On Wed, 2008-01-16 at 21:53 +0100, Michael Omotayo Akinde wrote:
> As far as I can see, the ulimits are set up as they should; and on a
> 64-bit machine with 16GB RAM, I don't see there should be a problem
> with allocating 2 GB maintenance work memory. In any case, I have
> serious difficulty believing that the ulimits can be the problem since
> PostgreSQL reacts the same even with maintenance work memory set very
> low.
>
> Basically, all of the tests we've run over the past couple of weeks
> end the same. VACUUM FULL on the tables runs out of memory and
> crashes. Ordinary VACUUM runs fine (albeit slowly) - but recommends
> that one runs VACUUM FULL.
Not surprising: AFAIK VACUUM FULL does not limit itself to
maintenance_work_mem. Or put another way, VF ignores both kinds of
work_mem completely.
-- Simon Riggs 2ndQuadrant http://www.2ndQuadrant.com