On Wed, Feb 15, 2012 at 4:42 PM, Khusro Jaleel <mailing-lists@xxxxxxxxxxxxxx> wrote:
Hi,
I have setup replication on Postgresql 9.1 in a simple scenario using a master and a slave. I now need a reliable way (and preferably fast way) to do backups of the database.
I tried using 'archive_command' on the master and created a script that simply bzips files and copies them to a directory, but after 1 day that directory was so huge that even a simple 'rm' complained that the 'argument list is too long'.
I can see that even though I have specified 30 minutes in the postgresql.conf, a new archive file is written every *minute* resulting in way too many files in the archive directory. This happens even when *nobody* is using the DB, or with just very light use. However I have noticed that sometimes this *does* work correctly and an archive is only written out every 30 mins.
Is there a way to force this to just write to the archive directory *only* every 30 mins as I only really need backups to be current to the last 30 mins or so?
No, there's no way to do this. If you have good number of transactions/minute (and/or the transactions are quite large) you will have a lot of stuff written to transaction log, so wal files would rotate quite often,
I want to avoid using pg_dump as I think that would require I pause writes to the DB until the backup is finished?
pg_dump won't block writes, thanks to MVCC. It may increase bloat and it will block DDL operations (ALTER TABLE/etc), but if your database is relatively small but have high load and you need frequent backups, this may be a way to go.
Vladimir Rusinov
http://greenmice.info/
