Richard Yen wrote:
> Hi all,
>
> Quick little tip here.  I've noticed that on postgres versions 8.1 and
> up, the autovacuum feature can somewhat conflict with slony-I.
>
> In the past, I've noticed that during random times in the day, and
> depending on machine hardware specs, slony would begin to build up a
> lag which peaks at around 20 sec, and then suddenly disappear.  It was
> pretty annoying because I'd get paged whenever the lag crosses a
> 10-second threshold (company "policy" states this as a service
> degradation).
>
> Just a couple weeks ago, I checked out my logs, and found that the
> autovacuum cycles often coincided with the "slony lagging" pages, and
> thought I'd try to do something about it.  I decided to put entries
> into pg_autovacuum with the 'enabled' column set to 'f'.  These
> entries were *all* the tables in the slony schema (the ones that begin
> with _schema.sl_*).
>
> I figured that since I've already set the cleanup cycles in the slon
> daemon, 1) it's pointless to vacuum twice, 2) slony is responsible
> enough to do its own vacuuming, and 3) when slony decides to vacuum,
> it won't contend with itself for table access.
>
> All the "slony lagging" pages have stopped now (I still get one every
> night when a resource-hogging cronjob runs every night).  So, I guess
> it was indeed some sort of contention between the postgres autovacuum
> and the slon daemon.
>
> Just thought I'd throw this out in case anyone would need this info,
> or perhaps provide some further input.
>
I'll see about adding that to the documentation...