Sweeps while using MPI and SLURM

Hello! I am attempting to perform a hyperparameter search on my project, which uses MPI under the hood to aggregate the results of multiple agents. I have 63 agents that run an episode, returning a total reward at the end. At the end, each worker node sends their results to the main node, which logs the total reward of every 5th training run.

I have tried to create a sweep with a custom command to use mpirun(as seen below) and running wandb agent sweepid --count 1 in the SLURM script. This results in using all the cores of the machine to start a sweep, effectively blocking my other agents from training.

program: src.sweep_mpi 
  - mpirun
  - "--mca" 
  - opal_warn_on_missing_libcuda
  - 0
  - python
  - "-m"
  - ${program}
  - ${args}

Next, I have tried setting up the sweeping agent inside the python code with a local controller, but this also led to issues regarding the parallelization. Currently, I need to initialize wandb using settings=wandb.Settings(start_method="fork"), but I cannot find any way to specify this as a sweep parameter. Therefore, each run crashes since it is not using the correct parallelization procedure.

Is there anything I can do in this case? Or should I implement my own parameter search?

Hi @vandrew,

You are right, the settings parameter can not be initialized through a sweep config, but it should be possible to place this parameter in the call to wandb.init(). Could you share the traceback you get when your run crashes? It will give us greater visibility into your issue and enable us in assisting you further.


Hi @ramit_goolry!

After a bit more tweaking, I made a secondary script that would intercept the configurations from the API with a local controller. Following this, I saved each configuration in a yaml file, which I then loaded using argparse.

Nevertheless, the issue regarding the parallelization sometimes persists, but is irregular. It will sometimes crash when using the exact same configuration file, without any changes to my code:

Traceback (most recent call last):
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/runpy.py", line 194, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/home/user/project/src/sweep_mpi.py", line 1, in <module>
  File "<frozen importlib._bootstrap>", line 991, in _find_and_load
  File "<frozen importlib._bootstrap>", line 971, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 914, in _find_spec
  File "<frozen importlib._bootstrap_external>", line 1342, in find_spec
  File "<frozen importlib._bootstrap_external>", line 1314, in _get_spec
  File "<frozen importlib._bootstrap_external>", line 1443, in find_spec
  File "<frozen importlib._bootstrap_external>", line 1483, in _fill_cache
BrokenPipeError: [Errno 108] Cannot send after transport endpoint shutdown: '/home/user/project'
Primary job  terminated normally, but 1 process returned
a non-zero exit code. Per user-direction, the job has been aborted.
mpirun detected that one or more processes exited with non-zero status, thus causing
the job to be terminated. The first process to do so was:

  Process name: [[19849,1],3]
  Exit code:    1

This seems to result from a read operation timeout, as indicated by the debug log:

2022-05-19 10:23:11,048 ERROR   SenderThread:7114 [retry.py:__call__():126] Retry attempt failed:
Traceback (most recent call last):
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/connectionpool.py", line 449, in _make_request
    six.raise_from(e, None)
  File "<string>", line 3, in raise_from
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/connectionpool.py", line 444, in _make_request
    httplib_response = conn.getresponse()
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/http/client.py", line 1347, in getresponse
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/http/client.py", line 307, in begin
    version, status, reason = self._read_status()
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/http/client.py", line 268, in _read_status
    line = str(self.fp.readline(_MAXLINE + 1), "iso-8859-1")
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/socket.py", line 669, in readinto
    return self._sock.recv_into(b)
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/ssl.py", line 1241, in recv_into
    return self.read(nbytes, buffer)
  File "/software/software/Python/3.8.6-GCCcore-10.2.0/lib/python3.8/ssl.py", line 1099, in read
    return self._sslobj.read(len, buffer)
socket.timeout: The read operation timed out

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/data/user/.envs/osim/lib/python3.8/site-packages/requests/adapters.py", line 440, in send
    resp = conn.urlopen(
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/connectionpool.py", line 785, in urlopen
    retries = retries.increment(
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/util/retry.py", line 550, in increment
    raise six.reraise(type(error), error, _stacktrace)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/packages/six.py", line 770, in reraise
    raise value
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/connectionpool.py", line 703, in urlopen
    httplib_response = self._make_request(
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/connectionpool.py", line 451, in _make_request
    self._raise_timeout(err=e, url=url, timeout_value=read_timeout)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/urllib3/connectionpool.py", line 340, in _raise_timeout
    raise ReadTimeoutError(
urllib3.exceptions.ReadTimeoutError: HTTPSConnectionPool(host='api.wandb.ai', port=443): Read timed out. (read timeout=10)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/data/user/.envs/osim/lib/python3.8/site-packages/wandb/sdk/lib/retry.py", line 102, in __call__
    result = self._call_fn(*args, **kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/wandb/sdk/internal/internal_api.py", line 140, in execute
    return self.client.execute(*args, **kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/wandb/vendor/gql-0.2.0/wandb_gql/client.py", line 52, in execute
    result = self._get_result(document, *args, **kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/wandb/vendor/gql-0.2.0/wandb_gql/client.py", line 60, in _get_result
    return self.transport.execute(document, *args, **kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/wandb/vendor/gql-0.2.0/wandb_gql/transport/requests.py", line 38, in execute
    request = requests.post(self.url, **post_args)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/requests/api.py", line 117, in post
    return request('post', url, data=data, json=json, **kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/requests/api.py", line 61, in request
    return session.request(method=method, url=url, **kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/requests/sessions.py", line 529, in request
    resp = self.send(prep, **send_kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/requests/sessions.py", line 645, in send
    r = adapter.send(request, **kwargs)
  File "/data/user/.envs/osim/lib/python3.8/site-packages/requests/adapters.py", line 532, in send
    raise ReadTimeout(e, request=request)
requests.exceptions.ReadTimeout: HTTPSConnectionPool(host='api.wandb.ai', port=443): Read timed out. (read timeout=10)

Hi @vandrew!

Apologies for the delay here. This error might be happening due to memory pressure on your computer. Were you able to successfully able to use fork mode for this? Additionally, it would be really helpful to us if you could provide the full debug.log and debug-internal.log associated to the run affected.

Some more information that could be helpful here:

  • What kind of machine are you running this experiment on?
  • What SDK version of wandb are you using?
  • Is the machine reliably connected to the internet?


Hi Andrei,

We wanted to follow up with you regarding your support request as we have not heard back from you. Please let us know if we can be of further assistance or if your issue has been resolved.

Weights & Biases


It is indeed an issue on my end, so I think it is safe to say that this issue can be disregarded. I no longer have access to the debug-internal log, but the cluster I was running my code on was heavily under use.

  • The system is based on a CentOS 7 distribution
  • Version 0.12.16
  • Yes, but the processes might take too long to run. I believe this was the cause of the timeouts in the previous case.

Thank you for the help so far! The way to go seems to be using a local controller in the main thread, which then supplies the received configuration to the other nodes.

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.