KB2147958 describes a situation that occurs because of the change in behavior of vNUMA in newer ESXi releases. The mitigation described is to set the numa.vcpu.followcorespersocket Advanced setting to 1 on any ESXi 6.5/6.7 hosts, which tells the host to essentially use the same behavior as existed in ESXi 6.0.
So here's my question, when you have gotten all hosts to ESXi 6.5/6.7, does removing this advanced setting from a host, which requires it to be in maintenance mode to set or remove, cause the behavior described in the KB to occur again? If so, what is the recommended course to get VMs to the new configuration without powering off any VM that has the Cores per Socket setting to > 1 and set these back to 1, then power on until the migration to ESXi 6.5/6.7 is complete for the entire cluster, then reconfigure the VMs again to the original Cores per Socket setting? This would assume that by reconfiguring the VM that the numa.vcpu.followcorespersocket Advanced setting doesn't have to be messed with.
Thoughts on this from the community or any support folks reading this?
