blob: 184619e19f275bc3d579ee7822b55943074211c2 [file] [log] [blame] [edit]
#!/usr/bin/python
#pylint: disable-msg=C0111
import gc, time
import common
from autotest_lib.frontend import setup_django_environment
from autotest_lib.frontend.afe import frontend_test_utils
from autotest_lib.client.common_lib.test_utils import mock
from autotest_lib.client.common_lib.test_utils import unittest
from autotest_lib.database import database_connection
from autotest_lib.frontend.afe import models
from autotest_lib.scheduler import agent_task
from autotest_lib.scheduler import monitor_db, drone_manager, email_manager
from autotest_lib.scheduler import pidfile_monitor
from autotest_lib.scheduler import scheduler_config, gc_stats
from autotest_lib.scheduler import scheduler_lib
from autotest_lib.scheduler import scheduler_models
_DEBUG = False
class DummyAgentTask(object):
num_processes = 1
owner_username = 'my_user'
def get_drone_hostnames_allowed(self):
return None
class DummyAgent(object):
started = False
_is_done = False
host_ids = ()
queue_entry_ids = ()
def __init__(self):
self.task = DummyAgentTask()
def tick(self):
self.started = True
def is_done(self):
return self._is_done
def set_done(self, done):
self._is_done = done
class IsRow(mock.argument_comparator):
def __init__(self, row_id):
self.row_id = row_id
def is_satisfied_by(self, parameter):
return list(parameter)[0] == self.row_id
def __str__(self):
return 'row with id %s' % self.row_id
class IsAgentWithTask(mock.argument_comparator):
def __init__(self, task):
self._task = task
def is_satisfied_by(self, parameter):
if not isinstance(parameter, monitor_db.Agent):
return False
tasks = list(parameter.queue.queue)
if len(tasks) != 1:
return False
return tasks[0] == self._task
def _set_host_and_qe_ids(agent_or_task, id_list=None):
if id_list is None:
id_list = []
agent_or_task.host_ids = agent_or_task.queue_entry_ids = id_list
class BaseSchedulerTest(unittest.TestCase,
frontend_test_utils.FrontendTestMixin):
_config_section = 'AUTOTEST_WEB'
def _do_query(self, sql):
self._database.execute(sql)
def _set_monitor_stubs(self):
# Clear the instance cache as this is a brand new database.
scheduler_models.DBObject._clear_instance_cache()
self._database = (
database_connection.TranslatingDatabase.get_test_database(
translators=scheduler_lib._DB_TRANSLATORS))
self._database.connect(db_type='django')
self._database.debug = _DEBUG
connection_manager = scheduler_lib.ConnectionManager(autocommit=False)
self.god.stub_with(connection_manager, 'db_connection', self._database)
self.god.stub_with(monitor_db, '_db_manager', connection_manager)
# These tests only make sense if hosts are acquired inline with the
# rest of the tick.
self.god.stub_with(monitor_db, '_inline_host_acquisition', True)
self.god.stub_with(monitor_db.BaseDispatcher,
'_get_pending_queue_entries',
self._get_pending_hqes)
self.god.stub_with(scheduler_models, '_db', self._database)
self.god.stub_with(drone_manager.instance(), '_results_dir',
'/test/path')
self.god.stub_with(drone_manager.instance(), '_temporary_directory',
'/test/path/tmp')
monitor_db.initialize_globals()
scheduler_models.initialize_globals()
def setUp(self):
self._frontend_common_setup()
self._set_monitor_stubs()
self._dispatcher = monitor_db.Dispatcher()
def tearDown(self):
self._database.disconnect()
self._frontend_common_teardown()
def _update_hqe(self, set, where=''):
query = 'UPDATE afe_host_queue_entries SET ' + set
if where:
query += ' WHERE ' + where
self._do_query(query)
def _get_pending_hqes(self):
query_string=('afe_jobs.priority DESC, '
'ifnull(nullif(host_id, NULL), host_id) DESC, '
'ifnull(nullif(meta_host, NULL), meta_host) DESC, '
'job_id')
return list(scheduler_models.HostQueueEntry.fetch(
joins='INNER JOIN afe_jobs ON (job_id=afe_jobs.id)',
where='NOT complete AND NOT active AND status="Queued"',
order_by=query_string))
class DispatcherSchedulingTest(BaseSchedulerTest):
_jobs_scheduled = []
def tearDown(self):
super(DispatcherSchedulingTest, self).tearDown()
def _set_monitor_stubs(self):
super(DispatcherSchedulingTest, self)._set_monitor_stubs()
def hqe__do_schedule_pre_job_tasks_stub(queue_entry):
"""Called by HostQueueEntry.run()."""
