1 #!/usr/bin/python 2 # pylint: disable=missing-docstring 3 4 import logging, os, signal, unittest 5 import common 6 import mock 7 from autotest_lib.client.common_lib import enum, global_config, host_protections 8 from autotest_lib.database import database_connection 9 from autotest_lib.frontend import setup_django_environment 10 from autotest_lib.frontend.afe import frontend_test_utils, models 11 from autotest_lib.frontend.afe import model_attributes 12 from autotest_lib.scheduler import drone_manager, email_manager 13 from autotest_lib.scheduler import monitor_db, scheduler_models 14 from autotest_lib.scheduler import scheduler_config 15 from autotest_lib.scheduler import scheduler_lib 16 17 HqeStatus = models.HostQueueEntry.Status 18 HostStatus = models.Host.Status 19 20 class NullMethodObject(object): 21 _NULL_METHODS = () 22 23 def __init__(self): 24 def null_method(*args, **kwargs): 25 pass 26 27 for method_name in self._NULL_METHODS: 28 setattr(self, method_name, null_method) 29 30 # the SpecialTask names here must match the suffixes used on the SpecialTask 31 # results directories 32 _PidfileType = enum.Enum('verify', 'cleanup', 'repair', 'job', 'gather', 33 'parse', 'archive', 'reset', 'provision') 34 35 36 _PIDFILE_TO_PIDFILE_TYPE = { 37 drone_manager.AUTOSERV_PID_FILE: _PidfileType.JOB, 38 drone_manager.CRASHINFO_PID_FILE: _PidfileType.GATHER, 39 drone_manager.PARSER_PID_FILE: _PidfileType.PARSE, 40 drone_manager.ARCHIVER_PID_FILE: _PidfileType.ARCHIVE, 41 } 42 43 44 _PIDFILE_TYPE_TO_PIDFILE = dict((value, key) for key, value 45 in _PIDFILE_TO_PIDFILE_TYPE.iteritems()) 46 47 48 class MockConnectionManager(object): 49 """docstring for MockConnectionManager""" 50 51 db = None 52 53 def __init__(self): 54 super(MockConnectionManager, self).__init__() 55 56 def get_connection(self): 57 assert MockConnectionManager.db 58 return MockConnectionManager.db 59 60 61 class MockDroneManager(NullMethodObject): 62 """ 63 Public attributes: 64 max_runnable_processes_value: value returned by max_runnable_processes(). 65 tests can change this to activate throttling. 66 """ 67 _NULL_METHODS = ('reinitialize_drones', 'copy_to_results_repository', 68 'copy_results_on_drone', 'trigger_refresh', 'sync_refresh') 69 70 class _DummyPidfileId(object): 71 """ 72 Object to represent pidfile IDs that is opaque to the scheduler code but 73 still debugging-friendly for us. 74 """ 75 def __init__(self, working_directory, pidfile_name, num_processes=None): 76 self._working_directory = working_directory 77 self._pidfile_name = pidfile_name 78 self._num_processes = num_processes 79 self._paired_with_pidfile = None 80 81 82 def key(self): 83 """Key for MockDroneManager._pidfile_index""" 84 return (self._working_directory, self._pidfile_name) 85 86 87 def __str__(self): 88 return os.path.join(self._working_directory, self._pidfile_name) 89 90 91 def __repr__(self): 92 return '<_DummyPidfileId: %s>' % str(self) 93 94 95 def __init__(self): 96 super(MockDroneManager, self).__init__() 97 self.process_capacity = 100 98 99 # maps result_dir to set of tuples (file_path, file_contents) 100 self._attached_files = {} 101 # maps pidfile IDs to PidfileContents 102 self._pidfiles = {} 103 # pidfile IDs that haven't been created yet 104 self._future_pidfiles = [] 105 # maps _PidfileType to the most recently created pidfile ID of that type 106 self._last_pidfile_id = {} 107 # maps (working_directory, pidfile_name) to pidfile IDs 108 self._pidfile_index = {} 109 # maps process to pidfile IDs 110 self._process_index = {} 111 # tracks pidfiles of processes that have been killed 112 self._pids_to_signals_received = {} 113 # pidfile IDs that have just been unregistered (so will disappear on the 114 # next cycle) 115 self._unregistered_pidfiles = set() 116 # Pids to write exit status for at end of tick 117 self._set_pidfile_exit_status_queue = [] 118 119 # utility APIs for use by the test 120 121 def finish_process(self, pidfile_type, exit_status=0): 122 pidfile_id = self._last_pidfile_id[pidfile_type] 123 self._set_pidfile_exit_status(pidfile_id, exit_status) 124 125 126 def finish_specific_process(self, working_directory, pidfile_name): 127 pidfile_id = self.pidfile_from_path(working_directory, pidfile_name) 128 self._set_pidfile_exit_status(pidfile_id, 0) 129 130 def finish_active_process_on_host(self, host_id): 131 match = 'hosts/host%d/' % host_id 132 for pidfile_id in self.nonfinished_pidfile_ids(): 133 if pidfile_id._working_directory.startswith(match): 134 self._set_pidfile_exit_status(pidfile_id, 0) 135 break 136 else: 137 raise KeyError('No active process matched %s' % match) 138 139 def _set_pidfile_exit_status(self, pidfile_id, exit_status): 140 assert pidfile_id is not None 141 contents = self._pidfiles[pidfile_id] 142 contents.exit_status = exit_status 143 contents.num_tests_failed = 0 144 145 146 def was_last_process_killed(self, pidfile_type, sigs): 147 pidfile_id = self._last_pidfile_id[pidfile_type] 148 return sigs == self._pids_to_signals_received[pidfile_id] 149 150 151 def nonfinished_pidfile_ids(self): 152 return [pidfile_id for pidfile_id, pidfile_contents 153 in self._pidfiles.iteritems() 154 if pidfile_contents.exit_status is None] 155 156 157 def running_pidfile_ids(self): 158 return [pidfile_id for pidfile_id in self.nonfinished_pidfile_ids() 159 if self._pidfiles[pidfile_id].process is not None] 160 161 162 def pidfile_from_path(self, working_directory, pidfile_name): 163 return self._pidfile_index[(working_directory, pidfile_name)] 164 165 166 def attached_files(self, working_directory): 167 """ 168 Return dict mapping path to contents for attached files with specified 169 paths. 