[DRBD-user] HVM windows servers with xen/drbd/ha failover

Scott Inderlied scott at northwestcomputer.biz
Tue Apr 14 11:39:00 CEST 2009

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

-----Original Message-----
From: Igor Neves [mailto:igor at 3gnt.net] 
Sent: Tuesday, April 14, 2009 4:08 AM
To: Scott Inderlied
Cc: drbd-user at lists.linbit.com
Subject: Re: [DRBD-user] HVM windows servers with xen/drbd/ha failover


Scott Inderlied wrote:
> I'm in the process of setting up a DRBD 2-node cluster with 2
> instances of windows server (one 2003, one 2008) virtualized in xen
> and a samba or possibly later iscsi data share volume. I've got CentOS
> 5.3 and DRBD 8.3.1 on LVM volumes. I've got DRBD up and working and
> almost have heartbeat (no hbstatus on one node, troubleshooting
> presently).
> My concern is what i've got to do to make live migration work with
> hvm. The documentation states:
> "Under these circumstances, you must use the traditional phy: device
> syntax and the DRBD device name that is associated with your resource,
> not the resource name. That, however, requires that you manage DRBD
> state transitions outside Xen, which is a less flexible approach than
> that provided by the drbd  resource type."

If you use something like this "disk = [ 'drbd:myresource,xvda,w']", and
you use like me, lvm as disk for your drbd resource, you will notice
that it will not work, even if you do dual-primary mode.

Have something to do with too fast initialization of the xen virtual
machine, if you do it by hand, and not with heartbeat, it will work,
with heartbeat don't, it's too fast when putting in primary and start
virtual machine, virtual machine starts but says it has no disk
attached, or disk without boot sector.

Anyway you can try setting up dual primary mode, with any of the
clustered filesystems on it, anyway i thought lvm was faster, without
file system stuff in the middle.

For anyone interested, my resource write/read performance of one drbd
resources under xen, was not very good, and i never found the problem.
It was something like in the host, direct to resource 85MB/sec writing,
inside virtual machine over the same resource, I get about 45MB/sec.
Yes I was with PV drivers inside windows, and yes I have tested with 3
different windows versions (XP, 2k3, 2k8). I have also tested the same
disks without on drbd resource, so my phy was directly sda, and i get
writing about 95MB/sec.
Something very strange, if i disconnect my resource, i get full speed,
connected again i become with 45MB/sec again, and the most important,
this same resource sync's at 90MB/sec. How weird? :) (never found out
the problem).

> Is this much more difficult to set up than with the drbd: disk
> resource? Can heartbeat handle the state and xen transitions? I'm
> using CRM so can i use the Xen OCF with the drbddisk agent? If anyone
> has experience with similar situations i'd welcome the advice.

Yes, i have one cluster exactly with this setup working so far so good.

I have used phy for the disk, drbd resources in primary/secondary, and
use drbddisk ocf to promote the drbd resource, and use xen resource to
promote the xen virtual machine. This is basically what drbd: does, so,
this is not a problem, you only have to manage one more resource. :)

> --
> TIA,
> Scott

PS: this is not drbd related but will help you improving your xen

Good Luck,

Igor Neves <igor.neves at 3gnt.net>
3GNTW - Tecnologias de Informação, Lda
 SIP: igor at 3gnt.net	JID: igor at 3gnt.net 
 ICQ: 249075444		MSN: igor at 3gnt.net
 TLM: 00351914503611	PSTN: 00351252377120

Thank you so much for saving me potentially hours of frustration learning
what you've discovered. I've found some websites mentioning pv win drivers
but none that were free/OS. Do you have any pointers there?

More information about the drbd-user mailing list