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', 'enable_archiving', True) 368 self.mock_config.set_config_value('SCHEDULER', 369 'clean_interval_minutes', 60) 370 self.mock_config.set_config_value('SCHEDULER', 371 'max_parse_processes', 50) 372 self.mock_config.set_config_value('SCHEDULER', 373 'max_transfer_processes', 50) 374 self.mock_config.set_config_value('SCHEDULER', 375 'clean_interval_minutes', 50) 376 self.mock_config.set_config_value('SCHEDULER', 377 'max_provision_retries', 1) 378 self.mock_config.set_config_value('SCHEDULER', 'max_repair_limit', 1) 379 self.mock_config.set_config_value( 380 'SCHEDULER', 'secs_to_wait_for_atomic_group_hosts', 600) 381 self.mock_config.set_config_value( 382 'SCHEDULER', 'inline_host_acquisition', True) 383 scheduler_config.config.read_config() 384 385 386 def _initialize_test(self): 387 self.dispatcher.initialize() 388 389 390 def _run_dispatcher(self): 391 for _ in xrange(self._A_LOT_OF_TICKS): 392 self.dispatcher.tick() 393 394 395 def test_idle(self): 396 self._initialize_test() 397 self._run_dispatcher() 398 399 400 def _assert_process_executed(self, working_directory, pidfile_name): 401 process_was_executed = self.mock_drone_manager.was_process_executed( 402 'hosts/host1/1-verify', drone_manager.AUTOSERV_PID_FILE) 403 self.assert_(process_was_executed, 404 '%s/%s not executed' % (working_directory, pidfile_name)) 405 406 407 def _update_instance(self, model_instance): 408 return type(model_instance).objects.get(pk=model_instance.pk) 409 410 411 def _check_statuses(self, queue_entry, queue_entry_status, 412 host_status=None): 413 self._check_entry_status(queue_entry, queue_entry_status) 414 if host_status: 415 self._check_host_status(queue_entry.host, host_status) 416 417 418 def _check_entry_status(self, queue_entry, status): 419 # update from DB 420 queue_entry = self._update_instance(queue_entry) 421 self.assertEquals(queue_entry.status, status) 422 423 424 def _check_host_status(self, host, status): 425 # update from DB 426 host = self._update_instance(host) 427 self.assertEquals(host.status, status) 428 429 430 def _run_pre_job_verify(self, queue_entry): 431 self._run_dispatcher() # launches verify 432 self._check_statuses(queue_entry, HqeStatus.VERIFYING, 433 HostStatus.VERIFYING) 434 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 435 436 437 def test_simple_job(self): 438 self._initialize_test() 439 job, queue_entry = self._make_job_and_queue_entry() 440 self._run_pre_job_verify(queue_entry) 441 self._run_dispatcher() # launches job 442 self._check_statuses(queue_entry, HqeStatus.RUNNING, HostStatus.RUNNING) 443 self._finish_job(queue_entry) 444 self._check_statuses(queue_entry, HqeStatus.COMPLETED, HostStatus.READY) 445 self._assert_nothing_is_running() 446 447 448 def _setup_for_pre_job_reset(self): 449 self._initialize_test() 450 job, queue_entry = self._make_job_and_queue_entry() 451 job.reboot_before = model_attributes.RebootBefore.ALWAYS 452 job.save() 453 return queue_entry 454 455 456 def _run_pre_job_reset_job(self, queue_entry): 457 self._run_dispatcher() # reset 458 self._check_statuses(queue_entry, HqeStatus.RESETTING, 459 HostStatus.RESETTING) 460 self.mock_drone_manager.finish_process(_PidfileType.RESET) 461 self._run_dispatcher() # job 462 self._finish_job(queue_entry) 463 464 465 def test_pre_job_reset(self): 466 queue_entry = self._setup_for_pre_job_reset() 467 self._run_pre_job_reset_job(queue_entry) 468 469 470 def _run_pre_job_reset_one_failure(self): 471 queue_entry = self._setup_for_pre_job_reset() 472 self._run_dispatcher() # reset 473 self.mock_drone_manager.finish_process(_PidfileType.RESET, 474 exit_status=256) 475 self._run_dispatcher() # repair 476 self._check_statuses(queue_entry, HqeStatus.QUEUED, 477 HostStatus.REPAIRING) 478 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 479 return queue_entry 480 481 482 def test_pre_job_reset_failure(self): 483 queue_entry = self._run_pre_job_reset_one_failure() 