Last updated:
0 purchases
luigisoftfailures 0.1.1
Provides a decorator for Luigi tasks that allows them to fail softly. Tasks that fail softly are seen by Luigi as complete, and thus don't prevent dependent tasks from running. The dependent task can check if it's dependencies actually ran successfully or failed softly.
Installation
pip install luigi_soft_failures
Usage
from luigi_soft_failures import softly_failing
Use as class decorator:
@softly_failing(catch_all=True, propagate=True)
class SomeTask(luigi.Task):
...
Or on demand in the requiring task:
def requires(self):
return softly_failing(catch_all=True)(SomeTask)(some_param=42)
The dependent task can check the status of it's dependencies using failed_softly:
def run(self):
if self.requires().failed_softly():
...
For a complete example see as_decorator.py.
API
softly_failing accepts the following parameters:
catch_all (bool, default False):
When True, any exception thrown in the task's run method will lead to a soft failure. Otherwise, soft failures can be generated manually by calling self.fail_softly('Some error message') from the task's run method and exiting the method without exception.
propagate (bool, default False):
When True, the task fails softly if any of it's dependencies failed softly, and run is never executed. Otherwise, run is executed as if the dependencies ran successfully.
output_dir (str, default None): Described below.
Storage of soft failure reports
Whan the wrapped task fails softly, it creates a report with the failure message or exception traceback to indicate this. These reports are stored in a directory specified in one of the following ways (in this order of precedence):
output_dir parameter passed to softly_failing
Specified in luigi.cfg:
[luigi_soft_failures.Config]
output_dir=/some/path
Environment variable LUIGI_SOFT_FAILURES_OUTPUT_DIR
Default ./soft_failures/
Limitations
Soft failure status is stored using a LocalTarget, so a local storage that all workers have access to is required.
In the Luigi visualizer, softly failed tasks are shown as complete.
Pull requests to address these are very welcome!
For personal and professional use. You cannot resell or redistribute these repositories in their original state.
There are no reviews.