NULL) AND 
((project_id)::text = 'proj2783'::text))
 Total runtime: 11.988 ms
(6 rows)

Time: 13.654 ms


Thanks,
Anne