Virtual Machine Management : Clustering

Issues when trying to add a single-node cluster under management in VMM

mlmich — Fri, 27 Nov 2009 19:33:00 GMT

A few folks have been experiencing errors and long delays in adding a single-node (i.e. a 1 node cluster) cluster under management in Virtual Machine Manager. This issue stems from the fact that VMM has an overcommitted property for each cluster. By default,...(read more)

Running the Virtual Machine Manager server component inside a Virtual Machine

mlmich — Wed, 21 Jan 2009 22:16:00 GMT

Running VMM inside a Virtual Machine is a fully supported way of running VMM. Our team tests this scenario and we have some customers deploying VMM in such an enviroment. If you would like to deploy VMM in such an environment, it is recommended to place the SQL server in a separate server, especially if your environment size will be substancial. SQL server can either be on a physical machine or on a virtual machine. Guidance on running SQL server as a virtual machine on Hyper-V can be found here: http://download.microsoft.com/download/d/9/4/d948f981-926e-40fa-a026-5bfcf076d9b9/SQL2008inHyperV2008.docx.

Because the VMM server component does not offer any high availability options, you can't make VMM a highly available application. However, there is a way to accomplish that by running VMM inside a Hyper-V Virtual Machine that is highly available. That way, when that VM fails over to another host, the VMM server fails over as well and can survive a hardware failure on the source host. This is a supported configuration for VMM. However, there are some gotchas that you have to be careful of in this scenario:

Make sure you don't migrate the Virtual Machine VMM is residing on from within VMM. This will fail the task and kill the VMM service in the process.
As per my comment above, the same restrictions on the DB apply here as well. Keep SQL in a remote server. SQL can also be clustered for high availability.

By keeping the VMM component inside a VM that is highly available, it means that on failover the VMM server's name and identity are left intact. Users can still connect to it and the host machines will authenticate the same server, so everything from a VMM perspective will work the same.

Update: With VMM 2008 R2 and Hyper-V R2 in Win2k8 R2, we are able to add two more scenarios to running the VMM server inside an HA VM:

Now the VM containing the VMM server can be live migrated from one host to another without any loss of service and no Jobs will fail
If you have installed the Administrator Console to a remote machine, it will not lose connection to the HA VM containing the VMM server during the live migration

Getting an incorrect SAN migration option during clone-vm in VMM

mlmich — Thu, 11 Dec 2008 20:20:00 GMT

If you have a Failover Cluster managed by VMM 2008, it is possible that you will get a placement page similar to the one below if you try to clone a Highly Available Virtual Machine that is part of the cluster. The placement page in the new-vm wizard below incorrectly states that the transfer type will be SAN. New Virtual Machines are always transfered (or created) over the LAN because we need to copy the binaries over to the target location for the VM. That copy is always done using BITS.

You can go ahead and ignore this transfer type and select the Host that best matches your needs for this Virtual Machine and complete the wizard. Our team is investigating a fix for this :)

Highly Available Virtual Networks and VMM

mlmich — Tue, 11 Nov 2008 03:46:00 GMT

We had a few customers ask us to clarify high availability (HA) with regards to Virtual Networks (VN).

In the VMM/Clustering case, there are a few considerations around virtual networks:

A VN to be considered HA has to have the same exact name in all nodes of the cluster (Virtual Network names are case-sensitive), the same location (the NICs to which the VN is attached to in each cluster node have the same location) and the same VN network tag
An HA VM in a cluster connected to a VN not marked as HA, would be marked as ‘Unsupported cluster configuration’ in VMM. Make sure the VN is HA in order to make the VM healthy again.

If you fail to properly configure your virtual networks in a cluster, a highly available VM might lose network connectivity upon failover to another cluster node.