Horton
Menu

Re: Rename max_parallel_degree?

Julien Rouhaud <julien.rouhaud@dalibo.com>

View thread
From: Julien Rouhaud <julien.rouhaud@dalibo.com>
To: Amit Kapila <amit.kapila16@gmail.com>
Cc: Robert Haas <robertmhaas@gmail.com>, Tom Lane <tgl@sss.pgh.pa.us>, Josh Berkus <josh@agliodbs.com>, "David G. Johnston" <david.g.johnston@gmail.com>, Peter Geoghegan <pg@heroku.com>, Noah Misch <noah@leadboat.com>, David Rowley <david.rowley@2ndquadrant.com>, Alvaro Herrera <alvherre@2ndquadrant.com>, Magnus Hagander <magnus@hagander.net>, Bruce Momjian <bruce@momjian.us>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2016-06-26T10:27:27Z
Lists: pgsql-hackers 
On 26/06/2016 08:37, Amit Kapila wrote:
> On Sat, Jun 25, 2016 at 2:27 PM, Julien Rouhaud
> <julien.rouhaud@dalibo.com> wrote:
>>>
>>
>> It's better thanks.  Should we document somewhere the link between this
>> parameter and custom dynamic background workers or is it pretty
>> self-explanatory?
>>
> 
> How about if add an additiona line like:
> Parallel workers are taken from the pool of processes established by
> guc-max-worker-processes.
> 
> I think one might feel some duplication of text between this and what
> we have for max_parallel_workers_per_gather, but it seems genuine to
> me.
> 

Ok, I'll add it.

> 
> @@ -370,6 +379,8 @@ ForgetBackgroundWorker(slist_mutable_iter *cur)
>   Assert(rw->rw_shmem_slot <
> max_worker_processes);
>   slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
>   slot->in_use =
> false;
> + if (slot->parallel)
> + BackgroundWorkerData->parallel_terminate_count++;
> 
> I think operations on parallel_terminate_count are not safe.
> ForgetBackgroundWorker() and RegisterDynamicBackgroundWorker() can try
> to read write at same time.  It seems you need to use atomic
> operations to ensure safety.
> 
> 

But operations on parallel_terminate_count are done by the postmaster,
and per Robert's previous email postmaster can't use atomics:

> We can't have the postmaster and the
> backends share the same counter, because then it would need locking,
> and the postmaster can't try to take spinlocks - can't even use
> atomics, because those might be emulated using spinlocks.

Do we support platforms on which 32bits operations are not atomic?

-- 
Julien Rouhaud
http://dalibo.com - http://dalibo.org