170 """ 171 return dict((path, contents) for path, contents 172 in self._attached_files.get(working_directory, []) 173 if path is not None) 174 175 176 # DroneManager emulation APIs for use by monitor_db 177 178 def get_orphaned_autoserv_processes(self): 179 return set() 180 181 182 def total_running_processes(self): 183 return sum(pidfile_id._num_processes 184 for pidfile_id in self.nonfinished_pidfile_ids()) 185 186 187 def max_runnable_processes(self, username, drone_hostnames_allowed): 188 return self.process_capacity - self.total_running_processes() 189 190 191 def refresh(self): 192 for pidfile_id in self._unregistered_pidfiles: 193 # intentionally handle non-registered pidfiles silently 194 self._pidfiles.pop(pidfile_id, None) 195 self._unregistered_pidfiles = set() 196 197 198 def execute_actions(self): 199 # executing an "execute_command" causes a pidfile to be created 200 for pidfile_id in self._future_pidfiles: 201 # Process objects are opaque to monitor_db 202 process = object() 203 self._pidfiles[pidfile_id].process = process 204 self._process_index[process] = pidfile_id 205 self._future_pidfiles = [] 206 207 for pidfile_id in self._set_pidfile_exit_status_queue: 208 self._set_pidfile_exit_status(pidfile_id, 271) 209 self._set_pidfile_exit_status_queue = [] 210 211 212 def attach_file_to_execution(self, result_dir, file_contents, 213 file_path=None): 214 self._attached_files.setdefault(result_dir, set()).add((file_path, 215 file_contents)) 216 return 'attach_path' 217 218 219 def _initialize_pidfile(self, pidfile_id): 220 if pidfile_id not in self._pidfiles: 221 assert pidfile_id.key() not in self._pidfile_index 222 self._pidfiles[pidfile_id] = drone_manager.PidfileContents() 223 self._pidfile_index[pidfile_id.key()] = pidfile_id 224 225 226 def _set_last_pidfile(self, pidfile_id, working_directory, pidfile_name): 227 if working_directory.startswith('hosts/'): 228 # such paths look like hosts/host1/1-verify, we'll grab the end 229 type_string = working_directory.rsplit('-', 1)[1] 230 pidfile_type = _PidfileType.get_value(type_string) 231 else: 232 pidfile_type = _PIDFILE_TO_PIDFILE_TYPE[pidfile_name] 233 self._last_pidfile_id[pidfile_type] = pidfile_id 234 235 236 def execute_command(self, command, working_directory, pidfile_name, 237 num_processes, log_file=None, paired_with_pidfile=None, 238 username=None, drone_hostnames_allowed=None): 239 logging.debug('Executing %s in %s', command, working_directory) 240 pidfile_id = self._DummyPidfileId(working_directory, pidfile_name) 241 if pidfile_id.key() in self._pidfile_index: 242 pidfile_id = self._pidfile_index[pidfile_id.key()] 243 pidfile_id._num_processes = num_processes 244 pidfile_id._paired_with_pidfile = paired_with_pidfile 245 246 self._future_pidfiles.append(pidfile_id) 247 self._initialize_pidfile(pidfile_id) 248 self._pidfile_index[(working_directory, pidfile_name)] = pidfile_id 249 self._set_last_pidfile(pidfile_id, working_directory, pidfile_name) 250 return pidfile_id 251 252 253 def get_pidfile_contents(self, pidfile_id, use_second_read=False): 254 if pidfile_id not in self._pidfiles: 255 logging.debug('Request for nonexistent pidfile %s', pidfile_id) 256 return self._pidfiles.get(pidfile_id, drone_manager.PidfileContents()) 257 258 259 def is_process_running(self, process): 260 return True 261 262 263 def register_pidfile(self, pidfile_id): 264 self._initialize_pidfile(pidfile_id) 265 266 267 def unregister_pidfile(self, pidfile_id): 268 self._unregistered_pidfiles.add(pidfile_id) 269 270 271 def declare_process_count(self, pidfile_id, num_processes): 272 pidfile_id.num_processes = num_processes 273 274 275 def absolute_path(self, path): 276 return 'absolute/' + path 277 278 279 def write_lines_to_file(self, file_path, lines, paired_with_process=None): 280 # TODO: record this 281 pass 282 283 284 def get_pidfile_id_from(self, execution_tag, pidfile_name): 285 default_pidfile = self._DummyPidfileId(execution_tag, pidfile_name, 286 num_processes=0) 287 return self._pidfile_index.get((execution_tag, pidfile_name), 288 default_pidfile) 289 290 291 def kill_process(self, process, sig=signal.SIGKILL): 292 pidfile_id = self._process_index[process] 293 294 if pidfile_id not in self._pids_to_signals_received: 295 self._pids_to_signals_received[pidfile_id] = set() 296 self._pids_to_signals_received[pidfile_id].add(sig) 297 298 if signal.SIGKILL == sig: 299 self._set_pidfile_exit_status_queue.append(pidfile_id) 300 301 302 class MockEmailManager(NullMethodObject): 303 _NULL_METHODS = ('send_queued_emails', 'send_email') 304 305 def enqueue_notify_email(self, subject, message): 306 