Hi, I'd appreciate any help on this one please. Lets say I have an ESXi 6.7u3 cluster with 3 nodes, and 1 dedicated failover host as node 3 for example.
I have a giant VM running on nodes 1 & 2 which take up all the resources and are fully reserved.
On an normal day if node 1 or 2 fail my giant VM will start on node 3.
However, I'm trying to find a way to utilize the resources on node 3 until there is a failure. Normally I would spread the workloads across the nodes and use a % of cluster resources admission control policy. However as these are giant VMs with reservations I can't do that- they take up a full node each.
Is there any way possible to run a (non-priority) VM on node 3 until the event of a failure and it is then powered off, or gets degraded possibly? I don't think I can use a degradation for the giant VMs with reservations?
Hope that makes sense, thanks in advance for the help
Steve
Ah sorry, I misread it. So you have 2 VMs with fully reserved resources and you take up ALL the resources of those two hosts. unfortunately, there's no way around this. I have proposed a solution fo this to the engineering team a while back, but so far it hasn't made it into the roadmap yet. I guess you need to ask yourself if a full reservation is absolutely required?
That is not possible. Why not use a different admission control policy? You could simply use the percentage based policy and then use all the hosts actively? I wrote about it in this book:
https://www.rubrik.com/en/resources/white-papers/19/clustering-deep-dive-ebook
Thanks Duncan, but if each VM is the size of a full node and has all memory reserved then that is 100% so that policy wouldn't help?
Regards
Steve
Ah sorry, I misread it. So you have 2 VMs with fully reserved resources and you take up ALL the resources of those two hosts. unfortunately, there's no way around this. I have proposed a solution fo this to the engineering team a while back, but so far it hasn't made it into the roadmap yet. I guess you need to ask yourself if a full reservation is absolutely required?
That's what I was thinking alright. These are SAP HANA workloads so we can't change the reservations unfortunately (SAP best practices). The "hope" was to utilize this very expensive HA node temporarily until the event of a failure. I suppose we could manually control it but that defeats the purpose. Would be a nice feature to see in the future.
Thanks for your help
Steve
Agreed, let me raise it again with the team and add the "SAP HANA" use case.
just realized I wrote an article on this admission control policy a while back:
http://www.yellow-bricks.com/2019/02/05/ha-admission-control-policy-dedicated-failover-hosts/
Thanks, I raised a feature request for it yesterday anyway and I've made our local solution engineer aware also. He's going to raise it with the SAP alliance team.
Hi, I've been wondering if I only keep the reservation on the critical VM and remove it from the others, would it be possible to use resource shares to support failover for the critical machine. i.e.:
Host 1 has one VM-A 6TB fully reserved all host memory. Shares set to High
Host 2 has one VM-B 6TB no reservation but all host memory allocated. Shares set to Low
Host 3 has one VM-C 6TB no reservation but all host memory allocated. Shares set to Low
If Host 1 fails, will the shares allow VM-A to start on another host? Or would I be right in thinking it depends on the memory demand on the VM at the time?
Thanks for any help on this.