Re: Missing clog, PITR - Mailing list pgsql-general

From Patryk Sidzina
Subject Re: Missing clog, PITR
Date
Msg-id 1267111262.4172.82.camel@ps-lap
Whole thread Raw
In response to Re: Missing clog, PITR  (Greg Smith <greg@2ndquadrant.com>)
List pgsql-general
> > 1) how do the clogs relate to wal shipping based replication? Clearly
> > the master doesn't need that clog but the slave does.
> >
> They should just be kept in sync.  There's some useful background on
> this topic at
> http://old.nabble.com/control-the-number-of-clog-files-and-xlog-files-td19173165.html
>

The thing is they are in sync but somehow the slave needs more clogs
than the master, otherwise it won't start up.


> > 3) is there a faster way to debug this problem? Clogs fill slowly. It
> > takes about a month on a very busy production server for a clog to be
> > removed by master DB.
> >
>
> You could create a bunch of transactions and then freeze things,
> following the ideas in the reference I suggested above.

Thanks for the info, I will try it as soon as I can.

> > More info:
> > PostgreSQL 8.2.14 64-bit (though this happened in older versions also)
> > pg_standby from PostgreSQL 8.3.6
> >
>
> There was a bug in this area fixed in 8.2.10:
> http://www.postgresql.org/docs/8.2/static/release-8-2-10.html
>
> "Fix potential miscalculation of datfrozenxid (Alvaro)
>
>     *
>
>       This error may explain some recent reports of failure to remove
>       old pg_clog data."
>
> If you were running this database with a version before that, I wonder
> if maybe there's still some junk left behind from that old, buggy
> version that's causing your issues.  You might try doing some manual
> VACUUM or VACUUM FREEZE work to remove any lingering issues and then
> re-create your standby systems afterwards.  I'm not quite familiar
> enough with this specific bug to suggest a clearer resolution path, or
> if in fact this is the same issue you're seeing.  It sure seems possible
> they're related though.


We run VACUUM on the master db every night (and VACUUM FULL on weekends)
and test the standby db using LVM snapshots. If our test generates the
missing clog error we have to recreate the standby from scratch. So I
doubt there is any junk left over, but I will look into this bug more
carefully anyway. Again, thank you for your suggestions.


--
Patryk Sidzina


pgsql-general by date:

Previous
From: hubert depesz lubaczewski
Date:
Subject: Re: postgres password change
Next
From: paragasu
Date:
Subject: Re: postgres password change