update to torch 1.11, python 3.10 (#423)

Summary:
Now that torch 1.11 is released we should update to it. Also adds python 3.10 support to the unit tests now that torch supports it.

Pull Request resolved: #423

Test Plan:
CI
```
torchx/runtime/container/build.sh
```

Reviewed By: PaliC

Differential Revision: D34903945

Pulled By: d4l3k

fbshipit-source-id: ba6be59f2aabac0accd94a9406cae31a43951f1c

.github/workflows/python-unittests.yaml

@@ -10,7 +10,7 @@ jobs:
   unittest:
     strategy:
       matrix:
-        python-version: [3.7, 3.8, 3.9]
+        python-version: [3.7, 3.8, 3.9, '3.10']
         platform: [ubuntu-18.04]
         include:
           - python-version: 3.9

dev-requirements.txt

@@ -12,12 +12,14 @@ kfp==1.8.9
 moto==3.0.2
 pyre-extensions==0.0.21
 pytest
-pytorch-lightning==1.5.6
-ray[default]==1.9.2
-torch-model-archiver==0.4.2
-torch==1.10.0
-torchserve==0.4.2
-torchtext==0.11.0
-torchvision==0.11.1
+pytorch-lightning==1.5.10
+torch-model-archiver>=0.4.2
+torch>=1.10.0
+torchserve>=0.4.2
+torchtext>=0.11.0
+torchvision>=0.11.1
 ts==0.5.1
 usort==0.6.4
+
+# Ray doesn't support Python 3.10
+ray[default]==1.11.0; python_version < '3.10'

torchx/runtime/container/Dockerfile

@@ -1,4 +1,4 @@
-FROM pytorch/pytorch:1.10.0-cuda11.3-cudnn8-runtime
+FROM pytorch/pytorch:1.11.0-cuda11.3-cudnn8-runtime
 
 WORKDIR /app
 

torchx/schedulers/ray/ray_driver.py

@@ -137,6 +137,7 @@ def create_command_actors(
 
 def main() -> None:  # pragma: no cover
     actors: List[RayActor] = load_actor_json("actors.json")
+    # pyre-fixme[16]: Module `worker` has no attribute `init`.
     ray.init(address="auto", namespace="torchx-ray")
     pgs: List[PlacementGroup] = create_placement_groups(actors)
     command_actors: List[CommandActor] = create_command_actors(actors, pgs)
@@ -148,6 +149,7 @@ def main() -> None:  # pragma: no cover
 
     # Await return result of remote ray function
     while len(active_workers) > 0:
+        # pyre-fixme[16]: Module `worker` has no attribute `wait`.
         completed_workers, active_workers = ray.wait(active_workers)
         # If a failure occurs the ObjectRef will be marked as completed.
         # Calling ray.get will expose the failure as a RayActorError.

torchx/schedulers/test/ray_scheduler_test.py

@@ -258,12 +258,14 @@ class RayClusterSetup:
         def __new__(cls):  # pyre-ignore[3]
             if cls._instance is None:
                 cls._instance = super(RayClusterSetup, cls).__new__(cls)
+                # pyre-fixme[16]: Module `worker` has no attribute `shutdown`.
                 ray.shutdown()
                 start_status: int = os.system("ray start --head")
                 if start_status != 0:
                     raise AssertionError(
                         "ray start --head command has failed. Cannot proceed with running tests"
                     )
+                # pyre-fixme[16]: Module `worker` has no attribute `init`.
                 ray.init(address="auto", ignore_reinit_error=True)
                 cls.reference_count: int = 2
             return cls._instance
@@ -274,6 +276,7 @@ def decrement_reference(cls) -> None:
                 cls.teardown_ray_cluster()
 
         def teardown_ray_cluster(cls) -> None:
+            # pyre-fixme[16]: Module `worker` has no attribute `shutdown`.
             ray.shutdown()
 
     class RayDriverTest(TestCase):
@@ -306,6 +309,7 @@ class RayIntegrationTest(TestCase):
         def test_ray_cluster(self) -> None:
             ray_cluster_setup = RayClusterSetup()
             ray_scheduler = self.setup_ray_cluster()
+            # pyre-fixme[16]: Module `worker` has no attribute `is_initialized`.
             assert ray.is_initialized() is True
 
             job_id = self.schedule_ray_job(ray_scheduler)

torchx/specs/api.py

@@ -494,7 +494,7 @@ def get_type_name(tp: Type[CfgVal]) -> str:
 
     Note: we use this method to print out generic typing like List[str].
     """
-    if hasattr(tp, "__name__"):
+    if tp.__module__ != "typing" and hasattr(tp, "__name__"):
         return tp.__name__
     else:
         return str(tp)