chore: formalize gRPC errors in case of UDF exceptions

src/main/java/io/numaproj/numaflow/batchmapper/Service.java

@@ -1,11 +1,16 @@
 package io.numaproj.numaflow.batchmapper;
 
+import com.google.protobuf.Any;
 import com.google.protobuf.ByteString;
 import com.google.protobuf.Empty;
+import com.google.rpc.Code;
+import com.google.rpc.DebugInfo;
 import io.grpc.Status;
+import io.grpc.protobuf.StatusProto;
 import io.grpc.stub.StreamObserver;
 import io.numaproj.numaflow.map.v1.MapGrpc;
 import io.numaproj.numaflow.map.v1.MapOuterClass;
+import io.numaproj.numaflow.shared.ExceptionUtils;
 import lombok.AllArgsConstructor;
 import lombok.extern.slf4j.Slf4j;
 
@@ -98,10 +103,15 @@ public void onNext(MapOuterClass.MapRequest mapRequest) {
                 } catch (Exception e) {
                     log.error("Encountered an error in batch map onNext", e);
                     shutdownSignal.completeExceptionally(e);
-                    responseObserver.onError(Status.INTERNAL
-                            .withDescription(e.getMessage())
-                            .withCause(e)
-                            .asException());
+                    // Build gRPC Status
+                    com.google.rpc.Status status = com.google.rpc.Status.newBuilder()
+                            .setCode(Code.INTERNAL.getNumber())
+                            .setMessage(ExceptionUtils.ERR_BATCH_MAP_EXCEPTION + ": " + (e.getMessage() != null ? e.getMessage() : ""))
+                            .addDetails(Any.pack(DebugInfo.newBuilder()
+                                    .setDetail(ExceptionUtils.getStackTrace(e))
+                                    .build()))
+                            .build();
+                    responseObserver.onError(StatusProto.toStatusRuntimeException(status));
                 }
             }
 

src/main/java/io/numaproj/numaflow/mapper/MapSupervisorActor.java

@@ -9,8 +9,13 @@
 import akka.japi.pf.DeciderBuilder;
 import akka.japi.pf.ReceiveBuilder;
 import io.grpc.Status;
+import com.google.protobuf.Any;
+import com.google.rpc.Code;
+import com.google.rpc.DebugInfo;
+import io.grpc.protobuf.StatusProto;
 import io.grpc.stub.StreamObserver;
 import io.numaproj.numaflow.map.v1.MapOuterClass;
+import io.numaproj.numaflow.shared.ExceptionUtils;
 import lombok.extern.slf4j.Slf4j;
 
 import java.util.Optional;
@@ -106,10 +111,15 @@ private void handleFailure(Exception e) {
             userException = e;
             // only send the very first exception to the client
             // one exception should trigger a container restart
-            responseObserver.onError(Status.INTERNAL
-                    .withDescription(e.getMessage())
-                    .withCause(e)
-                    .asException());
+            // Build gRPC Status
+            com.google.rpc.Status status = com.google.rpc.Status.newBuilder()
+                    .setCode(Code.INTERNAL.getNumber())
+                    .setMessage(ExceptionUtils.ERR_MAP_EXCEPTION + ": " + (e.getMessage() != null ? e.getMessage() : ""))
+                    .addDetails(Any.pack(DebugInfo.newBuilder()
+                            .setDetail(ExceptionUtils.getStackTrace(e))
+                            .build()))
+                    .build();
+            responseObserver.onError(StatusProto.toStatusRuntimeException(status));
         }
         activeMapperCount--;
     }

src/main/java/io/numaproj/numaflow/mapstreamer/Service.java

@@ -1,10 +1,15 @@
 package io.numaproj.numaflow.mapstreamer;
 
+import com.google.protobuf.Any;
 import com.google.protobuf.Empty;
+import com.google.rpc.Code;
+import com.google.rpc.DebugInfo;
 import io.grpc.Status;
+import io.grpc.protobuf.StatusProto;
 import io.grpc.stub.StreamObserver;
 import io.numaproj.numaflow.map.v1.MapGrpc;
 import io.numaproj.numaflow.map.v1.MapOuterClass;
+import io.numaproj.numaflow.shared.ExceptionUtils;
 import lombok.AllArgsConstructor;
 import lombok.extern.slf4j.Slf4j;
 
