Re: plan time of MASSIVE partitioning ...

Alvaro Herrera <alvherre@commandprompt.com> writes:
> Excerpts from Tom Lane's message of vie oct 29 14:15:55 -0300 2010:
>> samples  %        symbol name
>> 447433   47.1553  get_tabstat_entry

> Is there a reason for keeping the pgstat info in plain lists?

Yeah: anything else loses for small numbers of tables per query, which
is the normal case.  I'd guess you'd need ~100 tables touched in
a single transaction before a hashtable is even worth thinking about.

We could possibly adopt a solution similar to the planner's approach for
joinrels: start with a simple list, and switch over to hashing if the
list gets too long.  But I'm really doubtful that it's worth the code
space.  Even with Zoltan's 500-or-so-table case, this wasn't on the
radar screen.

			regards, tom lane