Cross language exception (#10023)

src/ray/core_worker/lib/java/jni_init.cc

@@ -167,7 +167,7 @@ jint JNI_OnLoad(JavaVM *vm, void *reserved) {
   java_system_class = LoadClass(env, "java/lang/System");
   java_system_gc = env->GetStaticMethodID(java_system_class, "gc", "()V");
 
-  java_ray_exception_class = LoadClass(env, "io/ray/api/exception/RayException");
+  java_ray_exception_class = LoadClass(env, "io/ray/runtime/exception/RayException");
 
   java_jni_exception_util_class = LoadClass(env, "io/ray/runtime/util/JniExceptionUtil");
   java_jni_exception_util_get_stack_trace = env->GetStaticMethodID(

src/ray/protobuf/common.proto

@@ -100,6 +100,50 @@ message FunctionDescriptor {
   }
 }
 
+// This enum type is used as object's metadata to indicate the object's
+// creating task has failed because of a certain error.
+// TODO(hchen): We may want to make these errors more specific. E.g., we may
+// want to distinguish between intentional and expected actor failures, and
+// between worker process failure and node failure.
+enum ErrorType {
+  // Indicates that a task failed because the worker died unexpectedly while
+  // executing it.
+  WORKER_DIED = 0;
+  // Indicates that a task failed because the actor died unexpectedly before
+  // finishing it.
+  ACTOR_DIED = 1;
+  // Indicates that an object is lost and cannot be restarted.
+  // Note, this currently only happens to actor objects. When the actor's
+  // state is already after the object's creating task, the actor cannot
+  // re-run the task.
+  // TODO(hchen): we may want to reuse this error type for more cases. E.g.,
+  // 1) A object that was put by the driver.
+  // 2) The object's creating task is already cleaned up from GCS (this
+  // currently crashes raylet).
+  OBJECT_UNRECONSTRUCTABLE = 2;
+  // Indicates that a task failed due to user code failure.
+  TASK_EXECUTION_EXCEPTION = 3;
+  // Indicates that the object has been placed in plasma. This error shouldn't
+  // ever be exposed to user code; it is only used internally to indicate the
+  // result of a direct call has been placed in plasma.
+  OBJECT_IN_PLASMA = 4;
+  // Indicates that an object has been cancelled.
+  TASK_CANCELLED = 5;
+  // Inidicates that creating the GCS service failed to create the actor.
+  ACTOR_CREATION_FAILED = 6;
+}
+
+/// The task exception encapsulates all information about task
+/// execution execeptions.
+message RayException {
+  // Language of this exception.
+  Language language = 1;
+  // The serialized exception.
+  bytes serialized_exception = 2;
+  // The formatted exception string.
+  string formatted_exception_string = 3;
+}
+
 /// The task specification encapsulates all immutable information about the
 /// task. These fields are determined at submission time, converse to the
 /// `TaskExecutionSpec` may change at execution time.

src/ray/protobuf/gcs.proto

@@ -422,37 +422,3 @@ message PubSubMessage {
   bytes id = 1;
   bytes data = 2;
 }
-
-// This enum type is used as object's metadata to indicate the object's
-// creating task has failed because of a certain error.
-// TODO(hchen): We may want to make these errors more specific. E.g., we may
-// want to distinguish between intentional and expected actor failures, and
-// between worker process failure and node failure.
-enum ErrorType {
-  // Indicates that a task failed because the worker died unexpectedly while
-  // executing it.
-  WORKER_DIED = 0;
-  // Indicates that a task failed because the actor died unexpectedly before
-  // finishing it.
-  ACTOR_DIED = 1;
-  // Indicates that an object is lost and cannot be restarted.
-  // Note, this currently only happens to actor objects. When the actor's
-  // state is already after the object's creating task, the actor cannot
-  // re-run the task.
-  // TODO(hchen): we may want to reuse this error type for more cases. E.g.,
-  // 1) A object that was put by the driver.
-  // 2) The object's creating task is already cleaned up from GCS (this
-  // currently
-  //    crashes raylet).
-  OBJECT_UNRECONSTRUCTABLE = 2;
-  // Indicates that a task failed due to user code failure.
-  TASK_EXECUTION_EXCEPTION = 3;
-  // Indicates that the object has been placed in plasma. This error shouldn't
-  // ever be exposed to user code; it is only used internally to indicate the
-  // result of a direct call has been placed in plasma.
-  OBJECT_IN_PLASMA = 4;
-  // Indicates that an object has been cancelled.
-  TASK_CANCELLED = 5;
-  // Inidicates that creating the GCS service failed to create the actor.
-  ACTOR_CREATION_FAILED = 6;
-}