VMware Cloud Community
primeX
Contributor
Contributor

vSAN 3-Node Cluster Upgrade

Hi. Is it possible to upgrade my 6.7 ESXi vSAN 3-Node cluster to ESXi 7 without any downtime? Any help will be much appreciated. Thank you.

Reply
0 Kudos
9 Replies
TheBobkin
Champion
Champion

@primeX Sure, just upgrade the vCenter to the relevant version (e.g. 7.0 U1 or 7.0 U2) and then update the hosts one at a time (to ESXi 7.0 U1 or 7.0 U2 (whichever vCenter was upgraded to) using Maintenance Mode 'Ensure Accessibility' option.

 

One thing to be mindful of is to not add, recreate or perform on-disk format upgrade of any Disk-Groups until all hosts have been updated to the later version.

primeX
Contributor
Contributor

Hi @TheBobkin . Great! Thank you for this.

After I upgrade the first node, could it be clustered to the old version that's left of one of the ESXi remaining two hosts since I will continue on upgrading the second node?

 

 

Reply
0 Kudos
TheBobkin
Champion
Champion

@primeX, Yes, the cluster remains formed when nodes of different versions are in the cluster and all data should remain available.

Thus rolling update of all nodes can be done without downtime for the VMs as they just get vMotioned between the nodes.

Reply
0 Kudos
primeX
Contributor
Contributor

This is noted @TheBobkin . Thank you.

I read that vSAN 7u1 has a larger usable capacity compared to 6.7. After I upgrade it to 7u1, will it automatically increase the usable capacity?

 

 

Reply
0 Kudos
TheBobkin
Champion
Champion

@primeX , It doesn't increase the vsanDatastore size - the changes made were to how data is structured so that deep-reconfiguations (e.g. a complete rebuild in a new layout) of data Objects requires less space, this was part of the slack-space guidance on vSAN of 25-30% which is no longer the case.

Reply
0 Kudos
primeX
Contributor
Contributor

So it means that no slack space is required anymore? Fab! Thank you so much for your help, @TheBobkin!

Reply
0 Kudos
TheBobkin
Champion
Champion

@primeX, So, slack space in vSAN was always required for things such as changing storage policies and it still is, but just a lot lot less as it can do deep reconfigurations a single component at a time as opposed to having to temporarily reserve the whole new layout space of Objects (before removing the old components from the old layout).

 

The original slack/reserve space advised for regaining data redundancy after a failure or for data evacuating a node being the same for all clusters of all sizes didn't make sense for every configuration - e.g. a 4-node vs 40-node cluster would need vastly different proportions of their free space to repair data following a node failure/removal.

Thus these recommendations have been revised and new parameters configurable for 'Host rebuild reserve' and 'Operations reserve' - these could be viewed as being similar to how HA failover reserve for compute resources works and similarly are entirely optional:

https://blogs.vmware.com/virtualblocks/2020/09/24/effective-capacity-management-with-vsan-7-update-1...

Reply
0 Kudos
primeX
Contributor
Contributor

It's a good read. Thank you @TheBobkin! Appreciate it.

Reply
0 Kudos
primeX
Contributor
Contributor

@TheBobkin, I'm also planning to have a DR infrastructure probably by early next year. Can I utilize vSphere replication by using Standard licenses vSAN HCI on my production and Essentials plus kit on my DR site, running on converged?

Reply
0 Kudos