logging.warning('enqueue_notify_email: %s', subject) 307 logging.warning(message) 308 309 310 class SchedulerFunctionalTest(unittest.TestCase, 311 frontend_test_utils.FrontendTestMixin): 312 # some number of ticks after which the scheduler is presumed to have 313 # stabilized, given no external changes 314 _A_LOT_OF_TICKS = 10 315 316 def setUp(self): 317 self._frontend_common_setup() 318 self._set_stubs() 319 self._set_global_config_values() 320 self._create_dispatcher() 321 322 logging.basicConfig(level=logging.DEBUG) 323 324 325 def _create_dispatcher(self): 326 self.dispatcher = monitor_db.Dispatcher() 327 328 329 def tearDown(self): 330 self._database.disconnect() 331 self._frontend_common_teardown() 332 333 334 def _set_stubs(self): 335 self.mock_config = global_config.FakeGlobalConfig() 336 self.god.stub_with(global_config, 'global_config', self.mock_config) 337 338 self.mock_drone_manager = MockDroneManager() 339 drone_manager._set_instance(self.mock_drone_manager) 340 341 self.mock_email_manager = MockEmailManager() 342 self.god.stub_with(email_manager, 'manager', self.mock_email_manager) 343 344 self._database = ( 345 database_connection.TranslatingDatabase.get_test_database( 346 translators=scheduler_lib._DB_TRANSLATORS)) 347 self._database.connect(db_type='django') 348 self.god.stub_with(monitor_db, '_db', self._database) 349 self.god.stub_with(scheduler_models, '_db', self._database) 350 351 MockConnectionManager.db = self._database 352 scheduler_lib.ConnectionManager = MockConnectionManager 353 354 monitor_db.initialize_globals() 355 scheduler_models.initialize_globals() 356 357 patcher = mock.patch( 358 'autotest_lib.scheduler.luciferlib.is_lucifer_enabled', 359 lambda: False) 360 patcher.start() 361 self.addCleanup(patcher.stop) 362 363 364 def _set_global_config_values(self): 365 self.mock_config.set_config_value('SCHEDULER', 'pidfile_timeout_mins', 366 1) 367 self.mock_config.set_config_value('SCHEDULER', 'gc_stats_interval_mins', 368 999999) 369 self.mock_config.set_config_value('SCHEDULER', 'enable_archiving', True) 370 self.mock_config.set_config_value('SCHEDULER', 371 'clean_interval_minutes', 60) 372 self.mock_config.set_config_value('SCHEDULER', 373 'max_parse_processes', 50) 374 self.mock_config.set_config_value('SCHEDULER', 375 'max_transfer_processes', 50) 376 self.mock_config.set_config_value('SCHEDULER', 377 'clean_interval_minutes', 50) 378 self.mock_config.set_config_value('SCHEDULER', 379 'max_provision_retries', 1) 380 self.mock_config.set_config_value('SCHEDULER', 'max_repair_limit', 1) 381 self.mock_config.set_config_value( 382 'SCHEDULER', 'secs_to_wait_for_atomic_group_hosts', 600) 383 self.mock_config.set_config_value( 384 'SCHEDULER', 'inline_host_acquisition', True) 385 scheduler_config.config.read_config() 386 387 388 def _initialize_test(self): 389 self.dispatcher.initialize() 390 391 392 def _run_dispatcher(self): 393 for _ in xrange(self._A_LOT_OF_TICKS): 394 self.dispatcher.tick() 395 396 397 def test_idle(self): 398 self._initialize_test() 399 self._run_dispatcher() 400 401 402 def _assert_process_executed(self, working_directory, pidfile_name): 403 process_was_executed = self.mock_drone_manager.was_process_executed( 404 'hosts/host1/1-verify', drone_manager.AUTOSERV_PID_FILE) 405 self.assert_(process_was_executed, 406 '%s/%s not executed' % (working_directory, pidfile_name)) 407 408 409 def _update_instance(self, model_instance): 410 return type(model_instance).objects.get(pk=model_instance.pk) 411 412 413 def _check_statuses(self, queue_entry, queue_entry_status, 414 host_status=None): 415 self._check_entry_status(queue_entry, queue_entry_status) 416 if host_status: 417 self._check_host_status(queue_entry.host, host_status) 418 419 420 def _check_entry_status(self, queue_entry, status): 421 # update from DB 422 queue_entry = self._update_instance(queue_entry) 423 self.assertEquals(queue_entry.status, status) 424 425 426 def _check_host_status(self, host, status): 427 # update from DB 428 host = self._update_instance(host) 429 self.assertEquals(host.status, status) 430 431 432 def _run_pre_job_verify(self, queue_entry): 433 self._run_dispatcher() # launches verify 434 self._check_statuses(queue_entry, HqeStatus.VERIFYING, 435 HostStatus.VERIFYING) 436 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 437 438 439 def test_simple_job(self): 440 self._initialize_test() 441 job, queue_entry = self._make_job_and_queue_entry() 442 self._run_pre_job_verify(queue_entry) 443 self._run_dispatcher() # launches job 444 self._check_statuses(queue_entry, HqeStatus.RUNNING, HostStatus.RUNNING) 445 self._finish_job(queue_entry) 446 self._check_statuses(queue_entry, HqeStatus.COMPLETED, HostStatus.READY) 447 self._assert_nothing_is_running() 448 449 450 def _setup_for_pre_job_reset(self): 451 self._initialize_test() 452 job, queue_entry = self._make_job_and_queue_entry() 453 job.reboot_before = model_attributes.RebootBefore.ALWAYS 454 job.save() 455 return queue_entry 456 457 458 def _run_pre_job_reset_job(self, queue_entry): 459 self._run_dispatcher() # reset 460 self._check_statuses(queue_entry, HqeStatus.RESETTING, 461 HostStatus.RESETTING) 