def test_complete_exception(self):
"Tests that a task is still scheduled if its sister task crashes in the complete() method"
class A(DummyTask):
def complete(self):
raise Exception("doh")
a = A()
class C(DummyTask):
pass
c = C()
class B(DummyTask):
def requires(self):
return a, c
b = B()
sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id="foo")
self.assertFalse(w.add(b))
self.assertTrue(w.run())
self.assertFalse(b.has_run)
self.assertTrue(c.has_run)
self.assertFalse(a.has_run)
w.stop()
def test_ping_retry(self):
""" Worker ping fails once. Ping continues to try to connect to scheduler
Kind of ugly since it uses actual timing with sleep to test the thread
"""
sch = CentralPlannerScheduler(
retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10,
)
self._total_pings = 0 # class var so it can be accessed from fail_ping
def fail_ping(worker):
# this will be called from within keep-alive thread...
self._total_pings += 1
raise Exception("Some random exception")
sch.ping = fail_ping
w = Worker(
scheduler=sch,
worker_id="foo",
ping_interval=0.01 # very short between pings to make test fast
)
# let the keep-alive thread run for a bit...
time.sleep(0.1) # yes, this is ugly but it's exactly what we need to test
w.stop()
self.assertTrue(
self._total_pings > 1,
msg="Didn't retry pings (%d pings performed)" % (self._total_pings,)
)
def test_requires_exception(self):
class A(DummyTask):
def requires(self):
raise Exception("doh")
a = A()
class C(DummyTask):
pass
c = C()
class B(DummyTask):
def requires(self):
return a, c
b = B()
sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id="foo")
self.assertFalse(w.add(b))
self.assertTrue(w.run())
self.assertFalse(b.has_run)
self.assertTrue(c.has_run)
self.assertFalse(a.has_run)
w.stop()
def test_interleaved_workers(self):
class A(DummyTask):
pass
a = A()
class B(DummyTask):
def requires(self):
return a
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEquals(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
w.add(b)
w2.add(eb)
logging.debug("RUNNING BROKEN WORKER")
w2.run()
self.assertFalse(a.complete())
self.assertFalse(b.complete())
logging.debug("RUNNING FUNCTIONAL WORKER")
w.run()
self.assertTrue(a.complete())
self.assertTrue(b.complete())
def setUp(self):
# InstanceCache.disable()
self.sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.w_raise = Worker(scheduler=self.sch, worker_id='X_raise', raise_on_error=True)
self.w2 = Worker(scheduler=self.sch, worker_id='Y')
self.time = time.time
def test_worker_executes_fork_handler(self):
class Stub(luigi.Task):
def complete(self):
return False
def run(self):
pass
s = CentralPlannerScheduler()
w = Worker(scheduler=s, worker_processes=2)
w.add(Stub())
w.run()
time.sleep(0.1)
worker_res = [self.worker_queue.get(), self.worker_queue.get(), self.worker_queue.get()]
self.assertIn("both", worker_res)
self.assertIn("master", worker_res)
child_res = [self.child_queue.get(), self.child_queue.get(), self.child_queue.get()]
self.assertIn("both", child_res)
self.assertIn("child", child_res)
self.assertTrue(self.worker_queue.empty())
self.assertTrue(self.child_queue.empty())
def test_allow_reschedule_with_many_missing_deps(self):
class A(Task):
""" Task that must run twice to succeed """
i = luigi.IntParameter()
runs = 0
def complete(self):
return self.runs >= 2
def run(self):
self.runs += 1
class B(Task):
done = False
def requires(self):
return map(A, range(20))
def complete(self):
return self.done
def run(self):
self.done = True
b = B()
w = Worker(scheduler=self.sch, worker_id='X', max_reschedules=1)
self.assertTrue(w.add(b))
self.assertFalse(w.run())
# For b to be done, we must have rescheduled its dependencies to run them twice
self.assertTrue(b.complete())
self.assertTrue(all(a.complete() for a in b.deps()))
def test_interleaved_workers2(self):
# two tasks without dependencies, one external, one not
class B(DummyTask):
pass
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEqual(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
self.assertTrue(w2.add(eb))
self.assertTrue(w.add(b))
self.assertTrue(w2.run())
self.assertFalse(b.complete())
self.assertTrue(w.run())
self.assertTrue(b.complete())
w.stop()
w2.stop()
def test_task_limit_exceeded(self):
w = Worker()
t = ForkBombTask(3, 2)
w.add(t)
w.run()
self.assertFalse(t.complete())
leaf_tasks = [ForkBombTask(3, 2, branch) for branch in [(0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1)]]
self.assertEquals(3, sum(t.complete() for t in leaf_tasks), "should have gracefully completed as much as possible even though the single last leaf didn't get scheduled")
def _run(self, dt, interval):
w = Worker(scheduler=self.sch)
tasks = self.gen_tasks(dt, interval)
map(w.add, tasks)
for t in filter(lambda x: isinstance(x, RunMetricRules), tasks):
if t.handler == 'save_to_mongodb':
if t.filter_rules:
filter_metrics_args = {"require": t, 'filter_rules': t.filter_rules}
# 过滤
filter_metrics = FilterMetrics(**filter_metrics_args)
require_task = filter_metrics
w.add(filter_metrics)
else:
require_task = t
# 保存
kwargs = dict(**self.mongo_conf)
kwargs.update({"require": require_task, "date": dt, 'app_id': self.app_id, 'interval': interval})
save_metrics = SaveMetrics(**kwargs)
w.add(save_metrics)
else:
kwargs = dict(**self.mongo_conf)
kwargs.update({"require": t,
"date": dt,
'app_id': self.app_id,
'interval': interval,
'handler_func': t.handler})
metric_handler = MetricsHandler(**kwargs)
w.add(metric_handler)
w.run()
def test_interleaved_workers3(self):
class A(DummyTask):
def run(self):
logging.debug('running A')
time.sleep(0.1)
super(A, self).run()
a = A()
class B(DummyTask):
def requires(self):
return a
def run(self):
logging.debug('running B')
super(B, self).run()
b = B()
sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X', keep_alive=True, count_uniques=True)
w2 = Worker(scheduler=sch, worker_id='Y', keep_alive=True, count_uniques=True, wait_interval=0.1)
self.assertTrue(w.add(a))
self.assertTrue(w2.add(b))
threading.Thread(target=w.run).start()
self.assertTrue(w2.run())
self.assertTrue(a.complete())
self.assertTrue(b.complete())
w.stop()
w2.stop()
class AssistantTest(unittest.TestCase):
def run(self, result=None):
self.sch = Scheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
self.assistant = Worker(scheduler=self.sch, worker_id='Y', assistant=True)
with Worker(scheduler=self.sch, worker_id='X') as w:
self.w = w
super(AssistantTest, self).run(result)
def test_get_work(self):
d = Dummy2Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assistant.run()
self.assertTrue(d.complete())
def test_bad_job_type(self):
class Dummy3Task(Dummy2Task):
task_family = 'UnknownTaskFamily'
d = Dummy3Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assertFalse(self.assistant.run())
self.assertFalse(d.complete())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()), [d.task_id])
def test_unimported_job_type(self):
MODULE_CONTENTS = b'''
import luigi
class UnimportedTask(luigi.Task):
def complete(self):
return False
'''
class NotImportedTask(luigi.Task):
task_family = 'UnimportedTask'
task_module = None
task = NotImportedTask()
# verify that it can't run the task without the module info necessary to import it
self.w.add(task)
self.assertFalse(self.assistant.run())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()), [task.task_id])
# check that it can import with the right module
with temporary_unloaded_module(MODULE_CONTENTS) as task.task_module:
self.w.add(task)
self.assertTrue(self.assistant.run())
self.assertEqual(list(self.sch.task_list('DONE', '').keys()), [task.task_id])
def test_wait_jitter_default(self, mock_sleep, mock_random):
""" verify default jitter is as expected """
mock_random.return_value = 1.0
w = Worker()
x = w._sleeper()
six.next(x)
mock_random.assert_called_with(0, 5.0)
mock_sleep.assert_called_with(2.0)
mock_random.return_value = 3.3
six.next(x)
mock_random.assert_called_with(0, 5.0)
mock_sleep.assert_called_with(4.3)
def test_disabled_shutdown_hook(self):
w = Worker(scheduler=self.sch, keep_alive=True, no_install_shutdown_handler=True)
with w:
try:
