def test_determine_error(self):
"""Tests that a pre-task successfully determines the correct error"""
scale_errors = [
ScaleDatabaseError(),
ScaleIOError(),
ScaleOperationalError(),
MissingSetting('')
]
for scale_error in scale_errors:
config = ExecutionConfiguration()
config.create_tasks(['pre'])
config.set_task_ids(self.job_exe.get_cluster_id())
task = PreTask('agent_1', self.job_exe, self.job_exe.job_type,
config)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
task.update(update)
update = job_test_utils.create_task_status_update(
task.id,
task.agent_id,
TaskStatusUpdate.FAILED,
now(),
exit_code=scale_error.exit_code)
error = task.determine_error(update)
self.assertEqual(scale_error.error_name, error.name)
def test_handle_failed_cleanup_task(self):
"""Tests handling failed cleanup task"""
when = now()
node = Node(self.node_agent, self.node)
node._last_heath_task = when
# Get initial cleanup task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
task_1_id = task.id
# Fail task after running and get different task next time
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FAILED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
# No new cleanup task right away
tasks = node.get_next_tasks(when + datetime.timedelta(seconds=5))
self.assertListEqual([], tasks)
self.assertFalse(node._is_initial_cleanup_completed)
# After error threshold, we should get new cleanup task
new_time = when + Node.CLEANUP_ERR_THRESHOLD + datetime.timedelta(
seconds=5)
node._last_heath_task = new_time # Get rid of health check task
task = node.get_next_tasks(new_time)[0]
self.assertNotEqual(task.id, task_1_id)
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
def test_lost_task(self):
"""Tests running through a job execution that has a task that gets lost"""
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
# Start, run, and complete pre-task
task = running_job_exe.start_next_task()
pre_task_started = now()
update = job_test_utils.create_task_status_update(task.id, 'agent', TaskStatusUpdate.RUNNING, pre_task_started)
running_job_exe.task_update(update)
pre_task_completed = pre_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(task.id, 'agent', TaskStatusUpdate.FINISHED,
pre_task_completed)
running_job_exe.task_update(update)
# Start job-task
task = running_job_exe.start_next_task()
job_task_id = task.id
job_task_started = pre_task_completed + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(task.id, 'agent', TaskStatusUpdate.RUNNING, job_task_started)
running_job_exe.task_update(update)
self.assertTrue(task.has_started)
# Lose task and make sure the same task is the next one to schedule again
when_lost = job_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(job_task_id, 'agent', TaskStatusUpdate.LOST, when_lost)
running_job_exe.task_update(update)
self.assertFalse(task.has_started)
task = running_job_exe.start_next_task()
self.assertEqual(job_task_id, task.id)
def test_lost_job_execution(self):
"""Tests running through a job execution that gets lost"""
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
# Start, run, and complete pre-task
task = running_job_exe.start_next_task()
pre_task_started = now()
update = job_test_utils.create_task_status_update(task.id, 'agent', TaskStatusUpdate.RUNNING, pre_task_started)
running_job_exe.task_update(update)
pre_task_completed = pre_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(task.id, 'agent', TaskStatusUpdate.FINISHED,
pre_task_completed)
running_job_exe.task_update(update)
# Start job-task and then execution gets lost
when_lost = pre_task_completed + timedelta(seconds=1)
job_task = running_job_exe.start_next_task()
lost_task = running_job_exe.execution_lost(when_lost)
self.assertEqual(job_task.id, lost_task.id)
self.assertTrue(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
job_exe = JobExecution.objects.get(id=self._job_exe_id)
self.assertEqual('FAILED', job_exe.status)
self.assertEqual(Error.objects.get_builtin_error('node-lost').id, job_exe.error_id)
self.assertEqual(when_lost, job_exe.ended)
def test_check_for_starvation(self):
"""Tests calling check_for_starvation() successfully"""
self.job_exe_mgr.schedule_job_exes([self.job_exe_1, self.job_exe_2], [])
# Start and complete first task of execution
task_1_launched = now()
task_1 = self.job_exe_1.start_next_task()
self.task_mgr.launch_tasks([task_1], task_1_launched)
task_1_started = task_1_launched + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.RUNNING, task_1_started)
self.task_mgr.handle_task_update(update)
self.job_exe_mgr.handle_task_update(update)
task_1_completed = task_1_started + timedelta(seconds=10)
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.FINISHED,
task_1_completed)
self.task_mgr.handle_task_update(update)
self.job_exe_mgr.handle_task_update(update)
# Check after the time threshold has passed and task 2 has still not been launched
check_time = task_1_completed + RESOURCE_STARVATION_THRESHOLD + timedelta(seconds=1)
finished_job_exes = self.job_exe_mgr.check_for_starvation(check_time)
# Check that execution 1 was failed for starvation
self.assertEqual(len(finished_job_exes), 1)
starved_job_exe = finished_job_exes[0]
self.assertEqual(starved_job_exe.id, self.job_exe_1.id)
self.assertEqual(starved_job_exe.status, 'FAILED')
self.assertEqual(starved_job_exe.error.name, 'resource-starvation')
self.assertEqual(starved_job_exe.finished, check_time)
def test_handle_failed_pull_task(self):
"""Tests handling failed Docker pull task"""
when = now()
node = Node(self.node_agent, self.node, self.scheduler)
node._last_health_task = when
node._initial_cleanup_completed()
node._update_state()
# Get Docker pull task
task = node.get_next_tasks(when)[0]
task_1_id = task.id
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
# Fail task after running
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.FAILED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
# No new pull task right away
tasks = node.get_next_tasks(when + datetime.timedelta(seconds=5))
self.assertListEqual([], tasks)
self.assertFalse(node._is_image_pulled)
# After error threshold, we should get new pull task
new_time = when + Node.IMAGE_PULL_ERR_THRESHOLD + datetime.timedelta(seconds=5)
node._last_health_task = new_time # Get rid of health check task
