On 05/09/2012 03:08 PM, Tom Lane wrote:
> I see no memory leak at all in this example, either in HEAD or 9.1
> branch tip. Perhaps whatever you're seeing is an already-fixed bug?
>
> Another likely theory is that you've changed settings from the 8.1
> installation. I would expect this example to eat about 10 times
> work_mem (due to one tuplestore for each generate_series invocation),
> and that's more or less what I see happening here. A large work_mem
> could look like a leak, but it isn't.
Good call -- of course that just means my contrived example fails to
duplicate the real issue :-(
In the real example, even with work_mem = 1 MB I see the same behavior
on 9.1.
> If you need further help in debugging, try launching the postmaster
> under a fairly restrictive memory ulimit, so that the backend will get a
> malloc failure before it starts to swap too badly. The memory map it
> will then print on stderr should point to where the memory is going.
Thanks -- will try that.
Joe
--
Joe Conway
credativ LLC: http://www.credativ.us
Linux, PostgreSQL, and general Open Source
Training, Service, Consulting, & 24x7 Support