add e2e test

.github/workflows/tests.yml

@@ -27,7 +27,12 @@ jobs:
     #needs: unit-and-integration # No use in running e2e tests if integration tests fail.
     strategy:
       matrix:
-        testcase: ["quickstart", "openai-python-client", "autoscaler-restart", "cache-shared-filesystem"]
+        testcase:
+        - "quickstart"
+        - "openai-python-client"
+        - "autoscaler-restart"
+        - "cache-shared-filesystem"
+        - "engine-vllm-pvc"
     steps:
       - name: Checkout code
         uses: actions/checkout@v4

Makefile

@@ -91,6 +91,10 @@ test-e2e-autoscaler-restart: skaffold
 test-e2e-cache-shared-filesystem: skaffold
 	./test/e2e/run.sh cache-shared-filesystem --profile e2e-test-default
 
+.PHONY: test-e2e-engine-vllm-pvc
+test-e2e-engine-vllm-pvc: skaffold
+	./test/e2e/run.sh engine-vllm-pvc --profile e2e-test-default
+
 .PHONY: test-e2e-engine
 test-e2e-engine: skaffold
 	CACHE_PROFILE=$(CACHE_PROFILE) ./test/e2e/run.sh engine-$(ENGINE) --profile e2e-test-default

test/e2e/engine-vllm-pvc/pvc.yaml

@@ -0,0 +1,11 @@
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: model-pvc
+spec:
+  accessModes:
+    - ReadWriteOnce
+  resources:
+    requests:
+      storage: 10Gi
+  volumeName: kind-hostpath

test/e2e/engine-vllm-pvc/test.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+
+source $REPO_DIR/test/e2e/common.sh
+
+models_release="kubeai-models"
+
+pip install -U "huggingface_hub[cli]"
+
+PV_HOST_PATH=/tmp/data
+
+huggingface-cli download facebook/opt-125m --local-dir ${PV_HOST_PATH} \
+  --exclude "tf_model.h5" --exclude "flax_model.msgpack"
+
+kubectl apply -f - <<EOF
+EOF
+
+helm install $models_release $REPO_DIR/charts/models -f - <<EOF
+catalog:
+  opt-125m-cpu:
+    enabled: true
+    url: pvc://model-pvc
+    minReplicas: 1
+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}'