The same again with 16.9 : was Re: PostgreSQL 16.6 , query stuck with STAT Ssl, wait_event_type : IPC , wait_event : ParallelFinish - Mailing list pgsql-admin

Hi ,

we had the same problem today again.

postgres@[local]/dynacom=# select * from pg_stat_activity where application_name~*'dbmirr';
-[ RECORD 1 ]----+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------------------------------------------------------
datid            | 207491653
datname          | dynacom
pid              | 1821681
leader_pid       |  
usesysid         | 10
usename          | postgres
application_name | DBMIRROR
client_addr      | 10.9.0.10
client_hostname  |  
client_port      | 45051
backend_start    | 2025-08-22 03:58:32.321683+03
xact_start       | 2025-08-22 04:06:08.897252+03
query_start      | 2025-08-22 04:06:09.254048+03
state_change     | 2025-08-22 04:06:09.254049+03
wait_event_type  | IPC
wait_event       | ParallelFinish
state            | active
backend_xid      |  
backend_xmin     | 222705697
query_id         | -3929522546936394707
query            | SELECT pd.XID,MAX(SeqId) FROM dbmirror_Pending pd LEFT JOIN dbmirror_MirroredTransaction mt INNER JOIN dbmirror_MirrorHost mh ON mt.MirrorHostId =  mh.MirrorHostId AND m
h.HostName= '192.168.145.1'  ON pd.XID = mt.XID WHERE mt.XID is null and (pd.slaveid is null or pd.slaveid = '4826')  GROUP BY pd.XID  ORDER BY MAX(pd.SeqId)
backend_type     | client backend

On 1/6/25 07:19, Tom Lane wrote:

Achilleas Mantzios <a.mantzios@cloud.gatewaynet.com> writes:

a query is stuck with the above, it seems it waits for parallel worker 
to finish, however , there are no parallel works running :

You didn't explain the subject about "STAT Ssl", but if you mean
that that was what ps was showing for the backend process, there's
something very wrong there.  According to "man ps", the "l" means
               l    is multi-threaded (using CLONE_THREAD, like NPTL pthreads                    do)

Yes. sorry, I didn't include this info, you are spot on, yes this the output of ps aux .

which is something that a Postgres backend should never be
(in existing releases anyway).  So I'm speculating that
the process somehow became multi-threaded and then some
wakeup signal went to the wrong thread.

We've had issues with perl or python introducing multi-threading
because of plperl or plpython functions doing things they
probably shouldn't.  Do you have any of those in your system?

Yes we have two perl functions only that I'd be happy to get rid off :

postgres@[local]/dynacom=# select p.proname, l.lanname from pg_language l, pg_proc p where p.prolang=l.oid and l.lanname ~* '.*perl.*';
proname  | lanname  
----------+---------
basename | plperlu
filetype | plperlu
(2 rows)

Nothing used in the app, just some two utility functions to help us batch insert some attachments, guess mimetype etc. However the calling client is Perl , based on libpg-perl (not DBI), basically this is a descendant of DBMirror.pl (we are still using it).

The strange thing is that we run pgsql 16.* since November, also we run our version of DBMirror since 2005 (and PostgreSQL since 2001) and we never had this problem before (at least from what I know).

			regards, tom lane