torchmonarch-nightly 2025.6.29__cp313-cp313-manylinux2014_x86_64.whl → 2025.7.1__cp313-cp313-manylinux2014_x86_64.whl

monarch/actor_mesh.py

@@ -521,8 +521,21 @@ class _Actor:
                 return None
 
             if self.instance is None:
+                # This could happen because of the following reasons. Both
+                # indicates a possible bug in the framework:
+                # 1. the execution of the previous message for "__init__" failed,
+                #    but that error is not surfaced to the caller.
+                #      - TODO(T229200522): there is a known bug. fix it.
+                # 2. this message is delivered to this actor before the previous
+                #    message of "__init__" is delivered. Out-of-order delivery
+                #    should never happen. It indicates either a bug in the
+                #    message delivery mechanism, or the framework accidentally
+                #    mixed the usage of cast and direct send.
                 raise AssertionError(
-                    "__init__ failed earlier and no Actor object is available"
+                    f"""
+                    actor object is missing when executing method {message.method}
+                    on actor {mailbox.actor_id}
+                    """
                 )
             the_method = getattr(self.instance, message.method)._method
 

tests/error_test_binary.py

@@ -9,6 +9,7 @@ import ctypes
 import sys
 
 import click
+from monarch._rust_bindings.monarch_extension.blocking import blocking_function
 
 from monarch._rust_bindings.monarch_extension.panic import panicking_function
 
@@ -36,6 +37,11 @@ class ErrorActor(Actor):
         """Endpoint that calls a Rust function that panics."""
         panicking_function()
 
+    @endpoint
+    async def cause_stuck(self) -> None:
+        """Endpoint that causes the process to hang indefinitely."""
+        blocking_function()
+
     @endpoint
     async def await_then_error(self) -> None:
         await asyncio.sleep(0.1)
@@ -171,6 +177,28 @@ async def _error_unmonitored():
     await asyncio.sleep(300)
 
 
+"""
+TODO: This test should be enabled when stop() is fully implemented.
+async def _error_unmonitored():
+    print("I actually ran")
+    sys.stdout.flush()
+
+    proc = await proc_mesh(gpus=1)
+    actor = await proc.spawn("error_actor", ErrorActor)
+
+    # fire and forget
+    send(actor.cause_stuck, (), {}, None, "all")
+    proc_mesh.stop()
+
+    # Wait. Eventually a supervision event will get propagated and the process
+    # will exit.
+    #
+    # If an event is not delivered, the test will time out before this sleep
+    # finishes.
+    await asyncio.sleep(300)
+"""
+
+
 @main.command("error-unmonitored")
 def error_unmonitored():
     asyncio.run(_error_unmonitored())

tests/test_python_actors.py

@@ -434,6 +434,7 @@ class DebugeeActor(Actor):
         return _debugee_actor_internal(rank)
 
 
+@pytest.mark.oss_skip  # pyre-ignore[56] TODO T229449782
 async def test_debug() -> None:
     input_mock = AsyncMock()
     input_mock.side_effect = [

{torchmonarch_nightly-2025.6.29.dist-info → torchmonarch_nightly-2025.7.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: torchmonarch-nightly
-Version: 2025.6.29
+Version: 2025.7.1
 Summary: Monarch: Single controller library
 Author: Meta
 Author-email: oncall+monarch@xmail.facebook.com

{torchmonarch_nightly-2025.6.29.dist-info → torchmonarch_nightly-2025.7.1.dist-info}/RECORD

