Changes to target\device, issue with datastore
Posted: Tue Sep 25, 2012 2:16 pm
Just wanted to run this past everyone to get some input, make sure I'm sane...and see if I'm missing anything.
Currently, I publish 2 300GB deduped targets to an ESX cluster (no StarWind HA...just a single server with deduped devices). I recently noticed that one of the spdata files had grown HUGE - around ~450GB due to several SVM actions to and from this datastore (without dedupe-delete). I wanted to recreate the device to get the space back down.
I SVM'd all the VMs off the datastore in question to another datastore (on a completely seperate server) and proceeded to unmount and delete the datastore with the huge spdata file. I then deleted the associated device files and recreated a new 300GB deduped device and assigned it to the same target it was previously assigned to. *Please note that I did NOT modify the second 300GB device AT ALL *
The problem occurred when the Rescan happend on the first ESX host - the new device showed up fine, but the other datastore (didn't modify this one) went into a missing\down state! I tried several things including re-assigning the device to the target, several storage rescans...the device was showing up with a completely different identifier in ESX and NOTHING was changed on this 2nd target\device in StarWind!
If I added this device as a datastore, a partition table was NOT found, and ESX was trying to create a new one thus wiping all data! I tried a few things to recreate the partition table, but it would not recreate. Long story short, I lost the VMs that I did not SVM off (but should not have needed to because I didn't modify this device\target).
Questions:
- Is this a known issue, has it happened before, etc? I have seen a few posts that are similar
- What is the correct procedure when modifying one of many targets on s StarWind server?
- Do the datastores need to be dismounted prior to making changes? Including recreating devices on the same targets or changing the cache size, etc?
- Same question for doing StarWind software updates?
I have no problem SVMing everything off the datastores that are about to be maintained, and I assume this is probably not as big of an issue when using HA....just seems like a bit of a long process, etc.
Thanks
Currently, I publish 2 300GB deduped targets to an ESX cluster (no StarWind HA...just a single server with deduped devices). I recently noticed that one of the spdata files had grown HUGE - around ~450GB due to several SVM actions to and from this datastore (without dedupe-delete). I wanted to recreate the device to get the space back down.
I SVM'd all the VMs off the datastore in question to another datastore (on a completely seperate server) and proceeded to unmount and delete the datastore with the huge spdata file. I then deleted the associated device files and recreated a new 300GB deduped device and assigned it to the same target it was previously assigned to. *Please note that I did NOT modify the second 300GB device AT ALL *
The problem occurred when the Rescan happend on the first ESX host - the new device showed up fine, but the other datastore (didn't modify this one) went into a missing\down state! I tried several things including re-assigning the device to the target, several storage rescans...the device was showing up with a completely different identifier in ESX and NOTHING was changed on this 2nd target\device in StarWind!
If I added this device as a datastore, a partition table was NOT found, and ESX was trying to create a new one thus wiping all data! I tried a few things to recreate the partition table, but it would not recreate. Long story short, I lost the VMs that I did not SVM off (but should not have needed to because I didn't modify this device\target).
Questions:
- Is this a known issue, has it happened before, etc? I have seen a few posts that are similar
- What is the correct procedure when modifying one of many targets on s StarWind server?
- Do the datastores need to be dismounted prior to making changes? Including recreating devices on the same targets or changing the cache size, etc?
- Same question for doing StarWind software updates?
I have no problem SVMing everything off the datastores that are about to be maintained, and I assume this is probably not as big of an issue when using HA....just seems like a bit of a long process, etc.
Thanks