task = node.get_next_tasks(new_time)[0]
self.assertNotEqual(task.id, task_1_id)
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
def test_sync_with_database(self):
"""Tests calling sync_with_database() successfully"""
self.job_exe_mgr.schedule_job_exes([self.job_exe_1, self.job_exe_2], [])
task_1 = self.job_exe_1.start_next_task()
task_1_started = now() - timedelta(minutes=5)
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.RUNNING, task_1_started)
self.job_exe_mgr.handle_task_update(update)
# Cancel job_exe_1 and have manager sync with database
Job.objects.update_jobs_to_canceled([self.job_exe_1.job_id], now())
tasks_to_kill = self.job_exe_mgr.sync_with_database()
self.assertEqual(self.job_exe_1.status, 'CANCELED')
self.assertEqual(len(tasks_to_kill), 1)
self.assertEqual(tasks_to_kill[0].id, task_1.id)
# No message yet since we wait for the canceled task to be killed
self.assertListEqual(self.job_exe_mgr.get_messages(), [])
# Task killed
task_1_killed = task_1_started + timedelta(minutes=5)
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.KILLED, task_1_killed)
self.job_exe_mgr.handle_task_update(update)
# Make sure a create_job_exe_ends message exists for the canceled job execution
message = self.job_exe_mgr.get_messages()[0]
self.assertEqual(message.type, 'create_job_exe_ends')
self.assertEqual(message._job_exe_ends[0].job_exe_id, self.job_exe_1.id)
def test_handle_task_update(self):
"""Tests calling handle_task_update() successfully"""
self.job_exe_mgr.schedule_job_exes([self.job_exe_1, self.job_exe_2], [])
# Start task
task_1 = self.job_exe_1.start_next_task()
task_1_started = now() - timedelta(minutes=5)
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.RUNNING, task_1_started)
# Job execution is not finished, so None should be returned and no message is available
result = self.job_exe_mgr.handle_task_update(update)
self.assertIsNone(result)
self.assertListEqual(self.job_exe_mgr.get_messages(), [])
# Fail task
task_1_failed = task_1_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.FAILED, task_1_failed,
exit_code=1)
# Job execution is finished, so it should be returned and a create_job_exe_ends message is available
result = self.job_exe_mgr.handle_task_update(update)
self.assertEqual(self.job_exe_1.id, result.id)
message = self.job_exe_mgr.get_messages()[0]
self.assertEqual(message.type, 'create_job_exe_ends')
self.assertEqual(message._job_exe_ends[0].job_exe_id, self.job_exe_1.id)
def test_handle_regular_cleanup_task(self):
"""Tests handling a regular cleanup task"""
when = now()
node = Node(self.node_agent, self.node)
node._last_heath_task = when
node._initial_cleanup_completed()
node._image_pull_completed()
node._update_state()
# No task since there are no job executions to clean
self.assertListEqual([], node.get_next_tasks(when))
# Add job execution and complete task to clean it up
job_exe = RunningJobExecution(self.job_exe)
node.add_job_execution(job_exe)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
self.assertFalse(task.is_initial_cleanup)
self.assertListEqual(task.job_exes, [job_exe])
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
# No task since all job executions have been cleaned
self.assertListEqual([], node.get_next_tasks(when))
def test_handle_failed_health_task_low_docker_space(self):
"""Tests handling a failed health task where Docker has low disk space"""
when = now()
node = Node(self.node_agent, self.node)
node._initial_cleanup_completed()
node._image_pull_completed()
node._update_state()
# Get health task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
# Fail task with low Docker space exit code
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id,
task.agent_id,
TaskStatusUpdate.FAILED,
now(),
exit_code=HealthTask.LOW_DOCKER_SPACE_CODE)
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
# Check node state
self.assertEqual(node._state, Node.DEGRADED)
self.assertTrue(NodeConditions.LOW_DOCKER_SPACE_ERR.name in
node._conditions._active_errors)
def test_handle_killed_pull_task(self):
"""Tests handling killed cleanup task"""
when = now()
node = Node(self.node_agent, self.node)
node._last_heath_task = when
node._initial_cleanup_completed()
node._update_state()
# Get pull task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
task_1_id = task.id
# Kill task after running and get different task next time
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.KILLED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertFalse(node._is_image_pulled)
def test_handle_failed_health_task_bad_logstash(self):
"""Tests handling a failed health task where logstash is unreachable"""
when = now()
node = Node(self.node_agent, self.node)
node._initial_cleanup_completed()
node._image_pull_completed()
node._update_state()
# Get health task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
# Fail task with bad logstash exit code
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id,
task.agent_id,
TaskStatusUpdate.FAILED,
now(),
exit_code=HealthTask.BAD_LOGSTASH_CODE)
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
# Check node state
self.assertEqual(node._state, Node.DEGRADED)
self.assertTrue(NodeConditions.BAD_LOGSTASH_ERR.name in
node._conditions._active_errors)
def test_handle_completed_db_update_task(self):
"""Tests handling completed database update task"""
# Get database update task
when = now()
self.assertFalse(self.system_task_mgr._is_db_update_completed)
task = self.system_task_mgr.get_tasks_to_schedule(when)[0]
self.assertTrue(task.id.startswith(DB_UPDATE_TASK_ID_PREFIX))
task_1_id = task.id
# Schedule database update task and make sure there are no more system tasks
task.agent_id = self.agent_id
self.task_mgr.launch_tasks([task], now())
self.assertListEqual([],
self.system_task_mgr.get_tasks_to_schedule(now()))
# Complete task, verify no new tasks
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
self.system_task_mgr.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
self.task_mgr.handle_task_update(update)
self.system_task_mgr.handle_task_update(update)
self.assertListEqual([],
self.system_task_mgr.get_tasks_to_schedule(now()))
self.assertTrue(self.system_task_mgr._is_db_update_completed)
def test_handle_regular_cleanup_task(self):