484 # from here the job should run as normal 485 self._run_pre_job_reset_job(queue_entry) 486 487 488 def test_pre_job_reset_double_failure(self): 489 # TODO (showard): this test isn't perfect. in reality, when the second 490 # reset fails, it copies its results over to the job directory using 491 # copy_results_on_drone() and then parses them. since we don't handle 492 # that, there appear to be no results at the job directory. the 493 # scheduler handles this gracefully, parsing gets effectively skipped, 494 # and this test passes as is. but we ought to properly test that 495 # behavior. 496 queue_entry = self._run_pre_job_reset_one_failure() 497 self._run_dispatcher() # second reset 498 self.mock_drone_manager.finish_process(_PidfileType.RESET, 499 exit_status=256) 500 self._run_dispatcher() 501 self._check_statuses(queue_entry, HqeStatus.FAILED, 502 HostStatus.REPAIR_FAILED) 503 # nothing else should run 504 self._assert_nothing_is_running() 505 506 507 def _assert_nothing_is_running(self): 508 self.assertEquals(self.mock_drone_manager.running_pidfile_ids(), []) 509 510 511 def _setup_for_post_job_cleanup(self): 512 self._initialize_test() 513 job, queue_entry = self._make_job_and_queue_entry() 514 job.reboot_after = model_attributes.RebootAfter.ALWAYS 515 job.save() 516 return queue_entry 517 518 519 def _run_post_job_cleanup_failure_up_to_repair(self, queue_entry, 520 include_verify=True): 521 if include_verify: 522 self._run_pre_job_verify(queue_entry) 523 self._run_dispatcher() # job 524 self.mock_drone_manager.finish_process(_PidfileType.JOB) 525 self._run_dispatcher() # parsing + cleanup 526 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 527 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP, 528 exit_status=256) 529 self._run_dispatcher() # repair, HQE unaffected 530 return queue_entry 531 532 533 def test_post_job_cleanup_failure(self): 534 queue_entry = self._setup_for_post_job_cleanup() 535 self._run_post_job_cleanup_failure_up_to_repair(queue_entry) 536 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 537 HostStatus.REPAIRING) 538 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 539 self._run_dispatcher() 540 self._check_statuses(queue_entry, HqeStatus.COMPLETED, HostStatus.READY) 541 542 543 def test_post_job_cleanup_failure_repair_failure(self): 544 queue_entry = self._setup_for_post_job_cleanup() 545 self._run_post_job_cleanup_failure_up_to_repair(queue_entry) 546 self.mock_drone_manager.finish_process(_PidfileType.REPAIR, 547 exit_status=256) 548 self._run_dispatcher() 549 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 550 HostStatus.REPAIR_FAILED) 551 552 553 def _ensure_post_job_process_is_paired(self, queue_entry, pidfile_type): 554 pidfile_name = _PIDFILE_TYPE_TO_PIDFILE[pidfile_type] 555 queue_entry = self._update_instance(queue_entry) 556 pidfile_id = self.mock_drone_manager.pidfile_from_path( 557 queue_entry.execution_path(), pidfile_name) 558 self.assert_(pidfile_id._paired_with_pidfile) 559 560 561 def _finish_job(self, queue_entry): 562 self._check_statuses(queue_entry, HqeStatus.RUNNING) 563 self.mock_drone_manager.finish_process(_PidfileType.JOB) 564 self._run_dispatcher() # launches parsing 565 self._check_statuses(queue_entry, HqeStatus.PARSING) 566 self._ensure_post_job_process_is_paired(queue_entry, _PidfileType.PARSE) 567 self._finish_parsing() 568 569 570 def _finish_parsing(self): 571 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 572 self._run_dispatcher() 573 574 575 def _create_reverify_request(self): 576 host = self.hosts[0] 577 models.SpecialTask.schedule_special_task( 578 host=host, task=models.SpecialTask.Task.VERIFY) 579 return host 580 581 582 def test_requested_reverify(self): 583 host = self._create_reverify_request() 584 self._run_dispatcher() 585 self._check_host_status(host, HostStatus.VERIFYING) 586 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 587 self._run_dispatcher() 588 self._check_host_status(host, HostStatus.READY) 