trouble with `PoolPartialDescriptor` - no `mro`

[andre-merzky]

Hey,

I am trying to catch up with the latest developments, but alas am stumbling over the following error:

$ python airflowHPC/scripts/base_runner.py get_mass
[2024-08-19T09:46:17.402+0200] {dagbag.py:587} INFO - Filling up the DagBag from /home/merzky/radical/airflowHPC/airflowHPC/airflowHPC/dags/
[2024-08-19T09:46:17.508+0200] {_logging.py:84} INFO - Importing gmxapi.
[2024-08-19T09:46:17.511+0200] {operation.py:99} INFO - Importing gmxapi.operation
[2024-08-19T09:46:17.516+0200] {commandline.py:62} INFO - Importing gmxapi.commandline
[2024-08-19T09:46:17.519+0200] {runtime.py:101} INFO - Importing gmxapi.runtime
[2024-08-19T09:46:17.522+0200] {workflow.py:56} INFO - Importing gmx.workflow
[2024-08-19T09:46:17.522+0200] {workflow.py:60} INFO - Using schema version gmxapi_workspec_0_1.
[2024-08-19T09:46:17.522+0200] {mdrun.py:66} INFO - Importing gmxapi.simulation.mdrun
[2024-08-19T09:46:17.524+0200] {read_tpr.py:56} INFO - Importing gmxapi.simulation.read_tpr
[2024-08-19T09:46:17.525+0200] {modify_input.py:60} INFO - Importing gmxapi.simulation.modify_input
[2024-08-19T09:46:18.107+0200] {dagbag.py:386} ERROR - Failed to import: /home/merzky/radical/airflowHPC/airflowHPC/airflowHPC/dags/test_mpi4py.py
Traceback (most recent call last):
  File "/home/merzky/radical/airflowHPC/ve3/lib/python3.10/site-packages/airflow/models/dagbag.py", line 382, in parse
    loader.exec_module(new_module)
  File "<frozen importlib._bootstrap_external>", line 883, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "/home/merzky/radical/airflowHPC/airflowHPC/airflowHPC/dags/test_mpi4py.py", line 33, in <module>
    multi = MPIExternalPythonOperator.partial(
  File "/home/merzky/radical/airflowHPC/airflowHPC/airflowHPC/operators/mpi_external_python_operator.py", line 32, in pool_slots_partial
    return airflow_partial(*args, **kwargs)
  File "/home/merzky/radical/airflowHPC/ve3/lib/python3.10/site-packages/airflow/models/baseoperator.py", line 284, in partial
    validate_mapping_kwargs(operator_class, "partial", kwargs)
  File "/home/merzky/radical/airflowHPC/ve3/lib/python3.10/site-packages/airflow/models/mappedoperator.py", line 98, in validate_mapping_kwargs
    for klass in op.mro():
AttributeError: 'PoolPartialDescriptor' object has no attribute 'mro'

This is using apache-airflow==2.10.0 and Python 3.10.13.

I don't really see why that happens - the PoolPartialDescriptor is a class after all. Any idea what's going on?
Thanks!

[ejjordan]

A few things to note here.

• I need to delete the base_runner.py script. The best way to run things is in the web interface, as documented in the readme. Barring that, if you want to run from the command line there are also instructions in the readme: airflow dags backfill -s YYYY-MM-DD --reset-dagruns -y run_gmx
  • I tested and both the methods work for the DAG you were trying to run
• The PoolPartialDescriptor class is "just a class" but the actual function of it is a bit more like a class factory that generates multiple instances of the operator in the case that the partial method is used to invoke the same operator with some arguments held fixed and some varying. Thus its behavior can be a little hard to reason about and also to debug
• I mostly left the MPIBashOperator in because I spent a lot of time trying to figure out how to get it to work. I subsequently developed the ResourceBashOperator (and children) which are the ones that should be used for development and testing purposes. Basically it was too much work to get limitations on both the number of MPI ranks and GPUs using the model of the MPIBashOperator, so I took a different course

[andre-merzky]

Thanks Joe. I also saw the error a couple of times when not using the command line tools but could not determine yet what triggers it. And yes, that PoolPartialDescriptor seems to be a funny :-)