On Tuesday 09 September 2008 04:37:09 Magnus Hagander wrote:
> Greg Smith wrote:
> > On Tue, 9 Sep 2008, Magnus Hagander wrote:
> >> As long as your SAN guarantees an atomic snapshot of all your data
> >> (which every SAN I've ever heard of guarantees if you're on a single
> >> volume - entry level SANs often don't have the functionality to do
> >> multi-volume atomic snapshots, though), you don't need to set up PITR
> >> for simple backups
> >
> > It's all those ifs in there that leave me still recommending it. It's
> > certainly possible to get a consistant snapshot with the right hardware
> > and setup. What concerns me about recommending that without a long list
> > of caveats is the kinds of corruption you'd get if all those conditions
> > aren't perfect will of course not ever happen during testing. Murphy
> > says that it will happen only when you find yourself really needing that
> > snapshot to work one day.
>
> Well, I agree one should be careful, but I don't see the risk if you
> just change all those ifs into a single one, which is "if all your data
> *and* WAL is on the same SAN LUN".
>
> (heck, you don't need hardware to do it, you can do software snapshot
> just fine - as long as you keep all your stuff on the same mountpoint
> there as well)
>
That's pretty key, but there can be advantages to doing it using the pitr
tools, and I think in most cases it would be hard to argue it isn't safer.
As a counter example to theo's zfs based post, I posted a linux/lvm script
that can work as the basis of a simple snapshot backup tool, available at
http://people.planetpostgresql.org/xzilla/index.php?/archives/344-ossdb-snapshot,-lvm-database-snapshot-tool.html
And yes, I prefer working on the zfs based one :-)
--
Robert Treat
http://www.omniti.com
Database: Scalability: Consulting:
