Server 2k8R2 Cluster Shared Volume Redirected access

[bubbrown]

I have worked with a new hyper-v win2k8r2 cluster the last few days, and am about to pull my hair out! I have two 2k8r2 enterprise servers connecting to a starwind san v5.4. I have three targets configured on the san (1 quorum and 2 cluster disks.) I configure each node to connect to the san targets and they show in the cluster storage as online. When I add them to the cluster shared volume, one will always show online (redirected access). I get the 1034 errors that the expected signature is "blah" for the redirected disk. Of course the signature of the cluster disk in question is exactly the signature that it is expecting. I've determined this using diskpart. I was able to get the CSV to show online for both data targets for a few hours, but now one is always back to always showing as redirected. I've removed and added the "favorites" iscsi targets on the 2k8r2 nodes, completely removed all settings in the iscsi applet, readded them, and I keep having the same redirected access issues. i've tried transferring ownership to the other node, but it never helps. as soon as i remove the cluster disk out of the CSV, it goes back online.
The starwind config is simple... i simply have targets specified to local hp raid 6 storage arrays with no starwind HA settings... simple single san only. I am about to throw this whole hyper-v thing in the trash and go buy vmware. I could have set up 15 vmware clusters in the time i've spent on this simple hyper-v mess. anyone have any ideas?

[Aitor_Ibarra]

I've never had this problem and I've been using R2 + CSV since it was released. Normally, there are only two reasons for a CSV going into redirected mode - iSCSI path failure of all paths on one hyper-v node, and doing a VSS based backup.

On the Starwind side, I'm sure you've enabled cluster access to the targets already, as otherwise you wouldn't get as far as you have. It might be worth upgrading to 5.5.

On the Hyper-V side, make sure that you are using good NICs with latest drivers, and that you have all the hyper-v patches applied (some of which don't come via Windows Update, you have to download and apply manually).

Maybe do a chkdsk on the cluster volumes. Or try removing them from cluster storage altogether, remove from iSCSI except on one hyper-v box, reformat, add back to iSCSI, add to cluster, add to CSV.

Hope this helps...

PS - I take it you are NOT putting iSCSI and SMB (CSV redirected mode relies on SMB networking) over the same network, right?

[hixont]

I've had redirection occur twice on my Hyper-V server farm running over a StarWind SAN. The first time I had to resort to creating a new volume and migrating my data over to that volume then reformatting the redirected one. The second time it was an iSCSI problem. Once I identified the machine that had the conflict in my farm I fixed the iSCSI issues by disconnecting the volume from the server causing the isue and then reconnected them after flushing the iSCSI cached disk information.

I suspect that my first occurrance could have been an iSCSI issue as well, but my experience with the technology was limited at that time so I resorted to the brute force method.

[Max (staff)]

Last time we have faced this issue the client had error with his VSS agents. The issue had no particular symptoms, the best way is to ask Microsoft support. They are more familiar with Windows core you know

[Aitor_Ibarra]

A while back a project I was involved with discovered a bug in VSS which had to go back to Microsoft for a hotfix... this was in a non-iSCSI, non-clustered environment, but might also cause problems elsewhere. Basically each time a VSS snapshot was taken by a backup operation, the registry would grow with a virtual device being added, and that would never be cleared up. On servers with frequent backups, the registry would swell so big that a reboot could take hours. It seemed to be a problem unique to 2008 R2, but Microsoft said the underlying bug was present since 2000!

[anton (staff)]

Any idea on registry size / VM usage? I mean - at OP environment...