Horton

RE: Connection limits/permissions, slotsync workers, etc

Zhijie Hou (Fujitsu) <houzj.fnst@fujitsu.com>

View thread View on postgr.es
From: "Zhijie Hou (Fujitsu)" <houzj.fnst@fujitsu.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>, Laurenz Albe <laurenz.albe@cybertec.at>
Date: 2024-12-28T03:16:56Z
Lists: pgsql-hackers 
On Saturday, December 28, 2024 1:31 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> 
> "Zhijie Hou (Fujitsu)" <houzj.fnst@fujitsu.com> writes:
> > On Thursday, December 26, 2024 3:50 AM Tom Lane <tgl@sss.pgh.pa.us>
> >> I wonder if the AV launcher and slotsync worker could be reclassified
> >> as "auxiliary processes" instead of being their own weird animal.
> 
> > It appears that the current aux processes do not run transactions as
> > stated in the comments[1], so we may need to somehow release this
> > restriction to achieve the goal.
> 
> Ah, right, I'd forgotten about that restriction.  I agree that removing it wouldn't
> be very reasonable.  However, I still would rather avoid making the slotsync
> worker be its very own special snowflake, because that offers no support for
> the next person who wants to invent a new sort of specialized
> transaction-capable process.
> 
> Attached is an alternative proposal that groups the autovac launcher and
> slotsync worker into a new category of "special workers" (better name
> welcome).  I chose to put them into the existing autovacFreeProcs freelist,
> partly because the autovac launcher lives there already but mostly because I
> don't want to add another freelist in a patch we need to put into v17.  (As
> written, your patch is an ABI break.

Right, thanks for pointing it out. The new version patch looks good to me.

Best Regards,
Hou zj

Commits

  1. Replace PGPROC.isBackgroundWorker with isRegularBackend.

  2. Exclude parallel workers from connection privilege/limit checks.

  3. Reserve a PGPROC slot and semaphore for the slotsync worker process.

  4. Try to avoid semaphore-related test failures on NetBSD/OpenBSD.