log endpoint name when supervision error happens

Summary:
The current error message logs the actor name and stacktrace when an exception happens (ActorError/SupervisionError). The endpoint name is included in the stacktrace.

But for cases like proc crash, the supervision error doesn't have a stacktrace, so it couldn't show the endpoint name.

This diff adds a error log message that proceeds the supervision error, it will print both the actor name and endpoint name of the call. It is very useful for users to figure out where the failure happened in all error cases.


github issue: #1899

Differential Revision: D87353113

Author: moonli

python/monarch/_src/actor/endpoint.py

@@ -7,6 +7,7 @@
 # pyre-strict
 
 import functools
+import logging
 import time
 from abc import ABC, abstractmethod
 from typing import (
@@ -233,7 +234,15 @@ def choose(self, *args: P.args, **kwargs: P.kwargs) -> Future[R]:
             1,
         )
         async def process() -> R:
-            result = await r.recv()
+            try:
+                result = await r.recv()
+            except Exception as e:
+                from monarch._src.actor.actor_mesh import current_actor_name
+
+                logging.error(
+                    f'Endpoint call "{method_name}" failed on actor mesh "{current_actor_name()}"'
+                )
+                raise e
             return result
 
         return Future(coro=process())
@@ -260,7 +269,15 @@ def call_one(self, *args: P.args, **kwargs: P.kwargs) -> Future[R]:
             1,
         )
         async def process() -> R:
-            result = await r.recv()
+            try:
+                result = await r.recv()
+            except Exception as e:
+                from monarch._src.actor.actor_mesh import current_actor_name
+
+                logging.error(
+                    f'Endpoint call "{method_name}" failed on actor mesh "{current_actor_name()}"'
+                )
+                raise e
             return result
 
         return Future(coro=process())
@@ -289,7 +306,15 @@ async def process() -> "ValueMesh[R]":
 
             results: List[R] = [None] * extent.nelements  # pyre-fixme[9]
             for _ in range(extent.nelements):
-                rank, value = await r._recv()
+                try:
+                    rank, value = await r._recv()
+                except Exception as e:
+                    from monarch._src.actor.actor_mesh import current_actor_name
+
+                    logging.error(
+                        f'Endpoint call "{method_name}" failed on actor mesh "{current_actor_name()}"'
+                    )
+                    raise e
                 results[rank] = value
             call_shape = Shape(
                 extent.labels,