torchmonarch-nightly 2025.7.30__cp313-cp313-manylinux2014_x86_64.whl → 2025.8.1__cp313-cp313-manylinux2014_x86_64.whl

monarch/_src/actor/proc_mesh.py

@@ -65,15 +65,19 @@ from monarch._src.actor.shape import MeshTrait
 
 HAS_TENSOR_ENGINE = False
 try:
+    # Torch is needed for tensor engine
+    import torch  # @manual
+
+    # Confirm that rust bindings were built with tensor engine enabled
     from monarch._rust_bindings.rdma import (  # type: ignore[import]
         _RdmaBuffer,
         _RdmaManager,
     )
 
     # type: ignore[16]
-    HAS_TENSOR_ENGINE = _RdmaBuffer.rdma_supported()
+    HAS_TENSOR_ENGINE = torch.cuda.is_available()
 except ImportError:
-    logging.warning("RDMA is not available on this platform")
+    logging.warning("Tensor engine is not available on this platform")
 
 
 if TYPE_CHECKING:
@@ -133,9 +137,10 @@ class ProcMesh(MeshTrait):
         setup: Callable[[], None] | None = None,
     ) -> "ProcMesh":
         _rdma_manager = (
-            # pyre-ignore
+            # type: ignore[16]
             await _RdmaManager.create_rdma_manager_nonblocking(self._proc_mesh)
-            if HAS_TENSOR_ENGINE
+            # type: ignore[16]
+            if HAS_TENSOR_ENGINE and _RdmaBuffer.rdma_supported()
             else None
         )
 

monarch/tools/utils.py

@@ -53,13 +53,10 @@ class conda:
         Returns the currently active conda environment name.
         `None` if run outside of a conda environment.
         """
-        env_name = os.getenv("CONDA_DEFAULT_ENV")
-
-        if not env_name:
-            # conda envs activated with metaconda doesn't set CODNA_DEFAULT_ENV so
-            # fallback to CONDA_PREFIX which points to the path of the currently active conda environment
-            # e.g./home/$USER/.conda/envs/{env_name}
-            if env_dir := conda.active_env_dir():
-                env_name = os.path.basename(env_dir)
+        # we do not check CODNA_DEFAULT_ENV as CONDA_PREFIX is a preferred way
+        # to get the active conda environment, e.g./home/$USER/.conda/envs/{env_name}
+        env_name: Optional[str] = None
+        if env_dir := conda.active_env_dir():
+            env_name = os.path.basename(env_dir)
 
         return env_name

tests/test_allocator.py

@@ -104,7 +104,9 @@ class TestActor(Actor):
 
 @contextlib.contextmanager
 def remote_process_allocator(
-    addr: Optional[str] = None, timeout: Optional[int] = None
+    addr: Optional[str] = None,
+    timeout: Optional[int] = None,
+    envs: Optional[dict[str, str]] = None,
 ) -> Generator[str, None, None]:
     """Start a remote process allocator on addr. If timeout is not None, have it
     timeout after that many seconds if no messages come in"""
@@ -120,16 +122,19 @@ def remote_process_allocator(
         if timeout is not None:
             args.append(f"--timeout-sec={timeout}")
 
+        env = {
+            # prefix PATH with this test module's directory to
+            # give 'process_allocator' and 'monarch_bootstrap' binary resources
+            # in this test module's directory precedence over the installed ones
+            # useful in BUCK where these binaries are added as 'resources' of this test target
+            "PATH": f"{package_path}:{os.getenv('PATH', '')}",
+            "RUST_LOG": "debug",
+        }
+        if envs:
+            env.update(envs)
         process_allocator = subprocess.Popen(
             args=args,
-            env={
-                # prefix PATH with this test module's directory to
-                # give 'process_allocator' and 'monarch_bootstrap' binary resources
-                # in this test module's directory precedence over the installed ones
-                # useful in BUCK where these binaries are added as 'resources' of this test target
-                "PATH": f"{package_path}:{os.getenv('PATH', '')}",
-                "RUST_LOG": "debug",
-            },
+            env=env,
         )
         try:
             yield addr
@@ -233,6 +238,26 @@ class TestRemoteAllocator(unittest.IsolatedAsyncioTestCase):
         computed_world_sizes = {p.rank: v for p, v in list(computed.flatten("rank"))}
         self.assertDictEqual(expected_world_sizes, computed_world_sizes)
 
