"AttributeError: 'DataRouter' object has no attribute 'pool' " while running the HTTP server on a machine without multiprocessing support

Rasa NLU version: 0.13.8

Operating system (windows, osx, …): Ubuntu 14.04

Content of model configuration file:

language: "en"

pipeline:
- name: "tokenizer_whitespace"
- name: "ner_crf"
- name: "ner_synonyms"
- name: "ner_duckling"
  dimensions: ["time"]
- name: "intent_featurizer_count_vectors"
- name: "intent_classifier_tensorflow_embedding"
  intent_tokenization_flag: true
  intent_split_symbol: "+"

Issue: On trying to start the rasa_nlu HTTP server, I am getting the following error.

(bot_env) [ps597689]$ python -m rasa_nlu.server --path nlu_models/
Traceback (most recent call last):
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/home/<username>/bot.<domain>/bot_env/lib/python3.6/site-packages/rasa_nlu/server.py", line 438, in <module>
    wait_time_between_pulls=cmdline_args.wait_time_between_pulls
  File "/home/<username>/bot.<domain>/bot_env/lib/python3.6/site-packages/rasa_nlu/data_router.py", line 119, in __init__
    self.pool = ProcessPool(self._training_processes)
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/concurrent/futures/process.py", line 390, in __init__
    EXTRA_QUEUED_CALLS)
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/multiprocessing/context.py", line 102, in Queue
    return Queue(maxsize, ctx=self.get_context())
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/multiprocessing/queues.py", line 42, in __init__
    self._rlock = ctx.Lock()
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/multiprocessing/context.py", line 67, in Lock
    return Lock(ctx=self.get_context())
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/multiprocessing/synchronize.py", line 163, in __init__
    SemLock.__init__(self, SEMAPHORE, 1, 1, ctx=ctx)
  File "/home/<username>/opt/python-3.6.2/lib/python3.6/multiprocessing/synchronize.py", line 60, in __init__
    unlink_now)
OSError: [Errno 38] Function not implemented
Exception ignored in: <bound method DataRouter.__del__ of <rasa_nlu.data_router.DataRouter object at 0x7f2a37903c18>>
Traceback (most recent call last):
  File "/home/<username>/bot.<domain>/bot_env/lib/python3.6/site-packages/rasa_nlu/data_router.py", line 123, in __del__
    self.pool.shutdown()
AttributeError: 'DataRouter' object has no attribute 'pool'

This error only occurs when I try to deploy the code on DreamHost VPS. The OSError: [Errno 38] Function not implemented is being thrown because DreamHost VPS doesn’t allow multiprocessing apparently (I tried running a simple multiprocessing code and it throws the same error).

However imo, ideally rasa_nlu should catch this and either:-

• Revert to doing the same job without using multiprocessing OR
• Give the user a more informative error message

Although I am not very experienced at Multiprocessing, Thread synchronization etc. I can try and write a PR to fix it if I am given some outline on what to do.

[ricwo]

Hi @parthsharma1996 , it seems that sklearn uses multiprocessing and there doesn’t seem to be a way to switch that off. Feel free to have another look into sklearn, but if it doesn’t work I’m afraid we can’t switch off multiprocessing in Rasa NLU and sklearn-crf. Sorry!

[ricwo]

Sure, so add a keyword argument to DataRouter called single_process, which is None by default. If it is set to True, do not initialise self.pool. In start_train_process(). Add a check for this flag to if six.PY2 and self._tf_in_pipeline(train_config), so that should become something like

if self.single_process or six.PY2 and self._tf_in_pipeline(train_config):

and you would have to update the logger.warning method that follows it. Make the single-process kwarg in DataRouter flag accessible in rasa_nlu.server by adding --single-process to the argument parser. Does that make sense?