@@ -1,7 +1,7 @@
 monarch/__init__.py,sha256=iUvWHc0-7Q2tovRoRxOIiA3TsefMXCbWl-jEfQ2djew,6897
-monarch/_rust_bindings.so,sha256=9cUwsCXdRE_hw101yv6QOQ3n2FwUKURjVkhqrVYqBTI,43742616
+monarch/_rust_bindings.so,sha256=U-ZaYneJGkTj3JtIL5JQAZ_PDlU2urIw6bLbtPaxZR0,44319160
 monarch/_testing.py,sha256=jOIOG6jcZBzvEvG_DwSnwCkaMVXvSun6sJAG6nXemww,7859
-monarch/actor_mesh.py,sha256=QqKHVTJk9H_I-v7GoxgOdOL8-ymnRpGvNFdda0-cNrE,24534
+monarch/actor_mesh.py,sha256=-IEXcIGdpVWzEV3CNeWhNJb4siMRBtnwDCtQydz7Db0,25369
 monarch/allocator.py,sha256=l0_mN43AH3K2aCchb5fk8ml95rvdgR31nRC_PqRmZWg,7865
 monarch/bootstrap_main.py,sha256=RCUQhJk07yMFiKp6HzQuqZFUpkgsT9kVEyimiwjn6_E,1827
 monarch/cached_remote_function.py,sha256=kYdB6r4OHx_T_uX4q3tCNcp1t2DJwF8tPTIahUiT2pU,8785
@@ -12,7 +12,7 @@ monarch/future.py,sha256=g1VYJl8ReBBS6VbikwWilnFqEr5qJDiSKid92AnWFV4,2058
 monarch/gradient_generator.py,sha256=Rl3dmXGceTdCc1mYBg2JciR88ywGPnW7TVkL86KwqEA,6366
 monarch/memory.py,sha256=ol86dBhFAJqg78iF25-BuK0wuwj1onR8FIioZ_B0gjw,1377
 monarch/mesh_controller.py,sha256=JIg2-MIGhqeFCXn793j8ivEu5-8ePksu--H_Pv0-Ltk,10377
-monarch/monarch_controller,sha256=6hzB3c6q4SEgxWIK_fAqANlfnCy4rcX0AwDkLcz9PYY,21723584
+monarch/monarch_controller,sha256=5E5TBqZB-_dRrKZqrKSvATydY65ujHHWE6dR5lfXlAY,22426416
 monarch/notebook.py,sha256=zu9MKDFKf1-rCM2TqFSRJjMBeiWuKcJSyUFLvoZRQzs,25949
 monarch/opaque_module.py,sha256=oajOu_WD1hD4hxE8HDdO-tvWY7KDHWd7VaAhJEa5L2I,10446
 monarch/opaque_object.py,sha256=IVpll4pyuKZMo_EnPh4s0qnx8RlAcJrJ1yoLX6E75wQ,2782
@@ -133,7 +133,7 @@ monarch_supervisor/python_executable.py,sha256=WfCiK3wdAvm9Jxx5jgjGF991NgGc9-oHU
 tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 tests/dispatch_bench.py,sha256=sU_m-8KAjQgYTsxI5khV664NdgLLutidni69Rtowk98,3933
 tests/dispatch_bench_helper.py,sha256=1ORgAMrRgjAjmmWeCHLLQd_bda9mJk0rS2ucEbRu28s,633
-tests/error_test_binary.py,sha256=BRj13wAROsUWx4jcxc07HYN2n-xyBNhnnRAhjqah-A0,5582
+tests/error_test_binary.py,sha256=vfbBI6fkWTPo-Gksi2YGPw8wSxfdOHlSALpHQ6e2MIE,6386
 tests/sleep_binary.py,sha256=XfLYaAfwm9xgzM-svs8fhAeFhwYIg6SyVEnx4e6wbUw,1009
 tests/test_actor_error.py,sha256=hMpaJDkvowBM_fKuitQdKmq3iyvySaom8KNLmxEsZ-o,9534
 tests/test_alloc.py,sha256=D6DdQbtOZEvvnnc7LV-WyWFMk0Xb77eblH6Oz90zJTA,745
@@ -146,7 +146,7 @@ tests/test_future.py,sha256=cXzaNi2YDwVyjR541ScXmgktX1YFsKzbl8wep0DMVbk,3032