self._record_job_scheduled(queue_entry.job.id, queue_entry.host.id)
queue_entry.set_status('Starting')
self.god.stub_with(scheduler_models.HostQueueEntry,
'_do_schedule_pre_job_tasks',
hqe__do_schedule_pre_job_tasks_stub)
def _record_job_scheduled(self, job_id, host_id):
record = (job_id, host_id)
self.assert_(record not in self._jobs_scheduled,
'Job %d scheduled on host %d twice' %
(job_id, host_id))
self._jobs_scheduled.append(record)
def _assert_job_scheduled_on(self, job_id, host_id):
record = (job_id, host_id)
self.assert_(record in self._jobs_scheduled,
'Job %d not scheduled on host %d as expected\n'
'Jobs scheduled: %s' %
(job_id, host_id, self._jobs_scheduled))
self._jobs_scheduled.remove(record)
def _assert_job_scheduled_on_number_of(self, job_id, host_ids, number):
"""Assert job was scheduled on exactly number hosts out of a set."""
found = []
for host_id in host_ids:
record = (job_id, host_id)
if record in self._jobs_scheduled:
found.append(record)
self._jobs_scheduled.remove(record)
if len(found) < number:
self.fail('Job %d scheduled on fewer than %d hosts in %s.\n'
'Jobs scheduled: %s' % (job_id, number, host_ids, found))
elif len(found) > number:
self.fail('Job %d scheduled on more than %d hosts in %s.\n'
'Jobs scheduled: %s' % (job_id, number, host_ids, found))
def _check_for_extra_schedulings(self):
if len(self._jobs_scheduled) != 0:
self.fail('Extra jobs scheduled: ' +
str(self._jobs_scheduled))
def _convert_jobs_to_metahosts(self, *job_ids):
sql_tuple = '(' + ','.join(str(i) for i in job_ids) + ')'
self._do_query('UPDATE afe_host_queue_entries SET '
'meta_host=host_id, host_id=NULL '
'WHERE job_id IN ' + sql_tuple)
def _lock_host(self, host_id):
self._do_query('UPDATE afe_hosts SET locked=1 WHERE id=' +
str(host_id))
def setUp(self):
super(DispatcherSchedulingTest, self).setUp()
self._jobs_scheduled = []
def _run_scheduler(self):
self._dispatcher._host_scheduler.tick()
for _ in xrange(2): # metahost scheduling can take two cycles
self._dispatcher._schedule_new_jobs()
def _test_basic_scheduling_helper(self, use_metahosts):
'Basic nonmetahost scheduling'
self._create_job_simple([1], use_metahosts)
self._create_job_simple([2], use_metahosts)
self._run_scheduler()
self._assert_job_scheduled_on(1, 1)
self._assert_job_scheduled_on(2, 2)
self._check_for_extra_schedulings()
def _test_priorities_helper(self, use_metahosts):
'Test prioritization ordering'
self._create_job_simple([1], use_metahosts)
self._create_job_simple([2], use_metahosts)
self._create_job_simple([1,2], use_metahosts)
self._create_job_simple([1], use_metahosts, priority=1)
self._run_scheduler()
self._assert_job_scheduled_on(4, 1) # higher priority
self._assert_job_scheduled_on(2, 2) # earlier job over later
self._check_for_extra_schedulings()
def _test_hosts_ready_helper(self, use_metahosts):
"""
Only hosts that are status=Ready, unlocked and not invalid get
scheduled.
"""
self._create_job_simple([1], use_metahosts)
self._do_query('UPDATE afe_hosts SET status="Running" WHERE id=1')
self._run_scheduler()
self._check_for_extra_schedulings()
self._do_query('UPDATE afe_hosts SET status="Ready", locked=1 '
'WHERE id=1')
self._run_scheduler()
self._check_for_extra_schedulings()
self._do_query('UPDATE afe_hosts SET locked=0, invalid=1 '
'WHERE id=1')
self._run_scheduler()
if not use_metahosts:
self._assert_job_scheduled_on(1, 1)
self._check_for_extra_schedulings()
def _test_hosts_idle_helper(self, use_metahosts):
'Only idle hosts get scheduled'
self._create_job(hosts=[1], active=True)
self._create_job_simple([1], use_metahosts)
self._run_scheduler()
self._check_for_extra_schedulings()
def _test_obey_ACLs_helper(self, use_metahosts):
self._do_query('DELETE FROM afe_acl_groups_hosts WHERE host_id=1')
self._create_job_simple([1], use_metahosts)
self._run_scheduler()
self._check_for_extra_schedulings()
def test_basic_scheduling(self):
self._test_basic_scheduling_helper(False)
def test_priorities(self):
self._test_priorities_helper(False)
def test_hosts_ready(self):
self._test_hosts_ready_helper(False)
def test_hosts_idle(self):
self._test_hosts_idle_helper(False)
def test_obey_ACLs(self):
self._test_obey_ACLs_helper(False)
def test_one_time_hosts_ignore_ACLs(self):
self._do_query('DELETE FROM afe_acl_groups_hosts WHERE host_id=1')
self._do_query('UPDATE afe_hosts SET invalid=1 WHERE id=1')
self._create_job_simple([1])
self._run_scheduler()
self._assert_job_scheduled_on(1, 1)
self._check_for_extra_schedulings()
def test_non_metahost_on_invalid_host(self):
"""
Non-metahost entries can get scheduled on invalid hosts (this is how
one-time hosts work).
