[DRBD-user] Hard disk upgrade

Matthias Zeichmann matthias at volltext.net
Fri Jan 30 08:17:18 CET 2004

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


<quote who="Lars Ellenberg" when="Thu, 29 Jan 2004 20:07:58 +0100"/>

> / 2004-01-29 16:47:55 +0100
> \ matthias zeichmann:
> > status quo is:
> > machine atem holds all the ha services and all drbd devices in a sane
> > state. [0]
> > machine kaelte has been freshly rebuild with new (larger) disks. some
> > drbd devices were partitioned to hold more diskspace [1]. drbd devices
> > are formatted ext3 and hold no data.
> 
> Well, the sync will void the formatting. This part of it was
> nonsense, unless you use plan B.

i was thinking along these lines too, tried it out on testmachines yesterday
and invalidated this plan.

> > Plan a:
> > bring up drbd on machine kaelte without changing drbd.conf (disk-size
> > still the same as on host atem)
> > meaning fullsync  atem -> kaelte
> > start heartbeat on host kaelte and let it assume its native services
> > (disksize of drbd devices should be as always)
> > stop heartbeat on host atem and let kaelte take over its services
> > stop drbd on host atem (-> hard disk upgrade)
> > 
> > and now the critical point which might as well be impossible to work
> > anyhow: change disk-size in drbd.conf to reflect the real disk-size of
> > the devices and restart drbd on host kaelte. will this blow my
> > machine/setup/cluster? 
> 
> plan C: hook up all disk in one box, an do
> dd if=/dev/good/data of=/dev/new/disk, then move one half of the
> disks to the other box.  its faster, you can skip two full syncs,
> but you'll have a downtime.
> 
> plan A:
> with minimal downtime I think it should work like this:
> bring up kaelte 
> 	The disk-size parameter should already correspond to the
> 	real (new) device size.
> 	As long as device size is larger than fs size, fs does not care.
> 	On connect drbd will "agree" on the minimum of device sizes.
> 	wait for the sync atem -> kaelte
> 	start heartbeat (and thus services)
> on atem stop heartbeat and services
> on atem stop drbd
> 	hard disk upgrade
> 	config (disk-size) upgrade
> 	rm /var/lib/drbd/*

this sounds interesting; i'll try this out in my testbed and report.

> Now you _could_ just start drbd on atem, and it _should_ work.
> I'd prefer to have a short downtime on kaelte first:
> stop heartbeat and all services, restart drbd, restart heartbeat
> and all services, just to be sure the bitmap and so on is
> correctly allocated.
> Then start drbd on atem, wait for the full sync kaelte -> atem,
> start heartbeat and services ...
> 
> plan B:
> of course you could just rsync it all over to the newly formatted
> disks, not using drbd. You have to make sure you get a consistent
> and uptodate view of your data.

that is the problem i had with this plan.

> Then first stop all on atem, second start all on kaelte,
> Now upgrade atem.
> Then start drbd on atem, wait for the full sync kaelte -> atem,
> start heartbeat and services ...
> 
> hth,

helps a lot, merci

m
-- 
siggen.pl: segmentation fault
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20040130/d3aa1e00/attachment.pgp>


More information about the drbd-user mailing list