Add nodeSelector, tolerations, and affinity support to Helm chart

[mirthroxmanu]

Summary

The Kusion Helm chart currently does not support Kubernetes pod scheduling configurations like nodeSelector, tolerations, and affinity. This limits deployment flexibility in production environments where workload placement needs to be controlled.

Use Case

In production Kubernetes clusters, it's common to:

• Segregate workloads to specific node pools (e.g., using Karpenter, spot instances, or GPU nodes)
• Use taints and tolerations to reserve nodes for specific applications
• Implement anti-affinity rules for high availability
• Separate control plane and data plane workloads

For example, when using Karpenter with labeled node pools:

apiVersion: karpenter.sh/v1
kind: NodePool
metadata:
  name: kusion-nodepool
spec:
  template:
    metadata:
      labels:
        app: kusion  # We want to schedule Kusion pods here

Currently, there's no way to schedule Kusion server and MySQL pods to these specific nodes through Helm values.

Proposed Solution

Add support for standard Kubernetes scheduling configurations in values.yaml:

server:
  # ... existing configuration
  nodeSelector: {}
  tolerations: []
  affinity: {}

mysql:
  # ... existing configuration
  nodeSelector: {}
  tolerations: []
  affinity: {}

Example Usage

server:
  nodeSelector:
    app: kusion
    workload-type: platform-engineering
  tolerations:
    - key: "dedicated"
      operator: "Equal"
      value: "kusion"
      effect: "NoSchedule"

mysql:
  nodeSelector:
    app: kusion
    storage-optimized: "true"
  affinity:
    podAntiAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
        - weight: 100
          podAffinityTerm:
            topologyKey: kubernetes.io/hostname

Implementation

The templates need minimal changes. For example, in templates/kusion-server.yml:

spec:
  template:
    spec:
      {{- with .Values.server.nodeSelector }}
      nodeSelector:
        {{- toYaml . | nindent 8 }}
      {{- end }}
      {{- with .Values.server.tolerations }}
      tolerations:
        {{- toYaml . | nindent 8 }}
      {{- end }}
      {{- with .Values.server.affinity }}
      affinity:
        {{- toYaml . | nindent 8 }}
      {{- end }}
      containers:
        - name: {{ .Values.server.name }}
          # ... rest of container spec

Similar changes would be needed in templates/mysql.yml for the StatefulSet.

Workarounds (Current)

Currently, users must either:

1. Manually patch deployments post-installation:

   kubectl patch deployment kusion-server -n kusion \
     -p '{"spec":{"template":{"spec":{"nodeSelector":{"app":"kusion"}}}}}'

2. Fork and modify the chart
3. Use helm template + manual edits before applying

All of these defeat the purpose of using Helm for lifecycle management.

References

This is a common pattern in popular Helm charts:

• Prometheus Operator
• Grafana
• ArgoCD

Benefits

• ✅ Better production deployment practices
• ✅ Cost optimization (e.g., spot instances with tolerations)
• ✅ Compliance with security policies (node isolation)
• ✅ Improved HA configurations
• ✅ Follows Kubernetes and Helm best practices

Additional Context

• Kusion version: [your version]
• Kubernetes version: [your version]
• Cluster setup: [EKS/GKE/AKS/self-managed]

Would the team be open to a PR adding this functionality? I'm happy to contribute if there's interest.