feat(components): Add yaml support to kfserving component (#4587)

components/kubeflow/kfserving/component.yaml

@@ -6,14 +6,15 @@ inputs:
   - {name: Default Model URI,               type: String, default: '',         description: 'Path of the S3 or GCS compatible directory containing default model.'}
   - {name: Canary Model URI,                type: String, default: '',         description: 'Optional Path of the S3 or GCS compatible directory containing canary model.'}
   - {name: Canary Model Traffic Percentage, type: String, default: '0',        description: 'Optional Traffic to be sent to default model'}
-  - {name: Namespace,                       type: String, default: 'kubeflow', description: 'Kubernetes namespace where the KFServing service is deployed.'}
+  - {name: Namespace,                       type: String, default: 'anonymous', description: 'Kubernetes namespace where the KFServing service is deployed.'}
   - {name: Framework,                       type: String, default: 'tensorflow', description: 'Machine Learning Framework for Model Serving.'}
   - {name: Default Custom Model Spec,       type: String, default: '{}', description: 'Custom runtime default custom model container spec.'}
   - {name: Canary Custom Model Spec,        type: String, default: '{}', description: 'Custom runtime canary custom model container spec.'}
   - {name: Autoscaling Target,              type: String, default: '0',        description: 'Autoscaling Target Number'}
   - {name: KFServing Endpoint,              type: String, default: '', description: 'KFServing remote deployer API endpoint'}
   - {name: Service Account,                 type: String, default: '', description: 'Model Service Account'}
   - {name: Enable Istio Sidecar,            type: Bool,   default: 'True',     description: 'Whether to enable istio sidecar injection'}
+  - {name: InferenceService YAML,           type: String, default: '{}',         description: 'Raw InferenceService serialized YAML for deployment'} 
 outputs:
   - {name: Service Endpoint URI, type: String,                      description: 'URI of the deployed prediction service..'}
 implementation:
@@ -35,5 +36,6 @@ implementation:
       --autoscaling-target,       {inputValue: Autoscaling Target},
       --service-account,          {inputValue: Service Account},
       --enable-istio-sidecar,     {inputValue: Enable Istio Sidecar},
-      --output-path,              {outputPath: Service Endpoint URI}
+      --output-path,              {outputPath: Service Endpoint URI},
+      --inferenceservice_yaml,    {inputValue: InferenceService YAML}
     ]

components/kubeflow/kfserving/src/app.py

@@ -40,7 +40,8 @@ def deploy_model_post():
             canary_custom_model_spec=request.json["canary_custom_model_spec"],
             autoscaling_target=request.json["autoscaling_target"],
             service_account=request.json["service_account"],
-            enable_istio_sidecar=request.json["enable_istio_sidecar"]
+            enable_istio_sidecar=request.json["enable_istio_sidecar"],
+            inferenceservice_yaml=request.json["inferenceservice_yaml"]
         )
     )
 

components/kubeflow/kfserving/src/kfservingdeployer.py

@@ -18,6 +18,7 @@
 import requests
 import re
 import time
+import yaml
 from distutils.util import strtobool
 
 from kubernetes import client
@@ -37,6 +38,13 @@
 from kfserving import V1alpha2InferenceService
 
 
+def yamlStr(str):
+    if str == "" or str == None:
+        return None
+    else:
+        return yaml.safe_load(str)
+
+
 def EndpointSpec(framework, storage_uri, service_account):
     if framework == "tensorflow":
         return V1alpha2EndpointSpec(
@@ -143,53 +151,62 @@ def deploy_model(
     canary_custom_model_spec,
     service_account,
     autoscaling_target=0,
-    enable_istio_sidecar=True
+    enable_istio_sidecar=True,
+    inferenceservice_yaml={}
 ):
-    # Create annotation
-    annotations = {}
-    if int(autoscaling_target) != 0:
-        annotations["autoscaling.knative.dev/target"] = str(autoscaling_target)
-    if not enable_istio_sidecar:
-        annotations["sidecar.istio.io/inject"] = 'false'
-    if not annotations:
-        annotations = None
-    metadata = client.V1ObjectMeta(
-        name=model_name, namespace=namespace, annotations=annotations
-    )
+    KFServing = KFServingClient()
 
-    # Create Default deployment if default model uri is provided.
-    if framework != "custom" and default_model_uri:
-        default_model_spec = EndpointSpec(framework, default_model_uri, service_account)
-    elif framework == "custom" and default_custom_model_spec:
-        default_model_spec = customEndpointSpec(
-            default_custom_model_spec, service_account
+    if inferenceservice_yaml:
+        # Overwrite name and namespace if exist
+        if namespace:
+            inferenceservice_yaml['metadata']['namespace'] = namespace
+        if model_name:
+            inferenceservice_yaml['metadata']['name'] = model_name
+        kfsvc = inferenceservice_yaml
+    else:
+        # Create annotation
+        annotations = {}
+        if int(autoscaling_target) != 0:
+            annotations["autoscaling.knative.dev/target"] = str(autoscaling_target)
+        if not enable_istio_sidecar:
+            annotations["sidecar.istio.io/inject"] = 'false'
+        if not annotations:
+            annotations = None
+        metadata = client.V1ObjectMeta(
+            name=model_name, namespace=namespace, annotations=annotations
         )
 
