# services/ai-llm/gpu-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: ollama-gpu
  namespace: ai
spec:
  replicas: 1
  revisionHistoryLimit: 2
  strategy:
    type: Recreate
  selector:
    matchLabels: {app: ollama-gpu}
  template:
    metadata:
      labels: {app: ollama-gpu}
      annotations:
        fluentbit.io/exclude: "true"
    spec:
      automountServiceAccountToken: false
      nodeSelector:
        kubernetes.io/hostname: titan-24
      runtimeClassName: nvidia
      securityContext:
        seccompProfile: {type: RuntimeDefault}
      containers:
        - name: ollama
          image: ollama/ollama@sha256:2c9595c555fd70a28363489ac03bd5bf9e7c5bdf2890373c3a830ffd7252ce6d
          command: [/bin/bash, -ec]
          args:
            - |
              while ! test -f /models/.pilot-v1-ready || ! test -f /reservation/reserved; do sleep 5; done
              # Backend parser diagnostics can contain submitted schema text.
              # Keep request status/timing at the gateway; retain no backend output.
              exec ollama serve >/dev/null 2>&1
          env:
            - {name: OLLAMA_HOST, value: "0.0.0.0:11434"}
            - {name: OLLAMA_MODELS, value: /models}
            - {name: OLLAMA_NO_CLOUD, value: "1"}
            - {name: OLLAMA_CONTEXT_LENGTH, value: "8192"}
            - {name: OLLAMA_MAX_LOADED_MODELS, value: "1"}
            - {name: OLLAMA_NUM_PARALLEL, value: "1"}
            - {name: OLLAMA_MAX_QUEUE, value: "1"}
            - {name: OLLAMA_KEEP_ALIVE, value: "20m"}
            - {name: OLLAMA_LOAD_TIMEOUT, value: "20m"}
            - {name: OLLAMA_DEBUG, value: "false"}
            - {name: OLLAMA_FLASH_ATTENTION, value: "1"}
            - {name: OLLAMA_KV_CACHE_TYPE, value: q8_0}
          ports:
            - {name: http, containerPort: 11434}
          readinessProbe:
            httpGet: {path: /api/version, port: http}
            periodSeconds: 10
            timeoutSeconds: 3
          securityContext:
            allowPrivilegeEscalation: false
            readOnlyRootFilesystem: true
            capabilities: {drop: [ALL]}
          resources:
            requests: {cpu: "4", memory: 16Gi, nvidia.com/gpu.shared: 4}
            limits: {cpu: "12", memory: 24Gi, nvidia.com/gpu.shared: 4}
          volumeMounts:
            - {name: models, mountPath: /models, readOnly: true}
            - {name: reservation, mountPath: /reservation, readOnly: true}
            - {name: tmp, mountPath: /tmp}
            - {name: identity, mountPath: /root/.ollama}
      volumes:
        - name: reservation
          hostPath: {path: /var/lib/atlas-maintenance/lan-inference-20260928, type: DirectoryOrCreate}
        - name: models
          persistentVolumeClaim:
            claimName: ollama-gpu-titan24
        - name: tmp
          emptyDir: {sizeLimit: 1Gi}
        - name: identity
          emptyDir: {medium: Memory, sizeLimit: 1Mi}