Autoscaling: preemptible instances if available, regular if not

[Zach Bjornson]

Hello!

I'm designing an internal load balancer+autoscaling group architecture for batch processing. I would like the target pool to use preemptible instances if they're available, but non-preemptible instances otherwise so that we save money and always satisfy our load. From the docs it sounds like if you specify a preemptible instance, the autoscaler will try repeatedly to launch instances until they're available. Is it possible to configure it to fallback to non-preemptible? I haven't tested to see how common it is to not have preemptible instances available, so maybe this isn't a problem.

Was looking at the failover feature (pool A is preemptible, B is not), but the failover rules and health check don't seem conducive to this.

Thanks,
Zach

[George]

Hello Zach,

You can use Prevmtable in order to manage a pool of preemtible GCE VMs across zones. You can have more information about this matter in this Github post.

I hope this helps.

Sincerely,

George