CPU pre-flights on auto-scaling clusters

Packaging an application
1

I’ve got a pre-flight defined in my kots-preflight.yaml that checks the number of cores available in a cluster.

    - nodeResources:
        checkName: Total CPU Cores
        outcomes:
          - fail:
              when: "sum(cpuCapacity) < 10"
              message: The cluster does not contain enough cores to run the app.
          - warn:
              when: "sum(cpuCapacity) < 14"
              message: The cluster can run the app, but may run into CPU allocation issues.
          - pass:
              message: The cluster contains the recommended number of cores.

The check appears to be based on the number of cores currently provisioned, and is failing frequently on auto-scaling clusters that have the ability to scale into the “pass” range. I would like for the test to pass in these scenarios. Is there any way to base this check on the cluster’s scaled-up limit instead of the currently-available core count?

1 Like
2

Great question @anyweez!
I think this isn’t currently possible with the nodeResources analyzer. I think it might also depend on what you use for the Cluster Auto Scaler. For example if you use the k8s Cluster Autoscaler, there’s an argument max-nodes-total in the faq. So maybe one way of doing this would be to use the run-pod to get to the max-nodes-total value?
I’m not sure if that value is maybe exposed via a svc or configmap.

3

Has there been any more thought or updates here? We are now thinking about the intersection of various autoscale methods and preflight checks.

4

Hey, Mitch. AFAIK this hasn’t changed yet. If it’s something that significant for you, feel free to open a feature request so we can track it internally and see if it’s something the team has considered.

Autoscaling has come up a few times recently with other customers. So I don’t know that this is something that will get implemented soon, but it’s good to see greater interest in it as well.