Behavior of vSphere HA VM Component Protection APD policies

Products

• VMware vSphere ESXi

Issue/Introduction

This article clarifies the behavior of vSphere HA's response to All Paths Down (APD) events. The behavior itself has not changed in 7.0 when compared to previous releases like 6.5 or 6.7.

To view the APD Failure Response section events from vCenter UI, go to Hosts and Cluster > <click on a cluster of choice> > Configure tab > vSphere Availability >  click Edit > Expand Datastore with APD item.

Currently, you have four options to choose:

1. Disabled
2. Issue Events

The Disabled and Issue Events do not alter anything for the VM itself. The former does nothing while the latter issues some events that are visible on the vCenter GUI.

3. Power off and restart VMs - Conservative restart policy
4. Power off and restart VMs - Aggressive restart policy

When an APD occurs for a datastore on a host, if either "Conservative" OR "Aggressive" options are selected, vSphere HA will first try to find a host where it can be powered on. If it can find one, the VM is terminated and attempted to power on at the selected host.

If a suitable host cannot be found is when there is a difference of behavior between "Conservative" and "Aggressive".
If the "Aggressive" behavior is chosen at the point where no suitable host was found to power on the VM, a further check is made to see if there is a network partition on the cluster. If partition exists, the VM is terminated (even when there was no suitable host). This is the key difference in behavior between "Conservative" and "Aggressive". "Conservative" will not try to terminate a VM even if there is a network partition.
For more information on these options and other reasons why a VM could not be failed over, see Datastore Inaccessibility is not resolved for a VM.

Environment

VMware vSphere 6.5.x
VMware vSphere 6.7.x
VMware vSphere 7.0.x