I have an issue where RDMs disconnected from the Guest OS (in this case, Windows 2008R2 64bit) using SnapDrive 6.3.1 are sometimes not fully removed from the virtual machine. The disk is not visible in the OS or in SnapDrive so appears to have been removed, but going to Edit Settings on the VM will still show the RDM as connected, and the LUN still exists and appears mapped on the filer. I've also noticed when viewing the storage paths on the host a number of 'dead' paths, which I assume are old RDM connections.
This happens on both ESXi4 and ESXi5 hosts.
This can cause problems if a LUN is removed from the filer, as it appears to be no longer connected to the Guest OS, when the VM still thinks it is connected.
Rescanning the datastores will remove the dead paths, however we have an automated process which maps/unmaps LUNS via snapdrive every hour to update some databases. Whenever a LUN is removed there's chance of a dead path remaining. Eventually these accumulate and seem to cause performance issues.
Of course I could manually rescan the datastores every week or so but a root cause would be nice!
This exact thing is also happening to us. The rescan drops the dead luns, but manually scanning isn't a good enough solution. We have had hosts disconnect because of the Snapdrive mounting the LUNS to the ESX's local datastore too when using Snapdrive 6.3.1. I've had opened many tickets with Netapp and Vmware but still no concrete solution.
At first it was the version of Snapdrive we were using, things clear up and then it happens again out of the blue. I have now been able to recreate the Host disconnection issue, and it happens during a Snapmanager backup for SQL. As soon as VMware scans the HBA's the Host disconnects. I haven't tried to connect Snapdrive to the host, because of HA, I wouldn't think you'd want to do that if the machine running Snapdrive migrated off that host, right?
Not sure if it's related but I have noticed that although the LUN is disconnected, the vmdk pointer file sometimes isn't removed from the datastore. All mounts/dismounts are being performed by Snapdrive CLI which I thought removed RDMs fully....?
I'll try raising a support case and see if they can help, but it's very hard to simulate!
Maybe still affect the ESX5 ? And the fact that VMDK pointer files is still there after is something that need to be investigating ? Maybe a interruption in the Snaprive / Vcenter execution of RDM lun remove ?
Documented in SDW 6.3.1R1 release notes under Known issues.