[ digging through some old email ]
I wrote:
> "Diehl, Jeffrey" <jdiehl@sandia.gov> writes:
>> Brick* implies 1334 tables
> Hmm ... I wonder if the query is somehow holding onto a buffer pin for
> the last block of each successively-accessed table? Will look into it.
I find that indeed 7.1 and older versions do that, but 7.2 does not
(due to simplifications in heap_getnext). So this problem should be
gone as of the current release.
regards, tom lane