+    async def test_allocate_failure_message(self) -> None:
+        spec = AllocSpec(AllocConstraints(), host=2, gpu=4)
+
+        with self.assertRaisesRegex(
+            Exception,
+            r"exited with code 1: Traceback \(most recent call last\).*",
+        ):
+            with remote_process_allocator(
+                envs={"MONARCH_ERROR_DURING_BOOTSTRAP_FOR_TESTING": "1"}
+            ) as host1, remote_process_allocator(
+                envs={"MONARCH_ERROR_DURING_BOOTSTRAP_FOR_TESTING": "1"}
+            ) as host2:
+                allocator = RemoteAllocator(
+                    world_id="test_remote_allocator",
+                    initializer=StaticRemoteAllocInitializer(host1, host2),
+                    heartbeat_interval=_100_MILLISECONDS,
+                )
+                alloc = await allocator.allocate(spec)
+                await ProcMesh.from_alloc(alloc)
+
     async def test_call_allocate_twice(self) -> None:
         class DeletingAllocInitializer(StaticRemoteAllocInitializer):
             """test initializer that removes the last address from the list each time initialize_alloc() is called

{torchmonarch_nightly-2025.7.30.dist-info → torchmonarch_nightly-2025.8.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: torchmonarch-nightly
-Version: 2025.7.30
+Version: 2025.8.1
 Summary: Monarch: Single controller library
 Author: Meta
 Author-email: oncall+monarch@xmail.facebook.com

{torchmonarch_nightly-2025.7.30.dist-info → torchmonarch_nightly-2025.8.1.dist-info}/RECORD

