Q. When does VMware High Availability (HA) VM restart priority become important?

A. Last time I told you that options exist when you can't afford the ESX cluster size you really want, when money is tight and cluster reserve represents wasted cash. One option is to simply disable admission control, but that option could create virtual machine (VM) performance problems down the road when a host fails.

A second option requires a change in policy. The admission control policy Percentage of cluster resources reserved as failover capacity provides a mechanism to reserve a smaller quantity of reserves than the more-common Host failures cluster tolerates policy.

That last article doesn't, however, address an extremely important facet of reducing your reserve. If you reduce reserved resources to a level lower than one host's contribution, some VMs won't have enough resources to power on after a failure. Admission control will specifically prevent those VMs from failing over.

I'll venture a guess that every ESX cluster has range of VMs. Some are extremely important, such as your mail server or production database. Others are less important, such as your IT file server or demo machines. You can probably tolerate these less-critical VMs not being automatically failed over should you lose a cluster host.

Figure: Cluster default settings

In this case, another page in the New Cluster Wizard becomes useful. As the figure above shows, the page defines the cluster default settings for VM restart priority and Host Isolation response. Many IT pros ignore this page during their first cluster creations, not realizing its impact on a cluster's health. VM restart policy is particularly useful when you've constrained your cluster's reserved resources to a quantity that's less than one host.

Three VM restart priorities are available: High, Medium, and Low. Setting the value here sets it as the default for all VMs that participate in the cluster. What most admins don't recognize is that this setting establishes a baseline so that you can tell which VMs should power on after a failover. Individual VMs can be set to a different restart policy than the cluster default, and if you're excessively constraining admission control, you'll absolutely want to set their policies differently.

My last Q&A discussed how an overly-constrained cluster must prevent some VMs from powering on. That's the case when an overly-constrained cluster hasn't set aside enough resources for every VM to fail over. When that happens, you surely want your high priority VMs to power on while relegating the lower-priority ones to a "power on if resources exist" status.

You can accomplish that by setting the cluster default to Medium. Then, after the cluster is created, go back into the cluster properties and reset individual VM priorities to High. The cluster will attempt to restart those VMs first.

If you want two levels of prioritization for VMs, consider setting your cluster default to Low. Then you can set individual VMs to Medium as well as High. Obviously, the High priority VMs start first, followed by the Medium priority ones, until your resources are fully consumed.

VM restart priority has other uses, but this is one that comes in handy until your budget, and thus your cluster size, gets a bit bigger.

Catch up with @ConcentratdGreg on Twitter!