Re: pg_background (and more parallelism infrastructure patches)

On 2014-11-12 11:15:26 -0500, Robert Haas wrote:
> On Mon, Nov 10, 2014 at 4:58 PM, Andres Freund <andres(at)2ndquadrant(dot)com> wrote:
> > No, that's not what I was thinking of. Consider:
> >
> > export pg_libdir=$(pg_config --libdir)
> > mkdir $pg_libdir/plugins
> > ln -s $pg_libdir/auto_explain.so $pg_libdir/plugins/
> > PG_OPTIONS='-c local_preload_libraries=auto_explain' psql
> >
> > and use pg_background() (or something else using this infrastructure) in
> > that context.
> > Without having reread your relevant code I'd expect a:
> > ERROR: 55P02: parameter "local_preload_libraries" cannot be set after connection start
> > because the code would try to import the "user session's"
> > local_preload_libraries values into the background process - after that
> > actually already has done its initialization.
>
> Thanks for these pointers. Your directions turn out to be wrong in
> detail, because it's PGOPTIONS not PG_OPTIONS, and the plugins
> directory needs to be under $(pg_config --libdir)/postgresql, not just
> $(pg_config --libdir). But it was enough to get me on the right
> track. Long story short, it seems to work:

Sorry, was extracting it from a larger local script...

> [rhaas ~]$ PGOPTIONS='-c local_preload_libraries=auto_explain' psql
> psql (9.5devel)
> Type "help" for help.
>
> rhaas=# show local_preload_libraries;
> local_preload_libraries
> -------------------------
> auto_explain
> (1 row)
>
> rhaas=# select * from pg_background_result(pg_background_launch('show
> local_preload_libraries')) as (x text);
> x
> --------------
> auto_explain
> (1 row)

The question is whether the library is actually loaded in that case?
Because that normally only happens early during startup - which is why
it's a PGC_BACKEND guc.

> > How do you define 'SAFE' here? Safe against concurrent SQL level
> > activity? Safe against receiving arbitrary data?
>
> The first one. The code powering the background worker is just as
> trustworthy as any other part of the backend, so it can be subverted
> if you load malicious C code into the backend, but not otherwise.

Ah, good. Because the latter sounds quite hard, if not impossible lest
we recreate send/recv ;)

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services