Support zero node clusters
The number of nodes in AKS can not be zero.
A same request has been opened on GitHub.
Managed Kubenetes services on both of AWS and GCP have the feature. Only Azure does not allow zero node clusters.
We can manually deallocate nodes to save money from the Azure portal but in case that we want the capacity increased/decreased automatically as a job scheduler, we expect that the number of nodes is set to zero when there is no use and the size is expanded systematically on demand. Manual operation is not acceptable.
100% needed. We cannot use Azure for most of our clients because of this. It would be completely fine to me if you needed to have at least one node group with 1 node in it, but at least make it so that any additional groups can scale to 0. Our clients wont want to pay for one expensive GPU node to be up all the time.
That's a show stopper.
Charles-emmanuel Frances commented
Very important features when competing with GCP or AWS.
Julien Senon commented
like Maxence, this is absolutely needed for GPU nodes or for batch with high CPU nodes
Maxence Boutet commented
Absolutely needed, especially for GPU nodes
Ondřej Hlaváček commented
Yes please! We run ad hoc tasks on expensive instances and need to be able to turn them off completely when not in use.
Florian Beier commented
This feature would be really useful!
Yes! we need this feature!!