[DRBD-user] drbd and lvm

[Lars Ellenberg]

please reply to the list!

/ 2004-03-19 10:44:15 -0800
\ Kees Cook:
> On Fri, Mar 19, 2004 at 06:30:18PM +0100, Lars Ellenberg wrote:
> > / 2004-03-19 08:35:20 -0800
> > \ Kees Cook:
> > > On Fri, Mar 19, 2004 at 04:01:56PM +0100, Baldur Norddahl wrote:
> > > > Do you have any advice of which filesystem is best suited for this? It should
> > > > support online resizing, work well with DRDB and PostgresQL.
> > > 
> > > I am personally a fan of reiserfs, but I've seen a lot of people rave 
> > > about XFS.  If you're sticking to the 0.6 DRBD, you can't use XFS (as far 
> > > as I understand), so if you need resizing-while-mounted, I think only 
> > > reiser will work for you.
> > > 
> > > (Hm, can jfs resize on the fly?)
> > 
> > since you *will* have a (very short) downtime resizing drbd,
> > resizing-while-mounted is a non-issue, so basically every
> > filesystem that can be resized with some tool will work.
> > ext3, reiserfs, jfs, whatever.
> 
> I would personally prefer running LVM on top of drbd.  That way you can
> just add more drbd-protected devices and chop up the space as you see fit
> with LVM, etc.  Much easier to deal with growing stuff.

But much harder for eventual disk failures.
Node failure: ok, no problem.
But if a single "PV" fails, you have to failover the full "VG".
Ok, you could just let the node in question commit suicide for
this, but...

> Bring down the secondary node, add a drive, bring up the node, bring up
> drbd on the new disk, fail services over to secondary node, add drive to
> primary node, finish other half of drbd for the drive.  On the secondary
> node, extend the volume group onto the new drbd drive, grow LVs and
> filesystems at will.  Tada, no service down-time!  :)

Yes, you can do so.
You need to use LVM2 for this to work.

	Lars Ellenberg