why does this use the wrong index?

Поиск
Список
Период
Сортировка
От Rainer Mager
Тема why does this use the wrong index?
Дата
Msg-id 000101c919ef$7c0b8e20$7422aa60$@com
обсуждение исходный текст
Ответы Re: why does this use the wrong index?
Список pgsql-performance
I have two identical queries except for the date range. In the first case,
with the wider date range, the correct (I believe) index is used. In the
second case where the date range is smaller a different index is used and a
less efficient plan is chosen. In the second query the problem seems to be
CPU resoures; while it is running 1 core of the CPU is 100% busy.

Note, if I drop the ad_log_date index then this query is always fast, but
some other queries I do require that index.

So, What can I do to encourage Postgres to use the first index even when the
date range is smaller.




# explain analyze SELECT name FROM players AS foo WHERE EXISTS (SELECT 1
FROM ad_log WHERE player = foo.id AND  date(start_time) BETWEEN
E'2008-09-14' AND E'2008-09-18' LIMIT 1) ORDER BY name;
                                                                       QUERY
PLAN
----------------------------------------------------------------------------
----------------------------------------------------------------------------
-
 Sort  (cost=1573.74..1574.31 rows=230 width=13) (actual time=28.421..28.505
rows=306 loops=1)
   Sort Key: foo.name
   Sort Method:  quicksort  Memory: 28kB
   ->  Seq Scan on players foo  (cost=0.00..1564.72 rows=230 width=13)
(actual time=0.104..27.876 rows=306 loops=1)
         Filter: (subplan)
         SubPlan
           ->  Limit  (cost=0.01..3.39 rows=1 width=0) (actual
time=0.058..0.058 rows=1 loops=460)
                 ->  Index Scan using ad_log_player_date on ad_log
(cost=0.01..34571.03 rows=10228 width=0) (actual time=0.056..0.056 rows=1
loops=460)
                       Index Cond: ((player = $0) AND (date(start_time) >=
'2008-09-14'::date) AND (date(start_time) <= '2008-09-18'::date))
 Total runtime: 28.623 ms
(10 rows)



# explain analyze SELECT name FROM players AS foo WHERE EXISTS (SELECT 1
FROM ad_log WHERE player = foo.id AND  date(start_time) BETWEEN
E'2008-09-18' AND E'2008-09-18' LIMIT 1) ORDER BY name;
                                                                 QUERY PLAN
----------------------------------------------------------------------------
-----------------------------------------------------------------
 Index Scan using players_name_key on players foo  (cost=0.00..8376.84
rows=230 width=13) (actual time=813.695..143452.810 rows=301 loops=1)
   Filter: (subplan)
   SubPlan
     ->  Limit  (cost=0.01..18.14 rows=1 width=0) (actual
time=311.846..311.846 rows=1 loops=460)
           ->  Index Scan using ad_log_date on ad_log  (cost=0.01..18.14
rows=1 width=0) (actual time=311.844..311.844 rows=1 loops=460)
                 Index Cond: ((date(start_time) >= '2008-09-18'::date) AND
(date(start_time) <= '2008-09-18'::date))
                 Filter: (player = $0)
 Total runtime: 143453.100 ms
(8 rows)



Thanks,

--Rainer


В списке pgsql-performance по дате отправления:

Предыдущее
От: "Scott Marlowe"
Дата:
Сообщение: Re: RAID arrays and performance
Следующее
От: Matthew Wakeling
Дата:
Сообщение: Re: RAID arrays and performance