update dataflow docs about windows (#1219)

docs/tutorial/efficient-dataflow.md

@@ -272,6 +272,17 @@ df = RemoteDataZMQ('ipc://@my-socket', 'tcp://0.0.0.0:8877')
 TestDataSpeed(df).start()
 ```
 
+
+## Common Issues on Windows:
+
+1. Windows does not support ZMQ. You can only use `MultiProcessRunner`,
+   `MultiThreadRunner`, and `MultiThreadMapData`. But you cannot use 
+   `MultiProcessRunnerZMQ` or `MultiProcessMapData` (which is an alias of `MultiProcessMapDataZMQ`).
+2. Windows needs to pickle your dataflow to run it in multiple processes.
+   As a result you cannot use lambda functions for mappings, like the examples above.
+   You need to write a new function in global scope that does the mapping.
+   This issue also exist on Linux if you do not use the 'fork' start method.
+
 [1]: #ref
 
 <div id=ref> </div>

tensorpack/dataflow/parallel.py

@@ -178,7 +178,8 @@ class MultiProcessRunner(ProxyDataFlow):
         Args:
             ds (DataFlow): input DataFlow.
             num_prefetch (int): size of the queue to hold prefetched datapoints.
-            num_proc (int): number of processes to use.
+                Required.
+            num_proc (int): number of processes to use. Required.
             nr_prefetch, nr_proc: deprecated argument names
         """
         if nr_prefetch is not None:
@@ -187,6 +188,8 @@ class MultiProcessRunner(ProxyDataFlow):
         if nr_proc is not None:
             log_deprecated("MultiProcessRunner(nr_proc)", "Renamed to 'num_proc'", "2020-01-01")
             num_proc = nr_proc
+        if num_prefetch is None or num_proc is None:
+            raise TypeError("Missing argument num_prefetch or num_proc in MultiProcessRunner!")
 
         # https://docs.python.org/3.6/library/multiprocessing.html?highlight=process#the-spawn-and-forkserver-start-methods
         if os.name == 'nt':
@@ -424,6 +427,9 @@ class MultiThreadRunner(DataFlow):
         if nr_thread is not None:
             log_deprecated("MultiThreadRunner(nr_thread)", "Renamed to 'num_thread'", "2020-01-01")
             num_thread = nr_thread
+        if num_prefetch is None or num_thread is None:
+            raise TypeError("Missing argument num_prefetch or num_thread in MultiThreadRunner!")
+
         assert num_thread > 0, num_thread
         assert num_prefetch > 0, num_prefetch
         self.num_thread = num_thread