"""
self._do_query('UPDATE afe_hosts SET invalid=1')
self._test_basic_scheduling_helper(False)
def test_metahost_scheduling(self):
"""
Basic metahost scheduling
"""
self._test_basic_scheduling_helper(True)
def test_metahost_priorities(self):
self._test_priorities_helper(True)
def test_metahost_hosts_ready(self):
self._test_hosts_ready_helper(True)
def test_metahost_hosts_idle(self):
self._test_hosts_idle_helper(True)
def test_metahost_obey_ACLs(self):
self._test_obey_ACLs_helper(True)
def test_nonmetahost_over_metahost(self):
"""
Non-metahost entries should take priority over metahost entries
for the same host
"""
self._create_job(metahosts=[1])
self._create_job(hosts=[1])
self._run_scheduler()
self._assert_job_scheduled_on(2, 1)
self._check_for_extra_schedulings()
# TODO: Revive this test.
# def test_HostScheduler_get_host_atomic_group_id(self):
# job = self._create_job(metahosts=[self.label6.id])
# queue_entry = scheduler_models.HostQueueEntry.fetch(
# where='job_id=%d' % job.id)[0]
# # Indirectly initialize the internal state of the host scheduler.
# self._dispatcher._refresh_pending_queue_entries()
#
# # Test the host scheduler
# host_scheduler = self._dispatcher._host_scheduler
#
#
# # Two labels each in a different atomic group. This should log an
# # error and continue.
# orig_logging_error = logging.error
# def mock_logging_error(message, *args):
# mock_logging_error._num_calls += 1
# # Test the logging call itself, we just wrapped it to count it.
# orig_logging_error(message, *args)
# mock_logging_error._num_calls = 0
# self.god.stub_with(logging, 'error', mock_logging_error)
# host_scheduler.refresh([])
# self.assertNotEquals(None, host_scheduler._get_host_atomic_group_id(
# [self.label4.id, self.label8.id], queue_entry))
# self.assertTrue(mock_logging_error._num_calls > 0)
# self.god.unstub(logging, 'error')
#
# # Two labels both in the same atomic group, this should not raise an
# # error, it will merely cause the job to schedule on the intersection.
# self.assertEquals(1, host_scheduler._get_host_atomic_group_id(
# [self.label4.id, self.label5.id]))
#
# self.assertEquals(None, host_scheduler._get_host_atomic_group_id([]))
# self.assertEquals(None, host_scheduler._get_host_atomic_group_id(
# [self.label3.id, self.label7.id, self.label6.id]))
# self.assertEquals(1, host_scheduler._get_host_atomic_group_id(
# [self.label4.id, self.label7.id, self.label6.id]))
# self.assertEquals(1, host_scheduler._get_host_atomic_group_id(
# [self.label7.id, self.label5.id]))
def test_only_schedule_queued_entries(self):
self._create_job(metahosts=[1])
self._update_hqe(set='active=1, host_id=2')
self._run_scheduler()
self._check_for_extra_schedulings()
def test_no_ready_hosts(self):
self._create_job(hosts=[1])
self._do_query('UPDATE afe_hosts SET status="Repair Failed"')
self._run_scheduler()
self._check_for_extra_schedulings()
def test_garbage_collection(self):
self.god.stub_with(self._dispatcher, '_seconds_between_garbage_stats',
999999)
self.god.stub_function(gc, 'collect')
self.god.stub_function(gc_stats, '_log_garbage_collector_stats')
gc.collect.expect_call().and_return(0)
gc_stats._log_garbage_collector_stats.expect_call()
# Force a garbage collection run
self._dispatcher._last_garbage_stats_time = 0
self._dispatcher._garbage_collection()
# The previous call should have reset the time, it won't do anything
# the second time. If it does, we'll get an unexpected call.
self._dispatcher._garbage_collection()
class DispatcherThrottlingTest(BaseSchedulerTest):
"""
Test that the dispatcher throttles:
* total number of running processes
* number of processes started per cycle
"""
_MAX_RUNNING = 3
_MAX_STARTED = 2
def setUp(self):
super(DispatcherThrottlingTest, self).setUp()
scheduler_config.config.max_processes_per_drone = self._MAX_RUNNING
scheduler_config.config.max_processes_started_per_cycle = (
self._MAX_STARTED)
def fake_max_runnable_processes(fake_self, username,
drone_hostnames_allowed):
running = sum(agent.task.num_processes
for agent in self._agents
if agent.started and not agent.is_done())
return self._MAX_RUNNING - running
self.god.stub_with(drone_manager.DroneManager, 'max_runnable_processes',
fake_max_runnable_processes)
def _setup_some_agents(self, num_agents):
self._agents = [DummyAgent() for i in xrange(num_agents)]
self._dispatcher._agents = list(self._agents)
def _run_a_few_cycles(self):
for i in xrange(4):
self._dispatcher._handle_agents()
def _assert_agents_started(self, indexes, is_started=True):
for i in indexes:
self.assert_(self._agents[i].started == is_started,
'Agent %d %sstarted' %
(i, is_started and 'not ' or ''))
def _assert_agents_not_started(self, indexes):
self._assert_agents_started(indexes, False)
def test_throttle_total(self):
self._setup_some_agents(4)
self._run_a_few_cycles()
self._assert_agents_started([0, 1, 2])
self._assert_agents_not_started([3])
def test_throttle_per_cycle(self):
self._setup_some_agents(3)
self._dispatcher._handle_agents()
self._assert_agents_started([0, 1])
self._assert_agents_not_started([2])
def test_throttle_with_synchronous(self):
self._setup_some_agents(2)
self._agents[0].task.num_processes = 3
self._run_a_few_cycles()
self._assert_agents_started([0])
self._assert_agents_not_started([1])
def test_large_agent_starvation(self):
"""
Ensure large agents don't get starved by lower-priority agents.
