diff --git a/test/e2e/common-manifests.yaml b/test/e2e/common-manifests.yaml index 1f2c8285..a24a03bb 100644 --- a/test/e2e/common-manifests.yaml +++ b/test/e2e/common-manifests.yaml @@ -3,13 +3,10 @@ kind: PersistentVolume metadata: name: kind-hostpath spec: - storageClassName: manual capacity: storage: 10Gi accessModes: - ReadWriteMany - - ReadOnlyMany - - ReadWriteOnce hostPath: path: /tmp/data type: DirectoryOrCreate diff --git a/test/e2e/engine-vllm-pvc/pvc.yaml b/test/e2e/engine-vllm-pvc/pvc.yaml index 64125a11..93e443c3 100644 --- a/test/e2e/engine-vllm-pvc/pvc.yaml +++ b/test/e2e/engine-vllm-pvc/pvc.yaml @@ -9,4 +9,4 @@ spec: resources: requests: storage: 10Gi - volumeName: kind-hostpath \ No newline at end of file + volumeName: kind-model-hostpath \ No newline at end of file diff --git a/test/e2e/engine-vllm-pvc/test.sh b/test/e2e/engine-vllm-pvc/test.sh index 30b73b5b..bcb2201c 100755 --- a/test/e2e/engine-vllm-pvc/test.sh +++ b/test/e2e/engine-vllm-pvc/test.sh @@ -6,7 +6,7 @@ models_release="kubeai-models" pip install -U "huggingface_hub[cli]" -PV_HOST_PATH=/tmp/data +PV_HOST_PATH=/tmp/model mkdir -p ${PV_HOST_PATH} @@ -20,12 +20,15 @@ catalog: opt-125m-cpu: enabled: true url: pvc://model-pvc - minReplicas: 1 + minReplicas: 2 EOF sleep 5 -curl http://localhost:8000/openai/v1/completions \ - --max-time 900 \ - -H "Content-Type: application/json" \ - -d '{"model": "opt-125m-cpu", "prompt": "Who was the first president of the United States?", "max_tokens": 40}' +for i in {1..10}; do + echo "Sending request $i" + curl http://localhost:8000/openai/v1/completions \ + --max-time 600 \ + -H "Content-Type: application/json" \ + -d '{"model": "opt-125m-cpu", "prompt": "Who was the first president of the United States?", "max_tokens": 40}' +done