This weekend I choose to decommission one of my old servers. This involved “evacuating” it of all its VMs to a new server. Evacuating seems to be a quite an unpleasant term, and sounds like some kind of painful clonic irrigation for ESX. Actually what I mean is “cold migration”. I had two errors. The first said “Invalid configuration for device 1″ and the other “The specified key, name, or identify already exists”. What was annoying about both errors is the failure came at 99% of the progress bar, and cause the VM remain on the source ESX host.

The first was cause by having a locally mounted ISO image – which ordinarily wouldn’t cause a problem at cold migration (but certainly does in VMotion) but at the point of powering on the VM once it had been moved because – the local mounted image wouldn’t be on the destination ESX host. However, in this case the error didn’t manifest itself at power on, as the cold migrate failed at 99%. The cold migration did warn me about this connected ISO, but as I the chance to ignore the warning (not an error) I did. More fool me! Disconnecting and removing the configuration to the local image stopped this problem occouring. Moral: Its worth resolving any errors or warnings rather than finding out they cause a problem later – despite the ability to blag & bluff through error messages.

The second was more complicated. It was like I was trying to move a VM called DC1, to an ESX host that already had a VM called DC1 – which it didn’t. There was no reference to the VM elsewhere in VirtualCenter and no files called DC1 on the destination VMFS volume. The work around which I found on the forums after a google-wack was to temporarily rename the VM before the cold migrate – cold migrate the VM – and then rename it back after the process was completed. I never got down to the bottom of why this happened but I believe it was caused by some database intregrity issue (always a good one to blame, as people just go “mmmmm” and node ignorantly when you say that!) and perhaps that object was still referenced in the database somewhere had become orphaned (another good one to use when explaining something you don’t really understand) and hadn’t been purged yet in the standard knowledge consistancy (even better!!!) check process in SQL…

This entry was posted on Monday, October 29th, 2007 at 10:32 am and is filed under ESX, vCenter. You can follow any responses to this entry through the RSS 2.0 feed. Responses are currently closed, but you can trackback from your own site.