benchmark_job.yaml

# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

apiVersion: batch/v1
kind: Job
metadata:
  name: dynamo-benchmark
spec:
  template:
    spec:
      serviceAccountName: dynamo-sa
      imagePullSecrets:
      - name: docker-imagepullsecret
      securityContext:
        runAsNonRoot: true
        runAsUser: 1000
        fsGroup: 1000
      containers:
      - name: benchmark-runner
        image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
        securityContext:
          allowPrivilegeEscalation: false
          capabilities:
            drop:
            - ALL
          readOnlyRootFilesystem: true
          runAsNonRoot: true
        resources:
          requests:
            cpu: "4"
            memory: "8Gi"
          limits:
            cpu: "8"
            memory: "16Gi"
        env:
          - name: HUGGING_FACE_HUB_TOKEN
            valueFrom:
              secretKeyRef:
                name: hf-token-secret
                key: HF_TOKEN
        command: ["python3", "-m", "benchmarks.utils.benchmark"]
        args:
          - --model
          - "Qwen/Qwen3-0.6B"
          - --isl
          - "2000"
          - --std
          - "10"
          - --osl
          - "256"
          - --output-dir
          - /data/results
          - --input
          - "qwen-vllm-agg=vllm-agg-frontend:8000"
          # add more copies of lines 58-59 for each additional service if you want to benchmark multiple services
          # - --input
          # - "name=service-url:port"
        volumeMounts:
          - name: data-volume
            mountPath: /data
      restartPolicy: Never
      volumes:
        - name: data-volume
          persistentVolumeClaim:
            claimName: dynamo-pvc
  backoffLimit: 0
  ttlSecondsAfterFinished: 3600  # Clean up job after 1 hour