@@ -1,5 +1,5 @@
 monarch/__init__.py,sha256=mgKiyD1kxky-1pvhMlNfF4VmxWnhi-FSYZNFzkW1BEM,7052
-monarch/_rust_bindings.so,sha256=r7qvxPiMHRlXvwxCM--hf3BgB0JYISYCRDpi64v5vO0,50331040
+monarch/_rust_bindings.so,sha256=prrqR30Gcgdga3OZoUTOxaZyt7r7ozzwXoTSeizn3JM,50340744
 monarch/_testing.py,sha256=_3MYNMq-_0T1qXCj2vxrW13GlWGdUuVFMskQF2Gsw_o,7877
 monarch/actor_mesh.py,sha256=VtPU9syi_vUdwDSJJ639Z4Y_EcWZUScyoj0lQ88RQPs,421
 monarch/bootstrap_main.py,sha256=39OZpNMrfvvNJf-iwuNzgslzYA_ItaRPHfXGn_V74N0,524
@@ -8,7 +8,7 @@ monarch/fetch.py,sha256=JMxC8HdWMvpik0T4E6e-gfxvmNnOkA0ul4eo4R3Jg_o,1712
 monarch/gradient_generator.py,sha256=Rl3dmXGceTdCc1mYBg2JciR88ywGPnW7TVkL86KwqEA,6366
 monarch/memory.py,sha256=ol86dBhFAJqg78iF25-BuK0wuwj1onR8FIioZ_B0gjw,1377
 monarch/mesh_controller.py,sha256=R9ZnVV89wYva0QTAwOgHi_PkjYPEj_7_yF9810NHPak,14675
-monarch/monarch_controller,sha256=Fzbm8eLjh8L_s83_DJN9vM0iaK0C1kaHF7jWBRANgcs,24036856
+monarch/monarch_controller,sha256=c5-zlhoJK5vpEVxY6DMwRXEmHSmBy0Qgv55Tbu-8vWI,24034032
 monarch/notebook.py,sha256=zu9MKDFKf1-rCM2TqFSRJjMBeiWuKcJSyUFLvoZRQzs,25949
 monarch/opaque_module.py,sha256=jCcg0DjbcEVXA9WNG0NhUzGteLHOJLTZEBvrIYJIAns,10436
 monarch/opaque_object.py,sha256=x1LoX6RIMGh4ux52xIfhPgoh6PhZHdkf9bMccHW3DW0,2808
@@ -35,7 +35,7 @@ monarch/_src/actor/event_loop.py,sha256=2i4fKIkemBzua_t47BqVa2roZ6fWB6sbmMFPNx2z
 monarch/_src/actor/future.py,sha256=7QDiPu6-CnTw7cN_GWomQa9qGxDo5yXqCSqgyCJ7roU,5195
 monarch/_src/actor/pdb_wrapper.py,sha256=3pjk-eTSc7_rctDiZl-vilqTXQoaERGqyi1LueyoQGg,4342
 monarch/_src/actor/pickle.py,sha256=jD_3E07OJmMIlcMOOrNFnIuRKZU2F_Q_BP-njDFXUNM,2044
-monarch/_src/actor/proc_mesh.py,sha256=FNBRl-F1HdvdLWwHt0YrWYQhzXK4y1y-nSKnfwkznwA,15531
+monarch/_src/actor/proc_mesh.py,sha256=5AKgjsskVe7blJjwROom31p1EXzx9ANg3ADzvcLcU5s,15746
 monarch/_src/actor/shape.py,sha256=E9kxf1RNym1LNJNXF18gNDmnAHR7SDcl3W4nXR65BPY,8293
 monarch/_src/actor/sync_state.py,sha256=GB6bTAGztkcN8fZ9K7zXklOzjYzv6cvkJeBje20xFkE,471
 monarch/_src/actor/tensor_engine_shim.py,sha256=hupavQ2rjPwECaTlDAhY-eeiEY18Wyyx59MZHcSEcYM,1622
@@ -50,7 +50,7 @@ monarch/builtins/__init__.py,sha256=QcfnHZGbc2qktBg7DyZt2ruE6VahnIt4S8lEZLHdJqU,
 monarch/builtins/log.py,sha256=H1QkuVzwxyi36Zyv-XR0VN0QsNimBWwxE1__fjs0_2o,554
 monarch/builtins/random.py,sha256=wPbvscg7u53EXpMFo885fO2XOlsyjrNAJ4rBxLzfxdg,1839
 monarch/common/_C.pyi,sha256=kHY2G3ksMAjQJ6IcPb4F1bBh5knzw5RVVNhhBlEmwFU,314
-monarch/common/_C.so,sha256=i0Mqno03fQrMtkAnBwsenC4EzrU-NQy7fUk3-cGfOcw,719904
+monarch/common/_C.so,sha256=WkoR6BuS6k6tfia5sGSmfQVsBNmdKqRXXQ8o9pqfpy0,719808
 monarch/common/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 monarch/common/_coalescing.py,sha256=HXf5cXAPSU_tpw9jFkzs2muytG_6sTZJSqSKV0XuFZE,10925
 monarch/common/_tensor_to_table.py,sha256=yRjCNwvtl188Z1Dwkx3ZU-Bh2mwYnQ0Lnue2RAztwvc,5753
@@ -89,7 +89,7 @@ monarch/controller/rust_backend/__init__.py,sha256=J8qjUOysmcMAek2KFN13mViOXZxTY
 monarch/controller/rust_backend/controller.py,sha256=8IYnVUiqEVKO9rGL3vKqcCSAhWJG1bYYQ0MoaMqsp78,9521
 monarch/gradient/__init__.py,sha256=kqmzwt16mMpk0M3GhpgP_f7da4DGnaV9chDzbt66k4Q,308
 monarch/gradient/_gradient_generator.pyi,sha256=6cX0UxaDt9NAlwgIhTgnweqGOf6qRhHiGnUzSWNCxdU,630
