We all are aware of DRS affinity and anti-affinity rules and we know the fact that HA doesn’t respect the VM-Host should rules and if any host fails then HA can restart the VM’s anywhere and not necessarily on the host to which VM’s has affinity rules tied up.
However on next invocation of DRS (default 5 minutes) this violation would be corrected by DRS itself and VM’s will return to the host according to the VM-Host affinity rules.
With vSphere 6.0 VMware has introduced one new advanced configurable option in HA. This setting is called “das.respectVmHostSoftAffinityRules”. As the name itself suggests, this setting will let HA to respect VM-Host affinity rules when it can and can violate it when HA doesn’t have much choice.
To make things clear let’s consider the below scenario.
We have 4 Esxi hosts and each host have 10 running VM’s. 3 VM (VM-a, VM-b, VM-c) on the host 2 has affinity rules configured to run only on host 2 and host 3. Now suppose host 2 fails and HA starts the failover. We have defined “das.respectVmHostSoftAffinityRules” in HA so HA ideally can restart VM-a, VM-b and VM-c on host 3 in accordance to the VM-Host affinity rules.
But suppose if the host 3 is already heavily loaded and running on nearly full capacity and it cannot accommodate any new incoming VM, then HA has no choice but to restart the VM’s elsewhere (may be on host 1 or host 4)
The second scenario can be none of the host where failed VM’s has affinity rules tied is available at the time of failover. In this case HA will behave just like it is doing till version 5.5 and can restart VM’s anywhere on the remaining nodes of the cluster.