Autonomous transaction, background worker

I try to setup a logging facility and want it to do its work
asynchronously in autonomous transactions. I have read
http://blog.dalibo.com/2016/08/19/Autonoumous_transactions_support_in_PostgreSQL.html
and chapter 47 of the documentation believing those articles are about
the same. However, pg_background_launch mentioned in the blog is not
available on my vanilla installation of Debian 10.3-2 from the
PostgreSQL repositories.
CHapter 47 makes me believe I need to code some C functions to use
background workers.

I've experimented with background workers a bit. Yes, normally you'd
write some C, but you could also install this (mentioned in the blog
post and the source of pg_background_launch):

https://github.com/vibhorkum/pg_background

Note that is not part of contrib and is not on pgxn, so to install it
you need to clone the repo and run `make && sudo make install`. Also I'm
not sure it is really production-level. In particular this is a serious
issue:

https://github.com/vibhorkum/pg_background/issues/1

(If the launching process exits, the child process won't be able to map
the shared memory segment.)

This is another example of background workers that might interest you,
just to see how they're doing it:

https://github.com/citusdata/pg_cron

In that case they start the worker upon database startup, so it doesn't
have problems with shared memory going away.

I took a stab recently at using background workers to run ad hoc async
SQL, but eventually I wound up doing something simpler.

I think background workers is a pretty interesting feature though, so if
you get something working I'm sure others would be happy to hear about
it. Good luck!

Yours,

--
Paul ~{:-)
pj@illuminatedcomputing.com

I've experimented with background workers a bit. Yes, normally you'd
write some C, but you could also install this (mentioned in the blog
post and the source of pg_background_launch):

Thanks for your answer. I want my logger to be used with the facilities
there are from PostgreSQL so I'll go the dblink way. The docs write that
fdw provides a more modern architecture for more ore less the same.
However, I do not think the execution of arbiträry stuff is possible by fdw.

Kind regards Thiemo

--
SIP/iptel.org: thiemo.kellner
Öffentlicher PGP-Schlüssel:
http://pgp.mit.edu/pks/lookup?op=get&search=0xCA167FB0E717AFFC