@@ -61,10 +66,15 @@ public void onNext(MapOuterClass.MapRequest request) {
                 } catch (Exception e) {
                     log.error("Encountered error in mapFn onNext", e);
                     shutdownSignal.completeExceptionally(e);
-                    responseObserver.onError(Status.INTERNAL
-                            .withDescription(e.getMessage())
-                            .withCause(e)
-                            .asException());
+                    // Build gRPC Status
+                    com.google.rpc.Status status = com.google.rpc.Status.newBuilder()
+                            .setCode(Code.INTERNAL.getNumber())
+                            .setMessage(ExceptionUtils.ERR_MAP_STREAM_EXCEPTION + ": " + (e.getMessage() != null ? e.getMessage() : ""))
+                            .addDetails(Any.pack(DebugInfo.newBuilder()
+                                    .setDetail(ExceptionUtils.getStackTrace(e))
+                                    .build()))
+                            .build();
+                    responseObserver.onError(StatusProto.toStatusRuntimeException(status));
                     return;
                 }
 

src/main/java/io/numaproj/numaflow/shared/ExceptionUtils.java

@@ -0,0 +1,31 @@
+package io.numaproj.numaflow.shared;
+
+import java.io.PrintWriter;
+import java.io.StringWriter;
+
+public class ExceptionUtils {
+    /**
+     * Formalized exception error strings
+     */
+    public static final String ERR_SOURCE_EXCEPTION = "UDF_EXECUTION_ERROR(source)";
+    public static final String ERR_TRANSFORMER_EXCEPTION = "UDF_EXECUTION_ERROR(transformer)";
+    public static final String ERR_SINK_EXCEPTION = "UDF_EXECUTION_ERROR(sink)";
+    public static final String ERR_MAP_STREAM_EXCEPTION = "UDF_EXECUTION_ERROR(mapstream)";
+    public static final String ERR_MAP_EXCEPTION = "UDF_EXECUTION_ERROR(map)";
+    public static final String ERR_BATCH_MAP_EXCEPTION = "UDF_EXECUTION_ERROR(batchmap)";
+
+    /**
+     * Converts the stack trace of an exception into a String.
+     *
+     * @param e the exception to extract the stack trace from
+     * @return the stack trace as a String
+     */
+    public static String getStackTrace(Throwable t) {
+        if (t == null) {
+            return "No exception provided.";
+        }
+        StringWriter sw = new StringWriter();
+        t.printStackTrace(new PrintWriter(sw));
+        return sw.toString();
+    }
+}

src/main/java/io/numaproj/numaflow/sideinput/Server.java

@@ -32,7 +32,7 @@ public Server(SideInputRetriever sideInputRetriever) {
     /**
      * constructor to create gRPC server with gRPC config.
      *
-     * @param grpcConfig to configure the max message size for grpc
+     * @param grpcConfig         to configure the max message size for grpc
      * @param sideInputRetriever to retrieve the side input
      */
     public Server(SideInputRetriever sideInputRetriever, GRPCConfig grpcConfig) {
@@ -41,7 +41,8 @@ public Server(SideInputRetriever sideInputRetriever, GRPCConfig grpcConfig) {
     }
 
     @VisibleForTesting
-    protected Server(GRPCConfig grpcConfig, SideInputRetriever service, ServerInterceptor interceptor, String serverName) {
+    protected Server(GRPCConfig grpcConfig, SideInputRetriever service, ServerInterceptor interceptor,
+            String serverName) {
         this.grpcConfig = grpcConfig;
         this.server = new GrpcServerWrapper(
                 interceptor,
@@ -67,8 +68,7 @@ public void start() throws Exception {
 
         log.info(
                 "server started, listening on {}",
-                this.grpcConfig.isLocal() ?
-                        "localhost:" + this.grpcConfig.getPort() : this.grpcConfig.getSocketPath());
+                this.grpcConfig.isLocal() ? "localhost:" + this.grpcConfig.getPort() : this.grpcConfig.getSocketPath());
 
         // register shutdown hook to gracefully shut down the server
         Runtime.getRuntime().addShutdownHook(new Thread(() -> {
@@ -83,11 +83,14 @@ public void start() throws Exception {
     }
 
