Fix DataLoader2 seed = 0 bug and clean up unused codes (#1098)

Summary: Pull Request resolved: #1098

Test Plan: Imported from OSS

Reviewed By: ejguan

Differential Revision: D44184683

Pulled By: NivekT

fbshipit-source-id: d8f5391f5aeb68ebb066133d0f04541ee5bfe89c

torchdata/dataloader2/dataloader2.py

@@ -7,7 +7,6 @@
 
 import warnings
 
-from dataclasses import dataclass
 from typing import Any, Dict, Generic, Iterable, Iterator, Optional, TypeVar, Union
 
 from torchdata.dataloader2.adapter import Adapter
@@ -22,14 +21,6 @@
 READING_SERVICE_STATE_KEY_NAME = "reading_service_state"
 
 
-@dataclass
-class ConcurrencySpec:
- num_workers: int
- timeout: Optional[int] = None
- prefetch_factor: int = 2
- persistent_workers: bool = False
-
-
 class DataLoader2Iterator(Iterator[T_co]):
  r"""
  An iterator wrapper returned by ``DataLoader2``'s ``__iter__` method. It delegates method/attribute calls
@@ -197,7 +188,7 @@ def __iter__(self) -> DataLoader2Iterator[T_co]:
  raise RuntimeError("Cannot iterate over the DataLoader as it has already been shut down")
 
  if self._reset_iter:
- if self._seed:
+ if self._seed is not None:
  if self._reset_seed:
  self._seed_generator.seed(self._seed)
  self._reset_seed = False

torchdata/dataloader2/reading_service.py

@@ -227,8 +227,6 @@ def initialize(self, datapipe: DataPipe) -> DataPipe:
  self._end_datapipe = datapipe
  return datapipe
 
- graph = traverse_dps(datapipe)
-
  ctx = mp.get_context(self.multiprocessing_context)
 
  # Launch dispatching process for the lowest common ancestor of non-replicable DataPipes
@@ -357,7 +355,7 @@ def finalize(self) -> None:
  req_queue.close()
 
  # Clean up dispatching process
- if self._dispatch_process:
+ if self._dispatch_process is not None:
  try:
  self._dispatch_process[0].join(default_dl2_worker_join_timeout_in_s)
  except TimeoutError: