Fwd: vacuum tx id wraparound issues

Sorry, meant to reply on-list. Unfortunately gmail has made it awkward
to stay on-list despite many people complaining about the change.

---------- Forwarded message ----------
From: Greg Stark <gsstark(at)mit(dot)edu>
Date: Thu, Jun 25, 2009 at 4:28 PM
Subject: Re: vacuum tx id wraparound issues
To: Kevin Kempter <kevink(at)consistentstate(dot)com>

On Thu, Jun 25, 2009 at 3:29 PM, Kevin
Kempter<kevink(at)consistentstate(dot)com> wrote:
>
> I did a vacuumdb last nite and already this am I see vacuums to prevent
> wraparound. Do you have any thoughts on increasing the
> autovacuum_freeze_max_age value?

You can safely raise this value substantially. The main cost is that
the clog will take extra space to record the status of all these old
transactions. Beware that if you set this to something like 2 billion
then the vacuums that kick in at that point have a limited amount of
time to complete before the whole database shuts down.

However it's unusual to run into this situation at all, let alone
after a just one day. That would be over 2,000 transactions per second
every second for 24 hours.

I suspect your vacuumdb didn't actually vacuum some tables. Do you
have multiple databases? Did you do vacuumdb -a? What database and
table is the autovacuum process kicking in for?

--
greg
http://mit.edu/~gsstark/resume.pdf

--
greg
http://mit.edu/~gsstark/resume.pdf