Shutdown VM on failover to free resources

[sne]

The following scenario:
Node 1 & 2 are a HA cluster.
Node 1 hosts VM 1
Node 2 hosts VM 2 & VM 3
-> Node 1 crashes. VM 1 is handed over to node 2 by HA job.

Is there a way to automatically shutdown VM 3 (low prio) to free resources on Node 2 for the downtime of Node 1 (duration of failover)?

 

[bbgeek17]

There is no built-in mechanism to what you want, simply because most people would not want to do such a thing.
It should be trivial to write a script that checks status of the nodes/vms and shutsdown/starts VM as needed. Put it cron and run every minute.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[VictorSTS]

Option 2: using a hook script on VM 1 which would stop VM 3 if the hostname were it is started is node2 and starts VM 3 again when VM 1 is stopped or migrated back to VM 1

Be careful about were you place the hook script, as it will have to be available from any node.

 

[bbgeek17]

Option 2: using a hook script on VM

Does hook script get executed on VM migration between nodes? The VM is not really stopped/started in that case.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[VictorSTS]

Good question!

Checked and it's a shame, but they dont trigger on migration, so my option isn't viable.
https://bugzilla.proxmox.com/show_bug.cgi?id=1996#c2

 

[t.lamprecht]

Does hook script get executed on VM migration between nodes? The VM is not really stopped/started in that case.

No, but even if there would be such a hook step it wouldn't trigger for HA recovery, as there we need to fence and "steal" the VMs/CTs to recover and freshly start them, which will trigger the pre-start hook where one could cover such semantics (see /usr/share/pve-docs/examples/guest-example-hookscript.pl on a PVE instance).

But, IMO you need more than two nodes for actual sensible HA anyway and if you're that over-commited in resources that you need to sacrifice a bunch of "less" important VMs it can get hairy fast and eats away time from any SLA guarantees one may need/want to uphold (stop isn't exactly zero cost/time).