[Linux-cluster] LVM Snapshots

[gordan at bobich.net]

On Wed, 6 Feb 2008, Jakub Suchy wrote:

> i am currently designing a cluster of XEN machines, having two nodes and
> one shared storage (SAS). We have Recovery Point Objective of 15
> minutes, so I am thinking about putting XEN machine on a LVM partition
> and making a snapshot of this partition every 15 minutes.
> Do anybody know any caveats of this? I wonder how fast would this be,
> i am afraid that every 15 minutes is a little bit too often. XEN machine
> will be Sybase server.

The caveat is that snapshot != backup. Not even close.
For a start, the underlying FS will be about as consistent as you'd expect 
it to be if you just power cycled the machine while it's working. A 
journalled FS will work around the need to fsck it, but your DB files (and 
in fact, any open files) are likely to end up not being internally 
consistent. You'd need to run check/repair on all your DB tables, for 
example. That may or may not be an acceptable solution for you, because it 
is not guaranteed that you will not lose any data.

Other than that, snapshotting frequency isn't a problem. It's just an undo 
log, not a complete copy. Bear in mind, though, that the more snapshots 
you concurrently have, the more times you multiply the writes, so the disk 
I/O preformance will degrade if you are running too many simultaneously.

Gordan