"""
self._setup_some_agents(3)
self._agents[1].task.num_processes = 3
self._run_a_few_cycles()
self._assert_agents_started([0])
self._assert_agents_not_started([1, 2])
self._agents[0].set_done(True)
self._run_a_few_cycles()
self._assert_agents_started([1])
self._assert_agents_not_started([2])
def test_zero_process_agent(self):
self._setup_some_agents(5)
self._agents[4].task.num_processes = 0
self._run_a_few_cycles()
self._assert_agents_started([0, 1, 2, 4])
self._assert_agents_not_started([3])
class PidfileRunMonitorTest(unittest.TestCase):
execution_tag = 'test_tag'
pid = 12345
process = drone_manager.Process('myhost', pid)
num_tests_failed = 1
def setUp(self):
self.god = mock.mock_god()
self.mock_drone_manager = self.god.create_mock_class(
drone_manager.DroneManager, 'drone_manager')
self.god.stub_with(pidfile_monitor, '_drone_manager',
self.mock_drone_manager)
self.god.stub_function(email_manager.manager, 'enqueue_notify_email')
self.god.stub_with(pidfile_monitor, '_get_pidfile_timeout_secs',
self._mock_get_pidfile_timeout_secs)
self.pidfile_id = object()
(self.mock_drone_manager.get_pidfile_id_from
.expect_call(self.execution_tag,
pidfile_name=drone_manager.AUTOSERV_PID_FILE)
.and_return(self.pidfile_id))
self.monitor = pidfile_monitor.PidfileRunMonitor()
self.monitor.attach_to_existing_process(self.execution_tag)
def tearDown(self):
self.god.unstub_all()
def _mock_get_pidfile_timeout_secs(self):
return 300
def setup_pidfile(self, pid=None, exit_code=None, tests_failed=None,
use_second_read=False):
contents = drone_manager.PidfileContents()
if pid is not None:
contents.process = drone_manager.Process('myhost', pid)
contents.exit_status = exit_code
contents.num_tests_failed = tests_failed
self.mock_drone_manager.get_pidfile_contents.expect_call(
self.pidfile_id, use_second_read=use_second_read).and_return(
contents)
def set_not_yet_run(self):
self.setup_pidfile()
def set_empty_pidfile(self):
self.setup_pidfile()
def set_running(self, use_second_read=False):
self.setup_pidfile(self.pid, use_second_read=use_second_read)
def set_complete(self, error_code, use_second_read=False):
self.setup_pidfile(self.pid, error_code, self.num_tests_failed,
use_second_read=use_second_read)
def _check_monitor(self, expected_pid, expected_exit_status,
expected_num_tests_failed):
if expected_pid is None:
self.assertEquals(self.monitor._state.process, None)
else:
self.assertEquals(self.monitor._state.process.pid, expected_pid)
self.assertEquals(self.monitor._state.exit_status, expected_exit_status)
self.assertEquals(self.monitor._state.num_tests_failed,
expected_num_tests_failed)
self.god.check_playback()
def _test_read_pidfile_helper(self, expected_pid, expected_exit_status,
expected_num_tests_failed):
self.monitor._read_pidfile()
self._check_monitor(expected_pid, expected_exit_status,
expected_num_tests_failed)
def _get_expected_tests_failed(self, expected_exit_status):
if expected_exit_status is None:
expected_tests_failed = None
else:
expected_tests_failed = self.num_tests_failed
return expected_tests_failed
def test_read_pidfile(self):
self.set_not_yet_run()
self._test_read_pidfile_helper(None, None, None)
self.set_empty_pidfile()
self._test_read_pidfile_helper(None, None, None)
self.set_running()
self._test_read_pidfile_helper(self.pid, None, None)
self.set_complete(123)
self._test_read_pidfile_helper(self.pid, 123, self.num_tests_failed)
def test_read_pidfile_error(self):
self.mock_drone_manager.get_pidfile_contents.expect_call(
self.pidfile_id, use_second_read=False).and_return(
drone_manager.InvalidPidfile('error'))
self.assertRaises(pidfile_monitor.PidfileRunMonitor._PidfileException,
self.monitor._read_pidfile)
self.god.check_playback()
def setup_is_running(self, is_running):
self.mock_drone_manager.is_process_running.expect_call(
self.process).and_return(is_running)
def _test_get_pidfile_info_helper(self, expected_pid, expected_exit_status,
expected_num_tests_failed):
self.monitor._get_pidfile_info()
self._check_monitor(expected_pid, expected_exit_status,
expected_num_tests_failed)
def test_get_pidfile_info(self):
"""
normal cases for get_pidfile_info
"""
# running
self.set_running()
self.setup_is_running(True)
self._test_get_pidfile_info_helper(self.pid, None, None)
# exited during check
self.set_running()
self.setup_is_running(False)
self.set_complete(123, use_second_read=True) # pidfile gets read again
self._test_get_pidfile_info_helper(self.pid, 123, self.num_tests_failed)
# completed
self.set_complete(123)
self._test_get_pidfile_info_helper(self.pid, 123, self.num_tests_failed)
def test_get_pidfile_info_running_no_proc(self):
"""
pidfile shows process running, but no proc exists
"""
# running but no proc
self.set_running()
self.setup_is_running(False)
self.set_running(use_second_read=True)
email_manager.manager.enqueue_notify_email.expect_call(
mock.is_string_comparator(), mock.is_string_comparator())
self._test_get_pidfile_info_helper(self.pid, 1, 0)
self.assertTrue(self.monitor.lost_process)
def test_get_pidfile_info_not_yet_run(self):
"""
pidfile hasn't been written yet
"""
self.set_not_yet_run()
self._test_get_pidfile_info_helper(None, None, None)
def test_process_failed_to_write_pidfile(self):
self.set_not_yet_run()
email_manager.manager.enqueue_notify_email.expect_call(
mock.is_string_comparator(), mock.is_string_comparator())
self.monitor._start_time = (time.time() -
pidfile_monitor._get_pidfile_timeout_secs() - 1)
self._test_get_pidfile_info_helper(None, 1, 0)
self.assertTrue(self.monitor.lost_process)
class AgentTest(unittest.TestCase):
def setUp(self):
self.god = mock.mock_god()
self._dispatcher = self.god.create_mock_class(monitor_db.Dispatcher,
'dispatcher')
def tearDown(self):
self.god.unstub_all()
def _create_mock_task(self, name):
task = self.god.create_mock_class(agent_task.AgentTask, name)
task.num_processes = 1
_set_host_and_qe_ids(task)
return task
def _create_agent(self, task):
agent = monitor_db.Agent(task)
agent.dispatcher = self._dispatcher
return agent
def _finish_agent(self, agent):
while not agent.is_done():
agent.tick()
def test_agent_abort(self):
task = self._create_mock_task('task')
task.poll.expect_call()
task.is_done.expect_call().and_return(False)
task.abort.expect_call()
task.aborted = True
agent = self._create_agent(task)
agent.tick()
agent.abort()
self._finish_agent(agent)
self.god.check_playback()
def _test_agent_abort_before_started_helper(self, ignore_abort=False):
task = self._create_mock_task('task')
task.abort.expect_call()
if ignore_abort:
task.aborted = False
task.poll.expect_call()
task.is_done.expect_call().and_return(True)
task.success = True
else:
task.aborted = True
agent = self._create_agent(task)
agent.abort()
self._finish_agent(agent)
self.god.check_playback()
def test_agent_abort_before_started(self):
self._test_agent_abort_before_started_helper()
self._test_agent_abort_before_started_helper(True)
class JobSchedulingTest(BaseSchedulerTest):
def _test_run_helper(self, expect_agent=True, expect_starting=False,
expect_pending=False):
if expect_starting:
expected_status = models.HostQueueEntry.Status.STARTING
elif expect_pending:
expected_status = models.HostQueueEntry.Status.PENDING
else:
expected_status = models.HostQueueEntry.Status.VERIFYING
job = scheduler_models.Job.fetch('id = 1')[0]
queue_entry = scheduler_models.HostQueueEntry.fetch('id = 1')[0]
assert queue_entry.job is job
job.run_if_ready(queue_entry)
self.god.check_playback()
self._dispatcher._schedule_delay_tasks()
self._dispatcher._schedule_running_host_queue_entries()
agent = self._dispatcher._agents[0]
actual_status = models.HostQueueEntry.smart_get(1).status
self.assertEquals(expected_status, actual_status)
if not expect_agent:
self.assertEquals(agent, None)
return
self.assert_(isinstance(agent, monitor_db.Agent))
self.assert_(agent.task)
return agent.task
def test_run_if_ready_delays(self):
# Also tests Job.run_with_ready_delay() on atomic group jobs.
django_job = self._create_job(hosts=[5, 6], atomic_group=1)
job = scheduler_models.Job(django_job.id)
self.assertEqual(1, job.synch_count)
django_hqes = list(models.HostQueueEntry.objects.filter(job=job.id))
self.assertEqual(2, len(django_hqes))
self.assertEqual(2, django_hqes[0].atomic_group.max_number_of_machines)
def set_hqe_status(django_hqe, status):
django_hqe.status = status
django_hqe.save()
scheduler_models.HostQueueEntry(django_hqe.id).host.set_status(status)
# An initial state, our synch_count is 1
set_hqe_status(django_hqes[0], models.HostQueueEntry.Status.VERIFYING)
set_hqe_status(django_hqes[1], models.HostQueueEntry.Status.PENDING)
# So that we don't depend on the config file value during the test.
self.assert_(scheduler_config.config
.secs_to_wait_for_atomic_group_hosts is not None)
self.god.stub_with(scheduler_config.config,
'secs_to_wait_for_atomic_group_hosts', 123456)
# Get the pending one as a scheduler_models.HostQueueEntry object.
hqe = scheduler_models.HostQueueEntry(django_hqes[1].id)
self.assert_(not job._delay_ready_task)
self.assertTrue(job.is_ready())
# Ready with one pending, one verifying and an atomic group should
# result in a DelayCallTask to re-check if we're ready a while later.
job.run_if_ready(hqe)
self.assertEquals('Waiting', hqe.status)
self._dispatcher._schedule_delay_tasks()
self.assertEquals('Pending', hqe.status)
agent = self._dispatcher._agents[0]
self.assert_(job._delay_ready_task)
self.assert_(isinstance(agent, monitor_db.Agent))
self.assert_(agent.task)
delay_task = agent.task
self.assert_(isinstance(delay_task, scheduler_models.DelayedCallTask))
self.assert_(not delay_task.is_done())
self.god.stub_function(delay_task, 'abort')
self.god.stub_function(job, 'run')
self.god.stub_function(job, '_pending_count')
self.god.stub_with(job, 'synch_count', 9)
self.god.stub_function(job, 'request_abort')
# Test that the DelayedCallTask's callback queued up above does the
# correct thing and does not call run if there are not enough hosts
# in pending after the delay.
job._pending_count.expect_call().and_return(0)
job.request_abort.expect_call()
delay_task._callback()
self.god.check_playback()
# Test that the DelayedCallTask's callback queued up above does the
# correct thing and returns the Agent returned by job.run() if
# there are still enough hosts pending after the delay.
job.synch_count = 4
job._pending_count.expect_call().and_return(4)
job.run.expect_call(hqe)
delay_task._callback()
self.god.check_playback()
job._pending_count.expect_call().and_return(4)
# Adjust the delay deadline so that enough time has passed.
job._delay_ready_task.end_time = time.time() - 111111
job.run.expect_call(hqe)
# ...the delay_expired condition should cause us to call run()
self._dispatcher._handle_agents()
self.god.check_playback()
delay_task.success = False
# Adjust the delay deadline back so that enough time has not passed.
job._delay_ready_task.end_time = time.time() + 111111
self._dispatcher._handle_agents()
self.god.check_playback()
# Now max_number_of_machines HQEs are in pending state. Remaining
# delay will now be ignored.
other_hqe = scheduler_models.HostQueueEntry(django_hqes[0].id)
self.god.unstub(job, 'run')
self.god.unstub(job, '_pending_count')
self.god.unstub(job, 'synch_count')
self.god.unstub(job, 'request_abort')
# ...the over_max_threshold test should cause us to call run()
delay_task.abort.expect_call()
other_hqe.on_pending()
self.assertEquals('Starting', other_hqe.status)
self.assertEquals('Starting', hqe.status)
self.god.stub_function(job, 'run')
self.god.unstub(delay_task, 'abort')
hqe.set_status('Pending')
other_hqe.set_status('Pending')
# Now we're not over the max for the atomic group. But all assigned
# hosts are in pending state. over_max_threshold should make us run().
hqe.atomic_group.max_number_of_machines += 1
hqe.atomic_group.save()
job.run.expect_call(hqe)
hqe.on_pending()
self.god.check_playback()
hqe.atomic_group.max_number_of_machines -= 1
hqe.atomic_group.save()
other_hqe = scheduler_models.HostQueueEntry(django_hqes[0].id)
self.assertTrue(hqe.job is other_hqe.job)
# DBObject classes should reuse instances so these should be the same.
self.assertEqual(job, other_hqe.job)
self.assertEqual(other_hqe.job, hqe.job)
# Be sure our delay was not lost during the other_hqe construction.
self.assertEqual(job._delay_ready_task, delay_task)
self.assert_(job._delay_ready_task)
self.assertFalse(job._delay_ready_task.is_done())
self.assertFalse(job._delay_ready_task.aborted)
# We want the real run() to be called below.
self.god.unstub(job, 'run')
# We pass in the other HQE this time the same way it would happen
# for real when one host finishes verifying and enters pending.
job.run_if_ready(other_hqe)
# The delayed task must be aborted by the actual run() call above.
self.assertTrue(job._delay_ready_task.aborted)
self.assertFalse(job._delay_ready_task.success)
self.assertTrue(job._delay_ready_task.is_done())
# Check that job run() and _finish_run() were called by the above:
self._dispatcher._schedule_running_host_queue_entries()
agent = self._dispatcher._agents[0]
self.assert_(agent.task)
task = agent.task
self.assert_(isinstance(task, monitor_db.QueueTask))
# Requery these hqes in order to verify the status from the DB.
django_hqes = list(models.HostQueueEntry.objects.filter(job=job.id))
for entry in django_hqes:
self.assertEqual(models.HostQueueEntry.Status.STARTING,
entry.status)
# We're already running, but more calls to run_with_ready_delay can
# continue to come in due to straggler hosts enter Pending. Make
# sure we don't do anything.
self.god.stub_function(job, 'run')
job.run_with_ready_delay(hqe)
self.god.check_playback()
self.god.unstub(job, 'run')
def test_run_synchronous_atomic_group_ready(self):
self._create_job(hosts=[5, 6], atomic_group=1, synchronous=True)
self._update_hqe("status='Pending', execution_subdir=''")
queue_task = self._test_run_helper(expect_starting=True)
self.assert_(isinstance(queue_task, monitor_db.QueueTask))
# Atomic group jobs that do not depend on a specific label in the
# atomic group will use the atomic group name as their group name.
self.assertEquals(queue_task.queue_entries[0].get_group_name(),
'atomic1')
def test_run_synchronous_atomic_group_with_label_ready(self):
job = self._create_job(hosts=[5, 6], atomic_group=1, synchronous=True)
job.dependency_labels.add(self.label4)
self._update_hqe("status='Pending', execution_subdir=''")
queue_task = self._test_run_helper(expect_starting=True)
self.assert_(isinstance(queue_task, monitor_db.QueueTask))
# Atomic group jobs that also specify a label in the atomic group
# will use the label name as their group name.
self.assertEquals(queue_task.queue_entries[0].get_group_name(),
'label4')
def test_run_synchronous_ready(self):
self._create_job(hosts=[1, 2], synchronous=True)
self._update_hqe("status='Pending', execution_subdir=''")
queue_task = self._test_run_helper(expect_starting=True)
self.assert_(isinstance(queue_task, monitor_db.QueueTask))
self.assertEquals(queue_task.job.id, 1)
hqe_ids = [hqe.id for hqe in queue_task.queue_entries]
self.assertEquals(hqe_ids, [1, 2])
def test_schedule_running_host_queue_entries_fail(self):
self._create_job(hosts=[2])
self._update_hqe("status='%s', execution_subdir=''" %
models.HostQueueEntry.Status.PENDING)
job = scheduler_models.Job.fetch('id = 1')[0]
queue_entry = scheduler_models.HostQueueEntry.fetch('id = 1')[0]
assert queue_entry.job is job
job.run_if_ready(queue_entry)
self.assertEqual(queue_entry.status,
models.HostQueueEntry.Status.STARTING)
self.assert_(queue_entry.execution_subdir)
self.god.check_playback()
class dummy_test_agent(object):
task = 'dummy_test_agent'
self._dispatcher._register_agent_for_ids(
self._dispatcher._host_agents, [queue_entry.host.id],
dummy_test_agent)
# Attempted to schedule on a host that already has an agent.
self.assertRaises(scheduler_lib.SchedulerError,
self._dispatcher._schedule_running_host_queue_entries)
def test_schedule_hostless_job(self):
job = self._create_job(hostless=True)
self.assertEqual(1, job.hostqueueentry_set.count())
hqe_query = scheduler_models.HostQueueEntry.fetch(
'id = %s' % job.hostqueueentry_set.all()[0].id)
self.assertEqual(1, len(hqe_query))
hqe = hqe_query[0]
self.assertEqual(models.HostQueueEntry.Status.QUEUED, hqe.status)
self.assertEqual(0, len(self._dispatcher._agents))
self._dispatcher._schedule_new_jobs()
self.assertEqual(models.HostQueueEntry.Status.STARTING, hqe.status)
self.assertEqual(1, len(self._dispatcher._agents))
self._dispatcher._schedule_new_jobs()
# No change to previously schedule hostless job, and no additional agent
self.assertEqual(models.HostQueueEntry.Status.STARTING, hqe.status)
self.assertEqual(1, len(self._dispatcher._agents))
class TopLevelFunctionsTest(unittest.TestCase):
def setUp(self):
self.god = mock.mock_god()
def tearDown(self):
self.god.unstub_all()
def test_autoserv_command_line(self):
machines = 'abcd12,efgh34'
extra_args = ['-Z', 'hello']
expected_command_line_base = set((monitor_db._autoserv_path, '-p',
'-m', machines, '-r',
drone_manager.WORKING_DIRECTORY))
expected_command_line = expected_command_line_base.union(
['--verbose']).union(extra_args)
command_line = set(
monitor_db._autoserv_command_line(machines, extra_args))
self.assertEqual(expected_command_line, command_line)
class FakeJob(object):
owner = 'Bob'
name = 'fake job name'
test_retry = 0
id = 1337
class FakeHQE(object):
job = FakeJob
expected_command_line = expected_command_line_base.union(
['-u', FakeJob.owner, '-l', FakeJob.name])
command_line = set(monitor_db._autoserv_command_line(
machines, extra_args=[], queue_entry=FakeHQE, verbose=False))
self.assertEqual(expected_command_line, command_line)
class AgentTaskTest(unittest.TestCase,
frontend_test_utils.FrontendTestMixin):
def setUp(self):
self._frontend_common_setup()
def tearDown(self):
self._frontend_common_teardown()
def _setup_drones(self):
self.god.stub_function(models.DroneSet, 'drone_sets_enabled')
models.DroneSet.drone_sets_enabled.expect_call().and_return(True)
drones = []
for x in xrange(4):
drones.append(models.Drone.objects.create(hostname=str(x)))
drone_set_1 = models.DroneSet.objects.create(name='1')
drone_set_1.drones.add(*drones[0:2])
drone_set_2 = models.DroneSet.objects.create(name='2')
drone_set_2.drones.add(*drones[2:4])
drone_set_3 = models.DroneSet.objects.create(name='3')
job_1 = self._create_job_simple([self.hosts[0].id],
drone_set=drone_set_1)
job_2 = self._create_job_simple([self.hosts[0].id],
drone_set=drone_set_2)
job_3 = self._create_job_simple([self.hosts[0].id],
drone_set=drone_set_3)
job_4 = self._create_job_simple([self.hosts[0].id])
job_4.drone_set = None
job_4.save()
hqe_1 = job_1.hostqueueentry_set.all()[0]
hqe_2 = job_2.hostqueueentry_set.all()[0]
hqe_3 = job_3.hostqueueentry_set.all()[0]
hqe_4 = job_4.hostqueueentry_set.all()[0]
return (hqe_1, hqe_2, hqe_3, hqe_4), agent_task.AgentTask()
def test_get_drone_hostnames_allowed_no_drones_in_set(self):
hqes, task = self._setup_drones()
task.queue_entry_ids = (hqes[2].id,)
self.assertEqual(set(), task.get_drone_hostnames_allowed())
self.god.check_playback()
def test_get_drone_hostnames_allowed_no_drone_set(self):
hqes, task = self._setup_drones()
hqe = hqes[3]
task.queue_entry_ids = (hqe.id,)
result = object()
self.god.stub_function(task, '_user_or_global_default_drone_set')
task._user_or_global_default_drone_set.expect_call(
hqe.job, hqe.job.user()).and_return(result)
self.assertEqual(result, task.get_drone_hostnames_allowed())
self.god.check_playback()
def test_get_drone_hostnames_allowed_success(self):
hqes, task = self._setup_drones()
task.queue_entry_ids = (hqes[0].id,)
self.assertEqual(set(('0','1')), task.get_drone_hostnames_allowed())
self.god.check_playback()
def test_get_drone_hostnames_allowed_multiple_jobs(self):
hqes, task = self._setup_drones()
task.queue_entry_ids = (hqes[0].id, hqes[1].id)
self.assertRaises(AssertionError,
task.get_drone_hostnames_allowed)
self.god.check_playback()
def test_get_drone_hostnames_allowed_no_hqe(self):
class MockSpecialTask(object):
requested_by = object()
class MockSpecialAgentTask(agent_task.SpecialAgentTask):
task = MockSpecialTask()
queue_entry_ids = []
def __init__(self, *args, **kwargs):
pass
task = MockSpecialAgentTask()
self.god.stub_function(models.DroneSet, 'drone_sets_enabled')
self.god.stub_function(task, '_user_or_global_default_drone_set')
result = object()
models.DroneSet.drone_sets_enabled.expect_call().and_return(True)
task._user_or_global_default_drone_set.expect_call(
task.task, MockSpecialTask.requested_by).and_return(result)
self.assertEqual(result, task.get_drone_hostnames_allowed())
self.god.check_playback()
def _setup_test_user_or_global_default_drone_set(self):
result = object()
class MockDroneSet(object):
def get_drone_hostnames(self):
return result
self.god.stub_function(models.DroneSet, 'get_default')
models.DroneSet.get_default.expect_call().and_return(MockDroneSet())
return result
def test_user_or_global_default_drone_set(self):
expected = object()
class MockDroneSet(object):
def get_drone_hostnames(self):
return expected
class MockUser(object):
drone_set = MockDroneSet()
self._setup_test_user_or_global_default_drone_set()
actual = agent_task.AgentTask()._user_or_global_default_drone_set(
None, MockUser())
self.assertEqual(expected, actual)
self.god.check_playback()
def test_user_or_global_default_drone_set_no_user(self):
expected = self._setup_test_user_or_global_default_drone_set()
actual = agent_task.AgentTask()._user_or_global_default_drone_set(
None, None)
self.assertEqual(expected, actual)
self.god.check_playback()
def test_user_or_global_default_drone_set_no_user_drone_set(self):
class MockUser(object):
drone_set = None
login = None
expected = self._setup_test_user_or_global_default_drone_set()
actual = agent_task.AgentTask()._user_or_global_default_drone_set(
None, MockUser())
self.assertEqual(expected, actual)
self.god.check_playback()
def test_abort_HostlessQueueTask(self):
hqe = self.god.create_mock_class(scheduler_models.HostQueueEntry,
'HostQueueEntry')
# If hqe is still in STARTING status, aborting the task should finish
# without changing hqe's status.
hqe.status = models.HostQueueEntry.Status.STARTING
hqe.job = None
hqe.id = 0
task = monitor_db.HostlessQueueTask(hqe)
task.abort()
# If hqe is in RUNNING status, aborting the task should change hqe's
# status to Parsing, so FinalReparseTask can be scheduled.
hqe.set_status.expect_call('Parsing')
hqe.status = models.HostQueueEntry.Status.RUNNING
hqe.job = None
hqe.id = 0
task = monitor_db.HostlessQueueTask(hqe)
task.abort()
if __name__ == '__main__':
unittest.main()