Microsoft Failover Clustering - Persistent Reservations issue

[thomps01]

Hi,

I'm trying to setup a fresh Win 2008 R2 failover cluster on my ESXi 5.0 hosts and sadly the cluster verification fails with a persistent reservation isssue.

I've followed all of the kb articles and whitepapers to correctly build the 2 guest VM's as follows:

Windows 2008 R2 Enterprise SP1

NIC for Public traffic

NIC for heartbeat private traffic

EagorThickZero disks for system

RDM added in physical compatibility mode

SCSI Controller (1:0) used in physical mode for both nodes

The storage array I'm using is a VNX 5300

I have Powerpath 5.7 running on each ESXi host

Does anyone have any idea where this issue is coming from?

I've attached the error message below.

[iw123 · Accepted solution]

Hi

Are the two guests running on the same host? Have you tried the validation when they are running on different hosts?

[thomps01]

Magic!

Indeed I hadn't noticed that the 2 nodes were on the same ESXi host and when I migrated one across to the other host and re-ran the test it completed without any issue.

Can you explain this?

I didn't expect this error.

[iw123]

Hi,

Yes, I'd had that same problem a while back. There's a bit of info in this KB - http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=103063...

The solution is to run the guests on different hosts, or to use virtual compatibility mode on the RDM, to run both guests on the same host.

[jdptechnc]

I'm not toally certain but it is likely due to the difference between how ESXi handles virtual RDM and Physical RDM.  With virtual RDM, ESXi is treating the RDM and resulting mapping file similarly to how it would handle a regular VMDK and that is evident in that it uses the same file locking mechanisms, you can do snapshots, etc.  The physical characteristics of the hardware are still unknown from the guest.

With physical RDM, you are directly attaching the LUN to the guest OS and bypassing the virtualization layer almost entirely.   The VM gets exclusive access to the SCSI device.  Two VM's on the same host could use the virtual mode RDM, but not physcial.  My guess is that the VM's on the same host would be accessing the LUN using the same physical HBA and virtual storage layer, and the vmkernel just doesn't know how to handle that, or is set up to refuse to handle it, but I can't give you a definitive answer.