This is my first post, please be gentle
We have a cluster of 4 hosts. HA is configured and working properly.
What I would like to achieve is the following:
If one or multiple host fails, VMs of our choice (that are still running on working hosts) would shut down automatically to accomodate the more important VMs.
Is this even possible?
Thanks!
>The feature you're talking about is for VMs that are on the host that failed, right?
correct
>I'm more concerned with shutting down VMs that are on hosts that are still online. Am I making any sense?
This is not possible as a configruatble feature in HA. Best you can do I suppose is to disable admission control on your HA Cluster. This way if a host fails because of an HA event you're guests will power on even if you violate resource contraints.
You can edit settings of your Cluster, go to Virtual Machine Options and set the restart priority. For those guests that you want to come up first set the restart priority to high
I guess I didn't explain myself properly, sorry!
The feature you're talking about is for VMs that are on the host that failed, right?
I'm more concerned with shutting down VMs that are on hosts that are still online. Am I making any sense?
This is not possible currently.
Elisha
>The feature you're talking about is for VMs that are on the host that failed, right?
correct
>I'm more concerned with shutting down VMs that are on hosts that are still online. Am I making any sense?
This is not possible as a configruatble feature in HA. Best you can do I suppose is to disable admission control on your HA Cluster. This way if a host fails because of an HA event you're guests will power on even if you violate resource contraints.
I already have admission control disabled. So I guess some manual intervention is in order here.
Hopefully vSphere 5 will have this feature!
Thanks
Troy, whether admission control is enabled or not has no effect on what HA will do when a host fails. Admission control only prevents a user from powering on an additional vm if there will not be enough failover resources.
Elisha
Nope, this feature is not in vSphere 5. It is an interesting enhancement we are considering for the future, though
hmmm... Then the definition needs to be more clearly defined. I was under the impression that admission control if disabled will allow guests of a failed host to power on even if they violate resource constraints. With admission control enabled, well I was under the impression that you could put yourself in a situation that not all guests power on because you have no available slots.
FYI: there has been at least one other posting on the forums asking for this feature (http://communities.vmware.com/message/1570941#276262 - about a year ago). So you're not alone . Folks reading this thread - please pipe up if you'd like a feature like this.
Elisha
It is not possible to power on vms if their resource constraints (ie. memory and cpu reservations) cannot be satisfied. This is true both for user-initiated power on operations and HA failover operations. It is independent of whether HA admission control is enabled or not.
You should be able to use a script combined with a vCenter alarm to shut down specific machines. Alarms can be used to trigger a script.
I guess I misunderstood/stand the definition and partial use case of Admission control.... Lesson learned!
This is for instance what SRM also does, it allows you to select VMs that will not be restarted or can even let you shutdown VMs based on prio etc.
I do wonder though if you are seriously overcommitting your servers to such an extend that powering them on isn't possible anymore? I haven't seen many environments where this was actually the case to be honest.
-duncan
Also Troy, Admission Control is a task of vCenter. HA failovers take place on the host side, hence the reason HA is not prone to HA Admission Control.
Hi Duncan,
I don't want to be too nitpicky but that's not the real reason HA failovers are not subject to admission control. Even if HA admission control was performed at the host level, failover would still bypass it. The whole point of admission control is to prevent users from reserving all the resources in the cluster and not leaving any buffer for failover. When a failure does happen and HA needs to restart some vms, it needs to be able to use those failover resources which are off-limits to users (because of admission control). Those failover resources were reserved precisely to be used in a case of failure, so it wouldn't make sense for HA admission control to prevent HA from using them.
Elisha