# services/ai-llm/gpu-seed-job.yaml
apiVersion: batch/v1
kind: Job
metadata:
  name: ollama-gpu-seed-v2
  namespace: ai
spec:
  backoffLimit: 2
  activeDeadlineSeconds: 7200
  template:
    metadata:
      labels:
        app: ollama-gpu-seed
    spec:
      automountServiceAccountToken: false
      restartPolicy: Never
      nodeSelector:
        kubernetes.io/hostname: titan-24
      containers:
        - name: seed
          image: ollama/ollama@sha256:2c9595c555fd70a28363489ac03bd5bf9e7c5bdf2890373c3a830ffd7252ce6d
          env:
            - {name: OLLAMA_HOST, value: "127.0.0.1:11434"}
            - {name: OLLAMA_MODELS, value: /models}
            - {name: OLLAMA_NO_CLOUD, value: "1"}
          command: [/bin/bash, -ec]
          args:
            - |
              ollama serve >/tmp/ollama.log 2>&1 &
              server_pid=$!
              trap 'kill "$server_pid"; wait "$server_pid" || true' EXIT
              for attempt in $(seq 1 60); do
                ollama list >/dev/null 2>&1 && break
                sleep 1
              done
              ollama pull qwen2.5:14b-instruct-q4_0 >/tmp/pull.log 2>&1 || { tail -c 1000 /tmp/pull.log; exit 1; }
              cd /models/manifests/registry.ollama.ai/library
              sha256sum -c <<'PINS'
              5449194ff8035ccb13a6409a5814de6c8f9c39f555f429e383ae0fb7137001bd  qwen2.5/14b-instruct-q4_0
              PINS
              # Runtime mounts this cache read-only, after registry digest checks.
              chmod -R a+rX /models
              touch /models/.pilot-v1-ready
          resources:
            requests: {cpu: 500m, memory: 512Mi}
            limits: {cpu: "2", memory: 2Gi}
          volumeMounts:
            - {name: models, mountPath: /models}
      volumes:
        - name: models
          persistentVolumeClaim:
            claimName: ollama-gpu-titan24