Use KServe component in tests instead of KFServing

tests/README.md

@@ -15,6 +15,8 @@ included in the tests.
 
 ### Test Suite
 
+**To use KFserving v0.5/0.6 instead of KServe v0.7, comment the 6,7th lines and uncomment the 10,11th lines here [e2e/utils/kserve.py](https://github.com/kubeflow/manifests/compare/master/tests/e2e/utils/kserve.py#L6-L11)**
+
 The e2e tests are completely independent of the underlying K8s cluster, as well
 as the platform of the cluster. These tests should be able to run in real
 world clusters, as well as ephemeral ones like KinD.

tests/e2e/README.md

@@ -6,7 +6,7 @@ This test is using the following Kubeflow CRDs:
 1. Kubeflow Pipelines
 2. Katib Experiments
 3. TFJobs
-4. KFServing InferenceServices
+4. KServe InferenceServices
 
 ## How to run
 

tests/e2e/mnist.py

@@ -1,4 +1,4 @@
-"""E2E Kubeflow test that tesst Pipelines, Katib, TFJobs and KFServing.
+"""E2E Kubeflow test that tesst Pipelines, Katib, TFJobs and KServe.
 
 Requires:
 pip install kfp==1.8.4
@@ -9,7 +9,7 @@
 from kubernetes import config
 
 import settings
-from utils import isvc, katib, kfserving, tfjob
+from utils import isvc, katib, kserve, tfjob
 
 config.load_kube_config()
 
@@ -38,8 +38,8 @@ def mnist_pipeline(name=settings.PIPELINE_NAME,
     tfjob_op = tfjob.create_tfjob_task(name, namespace, training_steps,
                                        katib_op, model_volume_op)
 
-    # Create the KFServing inference.
-    kfserving.create_kfserving_task(name, namespace, tfjob_op,
+    # Create the KServe Inference
+    kserve.create_serving_task(name, namespace, tfjob_op,
                                     model_volume_op)
 
 

tests/e2e/utils/isvc.py

@@ -1,6 +1,6 @@
 from . import watch
 
-GROUP = "serving.kubeflow.org"
+GROUP = "serving.kserve.io" #"serving.kubeflow.org" 
 PLURAL = "inferenceservices"
 VERSION = "v1beta1"
 

tests/e2e/utils/kserve.py

@@ -0,0 +1,28 @@
+from kfp import components
+
+
+def create_serving_task(model_name, model_namespace, tfjob_op, model_volume_op):
+
+    api_version = 'serving.kserve.io/v1beta1'
+    serving_component_url = 'https://raw.githubusercontent.com/kubeflow/pipelines/master/components/kserve/component.yaml'
+
+    # Uncomment the following two lines if you are using KFServing v0.6.x or v0.5.x
+    # api_version = 'serving.kubeflow.org/v1beta1'
+    # serving_component_url = 'https://raw.githubusercontent.com/kubeflow/pipelines/master/components/kubeflow/kfserving/component.yaml'
+
+    inference_service = '''
+apiVersion: "{}"
+kind: "InferenceService"
+metadata:
+  name: {}
+  namespace: {}
+  annotations:
+    "sidecar.istio.io/inject": "false"
+spec:
+  predictor:
+    tensorflow:
+      storageUri: "pvc://{}/"
+'''.format(api_version, model_name, model_namespace, str(model_volume_op.outputs["name"]))
+
+    serving_launcher_op = components.load_component_from_url(serving_component_url)
+    serving_launcher_op(action="apply", inferenceservice_yaml=inference_service).after(tfjob_op)