Creating an error when a SLURM variable isn't found, usually because …

[Tristanjmeyers]

Raising an environment error when the variable SLURM_LAUNCH_NODE_IPADDR is not found. This is usually because the user isn't running SLURM with srun, and instead using something like sbatch. So, instead of raising a TypeError, this will now raise an EnvironmentError and print some help.

PhysicsNeMo Pull Request

Description

• Line 359 of physicsnemo.distributed.manager now has a try: ... except TypeError: statement.
• Addresses issue 🐛[BUG]: physicsnemo.distributed.DistributedManager expects srun rather than sbatch #819

Checklist

• [ x] I am familiar with the Contributing Guidelines.
• [ x] New or existing tests cover these changes.
• [ x] The documentation is up to date with these changes.
• [ x] The CHANGELOG.md is up to date with these changes.
• [x ] An issue is linked to this pull request.

Dependencies

N/A

[coreyjadams]

Thanks @Tristanjmeyers for sending this in! Looks reasonable.

[coreyjadams]

/blossom-ci

[coreyjadams]

FYI this is failing CI due to formatting issues. It's nothing wrong really, black is just picky. I can't edit your branch, can you please apply this diff to the manager.py file?

❯ git diff
diff --git a/physicsnemo/distributed/manager.py b/physicsnemo/distributed/manager.py
index f9cc5be..2eb0c67 100644
--- a/physicsnemo/distributed/manager.py
+++ b/physicsnemo/distributed/manager.py
@@ -359,8 +359,9 @@ class DistributedManager(object):
         try:
             addr = os.environ.get("SLURM_LAUNCH_NODE_IPADDR")
         except TypeError:
-            raise EnvironmentError('SLURM variable "SLURM_LAUNCH_NODE_IPADDR" was not detected in the environment. Maybe you need to run with "srun"?')
-
+            raise EnvironmentError(
+                'SLURM variable "SLURM_LAUNCH_NODE_IPADDR" was not detected in the environment. Maybe you need to run with "srun"?'
+            )

         DistributedManager.setup(
             rank=rank,

(it's just moving the string to a new line...)

[Tristanjmeyers]

I've done that, sorry for not linting!

[coreyjadams]

/blossom-ci

[coreyjadams]

Sorry @Tristanjmeyers - I looked at this more closely and we need to adjust this. os.environ.get will return None for missing variables, and raise no error. I think you can either:

• Remove the try/except type error, and change to if addr is None: raise EnvErr...; or
• change from os.environ.get('SLURM_LAUNCH_NODE_IPADDR') to os.environ['SLURM_LAUNCH_NODE_IPADDR'] and TypeError to KeyError.