Hi all,
I've searched a bit and can't seem to come up with a definitive answer here.
We have a complex DRS problem; I'll do my best to describe it.
Our production cluster has 6 hosts in it. They are all blades, spread across 3 blade chassis like so:
Chassis 1
ESXi1
ESXi2
Chassis 2
ESXi3
ESXi4
Chassis 3
ESXi5
ESXi6
Our new exchange environment will consist of (among other things) 2 VMs that we do not want to exist on the same CHASSIS (this implies that they cannot exist on the same host).
I can't seem to figure out a set of DRS rules that will cover this. I have a feeling that the secret might lie in groups, but I just can't seem to quite nail it down.
Any thoughts out there?
Thanks.
Ryan
Hi,
I am afraid, I got your query correctly.
But You need to create Virtual Machine DRS group to place Exchange VMs in to the groups and Host DRS groups to place required hosts based on the chasis.
And need to set the DRS rule to move the exchange VMs only between the mentioned hosts (Host DRS group)
Hi there,
Thanks for your response.
My problem is I can't figure out the rules and groups. Here is the desired solution, I just can't figure out how to make it work:
IF ExchVM1 is on host in Chassis 1, ExchVM2 MUST NOT be on host in Chassis 1
IF ExchVM1 is on host in Chassis 2, ExchVM2 MUST NOT be on host in Chassis 2
IF ExchVM1 is on host in Chassis 3, ExchVM2 MUST NOT be on host in Chassis 3
The issue is the comparison. I can't figure out how to set anti-affinity based on the condition of another group; basic anti-affinity won't compare to a group, only to another VM; but a VM group can't be compared to another VM group. I can say "ExchVM 1 MUST BE on host in Chassis 1" (or I can say ExchVM2 MUST NOT be on host in Chassis 1), but that's as much as I can define; I don't want to lock ExchVM1 to chassis 1 and ExchVM2 to chassis 2, because that would mean that there are 2 hosts that are unavailable for use by the ExchVMs, which doesn't make sense.
Does that help clear up what I'm after?
Ryan
Unfortunately you've bumped up against the boundaries of just how granular DRS groups can be.
Not that I'm a fan of writing this many rules, but another slightly more flexible option would be:
This gives each VM 3 potential host options, and if the Chassis 3 blade is chosen for both, DRS will move one off.
In the event of a full chassis outage, HA is unaware of the "should" rule and will start the failed ExchVM on a surviving host.
Chris,
Thanks for that. I had exactly the same thought; it was actually TOO granular.
At any rate we did come up with this solution, but thank you for pointing out the ramifications of HA restarts, which we hadn't considered.
Here's hoping VMware allows this type of functionality in the future.
Ryan