"""Tests handling a regular cleanup task"""
node = Node(self.node_agent, self.node)
node.initial_cleanup_completed()
node_cleanup = NodeCleanup(node)
# No task since there are no job executions to clean
self.assertIsNone(node_cleanup.get_next_task())
# Add job execution and complete task to clean it up
job_exe = RunningJobExecution(self.job_exe)
node_cleanup.add_job_execution(job_exe)
task = node_cleanup.get_next_task()
self.assertIsNotNone(task)
self.assertFalse(task.is_initial_cleanup)
self.assertListEqual(task.job_exes, [job_exe])
task.launch(now())
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
node_cleanup.handle_task_update(update)
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
node_cleanup.handle_task_update(update)
# No task since all job executions have been cleaned
self.assertIsNone(node_cleanup.get_next_task())
def test_handle_task_update(self):
"""Tests calling TaskManager.handle_task_update()"""
task_id = 'task_1'
task_name = 'My Task'
agent_id = 'agent_1'
task_1 = ImplementedTask(task_id, task_name, agent_id)
when_launched = now()
manager = TaskManager()
manager.launch_tasks([task_1], when_launched)
when_finished = datetime.timedelta(seconds=1)
update_1 = job_test_utils.create_task_status_update(
task_1.id,
task_1.agent_id,
TaskStatusUpdate.FINISHED,
when=when_finished)
manager.handle_task_update(update_1)
self.assertTrue(task_1.has_ended)
self.assertEqual(task_1._ended, when_finished)
update_2 = job_test_utils.create_task_status_update(
'task_2', 'New Agent', TaskStatusUpdate.RUNNING, when=now())
manager.handle_task_update(update_2) # Should ignore, no error
def test_handle_killed_db_update_task(self):
"""Tests handling killed database update task"""
# Get database update task
when = now()
self.assertFalse(self.system_task_mgr._is_db_update_completed)
task = self.system_task_mgr.get_tasks_to_schedule(when)[0]
self.assertTrue(task.id.startswith(DB_UPDATE_TASK_ID_PREFIX))
task_1_id = task.id
# Kill task after running and get different task next time
task.agent_id = self.agent_id
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
self.system_task_mgr.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.KILLED, now())
self.task_mgr.handle_task_update(update)
self.system_task_mgr.handle_task_update(update)
task = self.system_task_mgr.get_tasks_to_schedule(when)[0]
self.assertTrue(task.id.startswith(DB_UPDATE_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertFalse(self.system_task_mgr._is_db_update_completed)
def test_handle_successful_health_task(self):
"""Tests handling the health task successfully"""
when = now()
node = Node(self.node_agent, self.node)
node._initial_cleanup_completed()
node._image_pull_completed()
node._update_state()
# Get health task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
self.assertEqual(task.agent_id, self.node_agent)
# Schedule health task and make sure no new task is ready
self.task_mgr.launch_tasks([task], now())
self.assertListEqual([], node.get_next_tasks(when))
self.assertTrue(node._conditions.is_health_check_normal)
# Complete pull task, verify no new task
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
self.assertListEqual([], node.get_next_tasks(when))
self.assertTrue(node._conditions.is_health_check_normal)
def test_handle_initial_cleanup_task(self):
"""Tests handling the initial cleanup task"""
when = now()
node = Node(self.node_agent, self.node)
node._last_heath_task = when
# Get initial cleanup task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
self.assertTrue(task.is_initial_cleanup)
self.assertEqual(task.agent_id, self.node_agent)
# Schedule initial cleanup and make sure no new task is ready
self.task_mgr.launch_tasks([task], now())
self.assertListEqual([], node.get_next_tasks(when))
self.assertFalse(node._is_initial_cleanup_completed)
# Complete initial clean up, verify no new cleanup task
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
for task in node.get_next_tasks(when):
self.assertFalse(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
self.assertTrue(node._is_initial_cleanup_completed)
def test_handle_successful_pull_task(self):
"""Tests handling the Docker pull task successfully"""
when = now()
node = Node(self.node_agent, self.node)
node._last_heath_task = when
node._initial_cleanup_completed()
node._update_state()
# Get Docker pull task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
self.assertEqual(task.agent_id, self.node_agent)
# Schedule pull task and make sure no new task is ready
self.task_mgr.launch_tasks([task], now())
self.assertListEqual([], node.get_next_tasks(when))
self.assertFalse(node._is_image_pulled)
# Complete pull task, verify no new task
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
self.assertListEqual([], node.get_next_tasks(when))
self.assertTrue(node._is_image_pulled)
# Node should now be ready
self.assertEqual(node._state, Node.READY)
def test_docker_terminated_error(self):
"""Tests running through a job execution where a Docker container terminates"""
# Clear error cache so test works correctly
CACHED_BUILTIN_ERRORS.clear()
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
# Start pre-task
task = running_job_exe.start_next_task()
pre_task_id = task.id
# Pre-task running
pre_task_started = now()
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.RUNNING,
pre_task_started)
running_job_exe.task_update(update)
# Pre-task Docker container terminates
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.FAILED, now(),
reason='REASON_EXECUTOR_TERMINATED')
running_job_exe.task_update(update)
# Check results
job_exe = JobExecution.objects.select_related().get(id=self._job_exe_id)
self.assertEqual(job_exe.status, 'FAILED')
self.assertEqual(job_exe.error.name, 'docker-terminated')
def test_lost_node(self):
"""Tests calling lost_node() successfully"""
self.job_exe_mgr.schedule_job_exes([self.job_exe_1, self.job_exe_2], [])
task_1 = self.job_exe_1.start_next_task()
task_1_started = now() - timedelta(minutes=5)
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.RUNNING, task_1_started)
self.job_exe_mgr.handle_task_update(update)
# Lose node and get lost task update
self.job_exe_mgr.lost_node(self.node_model_1.id, now())
update = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.LOST, task_1_started)