 tests/test_grad_generator.py,sha256=p4Pm4kMEeGldt2jUVAkGKCB0mLccKI28pltH6OTGbQA,3412
 tests/test_mock_cuda.py,sha256=5hisElxeLJ5MHw3KM9gwxBiXiMaG-Rm382u3AsQcDOI,3068
 tests/test_pdb_actor.py,sha256=5KJhuhcZDPWMdjC6eAtDdwnz1W7jNFXvIrMSFaCWaPw,3858
-tests/test_python_actors.py,sha256=0kF3LQpvPnAqT6xbNaBQxaG3gsMyBzzM4Ou7om9ZhoE,20069
+tests/test_python_actors.py,sha256=EXGiWBFscJtFbOk7hESJqSCSuW1SC6UlXIIzEOfwAYI,20126
 tests/test_remote_functions.py,sha256=5nxYB8dfA9NT9f9Od9O3htgQtPbiRNiXZ1Kgtn75sOQ,50056
 tests/test_rust_backend.py,sha256=94S3R995ZkyIhEiBsM5flcjf5X7bscEAHBtInbTRFe8,7776
 tests/test_signal_safe_block_on.py,sha256=bmal0XgzJowZXJV6T1Blow5a-vZluYWusCThLMGxyTE,3336
@@ -157,9 +157,9 @@ tests/simulator/test_profiling.py,sha256=TGYCfzTLdkpIwnOuO6KApprmrgPIRQe60KRX3wk
 tests/simulator/test_simulator.py,sha256=LO8lA0ssY-OGEBL5ipEu74f97Y765TEwfUOv-DtIptM,14568
 tests/simulator/test_task.py,sha256=ipqBDuDAysuo1xOB9S5psaFvwe6VATD43IovCTSs0t4,2327
 tests/simulator/test_worker.py,sha256=QrWWIJ3HDgDLkBPRc2mwYPlOQoXQcj1qRfc0WUfKkFY,3507
-torchmonarch_nightly-2025.6.29.dist-info/licenses/LICENSE,sha256=e0Eotbf_rHOYPuEUlppIbvwy4SN98CZnl_hqwvbDA4Q,1530
-torchmonarch_nightly-2025.6.29.dist-info/METADATA,sha256=pfxXahOAFVZWSQpbC-6VgvxpOnSxa8Acic1RsONWV6U,2780
-torchmonarch_nightly-2025.6.29.dist-info/WHEEL,sha256=OlISbtpDcfagPrLwG7WtpcZbPTUnoKPnwphA_26fNqE,104
-torchmonarch_nightly-2025.6.29.dist-info/entry_points.txt,sha256=sqfQ16oZqjEvttUI-uj9BBXIIE6jt05bYFSmy-2hyXI,106
-torchmonarch_nightly-2025.6.29.dist-info/top_level.txt,sha256=E-ZssZzyM17glpVrh-S9--qJ-w9p2EjuYOuNw9tQ4Eg,33
-torchmonarch_nightly-2025.6.29.dist-info/RECORD,,
+torchmonarch_nightly-2025.7.1.dist-info/licenses/LICENSE,sha256=e0Eotbf_rHOYPuEUlppIbvwy4SN98CZnl_hqwvbDA4Q,1530
+torchmonarch_nightly-2025.7.1.dist-info/METADATA,sha256=69r1r82Eei3iFcfZuQho0RtfcgLoS31nXUf3vg_R1XQ,2779
+torchmonarch_nightly-2025.7.1.dist-info/WHEEL,sha256=OlISbtpDcfagPrLwG7WtpcZbPTUnoKPnwphA_26fNqE,104
+torchmonarch_nightly-2025.7.1.dist-info/entry_points.txt,sha256=sqfQ16oZqjEvttUI-uj9BBXIIE6jt05bYFSmy-2hyXI,106
+torchmonarch_nightly-2025.7.1.dist-info/top_level.txt,sha256=E-ZssZzyM17glpVrh-S9--qJ-w9p2EjuYOuNw9tQ4Eg,33
+torchmonarch_nightly-2025.7.1.dist-info/RECORD,,