can not install HAMi on AWS EKS

[cataglyphis]

Please provide an in-depth description of the question you have:

What do you think about this question?:

Environment:

• HAMi version: 2.4.1
• Kubernetes version: v1.29.11-eks-56e63d8
• Others:

helm install hami hami-charts/hami --set scheduler.kubeScheduler.imageTag=v1.29.11 -n hami
Error: INSTALLATION FAILED: chart requires kubeVersion: >= 1.16.0 which is incompatible with Kubernetes v1.29.11-eks-56e63d8

[elrondwong]

The version v1.29.11-eks-56e63d8 is not a standard semantic version, so it cannot be recognized by Helm. To resolve this issue, there are two approaches:

• Directly remove or comment out the kubeVersion field in charts/hami/Chart.yaml

# kubeVersion: ">= 1.16.0"

• Use --set kubeVersionOverride=true when installing the Helm Chart to bypass the Kubernetes version check.

helm install hami hami-charts/hami --set kubeVersionOverride=true

[cataglyphis]

The version v1.29.11-eks-56e63d8 is not a standard semantic version, so it cannot be recognized by Helm. To resolve this issue, there are two approaches:

• Directly remove or comment out the kubeVersion field in charts/hami/Chart.yaml

# kubeVersion: ">= 1.16.0"

• Use --set kubeVersionOverride=true when installing the Helm Chart to bypass the Kubernetes version check.

helm install hami hami-charts/hami --set kubeVersionOverride=true

Since there is no user-managed kube-scheduler component in AWS EKS clusters, should I install the HAMi kube-scheduler or only the vgpu-scheduler-extender?

[elrondwong]

I have not installed HAMI on AWS EKS before. Normally, both components need to be installed:

• kube-scheduler will register a scheduler extender with the cluster.
• vgpu-scheduler-extender handles the actual scheduling requests.

[cataglyphis]

The version v1.29.11-eks-56e63d8 is not a standard semantic version, so it cannot be recognized by Helm. To resolve this issue, there are two approaches:

• Directly remove or comment out the kubeVersion field in charts/hami/Chart.yaml

# kubeVersion: ">= 1.16.0"

• Use --set kubeVersionOverride=true when installing the Helm Chart to bypass the Kubernetes version check.

helm install hami hami-charts/hami --set kubeVersionOverride=true

❯ helm install hami ./hami -n hami --set scheduler.kubeScheduler.imageTag=v1.29.11 --set kubeVersionOverride=true
Error: INSTALLATION FAILED: chart requires kubeVersion: >= 1.16.0 which is incompatible with Kubernetes v1.29.11-eks-56e63d8

[cataglyphis]

Looks like HAMi not works well with AWS EKS cluster.

NAME                              READY   STATUS             RESTARTS      AGE
hami-admission-patch-4shvn        0/1     CrashLoopBackOff   4 (22s ago)   113s
hami-scheduler-69fbc99df7-6x87p   1/2     Error              4 (64s ago)   119s

  ----     ------     ----                    ----               -------
  Normal   Scheduled  9m1s                    default-scheduler  Successfully assigned hami/hami-scheduler-69fbc99df7-6x87p to ip-10-170-85-172.us-west-2.compute.internal
  Normal   Pulling    9m1s                    kubelet            Pulling image "registry.cn-hangzhou.aliyuncs.com/google_containers/kube-scheduler:v1.29.11"
  Normal   Pulled     8m50s                   kubelet            Successfully pulled image "registry.cn-hangzhou.aliyuncs.com/google_containers/kube-scheduler:v1.29.11" in 10.5s (10.5s including waiting)
  Normal   Pulled     8m50s                   kubelet            Container image "projecthami/hami:v2.4.1" already present on machine
  Normal   Created    8m50s                   kubelet            Created container vgpu-scheduler-extender
  Normal   Started    8m50s                   kubelet            Started container vgpu-scheduler-extender
  Normal   Created    8m6s (x4 over 8m50s)    kubelet            Created container kube-scheduler
  Normal   Started    8m6s (x4 over 8m50s)    kubelet            Started container kube-scheduler
  Normal   Pulled     7m12s (x4 over 8m49s)   kubelet            Container image "registry.cn-hangzhou.aliyuncs.com/google_containers/kube-scheduler:v1.29.11" already present on machine
  Warning  BackOff    3m59s (x23 over 8m48s)  kubelet            Back-off restarting failed container kube-scheduler in pod hami-scheduler-69fbc99df7-6x87p_hami(93fb46b5-13a4-4cd6-9df4-124c311a45fb)

Defaulted container "kube-scheduler" out of: kube-scheduler, vgpu-scheduler-extender
Usage:
  kube-scheduler [flags]

Misc flags:

      --config string            The path to the configuration file.
      --master string            The address of the Kubernetes API server (overrides any value in kubeconfig)
      --write-config-to string   If set, write the configuration values to this file and exit.