Horton

Thread

Commits

  1. Add compute_query_id = regress

  1. installcheck fails when compute_query_id=on or pg_stat_statsement is loaded

    Anton A. Melnikov <aamelnikov@inbox.ru> — 2021-10-15T07:36:36Z

    Hello!
 
When extension  pg_stat_statsement is loaded into memory or compute_query_id=on in postgesql.conf
many of installcheck tests give errors.
The thing is that in *.out files appear lines "queryid = xxxxx" where xxxxx - some numeric value.
So 24 of 209 installcheck tests will fail.
It seems to be a normal behaviour as queryid calculation was moved into core in 
https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=5fd9dfa5f5 
but tests say that something wrong.
Create test variants is not the way, as queryid value varies randomly 
from test to test at the same enviroment.
I think this is a problem because these fake errors can mask a real errors in relevant tests.
What’s your opinion?
 
Best regards,
Anton Melnikov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company
 
 

  2. Re: installcheck fails when compute_query_id=on or pg_stat_statsement is loaded

    Julien Rouhaud <rjuju123@gmail.com> — 2021-10-15T08:35:25Z

    On Fri, Oct 15, 2021 at 3:36 PM Мельников Антон Андреевич
<aamelnikov@inbox.ru> wrote:
>
> When extension  pg_stat_statsement is loaded into memory or compute_query_id=on in postgesql.conf
> many of installcheck tests give errors.
> [...]
> I think this is a problem because these fake errors can mask a real errors in relevant tests.
> What’s your opinion?

This has been discussed previously (although I can't find the thread
right now).  Note that you don't really need to enable
pg_stat_statements, enabling compute_query_id is enough.  The query
identifier is only displayed for EXPLAIN (VERBOSE), so it's already a
bit filtered.  I don't see any simple way to entirely avoid the
problem though.

There are already many options that can break the regression tests, so
maybe it's ok to accept that this is yet another one.  If not, the
only alternative I see is to add a boolean QUERY_ID option to EXPLAIN
and make sure that all tests use it, but it seems like a big hammer,
error prone, for a maybe small problem.

  3. Re: installcheck fails when compute_query_id=on or pg_stat_statsement is loaded

    Tom Lane <tgl@sss.pgh.pa.us> — 2021-10-15T13:46:18Z

    Julien Rouhaud <rjuju123@gmail.com> writes:
> ... Note that you don't really need to enable
> pg_stat_statements, enabling compute_query_id is enough.  The query
> identifier is only displayed for EXPLAIN (VERBOSE), so it's already a
> bit filtered.  I don't see any simple way to entirely avoid the
> problem though.

> There are already many options that can break the regression tests, so
> maybe it's ok to accept that this is yet another one.

Yeah, that's my reaction.  We could add "compute_query_id = off" to
the database-level settings set up by pg_regress, but that cure seems
worse than the disease.  It would make it impossible to run the
regression tests with pg_stat_statements loaded, which you might wish
to do (just ignoring the bogus test failures) as a way of testing
pg_stat_statements.

			regards, tom lane