How to configure Luigi task retry correctly?
Asked Answered
B

1

6

I am trying to configure Luigi's retry mechanism so that failed tasks will be retried a few times. However, while the task is retried successfully, Luigi exits unsuccessfully:

===== Luigi Execution Summary =====

Scheduled 3 tasks of which:
* 2 ran successfully:
    - 1 FailOnceThenSucceed(path=/tmp/job-id-18.subtask)
    - 1 MasterTask(path=/tmp/job-id-18)
* 1 failed:
    - 1 FailOnceThenSucceed(path=/tmp/job-id-18.subtask)

This progress looks :( because there were failed tasks

So the question is: how do I configure Luigi (I have installed version 2.3.3 with pip install) so that when a task fails once, but is then retried with success, then Luigi will exit successfully with This progress looks :) instead of fail with This progress looks :(?

Here is a minimal scheduler and worker config I've come up with, as well as tasks to demonstrate the behavior:

[scheduler]
retry_count = 3
retry-delay = 1

[worker]
keep_alive=true

mytasks.py:

import luigi


class FailOnceThenSucceed(luigi.Task):
    path = luigi.Parameter()

    def output(self):
        return luigi.LocalTarget(self.path)

    def run(self):
        failmarker = luigi.LocalTarget(self.path + ".hasfailedonce")
        if failmarker.exists():
            with self.output().open('w') as target:
                target.write('OK')
        else:
            with failmarker.open('w') as marker:
                marker.write('Failed')
            raise RuntimeError("Failed once")


class MasterTask(luigi.Task):
    path = luigi.Parameter()

    def requires(self):
        return FailOnceThenSucceed(path=self.path + '.subtask')

    def output(self):
        return luigi.LocalTarget(self.path)

    def run(self):
        with self.output().open('w') as target:
            target.write('OK')

Example execution:

PYTHONPATH=. luigi --module mytasks MasterTask --workers=2 --path='/tmp/job-id-18'

Berg answered 24/10, 2016 at 12:26 Comment(1)
'retry-delay' should be 'retry_delay' luigi.readthedocs.io/en/stable/configuration.html#schedulerSteeplejack
T
2

This is an old issue of Luigi - where successful retried tasks were not marked as such when failed and then succeeded on retry: https://github.com/spotify/luigi/issues/1932

It was fixed in version 2.7.2: https://github.com/spotify/luigi/releases/tag/2.7.2

I suggest you upgrade to the latest Luigi version, i.e. by running pip install -U luigi.

Tantalizing answered 19/2, 2018 at 10:34 Comment(0)

© 2022 - 2024 — McMap. All rights reserved.