     /**
-     * Blocks until the server has terminated. If the server is already terminated, this method
-     * will return immediately. If the server is not yet terminated, this method will block the
+     * Blocks until the server has terminated. If the server is already terminated,
+     * this method
+     * will return immediately. If the server is not yet terminated, this method
+     * will block the
      * calling thread until the server has terminated.
      *
-     * @throws InterruptedException if the current thread is interrupted while waiting
+     * @throws InterruptedException if the current thread is interrupted while
+     *                              waiting
      */
     public void awaitTermination() throws InterruptedException {
         log.info("side input server is waiting for termination");
@@ -96,7 +99,8 @@ public void awaitTermination() throws InterruptedException {
     }
 
     /**
-     * Stop serving requests and shutdown resources. Await termination on the main thread since the
+     * Stop serving requests and shutdown resources. Await termination on the main
+     * thread since the
      * grpc library uses daemon threads.
      *
      * @throws InterruptedException if shutdown is interrupted

src/main/java/io/numaproj/numaflow/sideinput/Service.java

@@ -1,14 +1,13 @@
 package io.numaproj.numaflow.sideinput;
 
-import com.google.protobuf.ByteString;
 import com.google.protobuf.Empty;
 import io.grpc.stub.StreamObserver;
+import com.google.protobuf.ByteString;
 import io.numaproj.numaflow.sideinput.v1.SideInputGrpc;
 import io.numaproj.numaflow.sideinput.v1.Sideinput;
 import lombok.AllArgsConstructor;
 import lombok.extern.slf4j.Slf4j;
 
-
 @Slf4j
 @AllArgsConstructor
 class Service extends SideInputGrpc.SideInputImplBase {
@@ -29,11 +28,8 @@ public void retrieveSideInput(
                     responseObserver);
             return;
         }
-
-
         // process request
         Message message = sideInputRetriever.retrieveSideInput();
-
         // set response
         responseObserver.onNext(buildResponse(message));
         responseObserver.onCompleted();
@@ -50,8 +46,9 @@ public void isReady(Empty request, StreamObserver<Sideinput.ReadyResponse> respo
 
     private Sideinput.SideInputResponse buildResponse(Message message) {
         return Sideinput.SideInputResponse.newBuilder()
-                .setValue(message.getValue() == null ? ByteString.EMPTY : ByteString.copyFrom(
-                        message.getValue()))
+                .setValue(message.getValue() == null ? ByteString.EMPTY
+                        : ByteString.copyFrom(
+                                message.getValue()))
                 .setNoBroadcast(message.isNoBroadcast())
                 .build();
     }

src/main/java/io/numaproj/numaflow/sinker/Service.java

@@ -1,8 +1,13 @@
 package io.numaproj.numaflow.sinker;
 
+import com.google.protobuf.Any;
 import com.google.protobuf.Empty;
+import com.google.rpc.Code;
+import com.google.rpc.DebugInfo;
 import io.grpc.Status;
+import io.grpc.protobuf.StatusProto;
 import io.grpc.stub.StreamObserver;
+import io.numaproj.numaflow.shared.ExceptionUtils;
 import io.numaproj.numaflow.sink.v1.SinkGrpc;
 import io.numaproj.numaflow.sink.v1.SinkOuterClass;
 import lombok.AllArgsConstructor;
@@ -100,6 +105,15 @@
                     responseObserver.onError(Status.INTERNAL
                             .withDescription(e.getMessage())
                             .asException());
+                    // Build gRPC Status
+                    com.google.rpc.Status status = com.google.rpc.Status.newBuilder()
+                            .setCode(Code.INTERNAL.getNumber())
+                            .setMessage(ExceptionUtils.ERR_SINK_EXCEPTION + ": " + (e.getMessage() != null ? e.getMessage() : ""))
+                            .addDetails(Any.pack(DebugInfo.newBuilder()
+                                    .setDetail(ExceptionUtils.getStackTrace(e))
+                                    .build()))
+                            .build();
+                    responseObserver.onError(StatusProto.toStatusRuntimeException(status));
                 }
             }
 

src/main/java/io/numaproj/numaflow/sourcer/Service.java

@@ -1,8 +1,13 @@
 package io.numaproj.numaflow.sourcer;
 
+import com.google.protobuf.Any;
 import com.google.protobuf.Empty;
+import com.google.rpc.Code;
+import com.google.rpc.DebugInfo;
 import io.grpc.Status;
+import io.grpc.protobuf.StatusProto;
 import io.grpc.stub.StreamObserver;
+import io.numaproj.numaflow.shared.ExceptionUtils;
 import io.numaproj.numaflow.source.v1.SourceGrpc;
 import io.numaproj.numaflow.source.v1.SourceOuterClass;
 import lombok.AllArgsConstructor;
@@ -15,7 +20,6 @@
 
 import static io.numaproj.numaflow.source.v1.SourceGrpc.getPendingFnMethod;
 
