VDI on VSAN: Disconnected Replica (Node down)

[RyanH84]

Hi all,

I'm hoping that someone can help me with this one as I can't see a way around this other than the manual way.

We have a VDI cluster of 16 nodes, running 6.0 U1 and View 6.2.1.

We have had a host fail and it's currently down for maintenance due to hardware replacement. This has taken some of our linked clone VM's (not an issue) and also it has disconnected one of our replica VM's as this Host was where it was registered to.

http://imgur.com/z0EBopz

Now, luckily for us this replica is linked to a pool that is used for testing and not our 500+ desktop pool. If it were, I'm fairly confident that our provisioning actions would start to fail. The manual fix I would have for this would be to remove that replica object from the inventory and re-add it in from browsing the datastore.

Has anyone had this before? Is what I'm thinking correct? Finally is there anything that can negate this behaviour?

I appreciate any input!

[douglasarcidino · Accepted solution]

This shouldn't be a problem honestly. If a provisioning operation starts it should just clone a new replica. Removing the replica and registering to a new host is not the right answer. This will cause you many...many problems. If the host is never coming back then you can manually remove the replica from the DB and vCenter. The entire idea behind the replica VM is that it can be destroyed at will because your base image is somewhere else.

Can you give us some more details on your host, storage and pool configurations? Have you tried refreshing one of the linked clones in the pool just to see what happens?

Doug

[RyanH84]

Hi Douglas,

Thanks for the response. I guess you are right, removing the replica and adding it from the inventory again isn't the best of ideas. Allowing View to recreate it would be the best option.

The host was down for around 8 hours and the replica was off that entire time (but a belonged to a test pool so not really being accessed). It is all back and working as the host is online after maintenance. I didn't try refreshing a clone in the pool but might try that next time in order to prove that what you have described is the behaviour we see. I might take a host out in our pre-production cluster and replicate the issue, that would be a good piece of work to prove confidence in that process.

The question was more theoretical and a what-if rather than having a major issue with our pools. Thanks for your advice though, makes total sense.

Cheers!

[douglasarcidino]

Are you using shared storage for your view desktops or local storage?

[douglasarcidino]

I'm an idiot, you said that it was VSAN. What's your storage policy like? The replica should be stored on VSAN and accessible to your other hosts unless this pool is just running on local storage.

[RyanH84]

Hi,

Yes it's VSAN

The replica is on this with everything else and the View policies automatically create a policy for the replica which has FTT = 1. So yes, the objects are still visible. Hence I can remove the old replica (inventoried on the downed host) and add it in from the datastore again on another host.

Due to the replica VM being registered to the downed host it was unavailable.

[douglasarcidino]

I guess, I need to setup a virtual VSAN for a nested view environment in my lab to check this out. Rebalance or recompose would fix you if this came up again but I would love to come up with a more elegant solution to this. If you're using VSAN, it should be just as good as a real SAN. You've got me thinking now.

Side note, this is going to make me spend more time in my office which my girlfriend won't like. I would recommend you have someone else start your car for the next few weeks :smileygrin: