--- On Mon, 2/15/10, Kenneth Marshall <ktm@rice.edu> wrote:
> From: Kenneth Marshall <ktm@rice.edu>
> Subject: Re: [ADMIN] AIX - Out of Memory
> To: "Tom Lane" <tgl@sss.pgh.pa.us>
> Cc: "Thorne, Francis" <thornef@cromwell.co.uk>, pgsql-admin@postgresql.org
> Date: Monday, February 15, 2010, 11:18 AM
> On Mon, Feb 15, 2010 at 10:57:06AM
> -0500, Tom Lane wrote:
> > "Thorne, Francis" <thornef@cromwell.co.uk>
> writes:
> > > Looking for some help with regards to an 'Out of
> Memory' issue I have
> > > with our Postgresql install on AIX. When
> running large updates or
> > > select queries we get an out of memory error
> returned and details
> > > entered in the log file like below. This is
> a 64-bit install and I have
> > > set the ulimit for the postgres user to
> unlimited.
> >
> > The bloat seems to be here:
> >
> > > AfterTriggerEvents:
> 131063808 total in 26 blocks; 576 free (7
> > > chunks); 131063232 used
> >
> > but it's hard to believe you'd be getting "out of
> memory" after only
> > 130MB in a 64-bit build. Are you *sure* the
> postgres executable is
> > 64-bit? Are you *sure* the postmaster has been
> launched with
> > nonrestrictive ulimit? On lots of setups that
> takes modifying the
> > PG startup script, not just fooling with some user's
> .profile.
> >
> > > This is a 64-bit install (8.3) on AIX 5.3
> >
> > 8.3.what?
> >
> >
> regards, tom lane
>
> I no longer have an AIX box, but I had similar problems
> with other
> applications that needed large amounts of memory. Some OS
> specific
> steps needed to be taken to allow normal users to allocate
> large
> blocks of memory. The information needed was in their
> on-line docs
> as I recall, but I do not remember the details. The
> executables may
> need to be built with specific options/flags to work.
>
> Regards,
> Ken
>
Ken,
I recently saw a similar issue. It is two-fold:
1. I used "su -" to become the postgres user, and inherited the previous account's memory limits,
2. AfterTriggerEvents queues are caused by foreign key constraints, one per row. If you're loading data, dropping or
disablingthat constraint makes a world of difference. Just be sure to check afterwards if the RI has been violated
priorto recreating the FK constraint.
Bob