Serviceguard Lock LUN within VMware ESX

[Erich Freunberger]

hello,

i am trying to set up a sglx cluster for testing purposes, especially for a migration from a lock lun to a quorum server.

i created 2 virtual machines, a shared disk, verified that it is shared by writing data to it and read it on the other vm.

but when i do a cmcheckconf -C, i get:

Error: /dev/sdb1 is not a valid device. Validation failed.
Error: Invalid data for cluster lock LUN configuration

has anyone had success with this ?

[Steven E. Protter]

Shalom,

I'd be surprised if a SG lock disk configuration was supported within VMWARE at all. Perhaps I'm about to be surprised but I thought SG needs a little more physical control of the system than vmware gives it.

The error message seems to support this viewpoint.

RHCS as far as I know does not use disk locking and though probalbly also not supported might work.

SEP

[Steven E. Protter]

Sorry,

They just added a quorum lock disk in the late updates to RHCS 4 update 4. I don't know if this feature will work with vmware.

SEP

[Erich Freunberger]

hi,

i am sure that a vmware virtual disk is not supported as a cluster lock lun, serviceguard does not give any details on why it does not accept the disk.

if the problem is that not all scsi commands are supported on a virtual disk, maybe a raw lun mapping might help, but i could not establish a connection from both virtual machines to one EVA LUN. do you know how to map one LUN to 2 virtual guests ?

thanks,
erich

[Jarle Bjorgeengen]

Why not just use a 3rd VM as a quorum server , and avoid having the cluster lock at all ?

- Jarle

[Erich Freunberger]

hi,

i assumed that using a quorum server works - and it does, the cluster is running with a quorum server (3 VMs, as you suggested), so far i did not try creating a package, but can see the created PVs on both nodes.

but what i wanted to try at least once before doing it on a production system was to change an existing cluster from lock lun to quorum server.

erich

[Serviceguard for Linux]

As I think you are aware, this is not supported. That said, since you are trying to do this for "development purposes" I think I can help.

If you are running all of your virtual nodes on a single physical machine then LockLUN will not work. It is because of how the disks (and especially file systems used for create the shared disk) are handled by VMware in that case.

If you set up the virtual machines on two different servers AND use external shared storage, then it can be made to work (I have proof by someone else's example). I don't know if there are any special set up requirements. If you have this configuration, tell me and I may be able to get and post more details.

As you have discovered, setting up a QS, even when all virtual servers are on one physical server, does work.

I'll repeat again for everyone's benefit. This is not supported. I am describing it to allow you to experiment and learn more.

[Erich Freunberger]

i have only one esx server with access to an EVA storage system and i tried to use a shared virtual disk and a mapped raw device as well, with the scsi controller set to virtual and physical sharing.

as moving from a lock lun to a quorum server requires probably just a change of the ascii config file and applying that to the stopped cluster, it won't be a big risk to do it without testing directly on the production cluster.

using vmware guests was just meant to do a quick try on easily available test systems -nothing one would do in production systems, as already said - it is not supported.

thanks for your help!

[Serviceguard for Linux]

You are correct, the change is simple. One thing you can do, is test the quorum server from the two nodes on ESX. That will give you confidence that the QS is configured correctly.

Best regards

[Erich Freunberger]

thanks!