bug: StorageClass Not Applied to PVC in Helm Deployment

[brokedba]

Describe the bug

StorageClass Not Applied to PVC in Helm Deployment

Description

Hi team,

I am preparing a Terraform stack to deploy production-stack on EKS clusters. I noticed a misbehavior from the chart. The existing storage class gp2 should be chosen, but each time the PVC has a null value for the storage class after deployment. Please provide any fix for this bug.
Workaround: I am currently doing this manually, which defeats the purpose of automation.

Note: There is no default StorageClass in my cluster.

Helm Values

# This file is a Go template. Variables passed from Terraform are accessed with .VarName
servingEngineSpec:
  enableEngine: true
  runtimeClassName: ""  # Use default runtime for CPU
  nodeSelector:
    workload-type: cpu
    node-group: cpu-pool
  containerSecurityContext:
    privileged: true
  modelSpec:
  - name: "tinyllama-cpu"
    repository: "public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo"
    tag: "v0.8.5.post1"
    modelURL: "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
    replicaCount: 1
    requestCPU: 1
    requestMemory: "2Gi"
    requestGPU: 0
    limitCPU: "2"
    limitMemory: "4Gi"
    pvcStorage: "10Gi"
    storageClassName: "gp2"  # <------------- Storage Class
    vllmConfig:
      dtype: "bfloat16"
      extraArgs:
        - "--device"
        - "cpu"
    env:
      - name: VLLM_CPU_KVCACHE_SPACE
        value: "1"
      - name: VLLM_CPU_OMP_THREADS_BIND
        value: "0-2"
      - name: HUGGING_FACE_HUB_TOKEN
        valueFrom:
          secretKeyRef:
            name: hf-token-secret
            key: token

routerSpec:
  enableRouter: true
  routingLogic: "roundrobin"
  resources:
    requests:
      cpu: "500m"
      memory: "1Gi"
    limits:
      cpu: "1"
      memory: "2Gi"

Error Message

pod has unbound immediate PersistentVolumeClaims

To Reproduce

Use EKS with ebs-csi-driver and a storage of type gp2

Run the below helm values (through basic helm or using terraform)

see above shared values

Expected behavior

• The PVC should be created with the specified storageClassName: "gp2".

Additional context

• There is no default StorageClass in the cluster.
• The issue occurs consistently with the provided Helm values.
• The whole deployment is stuck because of the storageclassName value not being persistent .
• This forces users to patch the pvc that is supposed to be managed by the prod-stack helm

[zerofishnoodles]

Hi, thanks for bringing this up, We will look into it and fix ASAP.

[zerofishnoodles]

Hi, I think because of some legacy issue, this field is named as storageClass instead of storageClassName. Can you try using "storageClass" ? Lemme know if you think we really need to change it to align with k8s native field name.

[brokedba]

Thanks @zerofishnoodles will give it a shot. You'll probably right.
It matches what I curated in terms of table of values

I still think that adding a table in either helm readme of ArtifactHub page would help reduce confusion from users in the future.
see #444

[zerofishnoodles]

This table seems great! Can we add this to the artifacthub page?

[brokedba]

@zerofishnoodles yes please do !
I have suggested the same on issue #444 but kobe might not have seen it .

Note: You can close this issue . problem was a typo indeed.