Seed all workers when using DDP (#7942)

* Seed all workers when using DDP

* Fix to dataloader seeding

* Make argument name explicit

Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>

* Use f-strings when logging

* Removed a redundant log message

Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>

CHANGELOG.md

@@ -214,6 +214,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Fixed
 
+- Fixed setting `worker_init_fn` to seed dataloaders correctly when using DDP ([#7942](https://github.com/PyTorchLightning/pytorch-lightning/pull/7942))
+
+
 - Fixed `DataModule.prepare_data` could only be called on the global rank 0 process ([#7945](https://github.com/PyTorchLightning/pytorch-lightning/pull/7945))
 
 

pytorch_lightning/utilities/seed.py

@@ -84,8 +84,9 @@ def reset_seed() -> None:
     If :func:`pytorch_lightning.utilities.seed.seed_everything` is unused, this function will do nothing.
     """
     seed = os.environ.get("PL_GLOBAL_SEED", None)
+    workers = os.environ.get("PL_SEED_WORKERS", False)
     if seed is not None:
-        seed_everything(int(seed))
+        seed_everything(int(seed), workers=bool(workers))
 
 
 def pl_worker_init_function(worker_id: int, rank: Optional = None) -> None:  # pragma: no cover
@@ -100,6 +101,9 @@ def pl_worker_init_function(worker_id: int, rank: Optional = None) -> None:  # p
     process_seed = torch.initial_seed()
     # back out the base seed so we can use all the bits
     base_seed = process_seed - worker_id
+    log.debug(
+        f'Initializing random number generators of process {global_rank} worker {worker_id} with base seed {base_seed}'
+    )
     ss = np.random.SeedSequence([base_seed, worker_id, global_rank])
     # use 128 bits (4 x 32-bit words)
     np.random.seed(ss.generate_state(4))