Horton

Re: Normalization of utility queries in pg_stat_statements

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: jian he <jian.universality@gmail.com>
Cc: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com>, Postgres hackers <pgsql-hackers@lists.postgresql.org>
Date: 2023-08-17T23:31:37Z
Lists: pgsql-hackers 
On Wed, Aug 16, 2023 at 05:11:47PM +0800, jian he wrote:
> SELECT calls, toplevel, rows, query FROM pg_stat_statements ORDER BY
> query COLLATE "C";
> returns:
>  calls | toplevel | rows |               query
> -------+----------+------+------------------------------------
>      1 | t        |    0 | CALL ptest3($1)
>      2 | f        |    2 | INSERT INTO cp_test VALUES ($2, x)
>      1 | t        |    1 | SELECT pg_stat_statements_reset()
> 
> here, the intermediate CALL part is optimized away. or should I expect
> CALL ptest1($1) also in pg_stat_statements?

I would have guessed that ptest1() being called as part of ptest3()
should show up in the report if you use track = all, as all the nested
queries of a function, even if it is pure SQL, ought to show up.  Now
note that ptest1() not showing up is not a new behavior, ~15 does the
same thing by missing it.
--
Michael

Commits

  1. Reflect normalization of query strings for utilities in pg_stat_statements

  2. Refine query jumbling handling for CallStmt

  3. Ignore IntoClause.viewQuery in query jumbling

  4. Improve cleanup phases in regression tests of pg_stat_statements

  5. Refactor more the regression tests of pg_stat_statements

  6. doc: Update pg_stat_statements about query ID calculation of utilities

  7. Expand regression tests of pg_stat_statements for utility queries

  8. Refactor tests of pg_stat_statements for planning, utility and level tracking