ploomber.executors.Parallel¶

class ploomber.executors.Parallel(processes=None, print_progress=False, start_method=None)¶

Runs a DAG in parallel using multiprocessing

Parameters

processes (int, default=None) – The number of processes to use. If None, uses os.cpu_count
print_progress (bool, default=False) – Whether to print progress to stdout, otherwise just log it
start_method (str, default=None) – The method which should be used to start child processes. method can be ‘fork’, ‘spawn’ or ‘forkserver’. If None or empty then the default start_method is used.

Examples

Spec API:

# add at the top of your pipeline.yaml
executor: parallel

tasks:
  - source: script.py
    nb_product_key: [nb_ipynb, nb_html]
    product:
        nb_ipynb: nb.ipynb
        nb_html: report.html

Python API:

>>> from ploomber import DAG
>>> from ploomber.executors import Parallel
>>> dag = DAG(executor='parallel') # use with default values
>>> dag = DAG(executor=Parallel(processes=2)) # customize

DAG can exit gracefully on function tasks (PythonCallable):

>>> from ploomber.products import File
>>> from ploomber.tasks import PythonCallable
>>> from ploomber.exceptions import DAGBuildEarlyStop
>>> # A PythonCallable function that raises DAGBuildEarlyStop
>>> def early_stop_root(product):
...     raise DAGBuildEarlyStop('Ending gracefully')

>>> # Since DAGBuildEarlyStop is raised, DAG will exit gracefully.
>>> dag = DAG(executor='parallel')
>>> t = PythonCallable(early_stop_root, File('file.txt'), dag=dag)
>>> dag.build()  

DAG can also exit gracefully on notebook tasks:

>>> from pathlib import Path
>>> from ploomber.tasks import NotebookRunner
>>> from ploomber.products import File
>>> def early_stop():
...     raise DAGBuildEarlyStop('Ending gracefully')

>>> # Use task-level hook "on_finish" to exit DAG gracefully.
>>> dag = DAG(executor='parallel')
>>> t = NotebookRunner(Path('nb.ipynb'), File('report.html'), dag=dag)
>>> t.on_finish = early_stop
>>> dag.build()  

Notes

If any task crashes, downstream tasks execution is aborted, building continues until no more tasks can be executed

New in version 0.20: Added start_method argument

Contents

ploomber.executors.Parallel¶