lost_job_exe = self.job_exe_mgr.handle_task_update(update)
self.assertEqual(lost_job_exe.id, self.job_exe_1.id)
self.assertEqual(lost_job_exe.status, 'FAILED')
self.assertEqual(lost_job_exe.error.name, 'node-lost')
# Make sure a create_job_exe_ends message and failed_jobs message exists for the lost job execution
messages = self.job_exe_mgr.get_messages()
self.assertEqual(len(messages), 2)
job_exe_ends_msg = messages[0]
self.assertEqual(job_exe_ends_msg.type, 'create_job_exe_ends')
self.assertEqual(job_exe_ends_msg._job_exe_ends[0].job_exe_id, self.job_exe_1.id)
failed_jobs_msg = messages[1]
self.assertEqual(failed_jobs_msg.type, 'failed_jobs')
self.assertTrue(get_builtin_error('node-lost').id in failed_jobs_msg._failed_jobs)
self.assertEqual(failed_jobs_msg._failed_jobs.values()[0][0].job_id, self.job_exe_1.job_id)
def test_handle_task_update(self):
"""Tests calling handle_task_update() successfully"""
self.job_exe_mgr.schedule_job_exes([self.job_exe_1, self.job_exe_2], [])
# Start tasks
task_1 = self.job_exe_1.start_next_task()
task_1_started = now() - timedelta(minutes=5)
update_1 = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.RUNNING,
task_1_started)
task_2 = self.job_exe_2.start_next_task()
# Shortcut job exe 2 so that there is only one task to complete
self.job_exe_2._remaining_tasks = []
task_2_started = now() - timedelta(minutes=5)
update_2 = job_test_utils.create_task_status_update(task_2.id, 'agent', TaskStatusUpdate.RUNNING,
task_2_started)
# Job execution is not finished, so None should be returned and no message is available
result = self.job_exe_mgr.handle_task_update(update_1)
self.assertIsNone(result)
result = self.job_exe_mgr.handle_task_update(update_2)
self.assertIsNone(result)
self.assertListEqual(self.job_exe_mgr.get_messages(), [])
# Fail task 1 for job exe 1
task_1_failed = task_1_started + timedelta(seconds=1)
update_1 = job_test_utils.create_task_status_update(task_1.id, 'agent', TaskStatusUpdate.FAILED, task_1_failed,
exit_code=1)
# Complete task 2 for job exe 2
task_2_completed = task_2_started + timedelta(seconds=1)
update_2 = job_test_utils.create_task_status_update(task_2.id, 'agent', TaskStatusUpdate.FINISHED,
task_2_completed)
# Job executions are finished, so they should be returned and a create_job_exe_ends message, a failed_jobs
# message, and a completed_jobs message is available
result = self.job_exe_mgr.handle_task_update(update_1)
self.assertEqual(self.job_exe_1.id, result.id)
result = self.job_exe_mgr.handle_task_update(update_2)
self.assertEqual(self.job_exe_2.id, result.id)
messages = self.job_exe_mgr.get_messages()
self.assertEqual(len(messages), 3)
job_exe_ends_msg = messages[0]
self.assertEqual(job_exe_ends_msg.type, 'create_job_exe_ends')
self.assertEqual(job_exe_ends_msg._job_exe_ends[0].job_exe_id, self.job_exe_1.id)
self.assertEqual(job_exe_ends_msg._job_exe_ends[1].job_exe_id, self.job_exe_2.id)
completed_jobs_msg = messages[1]
self.assertEqual(completed_jobs_msg.type, 'completed_jobs')
self.assertEqual(completed_jobs_msg._completed_jobs[0].job_id, self.job_exe_2.job_id)
failed_jobs_msg = messages[2]
self.assertEqual(failed_jobs_msg.type, 'failed_jobs')
self.assertEqual(failed_jobs_msg._failed_jobs.values()[0][0].job_id, self.job_exe_1.job_id)
def test_post_task_launch_error(self):
"""Tests running through a job execution where a post-task fails to launch"""
# Clear error cache so test works correctly
CACHED_BUILTIN_ERRORS.clear()
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
# Start pre-task
task = running_job_exe.start_next_task()
pre_task_id = task.id
# Pre-task running
pre_task_started = now()
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.RUNNING,
pre_task_started)
running_job_exe.task_update(update)
# Complete pre-task
pre_task_completed = pre_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.FINISHED,
pre_task_completed)
running_job_exe.task_update(update)
# Start job-task
task = running_job_exe.start_next_task()
job_task_id = task.id
# Job-task running
job_task_started = now()
update = job_test_utils.create_task_status_update(job_task_id, 'agent', TaskStatusUpdate.RUNNING,
job_task_started)
running_job_exe.task_update(update)
# Complete job-task
job_task_completed = job_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(job_task_id, 'agent', TaskStatusUpdate.FINISHED,
job_task_completed)
running_job_exe.task_update(update)
# Start post-task
task = running_job_exe.start_next_task()
post_task_id = task.id
# Post-task fails to launch
update = job_test_utils.create_task_status_update(post_task_id, 'agent', TaskStatusUpdate.FAILED, now())
running_job_exe.task_update(update)
# Check results
job_exe = JobExecution.objects.select_related().get(id=self._job_exe_id)
self.assertEqual(job_exe.status, 'FAILED')
self.assertEqual(job_exe.error.name, 'docker-task-launch')
def test_handle_lost_health_task(self):
"""Tests handling lost health task"""
when = now()
node = Node(self.node_agent, self.node, self.scheduler)
node._initial_cleanup_completed()
node._image_pull_completed()
node._update_state()
# Get pull task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
task_1_id = task.id
self.assertIsNotNone(task)
# Lose task without scheduling and get different task next time
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertTrue(node._conditions.is_health_check_normal)
# Lose task with scheduling and get different task next time
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertTrue(node._conditions.is_health_check_normal)
# Lose task after running and get different task next time
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertTrue(node._conditions.is_health_check_normal)
def test_handle_lost_pull_task(self):
"""Tests handling lost pull task"""
when = now()
node = Node(self.node_agent, self.node)
node._last_heath_task = when
node._initial_cleanup_completed()
node._update_state()
# Get pull task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
task_1_id = task.id
self.assertIsNotNone(task)
# Lose task without scheduling and get same task again
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
self.assertEqual(task.id, task_1_id)
self.assertFalse(node._is_image_pulled)