# try to kill the worker!
os.kill(os.getpid(), signal.SIGUSR1)
except AttributeError:
raise unittest.SkipTest('signal.SIGUSR1 not found on this system')
# try to kill the worker... AGAIN!
t = SuicidalWorker(signal.SIGUSR1)
w.add(t)
w.run()
# task should have stepped away from the ledge, and completed successfully despite all the SIGUSR1 signals
self.assertEqual(list(self.sch.task_list('DONE', '').keys()), [t.task_id])
def test_die_for_non_unique_pending(self):
class A(DummyTask):
def run(self):
logging.debug('running A')
time.sleep(0.1)
super(A, self).run()
a = A()
class B(DummyTask):
def requires(self):
return a
def run(self):
logging.debug('running B')
super(B, self).run()
b = B()
sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X', keep_alive=True, count_uniques=True)
w2 = Worker(scheduler=sch, worker_id='Y', keep_alive=True, count_uniques=True, wait_interval=0.1)
self.assertTrue(w.add(b))
self.assertTrue(w2.add(b))
self.assertEqual(w._get_work()[0], 'A()')
self.assertTrue(w2.run())
self.assertFalse(a.complete())
self.assertFalse(b.complete())
w2.stop()
def test_wait_jitter(self, mock_sleep, mock_random):
""" verify configured jitter amount """
mock_random.return_value = 1.0
w = Worker()
x = w._sleeper()
six.next(x)
mock_random.assert_called_with(0, 10.0)
mock_sleep.assert_called_with(2.0)
mock_random.return_value = 2.0
six.next(x)
mock_random.assert_called_with(0, 10.0)
mock_sleep.assert_called_with(3.0)
class ParallelSchedulingTest(unittest.TestCase):
def setUp(self):
self.sch = mock.Mock()
self.w = Worker(scheduler=self.sch, worker_id='x')
def added_tasks(self, status):
return [kw['task_id'] for args, kw in self.sch.add_task.call_args_list if kw['status'] == status]
def test_multiprocess_scheduling_with_overlapping_dependencies(self):
self.w.add(OverlappingSelfDependenciesTask(5, 2), True)
self.assertEqual(15, self.sch.add_task.call_count)
self.assertEqual(set((
'OverlappingSelfDependenciesTask(n=1, k=1)',
'OverlappingSelfDependenciesTask(n=2, k=1)',
'OverlappingSelfDependenciesTask(n=2, k=2)',
'OverlappingSelfDependenciesTask(n=3, k=1)',
'OverlappingSelfDependenciesTask(n=3, k=2)',
'OverlappingSelfDependenciesTask(n=4, k=1)',
'OverlappingSelfDependenciesTask(n=4, k=2)',
'OverlappingSelfDependenciesTask(n=5, k=2)',
)), set(self.added_tasks('PENDING')))
self.assertEqual(set((
'OverlappingSelfDependenciesTask(n=0, k=0)',
'OverlappingSelfDependenciesTask(n=0, k=1)',
'OverlappingSelfDependenciesTask(n=1, k=0)',
'OverlappingSelfDependenciesTask(n=1, k=2)',
'OverlappingSelfDependenciesTask(n=2, k=0)',
'OverlappingSelfDependenciesTask(n=3, k=0)',
'OverlappingSelfDependenciesTask(n=4, k=0)',
)), set(self.added_tasks('DONE')))
@mock.patch('luigi.notifications.send_error_email')
def test_raise_exception_in_complete(self, send):
self.w.add(ExceptionCompleteTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(0, self.sch.add_task.call_count)
self.assertTrue('assert False' in send.call_args[0][1])
@mock.patch('luigi.notifications.send_error_email')
def test_raise_unpicklable_exception_in_complete(self, send):
# verify exception can't be pickled
self.assertRaises(Exception, UnpicklableExceptionTask().complete)
try:
UnpicklableExceptionTask().complete()
except Exception as e:
ex = e
self.assertRaises(pickle.PicklingError, pickle.dumps, ex)
# verify this can run async
self.w.add(UnpicklableExceptionTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(0, self.sch.add_task.call_count)
self.assertTrue('raise UnpicklableException()' in send.call_args[0][1])
@mock.patch('luigi.notifications.send_error_email')
def test_raise_exception_in_requires(self, send):
self.w.add(ExceptionRequiresTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(0, self.sch.add_task.call_count)
def setUp(self):
try:
from luigi.sqs_history import SqsHistory, SqsTaskHistory, SqsWorkerHistory
except ImportError as e:
raise unittest.SkipTest('Could not test WorkerTaskGlobalEventHandlerTests: %s' % e)
