"Scott Marlowe" <scott(dot)marlowe(at)gmail(dot)com> writes:
> On Wed, Sep 10, 2008 at 8:58 AM, Kevin Grittner
> <Kevin(dot)Grittner(at)wicourts(dot)gov> wrote:
>> Some of my high-volume databases don't quite go back to 0000, but this
>> does seem to be a problem. I have confirmed that VACUUM FREEZE on all
>> but template0 (which doesn't allow connections) does not clean them
>> up. No long running transactions are present.
> I have a pretty high volume server that's been online for one month
> and it had somewhere around 53, going back in order to 0000, and it
> was recently vacuumdb -az 'ed. Running another one. No long running
> transactions, etc...
The expected behavior (in 8.2 and newer) is to maintain about
autovacuum_freeze_max_age transactions' worth of clog; which is to say
about 50MB at the default settings. If you've got significantly more
than that then we should look more closely.
I don't remember what the truncation rule was in 8.1, so I can't speak
to the OP's complaint.
regards, tom lane
In response to
pgsql-performance by date
|Next:||From: Kevin Grittner||Date: 2008-09-10 17:47:54|
|Subject: Re: too many clog files|
|Previous:||From: Mark Wong||Date: 2008-09-10 17:38:30|
|Subject: Re: Effects of setting linux block device readahead size|