# Lose task with scheduling and get same task again
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
self.assertEqual(task.id, task_1_id)
self.assertFalse(node._is_image_pulled)
# Lose task after running and get same task again
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(PULL_TASK_ID_PREFIX))
self.assertEqual(task.id, task_1_id)
self.assertFalse(node._is_image_pulled)
def test_need_reconciliation(self):
"""Tests calling Task.need_reconciliation()"""
task_1 = ImplementedTask('task_1', 'Task 1', 'agent_id')
task_2 = ImplementedTask('task_2', 'Task 2', 'agent_id')
task_3 = ImplementedTask('task_3', 'Task 3', 'agent_id')
task_4 = ImplementedTask('task_4', 'Task 4', 'agent_id')
task_5 = ImplementedTask('task_5', 'Task 5', 'agent_id')
task_1_and_2_launch_time = now()
task_3_and_5_launch_time = task_1_and_2_launch_time + RUNNING_RECON_THRESHOLD
check_time = task_3_and_5_launch_time + datetime.timedelta(seconds=1)
check_time_2 = check_time + datetime.timedelta(seconds=1)
# Task 1 and 2 launch
task_1.launch(task_1_and_2_launch_time)
task_2.launch(task_1_and_2_launch_time)
# The reconciliation threshold has now expired
# Task 3 and 5 launches and a task update comes for task 2
task_3.launch(task_3_and_5_launch_time)
task_5.launch(task_3_and_5_launch_time)
update = job_test_utils.create_task_status_update(
task_2.id, 'agent_id', TaskStatusUpdate.RUNNING,
task_3_and_5_launch_time)
task_2.update(update)
# Task 5 gets force reconciliation call
task_5.force_reconciliation()
# A second later, we check for tasks needing reconciliation
# Task 1 was launched a while ago (exceeding threshold) so it should be reconciled
self.assertTrue(task_1.needs_reconciliation(check_time))
# Task 2 received an update 1 second ago so it should not be reconciled
self.assertFalse(task_2.needs_reconciliation(check_time))
# Task 3 was launched 1 second ago so it should not be reconciled
self.assertFalse(task_3.needs_reconciliation(check_time))
# Task 4 did not even launch so it should not be reconciled
self.assertFalse(task_4.needs_reconciliation(check_time))
# Task 5 had force_reconciliation() called so it should be reconciled
self.assertTrue(task_5.needs_reconciliation(check_time))
# Task 5 gets task update to clear force recon
update = job_test_utils.create_task_status_update(
task_5.id, 'agent_id', TaskStatusUpdate.RUNNING, check_time)
task_5.update(update)
# Task 5 received an update so force recon should be cleared and it should be not reconciled
self.assertFalse(task_5.needs_reconciliation(check_time_2))
def test_change_agent_id(self):
"""Tests the NodeManager where a node's agent ID changes"""
manager = CleanupManager()
node_1 = Node(self.node_agent_1, self.node_1)
node_2 = Node(self.node_agent_2, self.node_2)
manager.update_nodes([node_1, node_2])
tasks = manager.get_next_tasks()
task_1 = None
for task in tasks:
task.launch(now())
if task.agent_id == self.node_agent_1:
task_1 = task
# Node 1 changes agent ID
node_1.update_from_mesos(agent_id=self.node_agent_3)
manager.update_nodes([node_1, node_2])
# Should get new initial cleanup task for node 1
tasks = manager.get_next_tasks()
self.assertEqual(len(tasks), 1)
new_task_1 = tasks[0]
self.assertEqual(new_task_1.agent_id, self.node_agent_3)
# Task update comes back for original node 1 initial cleanup task, manager should ignore with no exception
update = job_test_utils.create_task_status_update(task_1.id, task_1.agent_id, TaskStatusUpdate.FAILED, now())
manager.handle_task_update(update)
def test_timed_out_system_job_task(self):
"""Tests running through a job execution where a system job task times out"""
ingest_job_type = Ingest.objects.get_ingest_job_type()
ingest_job_type.max_tries = 1
ingest_job_type.save()
job = job_test_utils.create_job(job_type=ingest_job_type, num_exes=1)
job_exe = job_test_utils.create_job_exe(job=job)
running_job_exe = RunningJobExecution(job_exe)
# Start job-task and then task times out
when_launched = now() + timedelta(seconds=1)
job_task_started = when_launched + timedelta(seconds=1)
when_timed_out = job_task_started + timedelta(seconds=1)
job_task = running_job_exe.start_next_task()
self.task_mgr.launch_tasks([job_task], when_launched)
update = job_test_utils.create_task_status_update(
job_task.id, 'agent', TaskStatusUpdate.RUNNING, job_task_started)
self.task_mgr.handle_task_update(update)
running_job_exe.task_update(update)
running_job_exe.execution_timed_out(job_task, when_timed_out)
self.assertTrue(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
job_exe = JobExecution.objects.get(id=job_exe.id)
self.assertEqual('FAILED', job_exe.status)
self.assertEqual('ingest-timeout', job_exe.error.name)
self.assertEqual(when_timed_out, job_exe.ended)
def test_job_exe_clean_task(self):
"""Tests the NodeManager where a cleanup task is returned to clean up a job execution"""
when = now()
node_mgr = NodeManager()
node_mgr.register_agents([self.agent_1, self.agent_2])
node_mgr.sync_with_database(scheduler_mgr.config)
cleanup_mgr = CleanupManager()
cleanup_mgr.update_nodes(node_mgr.get_nodes())
tasks = node_mgr.get_next_tasks(when)
task_mgr = TaskManager()
# Complete initial cleanup tasks
for task in tasks:
task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
task_mgr.handle_task_update(update)
node_mgr.handle_task_update(update)
# Mark image pull done to get rid of image tasks
for node in node_mgr.get_nodes():
node._image_pull_completed()
node._update_state()
job_exe = job_test_utils.create_running_job_exe(agent_id=self.agent_1,
node=self.node_1)
# Add a job execution to clean up and get the cleanup task for it
cleanup_mgr.add_job_execution(job_exe)
tasks = node_mgr.get_next_tasks(when)
self.assertEqual(len(tasks), 1)
task = tasks[0]
self.assertEqual(task.agent_id, self.agent_1.agent_id)
self.assertFalse(task.is_initial_cleanup)
self.assertEqual(len(task.job_exes), 1)
def test_get_tasks_to_kill(self):
"""Tests calling get_tasks_to_kill() successfully"""
# Start with 5 tasks
scheduler_mgr.config.num_message_handlers = 5
service = MessagingService()
tasks = service.get_tasks_to_schedule()
task_mgr.launch_tasks(tasks, now())
# Lower number of desired tasks to 3, should get 2 to kill
scheduler_mgr.config.num_message_handlers = 3
tasks_to_kill = service.get_tasks_to_kill()
