Open
Description
Description
Calling Scheduler.add_task() with mocked task dependencies causes a TypeError due to an implicit comparison between a MagicMock and int. This happens when task.priority is a MagicMock, and _update_priority() tries to call max(prio, task.priority), which results in TypeError: '>' not supported between instances of 'MagicMock' and 'int'.
A potential fix is to add a check or default initialization for task.priority before using it in a numerical comparison.
Steps to Reproduce
Minimal reproducible test:
from helpers import unittest
import mock
from luigi.scheduler import Scheduler, PENDING
WORKER = 'myworker'
class AddTaskTest(unittest.TestCase):
def setUp(self):
self.s = Scheduler()
self.s._state = mock.MagicMock()
self.s._config = mock.MagicMock()
self.s._config.stable_done_cooldown_secs = 300
self.s._update_worker = mock.MagicMock()
self.s._update_task_history = mock.MagicMock()
self.s._generate_retry_policy = mock.MagicMock()
self.worker = mock.MagicMock()
self.worker.enabled = True
self.s._update_worker.return_value = self.worker
def test_task_dependencies(self):
task = mock.MagicMock()
task.status = PENDING
task.deps = set()
self.s._state.get_task.return_value = task
self.s.add_task(worker=WORKER, task_id='test_task', deps=['dep1', 'dep2'], new_deps=['dep3'])
if __name__ == '__main__':
unittest.main()Expected behavior:
The task should perform type checking or raise a clear error if task.priority is not valid.
Actual behavior:
the stack trace is:
luigi/scheduler.py:927: in add_task
self._update_priority(task, priority, worker_id)
self = <luigi.scheduler.Scheduler object at 0x7f76f6666770> task = , prio = 0 worker = 'myworker'
def _update_priority(self, task, prio, worker):
"""
Update priority of the given task.
Priority can only be increased.
If the task doesn't exist, a placeholder task is created to preserve priority when the task is later scheduled.
"""
task.priority = prio = max(prio, task.priority)
E TypeError: '>' not supported between instances of 'MagicMock' and 'int'```
Metadata
Metadata
Assignees
Labels
No labels