---
# yaml-language-server: $schema=./values.schema.json

# Default pod-level options applied to every controller unless overridden.
# Set `runtimeClassName: nvidia` here if your cluster requires an explicit
# RuntimeClass to schedule on GPU nodes (e.g. containerd + nvidia toolkit
# without auto-injection).
defaultPodOptions: {}

# Main controller running the Draw Things gRPC server.
controllers:
  main:
    # A GPU is a singleton resource — only one replica makes sense.
    replicas: 1
    # Recreate avoids ever scheduling two pods that fight over the same GPU.
    strategy: Recreate
    containers:
      main:
        image:
          # Docker image for Draw Things gRPC server CLI
          repository: drawthingsai/draw-things-grpc-server-cli
          # Use the chart's appVersion for the image tag
          tag: ""
          pullPolicy: IfNotPresent
        # The image has no default CMD; we mirror the upstream docker example:
        #   gRPCServerCLI /grpc-models
        command: ["gRPCServerCLI"]
        args:
          - /grpc-models
        # Request one NVIDIA GPU. For `nvidia.com/gpu`, limits must equal
        # requests — Kubernetes will reject anything else.
        resources:
          limits:
            nvidia.com/gpu: 1
          requests:
            nvidia.com/gpu: 1

# Service exposing the gRPC interface on the upstream default port 7859.
service:
  main:
    controller: main
    type: ClusterIP
    ports:
      grpc:
        port: 7859
        targetPort: 7859
        protocol: TCP

# Persistent storage for the model library mounted at /grpc-models.
# 50 GiB is enough for a handful of SD 1.5 / SDXL checkpoints + LoRAs;
# bump it before installing if you plan to host more.
persistence:
  models:
    enabled: true
    type: persistentVolumeClaim
    accessMode: ReadWriteOnce
    size: 50Gi
    globalMounts:
      - path: /grpc-models

# Ingress is disabled by default — gRPC clients typically connect in-cluster
# or via `kubectl port-forward`. If you front the server with an
# ingress controller, make sure it supports gRPC (h2c) end-to-end.
ingress:
  main:
    enabled: false
    className: ""
    annotations: {}
    hosts:
      - host: draw-things.local
        paths:
          - path: /
            pathType: Prefix
            service:
              identifier: main
              port: grpc
    tls:
      - hosts:
          - draw-things.local
        secretName: draw-things-tls