commit oip model metadata endpoint

Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com>

.github/workflows/kserve_cpu_tests.yml

@@ -1,22 +1,14 @@
 name: KServe CPU Nightly Tests
 
 on:
-  pull_request:
-    branches: [master, release*]
-    paths:
-      - "**"
-      - "!.github/**"
-      - "!docs/**"
-      - "!**.md"
-      - ".github/workflows/kserve_cpu_tests.yml"
   workflow_dispatch:
   # runs everyday  at 5:15am
-  # schedule:
-  #   - cron:  '15 5 * * *'
+  schedule:
+    - cron:  '15 5 * * *'
 
 jobs:
   kserve-cpu-tests:
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, regression-test-gpu]
     steps:
       - name: Clean up previous run
         run: |
@@ -47,7 +39,7 @@ jobs:
         uses: actions/checkout@v4
         with:
           repository: kserve/kserve
-          ref: v0.11.1
+          ref: v0.11.2
           path: kserve
       - name: Validate torchserve-kfs and Open Inference Protocol
         run: ./kubernetes/kserve/tests/scripts/test_mnist.sh

frontend/server/src/main/java/org/pytorch/serve/http/api/rest/OpenInferenceProtocolRequestHandler.java

@@ -9,6 +9,7 @@
 import org.pytorch.serve.archive.DownloadArchiveException;
 import org.pytorch.serve.archive.model.ModelException;
 import org.pytorch.serve.archive.workflow.WorkflowException;
+import org.pytorch.serve.grpc.openinference.OpenInferenceGrpc.ModelMetadataResponse;
 import org.pytorch.serve.http.HttpRequestHandlerChain;
 import org.pytorch.serve.util.ConfigManager;
 import org.pytorch.serve.util.NettyUtils;
@@ -31,6 +32,7 @@ public class OpenInferenceProtocolRequestHandler extends HttpRequestHandlerChain
     private static final String SERVER_LIVE_API = "/v2/health/live";
     private static final String SERVER_READY_API = "/v2/health/ready";
     private static final String MODEL_READY_ENDPOINT_PATTERN = "^/v2/models/([^/]+)(?:/versions/([^/]+))?/ready$";
+    private static final String MODEL_METADATA_ENDPOINT_PATTERN = "^/v2/models/([^/]+)(?:/versions/([^/]+))?";
 
     /** Creates a new {@code OpenInferenceProtocolRequestHandler} instance. */
     public OpenInferenceProtocolRequestHandler() {}
@@ -83,6 +85,19 @@ public void handleRequest(
             response.addProperty("ready", isModelReady);
             NettyUtils.sendJsonResponse(ctx, response);
 
+        } else if (concatenatedSegments.matches(MODEL_METADATA_ENDPOINT_PATTERN)) {
+            String modelName = segments[3];
+            String modelVersion = null;
+            if (segments.length > 5) {
+                modelVersion = segments[5];
+            }
+
+            ModelManager modelManager = ModelManager.getInstance();
+
+            ModelMetadataResponse.Builder response = modelManager.modelMetadata(modelName, modelVersion);
+
+            NettyUtils.sendJsonResponse(ctx, response.build());
+
         } else if (segments.length > 5 && concatenatedSegments.contains("/versions")) {
             // As of now kserve not implemented versioning, we just throws not implemented.
             JsonObject response = new JsonObject();

frontend/server/src/main/java/org/pytorch/serve/wlm/ModelManager.java

@@ -1,6 +1,9 @@
 package org.pytorch.serve.wlm;
 
 import com.google.gson.JsonObject;
+
+import io.grpc.Status;
+
 import java.io.BufferedReader;
 import java.io.File;
 import java.io.IOException;
@@ -28,6 +31,9 @@
 import org.pytorch.serve.archive.model.ModelException;
 import org.pytorch.serve.archive.model.ModelNotFoundException;
 import org.pytorch.serve.archive.model.ModelVersionNotFoundException;
+import org.pytorch.serve.grpc.openinference.OpenInferenceGrpc.ModelMetadataResponse;
+import org.pytorch.serve.grpc.openinference.OpenInferenceGrpc.ModelMetadataResponse.TensorMetadata;
+import org.pytorch.serve.http.BadRequestException;
 import org.pytorch.serve.http.ConflictStatusException;
 import org.pytorch.serve.http.InvalidModelVersionException;
 import org.pytorch.serve.http.messages.RegisterModelRequest;
@@ -698,6 +704,38 @@ public boolean isModelReady(String modelName, String modelVersion)
     return numHealthy >= numScaled;
     }
 
+    public ModelMetadataResponse.Builder modelMetadata(String modelName, String modelVersion)
+        throws ModelVersionNotFoundException, ModelNotFoundException {
+
+        ModelManager modelManager = ModelManager.getInstance();
+        ModelMetadataResponse.Builder response = ModelMetadataResponse.newBuilder();
+        List<TensorMetadata> inputs = new ArrayList<>();
+        List<TensorMetadata> outputs = new ArrayList<>();
+        List<String> versions = new ArrayList<>();
+
+        if (modelVersion == null || "".equals(modelVersion)) {
+            modelVersion = null;
+        }
+
+        Model model = modelManager.getModel(modelName, modelVersion);
+        if (model == null) {
+            throw new ModelNotFoundException("Model not found: " + modelName);
+        }
+        modelManager
+                .getAllModelVersions(modelName)
+                .forEach(entry -> versions.add(entry.getKey()));
+        response.setName(modelName);
+        response.addAllVersions(versions);
+        response.setPlatform("");
+        response.addAllInputs(inputs);
+        response.addAllOutputs(outputs);
+
+        return response;
+
+    }
+
+    // return numHealthy >= numScaled;
+
     public void submitTask(Runnable runnable) {
         wlm.scheduleAsync(runnable);
     }

kubernetes/kserve/requirements.txt

@@ -1,4 +1,4 @@
-kserve[storage]>=0.11.0
+kserve[storage]==0.11.2
 transformers
 captum
 grpcio

kubernetes/kserve/tests/configs/mnist_oip_http.yaml

@@ -7,7 +7,7 @@ spec:
     pytorch:
       protocolVersion: v2
       storageUri: gs://kfserving-examples/models/torchserve/image_classifier/v2
-      image: muthuraji2i/torchserve-oip-test:latest
+      image: sivanantha/torchserve-oip-test:latest
       resources:
         limits:
           cpu: "1"

kubernetes/kserve/tests/scripts/test_mnist.sh

@@ -212,6 +212,12 @@ URL="http://${INGRESS_HOST}:${INGRESS_PORT}/v2/models/${MODEL_NAME}/ready"
 EXPECTED_OUTPUT='{"name":"mnist","ready":true}'
 make_cluster_accessible ${SERVICE_NAME} ${URL} "" ${EXPECTED_OUTPUT}
 
+# ModelMetadata
+echo "HTTP ModelMetadata method call"
+URL="http://${INGRESS_HOST}:${INGRESS_PORT}/v2/models/${MODEL_NAME}"
+EXPECTED_OUTPUT='{"name_":"mnist","versions_":["1.0"],"platform_":"","inputs_":[],"outputs_":[],"memoizedIsInitialized":1,"unknownFields":{"fields":{}},"memoizedSize":-1,"memoizedHashCode":0}'
+make_cluster_accessible ${SERVICE_NAME} ${URL} "" ${EXPECTED_OUTPUT}
+
 # delete oip http isvc
 kubectl delete inferenceservice ${SERVICE_NAME}