HugePages no resource allocation

[david-lumex-agency]

We are having an issue that our pod doesn't have acccess to HugePages feature in kubernetes:

Warning FailedMount 4s (x7 over 35s) kubelet MountVolume.SetUp failed for volume "hugepage-2mi" : medium HugePages-2Mi: hugePages storage requested, but there is no resource request for huge pages
Warning FailedMount 4s (x7 over 35s) kubelet MountVolume.SetUp failed for volume "hugepage-1gi" : medium HugePages-1Gi: hugePages storage requested, but there is no resource request for huge pages

https://kubernetes.io/docs/tasks/manage-hugepages/scheduling-hugepages/

We are following this guide from official kubernetes.

[sirishkr]

We saw your email come in earlier today - apologies we weren't able to get you an answer faster. Please stay tuned while we look into this.

[sahil-lakhwani]

It looks like the pod spec doesn't specify requests/limits for hugepages. Something like:

containers:
  - name: my-container
    image: your-image
    resources:
      limits:
        hugepages-2Mi: 64Mi
        hugepages-1Gi: 1Gi

("Huge page requests must equal the limits. This is the default if limits are specified, but requests are not.")

[david-lumex-agency]

When I run kubectl describe nodes on a generation 2 server, I get this, zero allocatable hugepages.

When I checked a generation 1 server, I also get zero allocatable hugepages

[david-lumex-agency]

Also get this insufficient huge pages error and pods refuse to schedule

[Q-efx]

Which Helm chart are you trying to deploy?

[david-lumex-agency]

It’s a proprietary C++ application, built on top of generic Ubuntu container. Not using helm, directly deployed as Kubernetes Deployment with kubectl.

The issue I see is with the nodes in the cluster, are not pre allocating any huge pages resources. This should be part of the node config in my opinion, like how the CNI and Kubernetes version can be chosen, this is a setting that is very important for high performance server tuning