Parameter initialization error in fix_get_action_develop branch

[Davidwhw]

I was trying to run an async workflow in fix_get_action_develop branch and had an error with parameter initialization.

In order to study the error of parameter initialization, I printed the parameter information in EXARL/exarl/utils/candleDriver.py and EXARL/exarl/utils/globals.py files respectively.

"""
EXARL/exarl/utils/candleDriver.py
"""

def initialize_parameters(params=None):
    ...
    print('initialize_parameters :: if params is None? ', params == None)
    ExaGlobals(params, candle.keras_default_config())

"""
EXARL/exarl/utils/globals.py
"""

def lookup_params(key):
    print(f'ExaGlobals Looking for {key} : run_params is None ? {ExaGlobals.__run_params == None}',)
    if ExaGlobals.__run_params is None:
        raise ExaGlobals.GlobalsNotInitialized(key)
    if key not in ExaGlobals.__run_params:
        raise ExaGlobals.GlobalDoesNotExist(ExaGlobals.__run_params, key)
    return ExaGlobals.__run_params[key]

For the 'agent' parameter shown below, the initialization in candleDriver.py is always normal. But in globals.py , sometimes the parameter is successfully found, and sometimes it is None.

initialize_parameters :: if params is None?  False
ExaGlobals Looking for log_level : run_params is None ? False
ExaGlobals Looking for agent : run_params is None ? False
ExaGlobals Looking for env : run_params is None ? False
ExaGlobals Looking for workflow : run_params is None ? False
ExaGlobals Looking for learner_procs : run_params is None ? False
ExaGlobals Looking for process_per_env : run_params is None ? False
ExaGlobals Looking for n_episodes : run_params is None ? False
ExaGlobals Looking for n_steps : run_params is None ? False
ExaGlobals Looking for action_type : run_params is None ? False
ExaGlobals Looking for output_dir : run_params is None ? False
ExaGlobals Looking for mpi4py_rc : run_params is None ? False
Turning mpi4py.rc.threads and mpi4py.rc.recv_mprobe to false!
ExaGlobals Looking for env : run_params is None ? False
ExaGlobals Looking for agent : run_params is None ? False
ExaGlobals Looking for env : run_params is None ? True
ExaGlobals Looking for agent : run_params is None ? False
ExaGlobals Looking for env : run_params is None ? False
ExaGlobals Looking for workflow : run_params is None ? False
ExaGlobals Looking for learner_procs : run_params is None ? False
ExaGlobals Looking for process_per_env : run_params is None ? False
ExaGlobals Looking for n_episodes : run_params is None ? False
ExaGlobals Looking for n_steps : run_params is None ? False
ExaGlobals Looking for action_type : run_params is None ? False
ExaGlobals Looking for output_dir : run_params is None ? False
ExaGlobals Looking for mpi4py_rc : run_params is None ? False
Turning mpi4py.rc.threads and mpi4py.rc.recv_mprobe to false!
ExaGlobals Looking for agent : run_params is None ? True
ExaGlobals Looking for env : run_params is None ? False
ExaGlobals Looking for agent : run_params is None ? False
ExaGlobals Looking for agent : run_params is None ? True
[Errno 2] No such file or directory: '/EXARL/results/EXP000/RUN012'
Could not load candle parameters.
Traceback (most recent call last):
  File "start.py", line 21, in <module>
    exa_learner = erl.ExaLearner(comm)
  File "/EXARL/exarl/base/learner_base.py", line 27, in __init__
    agent_id = ExaGlobals.lookup_params('agent')
  File "/EXARL/exarl/utils/globals.py", line 112, in lookup_params
    raise ExaGlobals.GlobalsNotInitialized(key)
exarl.utils.globals.GlobalsNotInitialized: ExaRL globals have not been set. Trying agent

How can I fix this mistake?

[Davidwhw]

I think I've found the reason for the error. Sometimes it try to delete folder that don't exist in candlelib. I don't know what causes this, but it does come up all the time. When I remove shutil.rmtree(outdir), it works fine.

"""
/EXARL/exarl/candlelib/default_utils.py
"""
def directory_from_parameters(params, commonroot='Output'):
    ...
    if not os.path.exists(outdir):
        os.makedirs(outdir, exist_ok=True)
    else:
        shutil.rmtree(outdir)
        os.makedirs(outdir, exist_ok=True)

    return outdir

Traceback (most recent call last):
  File "start.py", line 5, in <module>
    import exarl.utils.analyze_reward as ar
  File "/EXARL/exarl/__init__.py", line 3, in <module>
    candleDriver.initialize_parameters()
  File "/EXARL/exarl/utils/candleDriver.py", line 92, in initialize_parameters
    params = candle.finalize_parameters(driver)
  File "/EXARL/exarl/candlelib/default_utils.py", line 365, in finalize_parameters
    gParameters = args_overwrite_config(args, fileParameters)
  File "/EXARL/exarl/candlelib/default_utils.py", line 621, in args_overwrite_config
    params['output_dir'] = directory_from_parameters(params, params['output_dir'])
  File "/EXARL/exarl/candlelib/default_utils.py", line 730, in directory_from_parameters
    shutil.rmtree(outdir)
  File "/public/software/apps/DeepLearning/TensorFlow/tf2.7.0-dtk21.10-build/lib/python3.6/shutil.py", line 490, in rmtree
    onerror(os.rmdir, path, sys.exc_info())
  File "/public/software/apps/DeepLearning/TensorFlow/tf2.7.0-dtk21.10-build/lib/python3.6/shutil.py", line 488, in rmtree
    os.rmdir(path)
FileNotFoundError: [Errno 2] No such file or directory: '/EXARL/results/EXP000/RUN009'