Curt Sampson <cjs@cynic.net> writes:
> On Thu, 27 Jun 2002, Tom Lane wrote:
>> So you haven't really solved the problem --- somewhere there is a query
>> being issued that ramps the backend up to a lot of memory. All you've
>> done is ensured that the backend won't hang around very long. The
>> persistent connection isn't really at fault, except in that it causes
>> backends to keep being used after their memory usage has become bloated.
> Yeah, but if the queries after that are not using all of the mapped
> memory, that should be swapped out fairly quickly because the
> machine is short on memory.
And the swapping activity is exactly the problem, isn't it?
In any case, we can't make much progress until we identify the query
that is making the backend's address space grow.
(Markus, you don't happen to have SORT_MEM set to a large value,
do you?)
regards, tom lane