apiVersion: apps/v1
kind: Deployment
metadata:
  name: server
spec:
  replicas: 1
  strategy:
    type: Recreate
  selector:
    matchLabels:
      app.kubernetes.io/component: server
  template:
    metadata:
      labels:
        app.kubernetes.io/component: server
    spec:
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
            nodeSelectorTerms:
              - matchExpressions:
                  # Image only supports amd64
                  - key: kubernetes.io/arch
                    operator: In
                    values:
                      - amd64
      priorityClassName: high-priority
      containers:
        - name: server
          image: ghcr.io/mostlygeek/llama-swap:vulkan
          imagePullPolicy: Always
          args: ["--config", "/config/config.yaml"]
          resources:
            requests:
              memory: 4Gi
              cpu: 1000m
            limits:
              memory: 4Gi
              # nvidia.com/gpu: "1"
              amd.com/gpu: "1"
          ports:
            - name: http
              hostPort: 8080
              containerPort: 8080
          volumeMounts:
            - name: server-data
              mountPath: /root/.cache/llama.cpp
            - name: llama-swap-config
              mountPath: /config
      volumes:
        - name: server-data
          hostPath:
            path: /var/lib/llama.cpp
            type: DirectoryOrCreate
        - name: llama-swap-config
          configMap:
            name: llama-swap-config
---
apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: high-priority
value: 1000000
globalDefault: false
---
apiVersion: v1
kind: Service
metadata:
  name: server
  labels:
    app.kubernetes.io/component: server
spec:
  selector:
    app.kubernetes.io/component: server
  ports:
    - name: http
      port: 8080
      targetPort: http