Thread: Trigger functions and parallelism

Trigger functions and parallelism

From
"Karl O. Pinc"
Date:
Hi,

It's not entirely clear to me what the conditions
are for function parallel safety.  While the documentation
seems pretty clear in most cases, there still some
verbiage that requires knowledge of internals
in order to be useful.

What I'm looking for is some general statements about
parallel safety for trigger functions.  I don't know
enough about parallelism to even know if it applies
to data modification statements.

It seems to me that, because raising any sort of exception
affects transaction state, that triggers which 
validate data integrity and raise errors must be PARALLEL UNSAFE.

But are any sort of AFTER INSERT trigger on a generated sequential key
parallel UNSAFE because they access sequence state via NEW?

Can something generic be said, like: "Most trigger functions should
be declared PARALLEL UNSAFE (the default) with the possible exception
of BEFORE triggers that raise no exceptions and control system
behavior by their return value."?  

What can be said to provide some overview which gives
guidance on parallelism to trigger authors?

Regards,

Karl <kop@karlpinc.com>
Free Software:  "You don't pay back, you pay forward."
                 -- Robert A. Heinlein



Re: Trigger functions and parallelism

From
Tom Lane
Date:
"Karl O. Pinc" <kop@karlpinc.com> writes:
> What I'm looking for is some general statements about
> parallel safety for trigger functions.  I don't know
> enough about parallelism to even know if it applies
> to data modification statements.

It doesn't, which is why the docs are silent on the point.

            regards, tom lane



Re: Trigger functions and parallelism

From
"Karl O. Pinc"
Date:
On Mon, 02 Oct 2023 13:07:19 -0400
Tom Lane <tgl@sss.pgh.pa.us> wrote:

> "Karl O. Pinc" <kop@karlpinc.com> writes:
> > What I'm looking for is some general statements about
> > parallel safety for trigger functions.  I don't know
> > enough about parallelism to even know if it applies
> > to data modification statements.  
> 
> It doesn't, which is why the docs are silent on the point.

Thanks.

Regards,

Karl <kop@karlpinc.com>
Free Software:  "You don't pay back, you pay forward."
                 -- Robert A. Heinlein