warning about fork

examples/FasterRCNN/config.py

@@ -142,7 +142,7 @@ _C.TRAIN.NUM_GPUS = None         # by default, will be set from code
 _C.TRAIN.WEIGHT_DECAY = 1e-4
 _C.TRAIN.BASE_LR = 1e-2  # defined for total batch size=8. Otherwise it will be adjusted automatically
 _C.TRAIN.WARMUP = 1000   # in terms of iterations. This is not affected by #GPUs
-_C.TRAIN.WARMUP_INIT_LR = 0.  # defined for total batch size=8. Otherwise it will be adjusted automatically
+_C.TRAIN.WARMUP_INIT_LR = 1e-5  # defined for total batch size=8. Otherwise it will be adjusted automatically
 _C.TRAIN.STEPS_PER_EPOCH = 500
 _C.TRAIN.STARTING_EPOCH = 1  # the first epoch to start with, useful to continue a training
 

tensorpack/dataflow/parallel.py

@@ -223,7 +223,7 @@ lead of failure on some of the code.")
         self.procs = [MultiProcessRunner._Worker(self.ds, self.queue, idx)
                       for idx in range(self.num_proc)]
         ensure_proc_terminate(self.procs)
-        start_proc_mask_signal(self.procs)
+        self._reset_done = False
 
     def __iter__(self):
         for k in itertools.count():
@@ -233,8 +233,9 @@ lead of failure on some of the code.")
             yield dp
 
     def reset_state(self):
-        # do nothing. all ds are reset once and only once in spawned processes
-        pass
+        assert not self._reset_done, "reset_state() was called twice! This violates the API of DataFlow!"
+        self._reset_done = True
+        start_proc_mask_signal(self.procs)
 
 
 class MultiProcessRunnerZMQ(_MultiProcessZMQDataFlow):

tensorpack/utils/concurrency.py

@@ -232,8 +232,8 @@ def start_proc_mask_signal(proc):
             if isinstance(p, mp.Process):
                 if sys.version_info < (3, 4) or mp.get_start_method() == 'fork':
                     log_once("""
-Starting a process with 'fork' method is not safe and may consume unnecessary extra CPU memory.
-Use 'forkserver' or 'spawn' method (available after Py3.4) instead if you run into any issues.
+Starting a process with 'fork' method is efficient but not safe and may cause deadlock or crash.
+Use 'forkserver' or 'spawn' method instead if you run into such issues.
 See https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods on how to set them.
 """.replace("\n", ""),
 'warn')  # noqa