Re: [HACKERS] Perfomance bug in v10

Поиск
Список
Период
Сортировка
От Teodor Sigaev
Тема Re: [HACKERS] Perfomance bug in v10
Дата
Msg-id 2bc8a109-ca43-cdc0-551f-e88e1478b3a0@sigaev.ru
обсуждение исходный текст
Ответ на Re: [HACKERS] Perfomance bug in v10  (Tom Lane <tgl@sss.pgh.pa.us>)
Ответы Re: [HACKERS] Perfomance bug in v10  (Tom Lane <tgl@sss.pgh.pa.us>)
Список pgsql-hackers
>
> Teodor, could you check if this patch fixes your real-world problem?

It works fine with original query, thank you. But some other query slowdowns for 
~10% (9 secs vs 10 secs). Look at following part of plans of huge query:

without patch:
->  Nested Loop  (cost=34.82..50.91 rows=1 width=20)                 (actual time=0.017..0.061 rows=5 loops=24121)  ->
... ->  Materialize  (cost=0.56..15.69 rows=1 width=5)                   (actual time=0.003..0.004 rows=2 loops=109061)
  ->  Index Scan using ... (cost=0.56..15.68 rows=1 width=5)                             (actual time=0.013..0.014
rows=2loops=24121)
 

with patch:
->  Nested Loop  (cost=34.82..50.91 rows=1 width=20)                 (actual time=0.018..0.063 rows=5 loops=24121)  ->
... ->  Index Scan using ...  (cost=0.56..15.68 rows=1 width=5)                            (actual time=0.012..0.013
rows=2loops=109061)
 

(dots hidden the same parts)

As you said, it removes Materialize node, although it's useful here.

If you wish, I can do a test suite, its size will be around 10MB and send it by 
private  email.

-- 
Teodor Sigaev                                   E-mail: teodor@sigaev.ru
  WWW: http://www.sigaev.ru/
 



В списке pgsql-hackers по дате отправления:

Предыдущее
От: Robert Haas
Дата:
Сообщение: Re: [HACKERS] make check false success
Следующее
От: Hao Lee
Дата:
Сообщение: Re: [HACKERS] Do we need the gcc feature "__builtin_expect" topromote the branches prediction?