-monarch/gradient/_gradient_generator.so,sha256=Jw4MAcg1zcjYfm51N6EZWxTA8bedXaTcGMj3g45o9jQ,11534720
+monarch/gradient/_gradient_generator.so,sha256=kSqzNPJx7AeQj9mtAjqdg1U7gQ7Zc2_QCkOSOYqY3Gs,11755672
 monarch/parallel/__init__.py,sha256=6920kIkhiX7AiyjYvyc1ad8ccP-bStJJ1sS5KkeN2P0,352
 monarch/parallel/pipelining/__init__.py,sha256=J8qjUOysmcMAek2KFN13mViOXZxTYc5vCrF02t3VuFU,223
 monarch/parallel/pipelining/runtime.py,sha256=KK8TG1gUYEzSsquiZoPTWGSIC74mlncD7cYknKxfb3c,32470
@@ -119,7 +119,7 @@ monarch/tools/cli.py,sha256=b3mKZnK-MwP7JwskTxHI0KcJXxSU6498jEb2ntVr_VM,5001
 monarch/tools/commands.py,sha256=3xuvHcMwl0t6cWTVUxI_r8EqrJZnay0bkKxOijhlKrw,12126
 monarch/tools/mesh_spec.py,sha256=in6txNRmA-UvveVSMHCjX6mGpofd3K8vl2Plz1eD6rg,7935
 monarch/tools/network.py,sha256=mN8Fx9mervxM3VdFHRn4ZXt4z7yWxZp52BTxx2tfpus,2455
-monarch/tools/utils.py,sha256=p9zunV_OP_eYC2GQFZ_NKFj5Xl1tf7Tin1ZxjZqp3TY,2010
+monarch/tools/utils.py,sha256=gcZyalfoBC6Y3v65h-QMngwXsn24ejXh2TH8RxlgXkA,1888
 monarch/tools/components/__init__.py,sha256=J8qjUOysmcMAek2KFN13mViOXZxTYc5vCrF02t3VuFU,223
 monarch/tools/components/hyperactor.py,sha256=jbfC5J9oRzzMQFO2eIx9acRS8RPHp3GtKyJjYblcJFM,2169
 monarch/tools/config/__init__.py,sha256=HZjmRSC_R28WMDSvNAqNRlqhH7lMHFWPIrztcFFt8Us,890
@@ -147,7 +147,7 @@ tests/sleep_binary.py,sha256=XfLYaAfwm9xgzM-svs8fhAeFhwYIg6SyVEnx4e6wbUw,1009
 tests/test_actor_error.py,sha256=U7QL1jRn-YpS-o62imt7HFLPtaSbwMBu9xpD09Mb-Bc,20875
 tests/test_actor_shape.py,sha256=ph-RC9sMNHWptZOCwQqMfG4lIUEzhp_pEnfhITeYdHM,4533
 tests/test_alloc.py,sha256=IW7yJSaKxhOYc8SJtFyREakDUwiKWq9M0CGgYyBYHoc,743
-tests/test_allocator.py,sha256=4LcUB4QRNGDp0qBWAyLM6ektmoxpO922f-NcHZziJ_w,28762
+tests/test_allocator.py,sha256=DpQhygQ4jB19g-aY-BFR61J-gYbG-hDlTNPVuQ4Fmn0,29730
 tests/test_coalescing.py,sha256=JZ4YgQNlWWs7N-Z8KCCXQPANcuyyXEKjeHIXYbPnQhk,15606
 tests/test_controller.py,sha256=CIMb-ApmBcBj1eCqccDUAbVyyJWMGooAha5gQk0AoeY,31452
 tests/test_debugger.py,sha256=9opgQXCBuZ1Z-7uOKI-FuGB0jLbLLilmWQKq0sE-dgQ,21950
@@ -170,9 +170,9 @@ 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-2025.7.30.dist-info/licenses/LICENSE,sha256=e0Eotbf_rHOYPuEUlppIbvwy4SN98CZnl_hqwvbDA4Q,1530
-torchmonarch_nightly-2025.7.30.dist-info/METADATA,sha256=TLI6C3RGzWoCq2Ual0X4CEz_oUaMFvHdAxIsAI6yxaI,3852
-torchmonarch_nightly-2025.7.30.dist-info/WHEEL,sha256=OlISbtpDcfagPrLwG7WtpcZbPTUnoKPnwphA_26fNqE,104
-torchmonarch_nightly-2025.7.30.dist-info/entry_points.txt,sha256=60QVSpYVzkzS4iDOiLp0fsLxVp47X3J2l3v7W-59LMo,117
-torchmonarch_nightly-2025.7.30.dist-info/top_level.txt,sha256=E-ZssZzyM17glpVrh-S9--qJ-w9p2EjuYOuNw9tQ4Eg,33
-torchmonarch_nightly-2025.7.30.dist-info/RECORD,,
+torchmonarch_nightly-2025.8.1.dist-info/licenses/LICENSE,sha256=e0Eotbf_rHOYPuEUlppIbvwy4SN98CZnl_hqwvbDA4Q,1530
+torchmonarch_nightly-2025.8.1.dist-info/METADATA,sha256=gnu970IbH-M9RmjioXTmMbk_MXOCcf6iTTGa226YHdI,3851
+torchmonarch_nightly-2025.8.1.dist-info/WHEEL,sha256=OlISbtpDcfagPrLwG7WtpcZbPTUnoKPnwphA_26fNqE,104
+torchmonarch_nightly-2025.8.1.dist-info/entry_points.txt,sha256=60QVSpYVzkzS4iDOiLp0fsLxVp47X3J2l3v7W-59LMo,117
+torchmonarch_nightly-2025.8.1.dist-info/top_level.txt,sha256=E-ZssZzyM17glpVrh-S9--qJ-w9p2EjuYOuNw9tQ4Eg,33
+torchmonarch_nightly-2025.8.1.dist-info/RECORD,,