Fix torchelastic detection with non-distributed installations (#13142)

* Fix torchelastic detection under Mac

* CHANGELOG

CHANGELOG.md

@@ -223,6 +223,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Avoid redundant callback restore warning while tuning ([#13026](https://github.com/PyTorchLightning/pytorch-lightning/pull/13026))
 
 
+- Fixed torchelastic detection with non-distributed installations ([#13142](https://github.com/PyTorchLightning/pytorch-lightning/pull/13142))
+
+
 - Fixed an issue wrt unnecessary usage of habana mixed precision package for fp32 types ([#13028](https://github.com/PyTorchLightning/pytorch-lightning/pull/13028))
 
 

pytorch_lightning/plugins/environments/torchelastic_environment.py

@@ -62,7 +62,8 @@ class TorchElasticEnvironment(ClusterEnvironment):
     def detect() -> bool:
         """Returns ``True`` if the current process was launched using the torchelastic command."""
         if _TORCH_GREATER_EQUAL_1_9_1:
-            return torch.distributed.is_torchelastic_launched()
+            # if not available (for example on MacOS), `is_torchelastic_launched` is not defined
+            return torch.distributed.is_available() and torch.distributed.is_torchelastic_launched()
         required_env_vars = {"RANK", "GROUP_RANK", "LOCAL_RANK", "LOCAL_WORLD_SIZE"}
         return required_env_vars.issubset(os.environ.keys())