self.assertEqual(len(tasks_to_kill), 2)
# Kill the 2 tasks
for task in tasks_to_kill:
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.KILLED, now())
task_mgr.handle_task_update(update)
service.handle_task_update(update)
self.assertEqual(service.get_actual_task_count(), 3)
# Increase desired tasks to 10, should not get any to kill
scheduler_mgr.config.num_message_handlers = 10
tasks_to_kill = service.get_tasks_to_kill()
self.assertEqual(len(tasks_to_kill), 0)
def test_general_algorithm_error(self):
"""Tests running through a job execution where the job-task has a general algorithm error (non-zero exit code)
"""
# Clear error cache so test works correctly
CACHED_BUILTIN_ERRORS.clear()
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
# Start pre-task
task = running_job_exe.start_next_task()
pre_task_id = task.id
# Pre-task running
pre_task_started = now()
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.RUNNING,
pre_task_started)
running_job_exe.task_update(update)
# Complete pre-task
pre_task_completed = pre_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.FINISHED,
pre_task_completed)
running_job_exe.task_update(update)
# Start job-task
task = running_job_exe.start_next_task()
job_task_id = task.id
# Job-task running
job_task_started = now()
update = job_test_utils.create_task_status_update(job_task_id, 'agent', TaskStatusUpdate.RUNNING,
job_task_started)
running_job_exe.task_update(update)
# Fail job-task
job_task_failed = job_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(job_task_id, 'agent', TaskStatusUpdate.FAILED,
job_task_failed, exit_code=1)
running_job_exe.task_update(update)
# Check results
job_exe = JobExecution.objects.select_related().get(id=self._job_exe_id)
self.assertEqual(job_exe.status, 'FAILED')
self.assertEqual(job_exe.error.name, 'algorithm-unknown')
def test_handle_lost_cleanup_tasks(self):
"""Tests handling lost cleanup tasks"""
when = now()
node = Node(self.node_agent, self.node, self.scheduler)
# Get initial cleanup task
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
task_1_id = task.id
# Lose task without scheduling and get different task next time
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertFalse(node._is_initial_cleanup_completed)
# Lose task with scheduling and get different task next time
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertFalse(node._is_initial_cleanup_completed)
# Lose task after running and get different task next time
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.LOST, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
task = node.get_next_tasks(when)[0]
self.assertTrue(task.id.startswith(CLEANUP_TASK_ID_PREFIX))
self.assertNotEqual(task.id, task_1_id)
self.assertFalse(node._is_initial_cleanup_completed)
def test_sync_with_database(self):
"""Tests calling sync_with_database() successfully"""
self.job_exe_mgr.schedule_job_exes([self.job_exe_1, self.job_exe_2],
[])
task_1 = self.job_exe_1.start_next_task()
task_1_started = now() - timedelta(minutes=5)
update = job_test_utils.create_task_status_update(
task_1.id, 'agent', TaskStatusUpdate.RUNNING, task_1_started)
self.job_exe_mgr.handle_task_update(update)
# Cancel job_exe_1 and job_exe_2 and have manager sync with database
Job.objects.update_jobs_to_canceled_old(
[self.job_exe_1.job_id, self.job_exe_2.job_id], now())
finished_job_exes = self.job_exe_mgr.sync_with_database()
self.assertEqual(self.job_exe_1.status, 'CANCELED')
self.assertFalse(self.job_exe_1.is_finished())
self.assertEqual(self.job_exe_2.status, 'CANCELED')
self.assertTrue(self.job_exe_2.is_finished())
# Only job_exe_2 is finished, job_exe_1 has a task to kill
self.assertEqual(len(finished_job_exes), 1)
self.assertEqual(finished_job_exes[0].id, self.job_exe_2.id)
# Make sure a create_job_exe_ends message exists for job_exe_2
message = self.job_exe_mgr.get_messages()[0]
self.assertEqual(message.type, 'create_job_exe_ends')
self.assertEqual(message._job_exe_ends[0].job_exe_id,
self.job_exe_2.id)
# Task killed for job_exe_1
task_1_killed = task_1_started + timedelta(minutes=5)
update = job_test_utils.create_task_status_update(
task_1.id, 'agent', TaskStatusUpdate.KILLED, task_1_killed)
self.job_exe_mgr.handle_task_update(update)
# Make sure a create_job_exe_ends message exists for job_exe_1
self.assertTrue(self.job_exe_1.is_finished())
message = self.job_exe_mgr.get_messages()[0]
self.assertEqual(message.type, 'create_job_exe_ends')
self.assertEqual(message._job_exe_ends[0].job_exe_id,
self.job_exe_1.id)
def test_handle_killed_task(self):
"""Tests handling killed cleanup task"""
node = Node(self.node_agent, self.node)
node_cleanup = NodeCleanup(node)
# Get initial cleanup task
task = node_cleanup.get_next_task()
task_1_id = task.id
self.assertIsNotNone(task)
# Kill task after running and get different task next time
task.launch(now())
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
node_cleanup.handle_task_update(update)
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.KILLED, now())
node_cleanup.handle_task_update(update)
task = node_cleanup.get_next_task()
self.assertIsNotNone(task)
self.assertNotEqual(task.id, task_1_id)
self.assertFalse(node.is_initial_cleanup_completed)
def test_timed_out_system_job_task(self):
"""Tests running through a job execution where a system job task times out"""
ingest_job_type = Ingest.objects.get_ingest_job_type()
ingest_job_type.max_tries = 1
ingest_job_type.save()
running_job_exe = job_test_utils.create_running_job_exe(
agent_id='agent_1', job_type=ingest_job_type, num_exes=1)
# Start job-task and then task times out
when_launched = now() + timedelta(seconds=1)
job_task_started = when_launched + timedelta(seconds=1)
when_timed_out = job_task_started + timedelta(seconds=1)
job_task = running_job_exe.start_next_task()
self.task_mgr.launch_tasks([job_task], when_launched)
update = job_test_utils.create_task_status_update(
job_task.id, 'agent', TaskStatusUpdate.RUNNING, job_task_started)
self.task_mgr.handle_task_update(update)
running_job_exe.task_update(update)
running_job_exe.execution_timed_out(job_task, when_timed_out)
self.assertFalse(running_job_exe.is_finished()
) # Not finished until killed task update arrives
self.assertEqual(running_job_exe.status, 'FAILED')