-
 /**
  * Implementation of the gRPC service for the sourcer.
  */
@@ -31,7 +35,8 @@
      * @param responseObserver the response observer
      */
     @Override
-    public StreamObserver<SourceOuterClass.ReadRequest> readFn(final StreamObserver<SourceOuterClass.ReadResponse> responseObserver) {
+    public StreamObserver<SourceOuterClass.ReadRequest> readFn(
+            final StreamObserver<SourceOuterClass.ReadResponse> responseObserver) {
         return new StreamObserver<>() {
             private boolean handshakeDone = false;
 
@@ -80,10 +85,15 @@
                 } catch (Exception e) {
                     log.error("Encountered error in readFn onNext", e);
                     shutdownSignal.completeExceptionally(e);
-                    responseObserver.onError(Status.INTERNAL
-                            .withDescription(e.getMessage())
-                            .withCause(e)
-                            .asException());
+                    // Build gRPC Status
+                    com.google.rpc.Status status = com.google.rpc.Status.newBuilder()
+                            .setCode(Code.INTERNAL.getNumber())
+                            .setMessage(ExceptionUtils.ERR_SOURCE_EXCEPTION + ": " + (e.getMessage() != null ? e.getMessage() : ""))
+                            .addDetails(Any.pack(DebugInfo.newBuilder()
+                                    .setDetail(ExceptionUtils.getStackTrace(e))
+                                    .build()))
+                            .build();
+                    responseObserver.onError(StatusProto.toStatusRuntimeException(status));
                 }
             }
 
@@ -201,7 +211,8 @@
                 SourceOuterClass.PendingResponse.Result
                         .newBuilder()
                         .setCount(this.sourcer.getPending())
-                        .build()).build());
+                        .build())
+                .build());
         responseObserver.onCompleted();
     }
 
@@ -236,8 +247,8 @@
         responseObserver.onNext(SourceOuterClass.PartitionsResponse.newBuilder()
                 .setResult(
                         SourceOuterClass.PartitionsResponse.Result.newBuilder()
-                                .addAllPartitions(partitions)).
-                build());
+                                .addAllPartitions(partitions))
+                .build());
         responseObserver.onCompleted();
     }
 }

src/main/java/io/numaproj/numaflow/sourcetransformer/TransformSupervisorActor.java

@@ -8,8 +8,13 @@
 import akka.actor.SupervisorStrategy;
 import akka.japi.pf.DeciderBuilder;
 import akka.japi.pf.ReceiveBuilder;
-import io.grpc.Status;
 import io.grpc.stub.StreamObserver;
+import io.grpc.Status;
+import com.google.protobuf.Any;
+import com.google.rpc.Code;
+import com.google.rpc.DebugInfo;
+import io.grpc.protobuf.StatusProto;
+import io.numaproj.numaflow.shared.ExceptionUtils;
 import io.numaproj.numaflow.sourcetransformer.v1.Sourcetransformer;
 import lombok.extern.slf4j.Slf4j;
 
@@ -144,10 +149,16 @@ private void handleFailure(Exception e) {
             userException = e;
             // only send the very first exception to the client
             // one exception should trigger a container restart
-            responseObserver.onError(Status.INTERNAL
-                    .withDescription(e.getMessage())
-                    .withCause(e)
-                    .asException());
+
+            // Build gRPC Status
+            com.google.rpc.Status status = com.google.rpc.Status.newBuilder()
+                    .setCode(Code.INTERNAL.getNumber())
+                    .setMessage(ExceptionUtils.ERR_TRANSFORMER_EXCEPTION + ": " + (e.getMessage() != null ? e.getMessage() : ""))
+                    .addDetails(Any.pack(DebugInfo.newBuilder()
+                            .setDetail(ExceptionUtils.getStackTrace(e))
+                            .build()))
+                    .build();
+            responseObserver.onError(StatusProto.toStatusRuntimeException(status));
         }
         activeTransformersCount--;
     }
@@ -217,7 +228,6 @@ public SupervisorStrategy supervisorStrategy() {
                                     .asException());
                             return SupervisorStrategy.stop();
                         })
-                        .build()
-        );
+                        .build());
     }
 }