List Models v2 REST API (kubeflow#2963)

* List Models v2 REST API

Signed-off-by: Jagan Vujjini <jvujjini@Snehas-MacBook-Pro.local>
Signed-off-by: jvujjini <jagan120@gmail.com>

* Fixed lint error

Signed-off-by: jvujjini <jagan120@gmail.com>

---------

Signed-off-by: Jagan Vujjini <jvujjini@Snehas-MacBook-Pro.local>
Signed-off-by: jvujjini <jagan120@gmail.com>
Co-authored-by: Jagan Vujjini <jvujjini@Snehas-MacBook-Pro.local>

python/kserve/kserve/protocol/rest/server.py

@@ -37,7 +37,8 @@
 from .v1_endpoints import V1Endpoints
 from .v2_datamodels import (InferenceResponse, ModelMetadataResponse,
                             ModelReadyResponse, ServerLiveResponse,
-                            ServerMetadataResponse, ServerReadyResponse)
+                            ServerMetadataResponse, ServerReadyResponse,
+                            ListModelsResponse)
 from .v2_endpoints import V2Endpoints
 
 
@@ -100,6 +101,8 @@ def create_application(self) -> FastAPI:
                              response_model=ServerLiveResponse, tags=["V2"]),
                 FastAPIRoute(r"/v2/health/ready", v2_endpoints.ready,
                              response_model=ServerReadyResponse, tags=["V2"]),
+                FastAPIRoute(r"/v2/models", v2_endpoints.models,
+                             response_model=ListModelsResponse, tags=["V2"]),
                 FastAPIRoute(r"/v2/models/{model_name}",
                              v2_endpoints.model_metadata, response_model=ModelMetadataResponse, tags=["V2"]),
                 FastAPIRoute(r"/v2/models/{model_name}/versions/{model_version}",

python/kserve/kserve/protocol/rest/v2_datamodels.py

@@ -78,6 +78,17 @@ class MetadataTensor(BaseModel):
     shape: List[int]
 
 
+class ListModelsResponse(BaseModel):
+    """ListModelsResponse
+
+        $models_list_response =
+        {
+          "models" : [ $string, ... ]
+        }
+    """
+    models: List[str]
+
+
 class ModelMetadataResponse(BaseModel):
     """ModelMetadataResponse
 

python/kserve/kserve/protocol/rest/v2_endpoints.py

@@ -20,7 +20,7 @@
 from ..infer_type import InferInput, InferRequest
 from .v2_datamodels import (
     InferenceRequest, ServerMetadataResponse, ServerLiveResponse, ServerReadyResponse,
-    ModelMetadataResponse, InferenceResponse, ModelReadyResponse
+    ModelMetadataResponse, InferenceResponse, ModelReadyResponse, ListModelsResponse
 )
 from ..dataplane import DataPlane
 from ..model_repository_extension import ModelRepositoryExtension
@@ -61,6 +61,15 @@ async def ready() -> ServerReadyResponse:
         """
         return ServerReadyResponse(ready=True)
 
+    async def models(self) -> ListModelsResponse:
+        """Get a list of models in the model registry.
+
+        Returns:
+            ListModelsResponse: List of models object.
+        """
+        models = list(self.dataplane.model_registry.get_models().keys())
+        return ListModelsResponse.parse_obj({"models": models})
+
     async def model_metadata(self, model_name: str, model_version: Optional[str] = None) -> ModelMetadataResponse:
         """Model metadata handler. It provides information about a model.
 

python/kserve/test/test_server.py

@@ -233,6 +233,11 @@ def test_list_models(self, http_server_client):
         assert resp.status_code == 200
         assert resp.json() == {"models": ["TestModel"]}
 
+    def test_list_models_v2(self, http_server_client):
+        resp = http_server_client.get('/v2/models')
+        assert resp.status_code == 200
+        assert resp.json() == {"models": ["TestModel"]}
+
     def test_predict(self, http_server_client):
         resp = http_server_client.post('/v1/models/TestModel:predict',
                                        data=b'{"instances":[[1,2]]}')