2009/9/25 Jeff Janes <jeff.janes@gmail.com>:
> 2009/9/22 Grzegorz Jaśkiewicz <gryzman@gmail.com>:
>> On Tue, Sep 22, 2009 at 1:46 PM, Alan McKay <alan.mckay@gmail.com> wrote:
>>>> Best practice to avoid that, is to bump the work_mem temporarily
>>>> before the query, and than lower it again, lowers the chance of memory
>>>> exhaustion.
>>>
>>> Interesting - I can do that dynamically?
>>
>> you can do set work_mem=128M; select 1; set work_mem=64M;
>>
>> etc, in one query.
>
> But if all backends are running this one query at the same time, it
> won't help because they will all bump up their limits at the same
> time. If they are all running different queries, and just one of them
> really gets a big benefit from the extra memory, but the rest just use
> it because they think they have it even though it is only a small
> benefit, then bumping up just for the query that gets a big
> improvement could work.
This is, I think, a possible area for future optimizer work, but the
right design is far from clear.
...Robert