self.assertEqual(running_job_exe.error_category, 'SYSTEM')
self.assertEqual(running_job_exe.error.name, 'ingest-timeout')
self.assertEqual(running_job_exe.finished, when_timed_out)
self.assertFalse(running_job_exe.is_next_task_ready())
# Killed task update arrives, job execution is now finished
job_task_kill = when_timed_out + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(
job_task.id, 'agent', TaskStatusUpdate.KILLED, job_task_kill)
self.task_mgr.handle_task_update(update)
running_job_exe.task_update(update)
self.assertTrue(running_job_exe.is_finished())
self.assertEqual(running_job_exe.status, 'FAILED')
self.assertEqual(running_job_exe.error_category, 'SYSTEM')
self.assertEqual(running_job_exe.error.name, 'ingest-timeout')
self.assertEqual(running_job_exe.finished, when_timed_out)
self.assertFalse(running_job_exe.is_next_task_ready())
def test_handle_failed_health_task(self):
"""Tests handling failed health task"""
when = now()
node = Node(self.node_agent, self.node)
node._initial_cleanup_completed()
node._image_pull_completed()
node._update_state()
# Get health task
task = node.get_next_tasks(when)[0]
task_1_id = task.id
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
# Fail task after running
self.task_mgr.launch_tasks([task], now())
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.FAILED, now())
self.task_mgr.handle_task_update(update)
node.handle_task_update(update)
# Check node state
self.assertEqual(node._state, Node.DEGRADED)
self.assertTrue(NodeConditions.HEALTH_FAIL_ERR.name in
node._conditions._active_errors)
# No new health task right away
tasks = node.get_next_tasks(when + datetime.timedelta(seconds=5))
self.assertListEqual([], tasks)
self.assertFalse(node._conditions.is_health_check_normal)
# After error threshold, we should get new health task
new_time = when + Node.HEALTH_ERR_THRESHOLD + datetime.timedelta(
seconds=5)
task = node.get_next_tasks(new_time)[0]
self.assertNotEqual(task.id, task_1_id)
self.assertTrue(task.id.startswith(HEALTH_TASK_ID_PREFIX))
def test_canceled_job_execution(self):
"""Tests running through a job execution that gets canceled"""
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
# Start, run, and complete pre-task
task = running_job_exe.start_next_task()
pre_task_started = now()
update = job_test_utils.create_task_status_update(task.id, 'agent', TaskStatusUpdate.RUNNING, pre_task_started)
running_job_exe.task_update(update)
pre_task_completed = pre_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(task.id, 'agent', TaskStatusUpdate.FINISHED,
pre_task_completed)
running_job_exe.task_update(update)
# Start job-task and then execution gets canceled
job_task = running_job_exe.start_next_task()
canceled_task = running_job_exe.execution_canceled()
self.assertEqual(job_task.id, canceled_task.id)
self.assertTrue(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
def test_parsing_container_name(self):
"""Tests that a task successfully parses container name from a RUNNING task update"""
task_id = 'task_1'
task_name = 'My Task'
agent_id = 'agent_1'
container_name = 'container_1234'
data = {'Config': {'Env': ['DUMMY_ENV=DUMMY', 'MESOS_CONTAINER_NAME=' + container_name]}}
task = ImplementedTask(task_id, task_name, agent_id)
update = job_test_utils.create_task_status_update(task_id, agent_id, TaskStatusUpdate.RUNNING, now(), data=data)
task.update(update)
self.assertEqual(task.container_name, container_name)
def test_determine_error(self):
"""Tests that a pre-task successfully determines the correct error"""
scale_errors = [
ScaleDatabaseError(),
ScaleIOError(),
ScaleOperationalError(),
MissingSetting('')
]
for scale_error in scale_errors:
task = PreTask(self.job_exe)
update = job_test_utils.create_task_status_update(
task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
task.update(update)
update = job_test_utils.create_task_status_update(
task.id,
task.agent_id,
TaskStatusUpdate.FAILED,
now(),
exit_code=scale_error.exit_code)
error = task.determine_error(update)
self.assertEqual(scale_error.error_name, error.name)
def test_handle_lost_tasks(self):
"""Tests handling lost cleanup tasks"""
node = Node(self.node_agent, self.node)
node_cleanup = NodeCleanup(node)
# Get initial cleanup task
task = node_cleanup.get_next_task()
task_1_id = task.id
self.assertIsNotNone(task)
# Lose task without scheduling and get same task again
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.LOST, now())
node_cleanup.handle_task_update(update)
task = node_cleanup.get_next_task()
self.assertIsNotNone(task)
self.assertEqual(task.id, task_1_id)
self.assertFalse(node.is_initial_cleanup_completed)
# Lose task with scheduling and get same task again
task.launch(now())
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.LOST, now())
node_cleanup.handle_task_update(update)
task = node_cleanup.get_next_task()
self.assertIsNotNone(task)
self.assertEqual(task.id, task_1_id)
self.assertFalse(node.is_initial_cleanup_completed)
# Lose task after running and get same task again
task.launch(now())
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
node_cleanup.handle_task_update(update)
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.LOST, now())
node_cleanup.handle_task_update(update)
task = node_cleanup.get_next_task()
self.assertIsNotNone(task)
self.assertEqual(task.id, task_1_id)
self.assertFalse(node.is_initial_cleanup_completed)
def test_failed_normal_job_execution(self):
"""Tests running through a normal job execution that fails"""
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
self.assertFalse(running_job_exe.is_finished())
self.assertTrue(running_job_exe.is_next_task_ready())
# Start pre-task
task = running_job_exe.start_next_task()
pre_task_id = task.id
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Pre-task running
pre_task_started = now() - timedelta(minutes=5) # Lots of time so now() called at completion is in future
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.RUNNING,
pre_task_started)
running_job_exe.task_update(update)
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Fail pre-task
pre_task_failed = pre_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.FAILED,
pre_task_failed, exit_code=1)
running_job_exe.task_update(update)
