Two node VSAN disk balance questions

[MBrownWFP]

We recently deployed a 2node VSAN ROBO cluster at our corporate HQ. The witness lives in our primary datacenter with L3 connectivity back to the management and VSAN networks at the office site. Other than the known "Site latency health" false positive, all cluster health checks showed as green following deployment.

The new cluster recently began alerting that a re-balance is needed due to a single disk being out of balance by 17%. I'm somewhat familiar with the re-balancing process (and "gotchas" around it) as I've needed to run this numerous times at our two datacenters. I triggered re-balance for the site cluster and gave it 24 hours to complete and clear the task entry, but when I checked the next day the cluster was still throwing the same alert and the same single disk was still showing 17% out of balance.

I have run the same process twice since then with identical results. The process runs and the re-balance task clears but nothing appears to have actually changed with the data layout on VSAN.

Is balancing behaviour is different with 2 node cluster due to the smaller number of nodes contributing storage? If so, is there a workaround for a perpetual alert if there is any imbalance?

Thanks,

Matt

[dickinsonzach]

Good morning, are all of your VMs in compliance with their storage policies?  If not, re-apply them and do the re-balance again.  Thank you, Zach.

[depping · Accepted solution]

Keep in mind that in a smaller cluster with larger objects a rebalance may result in the same unbalanced situation as the objects are not split up in multiple smaller object in 6.5 or before. Unless you are reaching 80% full I wouldn't worry about this too much in smaller clusters

[MBrownWFP]

I did check that the VM in the cluster are compliant with their storage policy, and they are.

I had a quick chat with support about this behaviour and was provided with the same info that Duncan mentioned. When a cluster has low usage (less than 30% capacity consumed) with large components, it is expected that in some cases the data simply cannot be perfectly balanced. In these scenarios the only option for dealing with the balancing alert is to disable that alarm for the cluster.

We are going to leave the alarms enabled, its sufficient to understand why the alerts are present and that they are not of concern.

Thank you both for the inputs.

Matt