use-cases/inferencing/cost-optimization/gcsfuse/manifests/model-deployment-tuned-a100-dws.yaml

# Copyright 2025 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: vllm-openai-gcs-tuned-llama33-70b-a100
spec:
  replicas: 1
  selector:
    matchLabels:
      app: vllm-openai-gcs-tuned-llama33-70b-a100
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        app: vllm-openai-gcs-tuned-llama33-70b-a100
      annotations:
        gke-gcsfuse/volumes: "true"
        gke-gcsfuse/cpu-limit: "0"
        gke-gcsfuse/memory-limit: "0"
        gke-gcsfuse/ephemeral-storage-limit: "0"
        cluster-autoscaler.kubernetes.io/consume-provisioning-request: a100-storage-benchmark-tuned
        cluster-autoscaler.kubernetes.io/provisioning-class-name: "queued-provisioning.gke.io" 
    spec:
      containers:
      - name: fetch-safetensors
        image: busybox
        command: ["/bin/sh", "-c"]
        args:
          - |
            echo "########### $(date) - Starting parallel-fetch-safetensors"
            find /gcs/${MODEL_NAME}/${MODEL_VERSION}/*safetensors -type f | xargs -I {} -P 15 sh -c 'echo "########### $(date) - Fetching: {}"; dd if={} of=/dev/null'
            echo "########### $(date) - Finished parallel-fetch-safetensors"
            sleep infinity
        volumeMounts:
        - name: gcsfuse
          mountPath: /gcs/${MODEL_NAME}/${MODEL_VERSION}/
          readOnly: true      
      - name: inference-server
        args:
        - --model=$(MODEL)
        - --tensor-parallel-size=8
        - --trust-remote-code
        env:
        - name: MODEL
          value: /gcs/${MODEL_NAME}/${MODEL_VERSION}
        image: ${VLLM_IMAGE_NAME}
        imagePullPolicy: Always
        readinessProbe:
          failureThreshold: 3
          httpGet:
            path: /health
            port: 8000
            scheme: HTTP
          initialDelaySeconds: 60
          periodSeconds: 1
          successThreshold: 1
          timeoutSeconds: 1
        resources:
          requests:
            cpu: "64"
            memory: "192G"
            nvidia.com/gpu: "8"
          limits:
            cpu: "64"
            memory: "192G"
            nvidia.com/gpu: "8"
        volumeMounts:
        - mountPath: /dev/shm
          name: dshm
        - name: gcsfuse
          mountPath: /gcs/${MODEL_NAME}/${MODEL_VERSION}/
          readOnly: true
      nodeSelector:
        cloud.google.com/gke-accelerator: nvidia-tesla-a100
      serviceAccountName: ${MLP_STORAGE_BENCHMARKING_KSA}
      tolerations:
      - key: "nvidia.com/gpu"
        operator: "Exists"
        effect: "NoSchedule"
      - key: "on-demand"
        value: "true"
        operator: "Equal"
        effect: "NoSchedule" 
      volumes:
      - name: dshm
        emptyDir:
            medium: Memory
      - name: gcsfuse
        csi:
          driver: gcsfuse.csi.storage.gke.io
          volumeAttributes:
            bucketName: ${MLP_STORAGE_BENCHMARK_HIERARCHICAL_BUCKET}
            mountOptions: "metadata-cache:ttl-secs:-1,metadata-cache:stat-cache-max-size-mb:-1,metadata-cache:type-cache-max-size-mb:-1,file-cache:max-size-mb:-1,file-cache:cache-file-for-range-read:true,file-cache:enable-parallel-downloads:true,file-system:kernel-list-cache-ttl-secs:-1,only-dir:${MODEL_NAME}/${MODEL_VERSION}/"
            skipCSIBucketAccessCheck: "true"
      - name: gke-gcsfuse-cache
        emptyDir:
          medium: Memory
      - name: gke-gcsfuse-tmp
        emptyDir:
          medium: Memory
      - name: gke-gcsfuse-buffer
        emptyDir:
          medium: Memory      
---
apiVersion: v1
kind: Service
metadata:
  name: vllm-openai-gcs-tuned-llama33-70b-a100
spec:
  selector:
    app: vllm-openai-gcs-tuned-llama33-70b-a100
  type: ClusterIP
  ports:
  - protocol: TCP
    port: 8000
    targetPort: 8000