self.assertTrue(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
job_exe = JobExecution.objects.get(id=self._job_exe_id)
self.assertEqual(pre_task_started, job_exe.pre_started)
self.assertEqual(pre_task_failed, job_exe.pre_completed)
self.assertEqual(1, job_exe.pre_exit_code)
self.assertEqual('FAILED', job_exe.status)
self.assertIsNotNone(job_exe.error_id)
self.assertGreater(job_exe.ended, pre_task_failed)
def test_handle_initial_cleanup_task(self):
"""Tests handling the initial cleanup task"""
node = Node(self.node_agent, self.node)
node_cleanup = NodeCleanup(node)
# Get initial cleanup task
task = node_cleanup.get_next_task()
self.assertIsNotNone(task)
self.assertTrue(task.is_initial_cleanup)
self.assertEqual(task.agent_id, self.node_agent)
# Schedule initial cleanup and make sure no new task is ready
task.launch(now())
self.assertIsNone(node_cleanup.get_next_task())
self.assertFalse(node.is_initial_cleanup_completed)
# Complete initial clean up, verify no new task
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.RUNNING, now())
node_cleanup.handle_task_update(update)
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
node_cleanup.handle_task_update(update)
self.assertIsNone(node_cleanup.get_next_task())
self.assertTrue(node.is_initial_cleanup_completed)
def test_no_job_exes_to_clean(self):
"""Tests the NodeManager where no cleanup tasks are returned due to no job executions to clean"""
manager = CleanupManager()
node_1 = Node(self.node_agent_1, self.node_1)
node_2 = Node(self.node_agent_2, self.node_2)
manager.update_nodes([node_1, node_2])
tasks = manager.get_next_tasks()
# Complete initial cleanup tasks
for task in tasks:
task.launch(now())
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
manager.handle_task_update(update)
tasks = manager.get_next_tasks()
self.assertListEqual(tasks, []) # No tasks since there are no job executions to clean up
def test_job_exe_clean_task(self):
"""Tests the NodeManager where a cleanup task is returned to clean up a job execution"""
manager = CleanupManager()
node_1 = Node(self.node_agent_1, self.node_1)
node_2 = Node(self.node_agent_2, self.node_2)
manager.update_nodes([node_1, node_2])
tasks = manager.get_next_tasks()
# Complete initial cleanup tasks
for task in tasks:
task.launch(now())
update = job_test_utils.create_task_status_update(task.id, task.agent_id, TaskStatusUpdate.FINISHED, now())
manager.handle_task_update(update)
# Add a job execution to clean up and get the cleanup task for it
manager.add_job_execution(RunningJobExecution(self.job_exe_1))
tasks = manager.get_next_tasks()
self.assertEqual(len(tasks), 1)
task = tasks[0]
self.assertEqual(task.agent_id, self.node_agent_1)
self.assertFalse(task.is_initial_cleanup)
self.assertEqual(len(task.job_exes), 1)
def test_successful_normal_job_execution(self):
"""Tests running through a normal job execution successfully"""
job_exe = JobExecution.objects.get_job_exe_with_job_and_job_type(self._job_exe_id)
running_job_exe = RunningJobExecution(job_exe)
self.assertFalse(running_job_exe.is_finished())
self.assertTrue(running_job_exe.is_next_task_ready())
# Start pre-task
task = running_job_exe.start_next_task()
pre_task_id = task.id
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Pre-task running
pre_task_started = now() - timedelta(minutes=5) # Lots of time so now() called at completion is in future
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.RUNNING,
pre_task_started)
running_job_exe.task_update(update)
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Pre-task sets updated command arguments
updated_commands_args = '-arg updated'
JobExecution.objects.filter(id=self._job_exe_id).update(command_arguments=updated_commands_args)
# Complete pre-task
pre_task_completed = pre_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(pre_task_id, 'agent', TaskStatusUpdate.FINISHED,
pre_task_completed, exit_code=1)
running_job_exe.task_update(update)
self.assertFalse(running_job_exe.is_finished())
self.assertTrue(running_job_exe.is_next_task_ready())
# Start job-task
task = running_job_exe.start_next_task()
job_task_id = task.id
self.assertEqual(task._command_arguments, updated_commands_args) # Make sure job task has updated command args
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Job-task running
job_task_started = pre_task_completed + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(job_task_id, 'agent', TaskStatusUpdate.RUNNING,
job_task_started)
running_job_exe.task_update(update)
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Complete job-task
job_task_completed = job_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(job_task_id, 'agent', TaskStatusUpdate.FINISHED,
job_task_completed, exit_code=2)
running_job_exe.task_update(update)
self.assertFalse(running_job_exe.is_finished())
self.assertTrue(running_job_exe.is_next_task_ready())
# Start post-task
task = running_job_exe.start_next_task()
post_task_id = task.id
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Post-task running
post_task_started = job_task_completed + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(post_task_id, 'agent', TaskStatusUpdate.RUNNING,
post_task_started)
running_job_exe.task_update(update)
self.assertFalse(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
# Complete post-task
post_task_completed = post_task_started + timedelta(seconds=1)
update = job_test_utils.create_task_status_update(post_task_id, 'agent', TaskStatusUpdate.FINISHED,
post_task_completed, exit_code=3)
running_job_exe.task_update(update)
self.assertTrue(running_job_exe.is_finished())
self.assertFalse(running_job_exe.is_next_task_ready())
job_exe = JobExecution.objects.get(id=self._job_exe_id)
self.assertEqual(pre_task_started, job_exe.pre_started)
self.assertEqual(pre_task_completed, job_exe.pre_completed)
self.assertEqual(1, job_exe.pre_exit_code)
self.assertEqual(job_task_started, job_exe.job_started)
self.assertEqual(job_task_completed, job_exe.job_completed)
self.assertEqual(2, job_exe.job_exit_code)
self.assertEqual(post_task_started, job_exe.post_started)
self.assertEqual(post_task_completed, job_exe.post_completed)
self.assertEqual(3, job_exe.post_exit_code)
self.assertEqual('COMPLETED', job_exe.status)
self.assertGreater(job_exe.ended, post_task_completed)
