Add capability to run RawDeployment E2Es in OpenShift-ci

config/overlays/test/configmap/inferenceservice-openshift-ci-raw.yaml

@@ -0,0 +1,23 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: inferenceservice-config
+  namespace: kserve
+data:
+  deploy: |-
+    {
+      "defaultDeploymentMode": "RawDeployment"
+    }
+  ingress: |-
+    {
+        "ingressGateway" : "knative-serving/knative-ingress-gateway",
+        "ingressService" : "istio-ingressgateway.istio-system.svc.cluster.local",
+        "localGateway" : "knative-serving/knative-local-gateway",
+        "localGatewayService" : "knative-local-gateway.istio-system.svc.cluster.local",
+        "ingressDomain"  : "$OPENSHIFT_INGRESS_DOMAIN",
+        "ingressClassName" : "openshift-default",
+        "domainTemplate": "{{ .Name }}-{{ .Namespace }}.{{ .IngressDomain }}",
+        "urlScheme": "http",
+        "disableIstioVirtualHost": false,
+        "disableIngressCreation": false
+    }

python/kserve/kserve/models/v1beta1_inference_service.py

@@ -150,8 +150,6 @@ def metadata(self, metadata):
         :param metadata: The metadata of this V1beta1InferenceService.  # noqa: E501
         :type: V1ObjectMeta
         """
-        if metadata is not None:
-            metadata.annotations = {"serving.knative.openshift.io/enablePassthrough": "true"}
 
         self._metadata = metadata
 

test/e2e/predictor/test_raw_deployment.py

@@ -113,6 +113,7 @@ def test_raw_deployment_runtime_kserve():
 
 @pytest.mark.grpc
 @pytest.mark.raw
+@pytest.mark.skip("The custom-model-grpc image fails in OpenShift with a permission denied error")
 def test_raw_isvc_with_multiple_container_port():
     service_name = "raw-multiport-custom-model"
     model_name = "custom-model"

test/e2e/transformer/test_collocation.py

@@ -101,6 +101,7 @@ def test_transformer_collocation():
 
 
 @pytest.mark.raw
+@pytest.mark.skip("The torchserve container fails in OpenShift with permission denied errors")
 def test_raw_transformer_collocation():
     service_name = "raw-custom-model-collocation"
     model_name = "mnist"

test/e2e/transformer/test_raw_transformer.py

@@ -32,6 +32,7 @@
 
 
 @pytest.mark.raw
+@pytest.mark.skip("The torchserve container fails in OpenShift with permission denied errors")
 def test_transformer():
     service_name = 'raw-transformer'
     predictor = V1beta1PredictorSpec(

test/scripts/openshift-ci/run-e2e-tests.sh

@@ -63,10 +63,12 @@ pushd $PROJECT_ROOT/python/kserve >/dev/null
 popd
 
 # Install KServe stack
-echo "Installing OSSM"
-$MY_PATH/deploy.ossm.sh
-echo "Installing Serverless"
-$MY_PATH/deploy.serverless.sh
+if [ "$1" != "raw" ]; then
+  echo "Installing OSSM"
+  $MY_PATH/deploy.ossm.sh
+  echo "Installing Serverless"
+  $MY_PATH/deploy.serverless.sh
+fi
 
 echo "Installing KServe with Minio"
 kustomize build $PROJECT_ROOT/config/overlays/test | \
@@ -75,10 +77,21 @@ kustomize build $PROJECT_ROOT/config/overlays/test | \
   sed "s|kserve/router:latest|${KSERVE_ROUTER_IMAGE}|" | \
   sed "s|kserve/kserve-controller:latest|${KSERVE_CONTROLLER_IMAGE}|" | \
   oc apply -f -
+
+# Patch the inferenceservice-config ConfigMap, when running RawDeployment tests
+ if [ "$1" == "raw" ]; then
+  export OPENSHIFT_INGRESS_DOMAIN=$(oc get ingresses.config cluster -o jsonpath='{.spec.domain}')
+  oc patch configmap inferenceservice-config -n kserve --patch-file <(cat config/overlays/test/configmap/inferenceservice-openshift-ci-raw.yaml | envsubst)
+  oc delete pod -n kserve -l control-plane=kserve-controller-manager
+ fi
+
+# Wait until KServe starts
 oc wait --for=condition=ready pod -l control-plane=kserve-controller-manager -n kserve --timeout=300s
 
-echo "Installing odh-model-controller"
-oc apply -k $PROJECT_ROOT/test/scripts/openshift-ci
+if [ "$1" != "raw" ]; then
+  echo "Installing odh-model-controller"
+  oc apply -k $PROJECT_ROOT/test/scripts/openshift-ci
+fi
 
 echo "Add testing models to minio storage ..." # Reference: config/overlays/test/minio/minio-init-job.yaml
 curl -L https://storage.googleapis.com/kfserving-examples/models/sklearn/1.0/model/model.joblib -o /tmp/sklearn-model.joblib
@@ -96,7 +109,10 @@ apiVersion: v1
 kind: Namespace
 metadata:
   name: kserve-ci-e2e-test
----
+EOF
+
+if [ "$1" != "raw" ]; then
+  cat <<EOF | oc apply -f -
 apiVersion: maistra.io/v1
 kind: ServiceMeshMember
 metadata:
@@ -107,6 +123,7 @@ spec:
     namespace: istio-system
     name: basic
 EOF
+fi
 
 oc apply -f $PROJECT_ROOT/config/overlays/test/minio/minio-user-secret.yaml -n kserve-ci-e2e-test
 
@@ -115,9 +132,17 @@ kustomize build $PROJECT_ROOT/config/overlays/test/clusterresources | \
   sed "s|kserve/sklearnserver:latest|${SKLEARN_IMAGE}|" | \
   oc apply -n kserve-ci-e2e-test -f -
 
-#
+# Add the enablePassthrough annotation to the ServingRuntimes, to let Knative to
+# generate passthrough routes. If RawDeployment test are being run, this annotation would have
+# no effect, because of missing Knative
+oc annotate servingruntimes -n kserve-ci-e2e-test --all serving.knative.openshift.io/enablePassthrough=true
+
 echo "Run E2E tests: $1"
 pushd $PROJECT_ROOT >/dev/null
+  # Note: The following images are set by openshift-ci. Uncomment if you are running on your own machine.
+  # export CUSTOM_MODEL_GRPC_IMG_TAG=kserve/custom-model-grpc:latest
+  # export IMAGE_TRANSFORMER_IMG_TAG=kserve/image-transformer:latest
+
   export GITHUB_SHA=$(git rev-parse HEAD)
   export CI_USE_ISVC_HOST="1"
   ./test/scripts/gh-actions/run-e2e-tests.sh "$1"