462 self.mock_drone_manager.finish_process(_PidfileType.RESET) 463 self._run_dispatcher() # job 464 self._finish_job(queue_entry) 465 466 467 def test_pre_job_reset(self): 468 queue_entry = self._setup_for_pre_job_reset() 469 self._run_pre_job_reset_job(queue_entry) 470 471 472 def _run_pre_job_reset_one_failure(self): 473 queue_entry = self._setup_for_pre_job_reset() 474 self._run_dispatcher() # reset 475 self.mock_drone_manager.finish_process(_PidfileType.RESET, 476 exit_status=256) 477 self._run_dispatcher() # repair 478 self._check_statuses(queue_entry, HqeStatus.QUEUED, 479 HostStatus.REPAIRING) 480 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 481 return queue_entry 482 483 484 def test_pre_job_reset_failure(self): 485 queue_entry = self._run_pre_job_reset_one_failure() 486 # from here the job should run as normal 487 self._run_pre_job_reset_job(queue_entry) 488 489 490 def test_pre_job_reset_double_failure(self): 491 # TODO (showard): this test isn't perfect. in reality, when the second 492 # reset fails, it copies its results over to the job directory using 493 # copy_results_on_drone() and then parses them. since we don't handle 494 # that, there appear to be no results at the job directory. the 495 # scheduler handles this gracefully, parsing gets effectively skipped, 496 # and this test passes as is. but we ought to properly test that 497 # behavior. 498 queue_entry = self._run_pre_job_reset_one_failure() 499 self._run_dispatcher() # second reset 500 self.mock_drone_manager.finish_process(_PidfileType.RESET, 501 exit_status=256) 502 self._run_dispatcher() 503 self._check_statuses(queue_entry, HqeStatus.FAILED, 504 HostStatus.REPAIR_FAILED) 505 # nothing else should run 506 self._assert_nothing_is_running() 507 508 509 def _assert_nothing_is_running(self): 510 self.assertEquals(self.mock_drone_manager.running_pidfile_ids(), []) 511 512 513 def _setup_for_post_job_cleanup(self): 514 self._initialize_test() 515 job, queue_entry = self._make_job_and_queue_entry() 516 job.reboot_after = model_attributes.RebootAfter.ALWAYS 517 job.save() 518 return queue_entry 519 520 521 def _run_post_job_cleanup_failure_up_to_repair(self, queue_entry, 522 include_verify=True): 523 if include_verify: 524 self._run_pre_job_verify(queue_entry) 525 self._run_dispatcher() # job 526 self.mock_drone_manager.finish_process(_PidfileType.JOB) 527 self._run_dispatcher() # parsing + cleanup 528 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 529 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP, 530 exit_status=256) 531 self._run_dispatcher() # repair, HQE unaffected 532 return queue_entry 533 534 535 def test_post_job_cleanup_failure(self): 536 queue_entry = self._setup_for_post_job_cleanup() 537 self._run_post_job_cleanup_failure_up_to_repair(queue_entry) 538 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 539 HostStatus.REPAIRING) 540 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 541 self._run_dispatcher() 542 self._check_statuses(queue_entry, HqeStatus.COMPLETED, HostStatus.READY) 543 544 545 def test_post_job_cleanup_failure_repair_failure(self): 546 queue_entry = self._setup_for_post_job_cleanup() 547 self._run_post_job_cleanup_failure_up_to_repair(queue_entry) 548 self.mock_drone_manager.finish_process(_PidfileType.REPAIR, 549 exit_status=256) 550 self._run_dispatcher() 551 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 552 HostStatus.REPAIR_FAILED) 553 554 555 def _ensure_post_job_process_is_paired(self, queue_entry, pidfile_type): 556 pidfile_name = _PIDFILE_TYPE_TO_PIDFILE[pidfile_type] 557 queue_entry = self._update_instance(queue_entry) 558 pidfile_id = self.mock_drone_manager.pidfile_from_path( 559 queue_entry.execution_path(), pidfile_name) 560 self.assert_(pidfile_id._paired_with_pidfile) 561 562 563 def _finish_job(self, queue_entry): 564 self._check_statuses(queue_entry, HqeStatus.RUNNING) 565 self.mock_drone_manager.finish_process(_PidfileType.JOB) 566 self._run_dispatcher() # launches parsing 567 self._check_statuses(queue_entry, HqeStatus.PARSING) 568 self._ensure_post_job_process_is_paired(queue_entry, _PidfileType.PARSE) 569 self._finish_parsing() 570 571 572 def _finish_parsing(self): 573 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 574 self._run_dispatcher() 575 576 577 def _create_reverify_request(self): 578 host = self.hosts[0] 579 models.SpecialTask.schedule_special_task( 580 host=host, task=models.SpecialTask.Task.VERIFY) 581 return host 582 583 584 def test_requested_reverify(self): 585 host = self._create_reverify_request() 586 self._run_dispatcher() 587 self._check_host_status(host, HostStatus.VERIFYING) 588 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 589 self._run_dispatcher() 590 self._check_host_status(host, HostStatus.READY) 591 592 593 def test_requested_reverify_failure(self): 594 host = self._create_reverify_request() 595 self._run_dispatcher() 596 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 597 exit_status=256) 598 self._run_dispatcher() # repair 599 self._check_host_status(host, HostStatus.REPAIRING) 600 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 601 self._run_dispatcher() 