Re: SQL/JSON: JSON_TABLE - pg_stat_statements crash

Поиск
Список
Период
Сортировка
От Pavel Stehule
Тема Re: SQL/JSON: JSON_TABLE - pg_stat_statements crash
Дата
Msg-id CAFj8pRBsjZ_Oem8qA8KvBbS6rCPDfd-3dOUHq-Gs6WvXGspYYg@mail.gmail.com
обсуждение исходный текст
Ответ на Re: SQL/JSON: JSON_TABLE - pg_stat_statements crash  (Erik Rijkers <er@xs4all.nl>)
Список pgsql-hackers


út 14. 9. 2021 v 20:04 odesílatel Erik Rijkers <er@xs4all.nl> napsal:
On 9/14/21 2:53 PM, Andrew Dunstan wrote:
> On 9/13/21 5:41 AM, Erik Rijkers wrote:
>> On 9/2/21 8:52 PM, Andrew Dunstan wrote:
>>

 >> [0001-SQL-JSON-functions-v51.patch]
 >> [0002-JSON_TABLE-v51.patch]
 >> [0003-JSON_TABLE-PLAN-DEFAULT-clause-v51.patch]
 >> [0004-JSON_TABLE-PLAN-clause-v51.patch]

Thanks, builds fine now.

But I found that the server crashes on certain forms of SQL when
postgresql.conf has a 'shared_preload_libraries' that contains module
'pg_stat_statements' (my value was:
'pg_stat_statements,auth_delay,auto_explain,passwordcheck').  Only
pg_stat_statements seems to cause the problem.

The offending SQL (I took it from the jsonb_sqljson.sql test file):

testdb=# SELECT JSON_EXISTS(jsonb '{"a": 1, "b": 2}', '$.* ? (@ > $x &&
@ < $y)' PASSING 0 AS x, 2 AS y);
server closed the connection unexpectedly
         This probably means the server terminated abnormally
         before or while processing the request.
The connection to the server was lost. Attempting reset: Failed.
Time: 2.551 ms
!?>

(Of course, that SQL running during regression testing has no problems
as there is then no pg_stat_statements.)

The statement sometimes succeeds but never very often.

The same crash was there before but I only now saw the connection with
the 'shared_preload_libraries/pg_stat_statements'.

I seem to remember some things changed in pg_stat_statements but I
didn't follow and don't know who to CC for it.

These issues are easily debugged - you can run gdb in the outer terminal, and attach it to the psql session. Then you can run the query.

Probably it will be a problem in pg_stat_statements callbacks - maybe query processing there doesn't know some new nodes that this patch introduces.

Regards

Pavel



Thanks,

Erik Rijkers

В списке pgsql-hackers по дате отправления:

Предыдущее
От: "Bossart, Nathan"
Дата:
Сообщение: Re: .ready and .done files considered harmful
Следующее
От: Daniel Gustafsson
Дата:
Сообщение: Re: [HACKERS] GSoC 2017: Foreign Key Arrays