Re: RFC: Logging plan of the running query

Поиск
Список
Период
Сортировка
От Matthias van de Meent
Тема Re: RFC: Logging plan of the running query
Дата
Msg-id CAEze2WgqouNEQztyd70hD6ZEAf0_SAAiwMB-zTqK_XBLRq_KzA@mail.gmail.com
обсуждение исходный текст
Ответ на RFC: Logging plan of the running query  (torikoshia <torikoshia@oss.nttdata.com>)
Ответы Re: RFC: Logging plan of the running query  (torikoshia <torikoshia@oss.nttdata.com>)
Список pgsql-hackers
On Wed, 12 May 2021 at 13:24, torikoshia <torikoshia@oss.nttdata.com> wrote:
>
> Hi,
>
> During the discussion about memory contexts dumping[1], there
> was a comment that exposing not only memory contexts but also
> query plans and untruncated query string would be useful.
>
> I also feel that it would be nice when thinking about situations
> such as troubleshooting a long-running query on production
> environments where we cannot use debuggers.
>
> At that point of the above comment, I was considering exposing
> such information on the shared memory.
> However, since memory contexts are now exposed on the log by
> pg_log_backend_memory_contexts(PID), I'm thinking about
> defining a function that logs the plan of a running query and
> untruncated query string on the specified PID in the same way
> as below.
>
>    postgres=# SELECT * FROM pg_log_current_plan(2155192);
>     pg_log_current_plan
>    ---------------------
>     t
>    (1 row)
>
>    $ tail -f data/log/postgresql-2021-05-12.log
>
>    2021-05-12 17:37:19.481 JST [2155192] LOG:  logging the plan of
> running query on PID 2155192
>            Query Text: SELECT a.filler FROM pgbench_accounts a JOIN
> pgbench_accounts b ON a.aid = b.aid;
>            Merge Join  (cost=0.85..83357.85 rows=1000000 width=85)
>              Merge Cond: (a.aid = b.aid)
>              ->  Index Scan using pgbench_accounts_pkey on
> pgbench_accounts a  (cost=0.42..42377.43 rows=1000000 width=89)
>              ->  Index Only Scan using pgbench_accounts_pkey on
> pgbench_accounts b  (cost=0.42..25980.42 rows=1000000 width=4)
>
>
> Attached a PoC patch.
>
> Any thoughts?

Great idea. One feature I'd suggest would be adding a 'format' option
as well, if such feature would be feasable.


With regards,

Matthias van de Meent



В списке pgsql-hackers по дате отправления:

Предыдущее
От: wenjing
Дата:
Сообщение: Re: [Proposal] Global temporary tables
Следующее
От: David Rowley
Дата:
Сообщение: Do we need to rethink how to parallelize regression tests to speedup CLOBBER_CACHE_ALWAYS?