Merlin Moncure wrote:
> n 2/6/07, Alvaro Herrera <alvherre@commandprompt.com> wrote:
> >Merlin Moncure wrote:
> >
> >> ya, it doesn't seem to match, as this seems to be repeating quite
> >> regularly. interesting that my 'clog' files start at 06B6 and count
> >> up. 0207 is way off the charts.
> >>
> >> a lot of applications are hitting this database, and so far everything
> >> seems to be running ok (i found this log msg by accident), but I am
> >> now officially very nervous.
> >
> >I don't think there's much cause for concern here. If my theory is
> >correct, this is an autovacuum bug which was fixed in 8.1.7.
> >
> >What I'd do is create a 0207 clog file, fill it with 0x55 (which is
> >"transactions committed" for all transactions in that interval), and do
> >a VACUUM FREEZE on that database. You'll need to set
> >pg_database.datallowconn=true beforehand.
> >
> >Of course, I'd copy the files somewhere else and experiment on a scratch
> >postmaster, running on a different port, just to be sure ...
>
> thats a big help, database is actually fairly huge, so I may have to
> just go ahead and do it. I'm off to a meeting, but I'll check back
> when I'm done and assuming nobody else says 'don't do that', I'll try
> the fix and post back with the result.
Well, you don't need to copy all databases for the test area, just the
base/<oid> dir for template0 (along with all pg_xlog and pg_clog files,
etc, but these shouldn't be as big as all the other stuff in base/).
--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.