-    # Create Canary deployment if canary model uri is provided.
-    if framework != "custom" and canary_model_uri:
-        canary_model_spec = EndpointSpec(framework, canary_model_uri, service_account)
-        kfsvc = InferenceService(
-            metadata, default_model_spec, canary_model_spec, canary_model_traffic
-        )
-    elif framework == "custom" and canary_custom_model_spec:
-        canary_model_spec = customEndpointSpec(
-            canary_custom_model_spec, service_account
-        )
-        kfsvc = InferenceService(
-            metadata, default_model_spec, canary_model_spec, canary_model_traffic
-        )
-    else:
-        kfsvc = InferenceService(metadata, default_model_spec)
+        # Create Default deployment if default model uri is provided.
+        if framework != "custom" and default_model_uri:
+            default_model_spec = EndpointSpec(framework, default_model_uri, service_account)
+        elif framework == "custom" and default_custom_model_spec:
+            default_model_spec = customEndpointSpec(
+                default_custom_model_spec, service_account
+            )
 
-    KFServing = KFServingClient()
+        # Create Canary deployment if canary model uri is provided.
+        if framework != "custom" and canary_model_uri:
+            canary_model_spec = EndpointSpec(framework, canary_model_uri, service_account)
+            kfsvc = InferenceService(
+                metadata, default_model_spec, canary_model_spec, canary_model_traffic
+            )
+        elif framework == "custom" and canary_custom_model_spec:
+            canary_model_spec = customEndpointSpec(
+                canary_custom_model_spec, service_account
+            )
+            kfsvc = InferenceService(
+                metadata, default_model_spec, canary_model_spec, canary_model_traffic
+            )
+        else:
+            kfsvc = InferenceService(metadata, default_model_spec)
 
     def create(kfsvc, model_name, namespace):
-        KFServing.create(kfsvc)
+        KFServing.create(kfsvc, namespace=namespace)
         time.sleep(1)
         KFServing.get(model_name, namespace=namespace, watch=True, timeout_seconds=120)
 
     def update(kfsvc, model_name, namespace):
-        KFServing.patch(model_name, kfsvc)
+        KFServing.patch(model_name, kfsvc, namespace=namespace)
         time.sleep(1)
         KFServing.get(model_name, namespace=namespace, watch=True, timeout_seconds=120)
 
@@ -203,6 +220,8 @@ def update(kfsvc, model_name, namespace):
         except:
             update(kfsvc, model_name, namespace)
     elif action == "rollout":
+        if inferenceservice_yaml:
+            raise("Rollout is not supported for inferenceservice yaml")
         KFServing.rollout_canary(
             model_name,
             canary=canary_model_spec,
@@ -253,7 +272,7 @@ def update(kfsvc, model_name, namespace):
         "--namespace",
         type=str,
         help="Kubernetes namespace where the KFServing service is deployed.",
-        default="kubeflow",
+        default="anonymous",
     )
     parser.add_argument(
         "--framework", type=str, help="Model Serving Framework", default="tensorflow"
@@ -286,7 +305,16 @@ def update(kfsvc, model_name, namespace):
         default="",
     )
     parser.add_argument(
-        "--enable-istio-sidecar", type=strtobool, help="Whether to inject istio sidecar", default="True"
+        "--enable-istio-sidecar",
+        type=strtobool,
+        help="Whether to inject istio sidecar",
+        default="True"
+    )
+    parser.add_argument(
+        "--inferenceservice_yaml",
+        type=yamlStr,
+        help="Raw InferenceService serialized YAML for deployment",
+        default={}
     )
     parser.add_argument("--output-path", type=str, help="Path to store URI output")
     args = parser.parse_args()
@@ -307,6 +335,7 @@ def update(kfsvc, model_name, namespace):
     autoscaling_target = int(args.autoscaling_target)
     service_account = args.service_account
     enable_istio_sidecar = args.enable_istio_sidecar
+    inferenceservice_yaml = args.inferenceservice_yaml
 
     if kfserving_endpoint:
         formData = {
@@ -321,7 +350,8 @@ def update(kfsvc, model_name, namespace):
             "canary_custom_model_spec": canary_custom_model_spec,
             "autoscaling_target": autoscaling_target,
             "service_account": service_account,
-            "enable_istio_sidecar": enable_istio_sidecar
+            "enable_istio_sidecar": enable_istio_sidecar,
+            "inferenceservice_yaml": inferenceservice_yaml
         }
         response = requests.post(
             "http://" + kfserving_endpoint + "/deploy-model", json=formData
@@ -340,7 +370,8 @@ def update(kfsvc, model_name, namespace):
             canary_custom_model_spec=canary_custom_model_spec,
             autoscaling_target=autoscaling_target,
             service_account=service_account,
-            enable_istio_sidecar=enable_istio_sidecar
+            enable_istio_sidecar=enable_istio_sidecar,
+            inferenceservice_yaml=inferenceservice_yaml
         )
     print(model_status)
     # Check whether the model is ready