Re: PG process architecture

"Milen Kulev" <makulev@gmx.net> writes:
> My questions is:
> Where PG is storing data dictionary information (coming form system pg_*
> tables) while parsing the queries ? 

There's a limited-size "catalog cache" in each backend process, which
might be the closest analogy to this.  Offhand I think it's fixed at
5000 rows --- there's no user-visible knob to control it, anyway.

> I suppose each each background process is parsing (and eventually caching)
> the parsed SQL
> statements in his own memory (within each backend process),

There is no such cache.  If you tell us to PREPARE something, we save
the plan, otherwise not.

> If I have a system with 50 or 100 connection (and the corresponding 100
> backend processes),
> and one session  creates an index on a given table, how do the other 99
> processes 
> notice that they can use (or at least estimate the appropriatness of the
> usage of) the new index ?

There's a cache-invalidation signaling mechanism for the catcache, which
is what drives this sort of thing.
        regards, tom lane