Can i change a storage controller on a 4 node cluster without evacuating the host

[Mokshir]

I am running into an issue where i am unable to evacuate a host in a 4-node cluster. I have free space of 12Tb and used space of 7Tb but I still cannot evacuate the host. I need to change the storage controller from onboard (LSI2008) to Del PERC H730. Can i just replace the controller and hope the disks show up or VSAN will tell me to do something after it detects a new controller?

[TheBobkin]

Hello Mokshir​

It likely isn't a case of not enough free space in general but that you have either insufficient space on the available Fault Domains or you don't have enough Fault Domains.

E.g. if you are using a RAID5 Storage Policy(SP) then this requires 4-nodes and thus Full Data Migration of one node is not possible (as this would violate the rules of this SP). Other scenarios would include if this was a 2+2+1 cluster then you wouldn't be able to do Full Data Evacuation of one node unless the clusters capacity was <50% used as it would only be able to move the data from the evacuated node to the other node it that site.

If you are using RAID5 then you should take back-ups and then use Maintenance Mode with 'Ensure Accessibility' option, rebuild the hosts one at a time and allow the data to resync back to compliance between doing the changes on each node.

"Can i just replace the controller and hope the disks show up or VSAN will tell me to do something after it detects a new controller?"

Yes, provided the disks are in passthrough mode - if they are RAID0 on the controller this will not work.

Bob

[Mokshir]

Thanks Bob,

I do have a 4 Node-Cluster and a RAID5/6 Erasure coding VSAN Policy for the datastore. So if I understand your comment, when using RAID-5/6 Erasure Coding on a VSAN cluster of 4 nodes does allow me to evacuate the host?

the whole exercise for doing this was to be safe when changing storage controller which are in Pass-through (HBA) mode. If i can just replace the controller, then may be i can avoid evacuating the host. Hoping this will go through since i am not able to evacuate 1 host at time.

[TheBobkin]

Mokshir, FTT=1,FTM=RAID5/6 requires a minimum of 4 Fault Domains(FDs) - thus in a 4-node cluster using this you cannot do MM with Full Data Migration(FDM) as that would require 4 FDs to be available but you only have 3 available (as you are taking one out by placing a node in MM).

If you had 5 or more nodes in the cluster (and adequate space on each) this would be possible but not with 4-nodes, only MM with 'Ensure Accessibility'(EA) option which means your data is FTT=0 until the node is back in the cluster and the data resynced.

If you have adequate space then considering FTT=1,FTM=RAID1 would allow MM with FDM but this is not something that should be changed without understanding the space usage.

As I said above, if they are in passthrough access mode then you should be able to just put the host in MM EA and switch the controllers - do this with the servers powered-down unless you are 100% positive the controller and it's firmware+driver combination support hot-swap (or do it cold anyway to be sure).

Bob