3node vxrail, 1 node failes, what happens to vm's running on that node ?

3 Node VxRail cluster, it means VSAN cluster having 3 nodes.

By default, VSAN storage policy is RAID1, FTT=1 (failures to tolerate)

It means for a VM to be compliant, VM should have two mirror copies COPY1, COPY2, WITNESS

VSAN will decide the placement of these copies:

if there are only 3 nodes:

COPY1 on Host1

COPY2 on Host2

WITNESS on Host3

if one host failed, it still has more than 50% votes, so already powered on VM will keep running., VM storage policy compliance will be shown as non-compliant.

In 3 node cluster, if a host failed, VSAN will not have self-healing capability, meaning it cannot re-build the missing component on another host.

But when a host failed, and a customer wanted to create a new VM when there are only two hosts functioning, it will not be successful. because there is a host failed when you want to create a new VM, VSAN wanted to spread the copy1, copy2, witness to 3 hosts, when it searches for the 3rd host, it will not find, so VM cannot be created.

Hope this helps.

Apparantly all VM's that were running on the failed node will have to be restarted by HA on another node.
So you'll have downtime, but no data loss.

Is a 3 node cluster really a 2 node cluster with an off cluster witness? I can find no documentation that supports the witness running on the same cluster as the hosts.

3 node cluster is the 3 node cluster. it's different from 2node + witness.