# Replace _config method with one that uses our dummy queue.
def fake_config(s, *args):
s._queue = DummyQueue()
SqsHistory._config = fake_config
# InstanceCache.disable()
self.sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.w2 = Worker(scheduler=self.sch, worker_id='Y')
self.time = time.time
def test_purge_hung_worker_override_timeout_time(self, mock_time):
w = Worker(worker_processes=2, wait_interval=0.01, timeout=5)
mock_time.time.return_value = 0
w.add(HungWorker(10))
w._run_task('HungWorker(worker_timeout=10)')
mock_time.time.return_value = 10
w._handle_next_task()
self.assertEqual(1, len(w._running_tasks))
mock_time.time.return_value = 11
w._handle_next_task()
self.assertEqual(0, len(w._running_tasks))
def test_purge_hung_worker_default_timeout_time(self, mock_time):
w = Worker(worker_processes=2, wait_interval=0.01, worker_timeout=5)
mock_time.time.return_value = 0
w.add(HungWorker())
w._run_task('HungWorker(worker_timeout=None)')
mock_time.time.return_value = 5
w._handle_next_task()
self.assertEqual(1, len(w._running_tasks))
mock_time.time.return_value = 6
w._handle_next_task()
self.assertEqual(0, len(w._running_tasks))
def test_single_threaded_worker_doesnot_execute_fork_handler(self):
class Stub(luigi.Task):
def complete(self):
return False
def run(self):
pass
s = CentralPlannerScheduler()
w = Worker(scheduler=s, worker_processes=1)
w.add(Stub())
w.run()
time.sleep(0.1)
self.assertTrue(self.worker_queue.empty())
self.assertTrue(self.child_queue.empty())
def test_purge_multiple_workers(self):
w = Worker(worker_processes=2, wait_interval=0.01)
t1 = SuicidalWorker(signal.SIGTERM)
t2 = SuicidalWorker(signal.SIGKILL)
w.add(t1)
w.add(t2)
w._run_task(t1.task_id)
w._run_task(t2.task_id)
time.sleep(1.0)
w._handle_next_task()
w._handle_next_task()
w._handle_next_task()
class AssistantTest(unittest.TestCase):
def run(self, result=None):
self.sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
self.assistant = Worker(scheduler=self.sch, worker_id='Y', assistant=True)
with Worker(scheduler=self.sch, worker_id='X') as w:
self.w = w
super(AssistantTest, self).run(result)
def test_get_work(self):
d = Dummy2Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assistant.run()
self.assertTrue(d.complete())
def test_bad_job_type(self):
class Dummy3Task(Dummy2Task):
task_family = 'UnknownTaskFamily'
d = Dummy3Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assertFalse(self.assistant.run())
self.assertFalse(d.complete())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()), [str(d)])
def test_unimported_job_type(self):
class NotImportedTask(luigi.Task):
task_family = 'UnimportedTask'
task_module = None
task = NotImportedTask()
# verify that it can't run the task without the module info necessary to import it
self.w.add(task)
self.assertFalse(self.assistant.run())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()), ['UnimportedTask()'])
# check that it can import with the right module
task.task_module = 'dummy_test_module.not_imported'
self.w.add(task)
self.assertTrue(self.assistant.run())
self.assertEqual(list(self.sch.task_list('DONE', '').keys()), ['UnimportedTask()'])
def test_purge_hung_worker_default_timeout_time(self, mock_time):
w = Worker(worker_processes=2, wait_interval=0.01, timeout=5)
mock_time.time.return_value = 0
task = HangTheWorkerTask()
w.add(task)
w._run_task(task.task_id)
mock_time.time.return_value = 5
w._handle_next_task()
self.assertEqual(1, len(w._running_tasks))
mock_time.time.return_value = 6
w._handle_next_task()
self.assertEqual(0, len(w._running_tasks))
def _test_context_manager(self, force_multiprocessing):
CONTEXT_MANAGER_MODULE = b'''
class MyContextManager(object):
def __init__(self, task_process):
self.task = task_process.task
def __enter__(self):
assert not self.task.run_event.is_set(), "the task should not have run yet"
self.task.enter_event.set()
return self
def __exit__(self, exc_type=None, exc_value=None, traceback=None):
assert self.task.run_event.is_set(), "the task should have run"
self.task.exit_event.set()
'''
class DummyEventRecordingTask(luigi.Task):
def __init__(self, *args, **kwargs):
self.enter_event = multiprocessing.Event()
self.exit_event = multiprocessing.Event()
self.run_event = multiprocessing.Event()
super(DummyEventRecordingTask, self).__init__(*args, **kwargs)
def run(self):
assert self.enter_event.is_set(), "the context manager should have been entered"
assert not self.exit_event.is_set(), "the context manager should not have been exited yet"
assert not self.run_event.is_set(), "the task should not have run yet"
self.run_event.set()
def complete(self):
return self.run_event.is_set()
with temporary_unloaded_module(CONTEXT_MANAGER_MODULE) as module_name:
t = DummyEventRecordingTask()
w = Worker(task_process_context=module_name + '.MyContextManager',
force_multiprocessing=force_multiprocessing)
w.add(t)
self.assertTrue(w.run())
self.assertTrue(t.complete())
self.assertTrue(t.enter_event.is_set())
self.assertTrue(t.exit_event.is_set())
def test_stop_worker_kills_subprocesses(self):
w = Worker(worker_processes=2)
hung_task = HungWorker()
w.add(hung_task)
w._run_task(hung_task.task_id)
pids = [p.pid for p in w._running_tasks.values()]
self.assertEqual(1, len(pids))
pid = pids[0]
def is_running():
return pid in {p.pid for p in psutil.Process().children()}
self.assertTrue(is_running())
w.stop()
self.assertFalse(is_running())
def invoke_task(godzilla_task, task_id=None, **kwargs):
print godzilla_task, task_id, kwargs
task = tasks[godzilla_task]
kwargs['task_id'] = task_id
task = task.as_luigi(**kwargs)
w = Worker(scheduler=sch)
w.add(task)
try:
w.run()
if not w.run_succeeded:
raise Exception("job failed")
return task.output().path
finally:
w.stop()
class MultiprocessWorkerTest(unittest.TestCase):
def setUp(self):
self.scheduler = RemoteScheduler()
self.scheduler.add_worker = Mock()
self.scheduler.add_task = Mock()
self.worker = Worker(scheduler=self.scheduler, worker_id='X', worker_processes=2)
def tearDown(self):
self.worker.stop()
def test_positive_path(self):
a = DummyTask("a")
b = DummyTask("b")
class MultipleRequirementTask(DummyTask):
def requires(self):
return [a, b]
c = MultipleRequirementTask("C")
self.assertTrue(self.worker.add(c))
self.scheduler.get_work = Mock(side_effect=[(3, str(a)), (2, str(b)), (1, str(c)), (0, None), (0, None)])
self.assertTrue(self.worker.run())
self.assertTrue(c.has_run)
def test_path_with_task_failures(self):
class FailingTask(DummyTask):
def run(self):
raise Exception("I am failing")
a = FailingTask("a")
b = FailingTask("b")
class MultipleRequirementTask(DummyTask):
def requires(self):
return [a, b]
c = MultipleRequirementTask("C")
self.assertTrue(self.worker.add(c))
self.scheduler.get_work = Mock(side_effect=[(3, str(a)), (2, str(b)), (1, str(c)), (0, None), (0, None)])
self.assertFalse(self.worker.run())
def test_process_killed_handler(self, task_proc):
result = []
@HangTheWorkerTask.event_handler(Event.PROCESS_FAILURE)
def store_task(t, error_msg):
self.assertTrue(error_msg)
result.append(t)
w = Worker()
task = HangTheWorkerTask()
task_process = mock.MagicMock(is_alive=lambda: False, exitcode=-14, task=task)
task_proc.return_value = task_process
w.add(task)
w._run_task(task.task_id)
w._handle_next_task()
self.assertEqual(result, [task])
def test_timeout_handler(self, mock_time):
result = []
@HangTheWorkerTask.event_handler(Event.TIMEOUT)
def store_task(t, error_msg):
self.assertTrue(error_msg)
result.append(t)
w = Worker(worker_processes=2, wait_interval=0.01, timeout=5)
mock_time.time.return_value = 0
task = HangTheWorkerTask(worker_timeout=1)
w.add(task)
w._run_task(task.task_id)
mock_time.time.return_value = 3
w._handle_next_task()
self.assertEqual(result, [task])
def test_fails_registering_signal(self):
with mock.patch('luigi.worker.signal', spec=['signal']):
# mock will raise an attribute error getting signal.SIGUSR1
Worker()
def test_asserts_for_worker(self):
"""
Test that Worker() asserts that it's sanely configured
"""
Worker(wait_interval=1) # This shouldn't raise
self.assertRaises(AssertionError, Worker, wait_interval=0)
def test_interleaved_workers2(self):
# two tasks without dependencies, one external, one not
class B(DummyTask):
pass
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEquals(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
self.assertTrue(w2.add(eb))
self.assertTrue(w.add(b))
self.assertTrue(w2.run())
self.assertFalse(b.complete())
self.assertTrue(w.run())
self.assertTrue(b.complete())
w.stop()
w2.stop()
class WorkerEmailTest(EmailTest):
def setUp(self):
super(WorkerEmailTest, self).setUp()
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.worker = Worker(scheduler=sch, worker_id="foo")
def tearDown(self):
self.worker.stop()
@with_config(EMAIL_CONFIG)
def test_connection_error(self):
sch = RemoteScheduler(host="this_host_doesnt_exist",
port=1337,
connect_timeout=1)
worker = Worker(scheduler=sch)
self.waits = 0
def dummy_wait():
self.waits += 1
sch._wait = dummy_wait
class A(DummyTask):
pass
a = A()
self.assertEquals(self.last_email, None)
worker.add(a)
self.assertEquals(self.waits, 2) # should attempt to add it 3 times
self.assertNotEquals(self.last_email, None)
self.assertEquals(self.last_email[0],
"Luigi: Framework error while scheduling %s" % (a, ))
worker.stop()
@with_config(EMAIL_CONFIG)
def test_complete_error(self):
class A(DummyTask):
def complete(self):
raise Exception("b0rk")
a = A()
self.assertEquals(self.last_email, None)
self.worker.add(a)
self.assertEquals(("Luigi: %s failed scheduling" % (a, )),
self.last_email[0])
self.worker.run()
self.assertEquals(("Luigi: %s failed scheduling" % (a, )),
self.last_email[0])
self.assertFalse(a.has_run)
@with_config(EMAIL_CONFIG)
def test_complete_return_value(self):
class A(DummyTask):
def complete(self):
pass # no return value should be an error
a = A()
self.assertEquals(self.last_email, None)
self.worker.add(a)
self.assertEquals(("Luigi: %s failed scheduling" % (a, )),
self.last_email[0])
self.worker.run()
self.assertEquals(("Luigi: %s failed scheduling" % (a, )),
self.last_email[0])
self.assertFalse(a.has_run)
@with_config(EMAIL_CONFIG)
def test_run_error(self):
class A(luigi.Task):
def complete(self):
return False
def run(self):
raise Exception("b0rk")
a = A()
self.worker.add(a)
self.assertEquals(self.last_email, None)
self.worker.run()
self.assertEquals(("Luigi: %s FAILED" % (a, )), self.last_email[0])
def test_no_error(self):
class A(DummyTask):
pass
a = A()
self.assertEquals(self.last_email, None)
self.worker.add(a)
self.assertEquals(self.last_email, None)
self.worker.run()
self.assertEquals(self.last_email, None)
self.assertTrue(a.complete())
class WorkerTaskGlobalEventHandlerTests(unittest.TestCase):
@with_config(
dict(worker_history=dict(record_worker_history_sqs='true',
sqs_queue_name='name',
aws_access_key_id='key',
aws_secret_access_key='secret_key'),
worker_metadata=dict(meta1='data1')))
def setUp(self):
try:
from luigi.sqs_history import SqsHistory, SqsTaskHistory, SqsWorkerHistory
except ImportError as e:
raise unittest.SkipTest(
'Could not test WorkerTaskGlobalEventHandlerTests: %s' % e)
# Replace _config method with one that uses our dummy queue.
def fake_config(s, *args):
s._queue = DummyQueue()
SqsHistory._config = fake_config
# InstanceCache.disable()
self.sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.w2 = Worker(scheduler=self.sch, worker_id='Y')
self.time = time.time
def tearDown(self):
if time.time != self.time:
time.time = self.time
self.w.stop()
self.w2.stop()
def setTime(self, t):
time.time = lambda: t
def _parse_task_events(self, messages):
results = {}
for m in messages:
event = m.get('event')
if not event:
continue
messages = results.get(event, [])
messages.append(m)
results[event] = messages
return results
def test_dep(self):
class A(Task):
param_a = luigi.Parameter(default="a")
def run(self):
self.has_run = True
def complete(self):
return self.has_run
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.assertTrue(self.w.add(b))
self.assertTrue(self.w.run())
self.assertTrue(a.has_run)
self.assertTrue(b.has_run)
sent_messages = [
json.loads(m.get_body())
for m in self.w._worker_history_impl._queue.messages
]
event_messages = self._parse_task_events(sent_messages)
self.assertEquals(4, len(event_messages))
# Check started events:
started_events = event_messages.get(Event.START)
self.assertEquals(2, len(started_events))
self.assertEquals('A(param_a=a)', started_events[0]['task']['id'])
self.assertEquals('B()', started_events[1]['task']['id'])
# Check success events
success_events = event_messages.get(Event.SUCCESS)
self.assertEquals(2, len(success_events))
self.assertEquals('A(param_a=a)', success_events[0]['task']['id'])
self.assertEquals('B()', success_events[1]['task']['id'])
# Check processing time events
processing_events = event_messages.get(Event.PROCESSING_TIME)
self.assertEquals(2, len(processing_events))
self.assertEquals('A(param_a=a)', processing_events[0]['task']['id'])
self.assertTrue('processing_time' in processing_events[0])
self.assertEquals('B()', processing_events[1]['task']['id'])
self.assertTrue('processing_time' in processing_events[1])
# Check dependency event
dependency_event = event_messages.get(Event.DEPENDENCY_DISCOVERED)
self.assertEquals(1, len(dependency_event))
self.assertEquals('B()', dependency_event[0]['task']['id'])
self.assertEquals('A(param_a=a)',
dependency_event[0]['dependency_task']['id'])
def test_external_dep(self):
class A(ExternalTask):
def complete(self):
return False
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.assertTrue(self.w.add(b))
self.assertTrue(self.w.run())
self.assertFalse(a.has_run)
self.assertFalse(b.has_run)
sent_messages = [
json.loads(m.get_body())
for m in self.w._worker_history_impl._queue.messages
]
event_messages = self._parse_task_events(sent_messages)
self.assertEquals(2, len(event_messages))
# Check dependency event
dependency_event = event_messages.get(Event.DEPENDENCY_DISCOVERED)
self.assertEquals(1, len(dependency_event))
self.assertEquals('B()', dependency_event[0]['task']['id'])
self.assertEquals('A()', dependency_event[0]['dependency_task']['id'])
# Check dependency missing event
dependency_missing_event = event_messages.get(Event.DEPENDENCY_MISSING)
