On Tue, Oct 13, 2009 at 11:40 PM, Mitesh51 <mit_bca1@yahoo.com> wrote:
>
> I like that "why on earth are you messing about with the
> data directory when you don't understand what it does and how it works?"
>
> Actully I am not a DB person and having almost ZERO knowledge abt it but I
> am working on an application which takes backup(full & incremental) of diff
> DB like mysql, postgres...
>
> Now I dont have any support who helps me on DB side & still I need to move
> on with my java code so in current situation I am trying my hands on the DB
> backups as well since none else has that knowledge so I am doing it by
> searching on my own :)
>
> I had 2 approach in my mind...to sync up transaction log files with specific
> full backup
>
> 1) to keep only time relavent files in pg_xlog dir and move other files to
> archive dir with code which is not a good idea as u suggest
>
> 2) to copy files from pg_xlog & archive dir(which is used by
> archive_command) and move files from the archive dir & not from the pg_xlog
> & thus it will be a another direction for solution of inc backup.
>
> Moving of files is done my postgres as well so I guess, from the archive dir
> we can do that.
There are really three reliable ways to take a coherent backup.
1: pg_dump
2: PITR
3: Snapshots.
What you're trying seems closer to PITR (Point In Time Recovery).
Look it up in the docs see if it makes sense. OTOH, snapshots,
combined with some kind of diff utility (rdiff is nice) an provide
incrementals quite easily. The deltas may be large if your db changes
a lot over time.
The cool thing about rdiff is that the latest bu is a full backup, and
everything else is deltas going back in time. I.e. instead of just
storing a delta, it's applied to the most recent (i.e. full) backup to
bring it forward, then store that and the delta to go backwards
instead of forwards. Snapshotting methodology is important, it has to
make a coherent at an instant in time snapshops or they may not work
properly.
