Re: buildfarm: strange OOM failures on markhor (running CLOBBER_CACHE_RECURSIVELY)

Andres Freund <andres@2ndquadrant.com> writes:
> On 2014-05-19 23:40:32 +0200, Tomas Vondra wrote:
>> I was however wondering if this might be related to OOM errors a few
>> local users reported to us. IIRC they've been using temporary tables
>> quite heavily - not sure if that could be related.

> I've significant doubts that it is.

The main leak (the whole-Relation one) only applies to entries for certain
system catalogs.  The smaller leak would apply to any relation having
DEFAULT column values --- but it's fairly hard to see how any table would
get invalidated so many times in one query as to make that leak an issue,
unless you were doing CLOBBER_CACHE or similar stress testing.

> I suggest you disable memory overcommit - then you'll get nice memory
> dumps on stderr instead of oom kills.

Yeah, a context stats dump would help a lot here.
        regards, tom lane