self.assertEquals(1, len(dependency_missing_event))
self.assertEquals('A()', dependency_missing_event[0]['task']['id'])
def test_fail(self):
class A(Task):
def run(self):
self.has_run = True
raise Exception()
def complete(self):
return self.has_run
a = A()
a.has_run = False
self.assertTrue(self.w.add(a))
self.assertFalse(self.w.run())
self.assertTrue(a.has_run)
sent_messages = [
json.loads(m.get_body())
for m in self.w._worker_history_impl._queue.messages
]
event_messages = self._parse_task_events(sent_messages)
self.assertEquals(3, len(event_messages))
# Check failure event
failure_event = event_messages.get(Event.FAILURE)
self.assertEquals(1, len(failure_event))
self.assertEquals('A()', failure_event[0]['task']['id'])
self.assertEquals('Exception()', failure_event[0]['exception'])
def test_unknown_dep(self):
# see central_planner_test.CentralPlannerTest.test_remove_dep
class A(ExternalTask):
def complete(self):
return False
class C(Task):
def complete(self):
return True
def get_b(dep):
class B(Task):
def requires(self):
return dep
def run(self):
self.has_run = True
def complete(self):
return False
b = B()
b.has_run = False
return b
b_a = get_b(A())
b_c = get_b(C())
self.assertTrue(self.w.add(b_a))
self.assertTrue(self.w2.add(b_c))
sent_messages = [
json.loads(m.get_body())
for m in self.w._worker_history_impl._queue.messages
]
event_messages = self._parse_task_events(sent_messages)
# Verify missing event
dependency_missing_event = event_messages.get(Event.DEPENDENCY_MISSING)
self.assertEquals(1, len(dependency_missing_event))
self.assertEquals('A()', dependency_missing_event[0]['task']['id'])
sent_messages2 = [
json.loads(m.get_body())
for m in self.w2._worker_history_impl._queue.messages
]
event_messages2 = self._parse_task_events(sent_messages2)
# Verify present event
dependency_present_event = event_messages2.get(
Event.DEPENDENCY_PRESENT)
self.assertEquals(1, len(dependency_present_event))
self.assertEquals('C()', dependency_present_event[0]['task']['id'])
def test_broken_task(self):
# see central_planner_test.CentralPlannerTest.test_remove_dep
class A(object):
def complete(self):
return False
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = False
def complete(self):
return self.has_run
b = B()
b.has_run = False
try:
self.assertTrue(self.w.add(b))
except:
pass
sent_messages = [
json.loads(m.get_body())
for m in self.w._worker_history_impl._queue.messages
]
event_messages = self._parse_task_events(sent_messages)
# Verify broken event
broken_event = event_messages.get(Event.BROKEN_TASK)
self.assertEquals(1, len(broken_event))
self.assertEquals('B()', broken_event[0]['task']['id'])
self.assertEquals("Exception('requires() must return Task objects',)",
broken_event[0]['exception'])
class AssistantTest(unittest.TestCase):
def run(self, result=None):
self.sch = Scheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.assistant = Worker(scheduler=self.sch,
worker_id='Y',
assistant=True)
with Worker(scheduler=self.sch, worker_id='X') as w:
self.w = w
super(AssistantTest, self).run(result)
def test_get_work(self):
d = Dummy2Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assistant.run()
self.assertTrue(d.complete())
def test_bad_job_type(self):
class Dummy3Task(Dummy2Task):
task_family = 'UnknownTaskFamily'
d = Dummy3Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assertFalse(self.assistant.run())
self.assertFalse(d.complete())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()),
[d.task_id])
def test_unimported_job_type(self):
MODULE_CONTENTS = b'''
import luigi
class UnimportedTask(luigi.Task):
def complete(self):
return False
'''
class NotImportedTask(luigi.Task):
task_family = 'UnimportedTask'
task_module = None
task = NotImportedTask()
# verify that it can't run the task without the module info necessary to import it
self.w.add(task)
self.assertFalse(self.assistant.run())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()),
[task.task_id])
# check that it can import with the right module
with temporary_unloaded_module(MODULE_CONTENTS) as task.task_module:
self.w.add(task)
self.assertTrue(self.assistant.run())
self.assertEqual(list(self.sch.task_list('DONE', '').keys()),
[task.task_id])
def setUp(self):
self.sch = CentralPlannerScheduler()
self.w = Worker(scheduler=self.sch)
def setUp(self):
super(WorkerEmailTest, self).setUp()
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.worker = Worker(scheduler=sch, worker_id="foo")
def setUp(self):
self.sch = mock.Mock()
self.w = Worker(scheduler=self.sch, worker_id='x')
class ParallelSchedulingTest(unittest.TestCase):
def setUp(self):
self.sch = mock.Mock()
self.w = Worker(scheduler=self.sch, worker_id='x')
def added_tasks(self, status):
return [
kw['task_id'] for args, kw in self.sch.add_task.call_args_list
if kw['status'] == status
]
def test_number_of_processes(self):
import multiprocessing
real_pool = multiprocessing.Pool(1)
with mock.patch('multiprocessing.Pool') as mocked_pool:
mocked_pool.return_value = real_pool
self.w.add(OverlappingSelfDependenciesTask(n=1, k=1),
multiprocess=True,
processes=1234)
mocked_pool.assert_called_once_with(processes=1234)
def test_zero_processes(self):
import multiprocessing
real_pool = multiprocessing.Pool(1)
with mock.patch('multiprocessing.Pool') as mocked_pool:
mocked_pool.return_value = real_pool
self.w.add(OverlappingSelfDependenciesTask(n=1, k=1),
multiprocess=True,
processes=0)
mocked_pool.assert_called_once_with(processes=None)
def test_children_terminated(self):
before_children = running_children()
with pause_gc():
self.w.add(
OverlappingSelfDependenciesTask(5, 2),
multiprocess=True,
)
self.assertLessEqual(running_children(), before_children)
def test_multiprocess_scheduling_with_overlapping_dependencies(self):
self.w.add(OverlappingSelfDependenciesTask(5, 2), True)
self.assertEqual(15, self.sch.add_task.call_count)
self.assertEqual(
set((
OverlappingSelfDependenciesTask(n=1, k=1).task_id,
OverlappingSelfDependenciesTask(n=2, k=1).task_id,
OverlappingSelfDependenciesTask(n=2, k=2).task_id,
OverlappingSelfDependenciesTask(n=3, k=1).task_id,
OverlappingSelfDependenciesTask(n=3, k=2).task_id,
OverlappingSelfDependenciesTask(n=4, k=1).task_id,
OverlappingSelfDependenciesTask(n=4, k=2).task_id,
OverlappingSelfDependenciesTask(n=5, k=2).task_id,
)), set(self.added_tasks('PENDING')))
self.assertEqual(
set((
OverlappingSelfDependenciesTask(n=0, k=0).task_id,
OverlappingSelfDependenciesTask(n=0, k=1).task_id,
OverlappingSelfDependenciesTask(n=1, k=0).task_id,
OverlappingSelfDependenciesTask(n=1, k=2).task_id,
OverlappingSelfDependenciesTask(n=2, k=0).task_id,
OverlappingSelfDependenciesTask(n=3, k=0).task_id,
OverlappingSelfDependenciesTask(n=4, k=0).task_id,
)), set(self.added_tasks('DONE')))
@mock.patch('luigi.notifications.send_error_email')
def test_raise_exception_in_complete(self, send):
self.w.add(ExceptionCompleteTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(UNKNOWN, self.sch.add_task.call_args[1]['status'])
self.assertFalse(self.sch.add_task.call_args[1]['runnable'])
self.assertTrue('assert False' in send.call_args[0][1])
@mock.patch('luigi.notifications.send_error_email')
def test_raise_unpicklable_exception_in_complete(self, send):
# verify exception can't be pickled
self.assertRaises(Exception, UnpicklableExceptionTask().complete)
try:
UnpicklableExceptionTask().complete()
except Exception as e:
ex = e
self.assertRaises((pickle.PicklingError, AttributeError), pickle.dumps,
ex)
# verify this can run async
self.w.add(UnpicklableExceptionTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(UNKNOWN, self.sch.add_task.call_args[1]['status'])
self.assertFalse(self.sch.add_task.call_args[1]['runnable'])
self.assertTrue('raise UnpicklableException()' in send.call_args[0][1])
@mock.patch('luigi.notifications.send_error_email')
def test_raise_exception_in_requires(self, send):
self.w.add(ExceptionRequiresTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(UNKNOWN, self.sch.add_task.call_args[1]['status'])
self.assertFalse(self.sch.add_task.call_args[1]['runnable'])
def test_default_multiprocessing_behavior(self):
with Worker(worker_processes=1) as worker:
task = DummyTask("a")
task_process = worker._create_task_process(task)
self.assertFalse(task_process.use_multiprocessing)
def test_force_multiprocessing(self):
with Worker(worker_processes=1, force_multiprocessing=True) as worker:
task = DummyTask("a")
task_process = worker._create_task_process(task)
self.assertTrue(task_process.use_multiprocessing)
def test_interleaved_workers(self):
class A(DummyTask):
pass
a = A()
class B(DummyTask):
def requires(self):
return a
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEquals(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
self.assertTrue(w.add(b))
self.assertTrue(w2.add(eb))
logging.debug("RUNNING BROKEN WORKER")
self.assertTrue(w2.run())
self.assertFalse(a.complete())
self.assertFalse(b.complete())
logging.debug("RUNNING FUNCTIONAL WORKER")
self.assertTrue(w.run())
self.assertTrue(a.complete())
self.assertTrue(b.complete())
w.stop()
w2.stop()
def test_no_task_limit(self):
w = Worker()
t = ForkBombTask(4, 2)
w.add(t)
w.run()
self.assertTrue(t.complete())
class ExceptionFormatTest(unittest.TestCase):
def setUp(self):
self.sch = CentralPlannerScheduler()
self.w = Worker(scheduler=self.sch)
def tear_down(self):
self.w.stop()
def test_fail_run(self):
task = FailRunTask(foo='foo', bar='bar')
self._run_task(task)
def test_fail_run_html(self):
task = FailRunTask(foo='foo', bar='bar')
self._run_task_html(task)
def test_fail_schedule(self):
task = FailSchedulingTask(foo='foo', bar='bar')
self._run_task(task)
def test_fail_schedule_html(self):
task = FailSchedulingTask(foo='foo', bar='bar')
self._run_task_html(task)
@with_config({
'core': {
'error-email': '*****@*****.**',
'email-prefix': '[TEST] '
}
})
@mock.patch('luigi.notifications.send_error_email')
def _run_task(self, task, mock_send):
self.w.add(task)
self.w.run()
self.assertEqual(mock_send.call_count, 1)
args, kwargs = mock_send.call_args
self._check_subject(args[0], task)
self._check_body(args[1], task, html=False)
@with_config({
'core': {
'error-email': '*****@*****.**',
'email-prefix': '[TEST] ',
'email-type': 'html'
}
})
@mock.patch('luigi.notifications.send_error_email')
def _run_task_html(self, task, mock_send):
self.w.add(task)
self.w.run()
self.assertEqual(mock_send.call_count, 1)
args, kwargs = mock_send.call_args
self._check_subject(args[0], task)
self._check_body(args[1], task, html=True)
def _check_subject(self, subject, task):
self.assertIn(task.task_id, subject)
def _check_body(self, body, task, html=False):
if html:
self.assertIn('<th>name</th><td>{}</td>'.format(task.task_family),
body)
self.assertIn('<div class="highlight"', body)
self.assertIn('Oops!', body)
for param, value in task.param_kwargs.items():
self.assertIn('<th>{}</th><td>{}</td>'.format(param, value),
body)
else:
self.assertIn('Name: {}\n'.format(task.task_family), body)
self.assertIn('Parameters:\n', body)
self.assertIn('TestException: Oops!', body)
for param, value in task.param_kwargs.items():
self.assertIn('{}: {}\n'.format(param, value), body)
@with_config({"core": {"error-email": "[email protected]"}})
def testEmailRecipients(self):
six.assertCountEqual(self, notifications._email_recipients(),
["[email protected]"])
six.assertCountEqual(self, notifications._email_recipients("[email protected]"),
["[email protected]", "[email protected]"])
six.assertCountEqual(
self, notifications._email_recipients(["[email protected]", "[email protected]"]),
["[email protected]", "[email protected]", "[email protected]"])
@with_config({"core": {}}, replace_sections=True)
def testEmailRecipientsNoConfig(self):
six.assertCountEqual(self, notifications._email_recipients(), [])
six.assertCountEqual(self, notifications._email_recipients("[email protected]"),
["[email protected]"])
six.assertCountEqual(
self, notifications._email_recipients(["[email protected]", "[email protected]"]),
["[email protected]", "[email protected]"])
def test_task_limit_not_exceeded(self):
w = Worker()
t = ForkBombTask(3, 2)
w.add(t)
w.run()
self.assertTrue(t.complete())
class ParallelSchedulingTest(unittest.TestCase):
def setUp(self):
self.sch = mock.Mock()
self.w = Worker(scheduler=self.sch, worker_id='x')
def added_tasks(self, status):
return [
kw['task_id'] for args, kw in self.sch.add_task.call_args_list
if kw['status'] == status
]
def test_multiprocess_scheduling_with_overlapping_dependencies(self):
self.w.add(OverlappingSelfDependenciesTask(5, 2), True)
self.assertEqual(15, self.sch.add_task.call_count)
self.assertEqual(
set((
'OverlappingSelfDependenciesTask(n=1, k=1)',
'OverlappingSelfDependenciesTask(n=2, k=1)',
'OverlappingSelfDependenciesTask(n=2, k=2)',
'OverlappingSelfDependenciesTask(n=3, k=1)',
'OverlappingSelfDependenciesTask(n=3, k=2)',
'OverlappingSelfDependenciesTask(n=4, k=1)',
'OverlappingSelfDependenciesTask(n=4, k=2)',
'OverlappingSelfDependenciesTask(n=5, k=2)',
)), set(self.added_tasks('PENDING')))
self.assertEqual(
set((
'OverlappingSelfDependenciesTask(n=0, k=0)',
'OverlappingSelfDependenciesTask(n=0, k=1)',
'OverlappingSelfDependenciesTask(n=1, k=0)',
'OverlappingSelfDependenciesTask(n=1, k=2)',
'OverlappingSelfDependenciesTask(n=2, k=0)',
'OverlappingSelfDependenciesTask(n=3, k=0)',
'OverlappingSelfDependenciesTask(n=4, k=0)',
)), set(self.added_tasks('DONE')))
@mock.patch('luigi.notifications.send_error_email')
def test_raise_exception_in_complete(self, send):
self.w.add(ExceptionCompleteTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(0, self.sch.add_task.call_count)
self.assertTrue('assert False' in send.call_args[0][1])
@mock.patch('luigi.notifications.send_error_email')
def test_raise_unpicklable_exception_in_complete(self, send):
# verify exception can't be pickled
self.assertRaises(Exception, UnpicklableExceptionTask().complete)
try:
UnpicklableExceptionTask().complete()
except Exception as e:
ex = e
self.assertRaises(pickle.PicklingError, pickle.dumps, ex)
# verify this can run async
self.w.add(UnpicklableExceptionTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(0, self.sch.add_task.call_count)
self.assertTrue('raise UnpicklableException()' in send.call_args[0][1])
@mock.patch('luigi.notifications.send_error_email')