602 self._check_host_status(host, HostStatus.READY) 603 604 605 def _setup_for_do_not_verify(self): 606 self._initialize_test() 607 job, queue_entry = self._make_job_and_queue_entry() 608 queue_entry.host.protection = host_protections.Protection.DO_NOT_VERIFY 609 queue_entry.host.save() 610 return queue_entry 611 612 613 def test_do_not_verify_job(self): 614 queue_entry = self._setup_for_do_not_verify() 615 self._run_dispatcher() # runs job directly 616 self._finish_job(queue_entry) 617 618 619 def test_do_not_verify_job_with_cleanup(self): 620 queue_entry = self._setup_for_do_not_verify() 621 queue_entry.job.reboot_before = model_attributes.RebootBefore.ALWAYS 622 queue_entry.job.save() 623 624 self._run_dispatcher() # cleanup 625 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP) 626 self._run_dispatcher() # job 627 self._finish_job(queue_entry) 628 629 630 def test_do_not_verify_pre_job_cleanup_failure(self): 631 queue_entry = self._setup_for_do_not_verify() 632 queue_entry.job.reboot_before = model_attributes.RebootBefore.ALWAYS 633 queue_entry.job.save() 634 635 self._run_dispatcher() # cleanup 636 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP, 637 exit_status=256) 638 self._run_dispatcher() # failure ignored; job runs 639 self._finish_job(queue_entry) 640 641 642 def test_do_not_verify_post_job_cleanup_failure(self): 643 queue_entry = self._setup_for_do_not_verify() 644 queue_entry.job.reboot_after = model_attributes.RebootAfter.ALWAYS 645 queue_entry.job.save() 646 647 self._run_post_job_cleanup_failure_up_to_repair(queue_entry, 648 include_verify=False) 649 # failure ignored, host still set to Ready 650 self._check_statuses(queue_entry, HqeStatus.COMPLETED, HostStatus.READY) 651 self._run_dispatcher() # nothing else runs 652 self._assert_nothing_is_running() 653 654 655 def test_do_not_verify_requested_reverify_failure(self): 656 host = self._create_reverify_request() 657 host.protection = host_protections.Protection.DO_NOT_VERIFY 658 host.save() 659 660 self._run_dispatcher() 661 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 662 exit_status=256) 663 self._run_dispatcher() 664 self._check_host_status(host, HostStatus.READY) # ignore failure 665 self._assert_nothing_is_running() 666 667 668 def test_job_abort_in_verify(self): 669 self._initialize_test() 670 job = self._create_job(hosts=[1]) 671 queue_entries = list(job.hostqueueentry_set.all()) 672 self._run_dispatcher() # launches verify 673 self._check_statuses(queue_entries[0], HqeStatus.VERIFYING) 674 job.hostqueueentry_set.update(aborted=True) 675 self._run_dispatcher() # kills verify, launches cleanup 676 self.assert_(self.mock_drone_manager.was_last_process_killed( 677 _PidfileType.VERIFY, set([signal.SIGKILL]))) 678 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP) 679 self._run_dispatcher() 680 681 682 def test_job_abort(self): 683 self._initialize_test() 684 job = self._create_job(hosts=[1]) 685 job.run_reset = False 686 job.save() 687 queue_entries = list(job.hostqueueentry_set.all()) 688 689 self._run_dispatcher() # launches job 690 691 self._check_statuses(queue_entries[0], HqeStatus.RUNNING) 692 693 job.hostqueueentry_set.update(aborted=True) 694 695 self._run_dispatcher() # kills job, launches gathering 696 697 self._check_statuses(queue_entries[0], HqeStatus.GATHERING) 698 self.mock_drone_manager.finish_process(_PidfileType.GATHER) 699 self._run_dispatcher() # launches parsing + cleanup 700 queue_entry = job.hostqueueentry_set.all()[0] 701 self._finish_parsing() 702 # The abort will cause gathering to launch a cleanup. 703 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP) 704 self._run_dispatcher() 705 self.mock_drone_manager.finish_process(_PidfileType.RESET) 706 self._run_dispatcher() 707 708 709 def test_job_abort_queued_synchronous(self): 710 self._initialize_test() 711 job = self._create_job(hosts=[1,2]) 712 job.synch_count = 2 713 job.save() 714 715 job.hostqueueentry_set.update(aborted=True) 716 self._run_dispatcher() 717 for host_queue_entry in job.hostqueueentry_set.all(): 718 self.assertEqual(host_queue_entry.status, 719 HqeStatus.ABORTED) 720 721 722 def test_no_pidfile_leaking(self): 723 self._initialize_test() 724 725 self.test_simple_job() 726 self.mock_drone_manager.refresh() 727 self.assertEquals(self.mock_drone_manager._pidfiles, {}) 728 729 self.test_job_abort_in_verify() 730 self.mock_drone_manager.refresh() 731 self.assertEquals(self.mock_drone_manager._pidfiles, {}) 732 733 self.test_job_abort() 734 self.mock_drone_manager.refresh() 735 self.assertEquals(self.mock_drone_manager._pidfiles, {}) 736 737 738 def _make_job_and_queue_entry(self): 739 job = self._create_job(hosts=[1]) 740 queue_entry = job.hostqueueentry_set.all()[0] 741 return job, queue_entry 742 743 744 def test_recover_running_no_process(self): 745 # recovery should re-execute a Running HQE if no process is found 746 _, queue_entry = self._make_job_and_queue_entry() 747 queue_entry.status = HqeStatus.RUNNING 748 queue_entry.execution_subdir = '1-myuser/host1' 749 queue_entry.save() 750 queue_entry.host.status = HostStatus.RUNNING 751 queue_entry.host.save() 752 753 self._initialize_test() 754 self._run_dispatcher() 755 self._finish_job(queue_entry) 756 757 758 def test_recover_verifying_hqe_no_special_task(self): 759 # recovery should move a Resetting HQE with no corresponding 760 # Verify or Reset SpecialTask back to Queued. 