Dependency between bgw_notify_pid and bgw_flags - Mailing list pgsql-hackers

From Ashutosh Bapat
Subject Dependency between bgw_notify_pid and bgw_flags
Date
Msg-id CAFjFpRfwS-xCcTnou5TLjNuayWB6g5RJPkkv0dSKRSf1WXeHTg@mail.gmail.com
Whole thread Raw
Responses Re: Dependency between bgw_notify_pid and bgw_flags  (Robert Haas <robertmhaas@gmail.com>)
List pgsql-hackers
Hi,
Documentation here http://www.postgresql.org/docs/devel/static/bgworker.html does not indicate any relation between the fields bgw_notify_pid and bgw_flags of BackgroundWorker structure. But in one has to set BGWORKER_BACKEND_DATABASE_CONNECTION in order to use bgw_notify_pid feature.

In BackgroundWorkerStateChange
 318         /*
 319          * Copy the PID to be notified about state changes, but only if the
 320          * postmaster knows about a backend with that PID.  It isn't an error
 321          * if the postmaster doesn't know about the PID, because the backend
 322          * that requested the worker could have died (or been killed) just
 323          * after doing so.  Nonetheless, at least until we get some experience
 324          * with how this plays out in the wild, log a message at a relative
 325          * high debug level.
 326          */
 327         rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
 328         if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
 329         {
 330             elog(DEBUG1, "worker notification PID %lu is not valid",
 331                  (long) rw->rw_worker.bgw_notify_pid);
 332             rw->rw_worker.bgw_notify_pid = 0;
 333         }

bgw_notify_pid gets wiped out (and that too silently) if PostmasterMarkPIDForWorkerNotify() returns false. PostmasterMarkPIDForWorkerNotify() only looks at the BackendList, which does not contain all the background worker created by Register*BackgroundWorker() calls. Only a baground worker which has set BGWORKER_BACKEND_DATABASE_CONNECTION, gets added into BackendList in maybe_start_bgworker()

5629             if (rw->rw_worker.bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
5630             {
5631                 if (!assign_backendlist_entry(rw))
5632                     return;
5633             }
5634             else
5635                 rw->rw_child_slot = MyPMChildSlot = AssignPostmasterChildSlot();
5636
5637             do_start_bgworker(rw);      /* sets rw->rw_pid */
5638
5639             if (rw->rw_backend)
5640             {
5641                 dlist_push_head(&BackendList, &rw->rw_backend->elem);

Should we change the documentation to say "one needs to set BGWORKER_BACKEND_DATABASE_CONNECTION" in order to use bgw_notify_pid feature? OR we should fix the code not to wipe out bgw_notify_pid in the code above (may be change PostmasterMarkPIDForWorkerNotify() to scan the list of other background workers as well)?
--
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company

pgsql-hackers by date:

Previous
From: Stephen Frost
Date:
Subject: Re: [CORE] Restore-reliability mode
Next
From: Andrew Dunstan
Date:
Subject: Re: Re: [COMMITTERS] pgsql: Map basebackup tablespaces using a tablespace_map file