def test_raise_exception_in_requires(self, send):
self.w.add(ExceptionRequiresTask(), multiprocess=True)
send.check_called_once()
self.assertEqual(0, self.sch.add_task.call_count)
class MultiprocessWorkerTest(unittest.TestCase):
def setUp(self):
self.scheduler = RemoteScheduler()
self.scheduler.add_worker = Mock()
self.scheduler.add_task = Mock()
self.worker = Worker(scheduler=self.scheduler,
worker_id='X',
worker_processes=2)
def tearDown(self):
self.worker.stop()
def gw_res(self, pending, task_id):
return dict(n_pending_tasks=pending,
task_id=task_id,
running_tasks=0,
n_unique_pending=0)
def test_positive_path(self):
a = DummyTask("a")
b = DummyTask("b")
class MultipleRequirementTask(DummyTask):
def requires(self):
return [a, b]
c = MultipleRequirementTask("C")
self.assertTrue(self.worker.add(c))
self.scheduler.get_work = Mock(side_effect=[
self.gw_res(3, str(a)),
self.gw_res(2, str(b)),
self.gw_res(1, str(c)),
self.gw_res(0, None),
self.gw_res(0, None)
])
self.assertTrue(self.worker.run())
self.assertTrue(c.has_run)
def test_path_with_task_failures(self):
class FailingTask(DummyTask):
def run(self):
raise Exception("I am failing")
a = FailingTask("a")
b = FailingTask("b")
class MultipleRequirementTask(DummyTask):
def requires(self):
return [a, b]
c = MultipleRequirementTask("C")
self.assertTrue(self.worker.add(c))
self.scheduler.get_work = Mock(side_effect=[
self.gw_res(3, str(a)),
self.gw_res(2, str(b)),
self.gw_res(1, str(c)),
self.gw_res(0, None),
self.gw_res(0, None)
])
self.assertFalse(self.worker.run())
def setUp(self):
# InstanceCache.disable()
self.sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.w2 = Worker(scheduler=self.sch, worker_id='Y')
self.time = time.time
class AssistantTest(unittest.TestCase):
def setUp(self):
self.sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.assistant = Worker(scheduler=self.sch,
worker_id='Y',
assistant=True)
def test_get_work(self):
d = Dummy2Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assistant.run()
self.assertTrue(d.complete())
def test_bad_job_type(self):
class Dummy3Task(Dummy2Task):
task_family = 'UnknownTaskFamily'
d = Dummy3Task('123')
self.w.add(d)
self.assertFalse(d.complete())
self.assertFalse(self.assistant.run())
self.assertFalse(d.complete())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()),
[str(d)])
def test_unimported_job_type(self):
class NotImportedTask(luigi.Task):
task_family = 'UnimportedTask'
task_module = None
task = NotImportedTask()
# verify that it can't run the task without the module info necessary to import it
self.w.add(task)
self.assertFalse(self.assistant.run())
self.assertEqual(list(self.sch.task_list('FAILED', '').keys()),
['UnimportedTask()'])
# check that it can import with the right module
task.task_module = 'dummy_test_module.not_imported'
self.w.add(task)
self.assertTrue(self.assistant.run())
self.assertEqual(list(self.sch.task_list('DONE', '').keys()),
['UnimportedTask()'])
class WorkerTest(unittest.TestCase):
def setUp(self):
# InstanceCache.disable()
self.sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.w2 = Worker(scheduler=self.sch, worker_id='Y')
self.time = time.time
def tearDown(self):
if time.time != self.time:
time.time = self.time
def setTime(self, t):
time.time = lambda: t
def test_dep(self):
class A(Task):
def run(self):
self.has_run = True
def complete(self):
return self.has_run
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.w.add(b)
self.w.run()
self.assertTrue(a.has_run)
self.assertTrue(b.has_run)
def test_external_dep(self):
class A(ExternalTask):
def complete(self):
return False
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.w.add(b)
self.w.run()
self.assertFalse(a.has_run)
self.assertFalse(b.has_run)
def test_fail(self):
class A(Task):
def run(self):
self.has_run = True
raise Exception()
def complete(self):
return self.has_run
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.w.add(b)
self.w.run()
self.assertTrue(a.has_run)
self.assertFalse(b.has_run)
def test_unknown_dep(self):
# see central_planner_test.CentralPlannerTest.test_remove_dep
class A(ExternalTask):
def complete(self):
return False
class C(Task):
def complete(self):
return True
def get_b(dep):
class B(Task):
def requires(self):
return dep
def run(self):
self.has_run = True
def complete(self):
return False
b = B()
b.has_run = False
return b
b_a = get_b(A())
b_c = get_b(C())
self.w.add(b_a)
# So now another worker goes in and schedules C -> B
# This should remove the dep A -> B but will screw up the first worker
self.w2.add(b_c)
self.w.run(
) # should not run anything - the worker should detect that A is broken
self.assertFalse(b_a.has_run)
# not sure what should happen??
# self.w2.run() # should run B since C is fulfilled
# self.assertTrue(b_c.has_run)
def test_interleaved_workers(self):
class A(DummyTask):
pass
a = A()
class B(DummyTask):
def requires(self):
return a
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEquals(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
w.add(b)
w2.add(eb)
logging.debug("RUNNING BROKEN WORKER")
w2.run()
self.assertFalse(a.complete())
self.assertFalse(b.complete())
logging.debug("RUNNING FUNCTIONAL WORKER")
w.run()
self.assertTrue(a.complete())
self.assertTrue(b.complete())
def test_interleaved_workers2(self):
# two tasks without dependencies, one external, one not
class B(DummyTask):
pass
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEquals(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
w2.add(eb)
w.add(b)
w2.run()
self.assertFalse(b.complete())
w.run()
self.assertTrue(b.complete())
def setUp(self):
self.sch = CentralPlannerScheduler(retry_delay=100, remove_delay=1000, worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.assistant = Worker(scheduler=self.sch, worker_id='Y', assistant=True)
def test_interleaved_workers3(self):
class A(DummyTask):
def run(self):
logging.debug('running A')
time.sleep(0.1)
super(A, self).run()
a = A()
class B(DummyTask):
def requires(self):
return a
def run(self):
logging.debug('running B')
super(B, self).run()
b = B()
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X', keep_alive=True)
w2 = Worker(scheduler=sch,
worker_id='Y',
keep_alive=True,
wait_interval=0.1)
self.assertTrue(w.add(a))
self.assertTrue(w2.add(b))
threading.Thread(target=w.run).start()
self.assertTrue(w2.run())
self.assertTrue(a.complete())
self.assertTrue(b.complete())
w.stop()
w2.stop()
class WorkerEmailTest(unittest.TestCase):
def setUp(self):
super(WorkerEmailTest, self).setUp()
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.worker = Worker(scheduler=sch, worker_id="foo")
def tearDown(self):
self.worker.stop()
@email_patch
def test_connection_error(self, emails):
sch = RemoteScheduler('http://tld.invalid:1337', connect_timeout=1)
worker = Worker(scheduler=sch)
self.waits = 0
def dummy_wait():
self.waits += 1
sch._wait = dummy_wait
class A(DummyTask):
pass
a = A()
self.assertEqual(emails, [])
worker.add(a)
self.assertEqual(self.waits, 2) # should attempt to add it 3 times
self.assertNotEquals(emails, [])
self.assertTrue(
emails[0].find("Luigi: Framework error while scheduling %s" %
(a, )) != -1)
worker.stop()
@email_patch
def test_complete_error(self, emails):
class A(DummyTask):
def complete(self):
raise Exception("b0rk")
a = A()
self.assertEqual(emails, [])
self.worker.add(a)
self.assertTrue(emails[0].find("Luigi: %s failed scheduling" %
(a, )) != -1)
self.worker.run()
self.assertTrue(emails[0].find("Luigi: %s failed scheduling" %
(a, )) != -1)
self.assertFalse(a.has_run)
@email_patch
def test_complete_return_value(self, emails):
class A(DummyTask):
def complete(self):
pass # no return value should be an error
a = A()
self.assertEqual(emails, [])
self.worker.add(a)
self.assertTrue(emails[0].find("Luigi: %s failed scheduling" %
(a, )) != -1)
self.worker.run()
self.assertTrue(emails[0].find("Luigi: %s failed scheduling" %
(a, )) != -1)
self.assertFalse(a.has_run)
@email_patch
def test_run_error(self, emails):
class A(luigi.Task):
def complete(self):
return False
def run(self):
raise Exception("b0rk")
a = A()
self.worker.add(a)
self.assertEqual(emails, [])
self.worker.run()
self.assertTrue(emails[0].find("Luigi: %s FAILED" % (a, )) != -1)
@email_patch
def test_no_error(self, emails):
class A(DummyTask):
pass
a = A()
self.assertEqual(emails, [])
self.worker.add(a)
self.assertEqual(emails, [])
self.worker.run()
self.assertEqual(emails, [])
self.assertTrue(a.complete())
def test_ping_thread_shutdown(self):
with Worker(ping_interval=0.01) as w:
self.assertTrue(w._keep_alive_thread.is_alive())
self.assertFalse(w._keep_alive_thread.is_alive())
def test_ping_thread_shutdown(self):
w = Worker(ping_interval=0.01)
self.assertTrue(w._keep_alive_thread.is_alive())
w.stop() # should stop within 0.01 s
self.assertFalse(w._keep_alive_thread.is_alive())
class WorkerTest(unittest.TestCase):
def setUp(self):
# InstanceCache.disable()
self.sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
self.w = Worker(scheduler=self.sch, worker_id='X')
self.w2 = Worker(scheduler=self.sch, worker_id='Y')
self.time = time.time
def tearDown(self):
if time.time != self.time:
time.time = self.time
self.w.stop()
self.w2.stop()
def setTime(self, t):
time.time = lambda: t
def test_dep(self):
class A(Task):
def run(self):
self.has_run = True
def complete(self):
return self.has_run
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.assertTrue(self.w.add(b))
self.assertTrue(self.w.run())
self.assertTrue(a.has_run)
self.assertTrue(b.has_run)
def test_external_dep(self):
class A(ExternalTask):
def complete(self):
return False
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.assertTrue(self.w.add(b))
self.assertTrue(self.w.run())
self.assertFalse(a.has_run)
self.assertFalse(b.has_run)
def test_fail(self):
class A(Task):
def run(self):
self.has_run = True
raise Exception()
def complete(self):
return self.has_run
a = A()
class B(Task):
def requires(self):
return a
def run(self):
self.has_run = True
def complete(self):
return self.has_run
b = B()
a.has_run = False
b.has_run = False
self.assertTrue(self.w.add(b))
self.assertFalse(self.w.run())
self.assertTrue(a.has_run)
self.assertFalse(b.has_run)
def test_unknown_dep(self):
# see central_planner_test.CentralPlannerTest.test_remove_dep
class A(ExternalTask):
def complete(self):
return False
class C(Task):
def complete(self):
return True
def get_b(dep):
class B(Task):
def requires(self):
return dep
def run(self):
self.has_run = True
def complete(self):
return False
b = B()
b.has_run = False
return b
b_a = get_b(A())
b_c = get_b(C())
self.assertTrue(self.w.add(b_a))
# So now another worker goes in and schedules C -> B
# This should remove the dep A -> B but will screw up the first worker
self.assertTrue(self.w2.add(b_c))
self.assertFalse(
self.w.run()
) # should not run anything - the worker should detect that A is broken
self.assertFalse(b_a.has_run)
# not sure what should happen??
# self.w2.run() # should run B since C is fulfilled
# self.assertTrue(b_c.has_run)
def test_unfulfilled_dep(self):
class A(Task):
def complete(self):
return self.done
def run(self):
self.done = True
def get_b(a):
class B(A):
def requires(self):
return a
b = B()
b.done = False
a.done = True
return b
a = A()
b = get_b(a)
self.assertTrue(self.w.add(b))
a.done = False
self.w.run()
self.assertTrue(a.complete())
self.assertTrue(b.complete())
def test_avoid_infinite_reschedule(self):
class A(Task):
def complete(self):
return False
class B(Task):
def complete(self):
return False
def requires(self):
return A()
self.assertTrue(self.w.add(B()))
self.assertFalse(self.w.run())
def test_interleaved_workers(self):
class A(DummyTask):
pass
a = A()
class B(DummyTask):
def requires(self):
return a
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEquals(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
self.assertTrue(w.add(b))
self.assertTrue(w2.add(eb))
logging.debug("RUNNING BROKEN WORKER")
self.assertTrue(w2.run())
self.assertFalse(a.complete())
self.assertFalse(b.complete())
logging.debug("RUNNING FUNCTIONAL WORKER")
self.assertTrue(w.run())
self.assertTrue(a.complete())
self.assertTrue(b.complete())
w.stop()
w2.stop()
def test_interleaved_workers2(self):
# two tasks without dependencies, one external, one not
class B(DummyTask):
pass
class ExternalB(ExternalTask):
task_family = "B"
def complete(self):
return False
b = B()
eb = ExternalB()
self.assertEquals(eb.task_id, "B()")
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X')
w2 = Worker(scheduler=sch, worker_id='Y')
self.assertTrue(w2.add(eb))
self.assertTrue(w.add(b))
self.assertTrue(w2.run())
self.assertFalse(b.complete())
self.assertTrue(w.run())
self.assertTrue(b.complete())
w.stop()
w2.stop()
def test_interleaved_workers3(self):
class A(DummyTask):
def run(self):
logging.debug('running A')
time.sleep(0.1)
super(A, self).run()
a = A()
class B(DummyTask):
def requires(self):
return a
def run(self):
logging.debug('running B')
super(B, self).run()
b = B()
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id='X', keep_alive=True)
w2 = Worker(scheduler=sch,
worker_id='Y',
keep_alive=True,
wait_interval=0.1)
self.assertTrue(w.add(a))
self.assertTrue(w2.add(b))
threading.Thread(target=w.run).start()
self.assertTrue(w2.run())
self.assertTrue(a.complete())
self.assertTrue(b.complete())
w.stop()
w2.stop()
def test_complete_exception(self):
"Tests that a task is still scheduled if its sister task crashes in the complete() method"
class A(DummyTask):
def complete(self):
raise Exception("doh")
a = A()
class C(DummyTask):
pass
c = C()
class B(DummyTask):
def requires(self):
return a, c
b = B()
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id="foo")
self.assertFalse(w.add(b))
self.assertTrue(w.run())
self.assertFalse(b.has_run)
self.assertTrue(c.has_run)
self.assertFalse(a.has_run)
w.stop()
def test_requires_exception(self):
class A(DummyTask):
def requires(self):
raise Exception("doh")
a = A()
class C(DummyTask):
pass
c = C()
class B(DummyTask):
def requires(self):
return a, c
b = B()
sch = CentralPlannerScheduler(retry_delay=100,
remove_delay=1000,
worker_disconnect_delay=10)
w = Worker(scheduler=sch, worker_id="foo")
self.assertFalse(w.add(b))
self.assertTrue(w.run())
self.assertFalse(b.has_run)
self.assertTrue(c.has_run)
self.assertFalse(a.has_run)
w.stop()