761 _, queue_entry = self._make_job_and_queue_entry() 762 queue_entry.status = HqeStatus.RESETTING 763 queue_entry.save() 764 765 # make some dummy SpecialTasks that shouldn't count 766 models.SpecialTask.objects.create( 767 host=queue_entry.host, 768 task=models.SpecialTask.Task.RESET, 769 requested_by=models.User.current_user()) 770 models.SpecialTask.objects.create( 771 host=queue_entry.host, 772 task=models.SpecialTask.Task.CLEANUP, 773 queue_entry=queue_entry, 774 is_complete=True, 775 requested_by=models.User.current_user()) 776 777 self._initialize_test() 778 self._check_statuses(queue_entry, HqeStatus.QUEUED) 779 780 781 def _test_recover_verifying_hqe_helper(self, task, pidfile_type): 782 _, queue_entry = self._make_job_and_queue_entry() 783 queue_entry.status = HqeStatus.VERIFYING 784 queue_entry.save() 785 786 special_task = models.SpecialTask.objects.create( 787 host=queue_entry.host, task=task, queue_entry=queue_entry) 788 789 self._initialize_test() 790 self._run_dispatcher() 791 self.mock_drone_manager.finish_process(pidfile_type) 792 self._run_dispatcher() 793 # don't bother checking the rest of the job execution, as long as the 794 # SpecialTask ran 795 796 797 def test_recover_verifying_hqe_with_cleanup(self): 798 # recover an HQE that was in pre-job cleanup 799 self._test_recover_verifying_hqe_helper(models.SpecialTask.Task.CLEANUP, 800 _PidfileType.CLEANUP) 801 802 803 def test_recover_verifying_hqe_with_verify(self): 804 # recover an HQE that was in pre-job verify 805 self._test_recover_verifying_hqe_helper(models.SpecialTask.Task.VERIFY, 806 _PidfileType.VERIFY) 807 808 809 def test_recover_parsing(self): 810 self._initialize_test() 811 job, queue_entry = self._make_job_and_queue_entry() 812 job.run_verify = False 813 job.run_reset = False 814 job.reboot_after = model_attributes.RebootAfter.NEVER 815 job.save() 816 817 self._run_dispatcher() # launches job 818 self.mock_drone_manager.finish_process(_PidfileType.JOB) 819 self._run_dispatcher() # launches parsing 820 821 # now "restart" the scheduler 822 self._create_dispatcher() 823 self._initialize_test() 824 self._run_dispatcher() 825 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 826 self._run_dispatcher() 827 828 829 def test_recover_parsing__no_process_already_aborted(self): 830 _, queue_entry = self._make_job_and_queue_entry() 831 queue_entry.execution_subdir = 'host1' 832 queue_entry.status = HqeStatus.PARSING 833 queue_entry.aborted = True 834 queue_entry.save() 835 836 self._initialize_test() 837 self._run_dispatcher() 838 839 840 def test_job_scheduled_just_after_abort(self): 841 # test a pretty obscure corner case where a job is aborted while queued, 842 # another job is ready to run, and throttling is active. the post-abort 843 # cleanup must not be pre-empted by the second job. 844 # This test kind of doesn't make sense anymore after verify+cleanup 845 # were merged into reset. It should maybe just be removed. 846 job1, queue_entry1 = self._make_job_and_queue_entry() 847 queue_entry1.save() 848 job2, queue_entry2 = self._make_job_and_queue_entry() 849 job2.reboot_before = model_attributes.RebootBefore.IF_DIRTY 850 job2.save() 851 852 self.mock_drone_manager.process_capacity = 0 853 self._run_dispatcher() # schedule job1, but won't start verify 854 job1.hostqueueentry_set.update(aborted=True) 855 self.mock_drone_manager.process_capacity = 100 856 self._run_dispatcher() # reset must run here, not verify for job2 857 self._check_statuses(queue_entry1, HqeStatus.ABORTED, 858 HostStatus.RESETTING) 859 self.mock_drone_manager.finish_process(_PidfileType.RESET) 860 self._run_dispatcher() # now verify starts for job2 861 self._check_statuses(queue_entry2, HqeStatus.RUNNING, 862 HostStatus.RUNNING) 863 864 865 def test_reverify_interrupting_pre_job(self): 866 # ensure things behave sanely if a reverify is scheduled in the middle 867 # of pre-job actions 868 _, queue_entry = self._make_job_and_queue_entry() 869 870 self._run_dispatcher() # pre-job verify 871 self._create_reverify_request() 872 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 873 exit_status=256) 874 self._run_dispatcher() # repair 875 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 876 self._run_dispatcher() # reverify runs now 877 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 878 self._run_dispatcher() # pre-job verify 879 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 880 self._run_dispatcher() # and job runs... 881 self._check_statuses(queue_entry, HqeStatus.RUNNING, HostStatus.RUNNING) 882 self._finish_job(queue_entry) # reverify has been deleted 883 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 884 HostStatus.READY) 885 self._assert_nothing_is_running() 886 887 888 def test_reverify_while_job_running(self): 889 # once a job is running, a reverify must not be allowed to preempt 890 # Gathering 891 _, queue_entry = self._make_job_and_queue_entry() 892 self._run_pre_job_verify(queue_entry) 893 self._run_dispatcher() # job runs 894 self._create_reverify_request() 895 # make job end with a signal, so gathering will run 896 self.mock_drone_manager.finish_process(_PidfileType.JOB, 897 exit_status=271) 898 self._run_dispatcher() # gathering must start 899 self.mock_drone_manager.finish_process(_PidfileType.GATHER) 900 self._run_dispatcher() # parsing and cleanup 901 self._finish_parsing() 902 self._run_dispatcher() # now reverify runs 903 self._check_statuses(queue_entry, HqeStatus.FAILED, 904 HostStatus.VERIFYING) 905 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 906 self._run_dispatcher() 907 self._check_host_status(queue_entry.host, HostStatus.READY) 908 909 910 def test_reverify_while_host_pending(self): 911 # ensure that if a reverify is scheduled while a host is in Pending, it 912 # won't run until the host is actually free 913 job = self._create_job(hosts=[1,2]) 914 queue_entry = job.hostqueueentry_set.get(host__hostname='host1') 915 job.synch_count = 2 916 job.save() 917 918 host2 = self.hosts[1] 919 host2.locked = True 920 host2.save() 921 922 self._run_dispatcher() # verify host1 923 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 924 self._run_dispatcher() # host1 Pending 925 self._check_statuses(queue_entry, HqeStatus.PENDING, HostStatus.PENDING) 926 self._create_reverify_request() 927 self._run_dispatcher() # nothing should happen here 928 self._check_statuses(queue_entry, HqeStatus.PENDING, HostStatus.PENDING) 929 930 # now let the job run 931 host2.locked = False 932 host2.save() 933 self._run_dispatcher() # verify host2 934 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 935 self._run_dispatcher() # run job 936 self._finish_job(queue_entry) 937 # the reverify should now be running 938 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 939 HostStatus.VERIFYING) 940 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 941 self._run_dispatcher() 942 self._check_host_status(queue_entry.host, HostStatus.READY) 943 944 945 def test_throttling(self): 946 job = self._create_job(hosts=[1,2,3]) 947 job.synch_count = 3 948 job.save() 949 950 queue_entries = list(job.hostqueueentry_set.all()) 951 def _check_hqe_statuses(*statuses): 952 for queue_entry, status in zip(queue_entries, statuses): 953 self._check_statuses(queue_entry, status) 954 955 self.mock_drone_manager.process_capacity = 2 956 self._run_dispatcher() # verify runs on 1 and 2 957 queue_entries = list(job.hostqueueentry_set.all()) 958 _check_hqe_statuses(HqeStatus.QUEUED, 959 HqeStatus.VERIFYING, HqeStatus.VERIFYING) 960 self.assertEquals(len(self.mock_drone_manager.running_pidfile_ids()), 2) 961 962 self.mock_drone_manager.finish_specific_process( 963 'hosts/host3/1-verify', drone_manager.AUTOSERV_PID_FILE) 964 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 965 self._run_dispatcher() # verify runs on 3 966 _check_hqe_statuses(HqeStatus.VERIFYING, HqeStatus.PENDING, 967 HqeStatus.PENDING) 968 969 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 970 self._run_dispatcher() # job won't run due to throttling 971 _check_hqe_statuses(HqeStatus.STARTING, HqeStatus.STARTING, 972 HqeStatus.STARTING) 973 self._assert_nothing_is_running() 974 975 self.mock_drone_manager.process_capacity = 3 976 self._run_dispatcher() # now job runs 977 _check_hqe_statuses(HqeStatus.RUNNING, HqeStatus.RUNNING, 978 HqeStatus.RUNNING) 979 980 self.mock_drone_manager.process_capacity = 2 981 self.mock_drone_manager.finish_process(_PidfileType.JOB, 982 exit_status=271) 983 self._run_dispatcher() # gathering won't run due to throttling 984 _check_hqe_statuses(HqeStatus.GATHERING, HqeStatus.GATHERING, 985 HqeStatus.GATHERING) 986 self._assert_nothing_is_running() 987 988 self.mock_drone_manager.process_capacity = 3 989 self._run_dispatcher() # now gathering runs 990 991 self.mock_drone_manager.process_capacity = 0 992 self.mock_drone_manager.finish_process(_PidfileType.GATHER) 993 self._run_dispatcher() # parsing runs despite throttling 994 _check_hqe_statuses(HqeStatus.PARSING, HqeStatus.PARSING, 995 HqeStatus.PARSING) 996 997 998 def test_abort_starting_while_throttling(self): 999 self._initialize_test() 1000 job = self._create_job(hosts=[1,2], synchronous=True) 1001 queue_entry = job.hostqueueentry_set.all()[0] 1002 job.run_verify = False 1003 job.run_reset = False 1004 job.reboot_after = model_attributes.RebootAfter.NEVER 1005 job.save() 1006 1007 self.mock_drone_manager.process_capacity = 0 1008 self._run_dispatcher() # go to starting, but don't start job 1009 self._check_statuses(queue_entry, HqeStatus.STARTING, 1010 HostStatus.PENDING) 1011 1012 job.hostqueueentry_set.update(aborted=True) 1013 self._run_dispatcher() 1014 self._check_statuses(queue_entry, HqeStatus.GATHERING, 1015 HostStatus.RUNNING) 1016 1017 self.mock_drone_manager.process_capacity = 5 1018 self._run_dispatcher() 1019 self._check_statuses(queue_entry, HqeStatus.ABORTED, 1020 HostStatus.CLEANING) 1021 1022 1023 def test_simple_metahost_assignment(self): 1024 job = self._create_job(metahosts=[1]) 1025 self._run_dispatcher() 1026 entry = job.hostqueueentry_set.all()[0] 1027 self.assertEquals(entry.host.hostname, 'host1') 1028 self._check_statuses(entry, HqeStatus.VERIFYING, HostStatus.VERIFYING) 1029 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 1030 self._run_dispatcher() 1031 self._check_statuses(entry, HqeStatus.RUNNING, HostStatus.RUNNING) 1032 # rest of job proceeds normally 1033 1034 1035 def test_metahost_fail_verify(self): 1036 self.hosts[1].labels.add(self.labels[0]) # put label1 also on host2 1037 job = self._create_job(metahosts=[1]) 1038 self._run_dispatcher() # assigned to host1 1039 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 1040 exit_status=256) 1041 self._run_dispatcher() # host1 failed, gets reassigned to host2 1042 entry = job.hostqueueentry_set.all()[0] 1043 self.assertEquals(entry.host.hostname, 'host2') 1044 self._check_statuses(entry, HqeStatus.VERIFYING, HostStatus.VERIFYING) 1045 self._check_host_status(self.hosts[0], HostStatus.REPAIRING) 1046 1047 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 1048 self._run_dispatcher() 1049 self._check_statuses(entry, HqeStatus.RUNNING, HostStatus.RUNNING) 1050 1051 1052 def test_hostless_job(self): 1053 job = self._create_job(hostless=True) 1054 entry = job.hostqueueentry_set.all()[0] 1055 1056 self._run_dispatcher() 1057 self._check_entry_status(entry, HqeStatus.RUNNING) 1058 1059 self.mock_drone_manager.finish_process(_PidfileType.JOB) 1060 self._run_dispatcher() 1061 self._check_entry_status(entry, HqeStatus.PARSING) 1062 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1063 self._run_dispatcher() 1064 self._check_entry_status(entry, HqeStatus.COMPLETED) 1065 1066 1067 def test_pre_job_keyvals(self): 1068 job = self._create_job(hosts=[1]) 1069 job.run_verify = False 1070 job.run_reset = False 1071 job.reboot_before = model_attributes.RebootBefore.NEVER 1072 job.save() 1073 models.JobKeyval.objects.create(job=job, key='mykey', value='myvalue') 1074 1075 self._run_dispatcher() 1076 self._finish_job(job.hostqueueentry_set.all()[0]) 1077 1078 attached_files = self.mock_drone_manager.attached_files( 1079 '1-autotest_system/host1') 1080 job_keyval_path = '1-autotest_system/host1/keyval' 1081 self.assert_(job_keyval_path in attached_files, attached_files) 1082 keyval_contents = attached_files[job_keyval_path] 1083 keyval_dict = dict(line.strip().split('=', 1) 1084 for line in keyval_contents.splitlines()) 1085 self.assert_('job_queued' in keyval_dict, keyval_dict) 1086 self.assertEquals(keyval_dict['mykey'], 'myvalue') 1087 1088 1089 # This tests the scheduler functions with archiving step disabled 1090 class SchedulerFunctionalTestNoArchiving(SchedulerFunctionalTest): 1091 def _set_global_config_values(self): 1092 super(SchedulerFunctionalTestNoArchiving, self 1093 )._set_global_config_values() 1094 self.mock_config.set_config_value('SCHEDULER', 'enable_archiving', 1095 False) 1096 1097 1098 def _finish_parsing(self): 1099 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1100 self._run_dispatcher() 1101 1102 1103 def _run_post_job_cleanup_failure_up_to_repair(self, queue_entry, 1104 include_verify=True): 1105 if include_verify: 1106 self._run_pre_job_verify(queue_entry) 1107 self._run_dispatcher() # job 1108 self.mock_drone_manager.finish_process(_PidfileType.JOB) 1109 self._run_dispatcher() # parsing + cleanup 1110 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1111 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP, 1112 exit_status=256) 1113 self._run_dispatcher() # repair, HQE unaffected 1114 return queue_entry 1115 1116 1117 def test_hostless_job(self): 1118 job = self._create_job(hostless=True) 1119 entry = job.hostqueueentry_set.all()[0] 1120 1121 self._run_dispatcher() 1122 self._check_entry_status(entry, HqeStatus.RUNNING) 1123 1124 self.mock_drone_manager.finish_process(_PidfileType.JOB) 1125 self._run_dispatcher() 1126 self._check_entry_status(entry, HqeStatus.PARSING) 1127 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1128 self._run_dispatcher() 1129 self._check_entry_status(entry, HqeStatus.COMPLETED) 1130 1131 def test_synchronous_with_reset(self): 1132 # For crbug/621257. 1133 job = self._create_job(hosts=[1, 2]) 1134 job.synch_count = 2 1135 job.reboot_before = model_attributes.RebootBefore.ALWAYS 1136 job.save() 1137 1138 hqe1 = job.hostqueueentry_set.get(host__hostname='host1') 1139 hqe2 = job.hostqueueentry_set.get(host__hostname='host2') 1140 1141 self._run_dispatcher() 1142 1143 self._check_statuses(hqe1, HqeStatus.RESETTING, HostStatus.RESETTING) 1144 self._check_statuses(hqe2, HqeStatus.RESETTING, HostStatus.RESETTING) 1145 1146 self.mock_drone_manager.finish_active_process_on_host(1) 1147 self._run_dispatcher() 1148 1149 self._check_statuses(hqe1, HqeStatus.PENDING, HostStatus.PENDING) 1150 self._check_statuses(hqe2, HqeStatus.RESETTING, HostStatus.RESETTING) 1151 1152 self.mock_drone_manager.finish_active_process_on_host(2) 1153 self._run_dispatcher() 1154 1155 self._check_statuses(hqe1, HqeStatus.RUNNING, HostStatus.RUNNING) 1156 self._check_statuses(hqe2, HqeStatus.RUNNING, HostStatus.RUNNING) 1157 1158 1159 if __name__ == '__main__': 1160 unittest.main() 1161