589 590 591 def test_requested_reverify_failure(self): 592 host = self._create_reverify_request() 593 self._run_dispatcher() 594 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 595 exit_status=256) 596 self._run_dispatcher() # repair 597 self._check_host_status(host, HostStatus.REPAIRING) 598 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 599 self._run_dispatcher() 600 self._check_host_status(host, HostStatus.READY) 601 602 603 def _setup_for_do_not_verify(self): 604 self._initialize_test() 605 job, queue_entry = self._make_job_and_queue_entry() 606 queue_entry.host.protection = host_protections.Protection.DO_NOT_VERIFY 607 queue_entry.host.save() 608 return queue_entry 609 610 611 def test_do_not_verify_job(self): 612 queue_entry = self._setup_for_do_not_verify() 613 self._run_dispatcher() # runs job directly 614 self._finish_job(queue_entry) 615 616 617 def test_do_not_verify_job_with_cleanup(self): 618 queue_entry = self._setup_for_do_not_verify() 619 queue_entry.job.reboot_before = model_attributes.RebootBefore.ALWAYS 620 queue_entry.job.save() 621 622 self._run_dispatcher() # cleanup 623 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP) 624 self._run_dispatcher() # job 625 self._finish_job(queue_entry) 626 627 628 def test_do_not_verify_pre_job_cleanup_failure(self): 629 queue_entry = self._setup_for_do_not_verify() 630 queue_entry.job.reboot_before = model_attributes.RebootBefore.ALWAYS 631 queue_entry.job.save() 632 633 self._run_dispatcher() # cleanup 634 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP, 635 exit_status=256) 636 self._run_dispatcher() # failure ignored; job runs 637 self._finish_job(queue_entry) 638 639 640 def test_do_not_verify_post_job_cleanup_failure(self): 641 queue_entry = self._setup_for_do_not_verify() 642 queue_entry.job.reboot_after = model_attributes.RebootAfter.ALWAYS 643 queue_entry.job.save() 644 645 self._run_post_job_cleanup_failure_up_to_repair(queue_entry, 646 include_verify=False) 647 # failure ignored, host still set to Ready 648 self._check_statuses(queue_entry, HqeStatus.COMPLETED, HostStatus.READY) 649 self._run_dispatcher() # nothing else runs 650 self._assert_nothing_is_running() 651 652 653 def test_do_not_verify_requested_reverify_failure(self): 654 host = self._create_reverify_request() 655 host.protection = host_protections.Protection.DO_NOT_VERIFY 656 host.save() 657 658 self._run_dispatcher() 659 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 660 exit_status=256) 661 self._run_dispatcher() 662 self._check_host_status(host, HostStatus.READY) # ignore failure 663 self._assert_nothing_is_running() 664 665 666 def test_job_abort_in_verify(self): 667 self._initialize_test() 668 job = self._create_job(hosts=[1]) 669 queue_entries = list(job.hostqueueentry_set.all()) 670 self._run_dispatcher() # launches verify 671 self._check_statuses(queue_entries[0], HqeStatus.VERIFYING) 672 job.hostqueueentry_set.update(aborted=True) 673 self._run_dispatcher() # kills verify, launches cleanup 674 self.assert_(self.mock_drone_manager.was_last_process_killed( 675 _PidfileType.VERIFY, set([signal.SIGKILL]))) 676 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP) 677 self._run_dispatcher() 678 679 680 def test_job_abort(self): 681 self._initialize_test() 682 job = self._create_job(hosts=[1]) 683 job.run_reset = False 684 job.save() 685 queue_entries = list(job.hostqueueentry_set.all()) 686 687 self._run_dispatcher() # launches job 688 689 self._check_statuses(queue_entries[0], HqeStatus.RUNNING) 690 691 job.hostqueueentry_set.update(aborted=True) 692 693 self._run_dispatcher() # kills job, launches gathering 694 695 self._check_statuses(queue_entries[0], HqeStatus.GATHERING) 696 self.mock_drone_manager.finish_process(_PidfileType.GATHER) 697 self._run_dispatcher() # launches parsing + cleanup 698 queue_entry = job.hostqueueentry_set.all()[0] 699 self._finish_parsing() 700 # The abort will cause gathering to launch a cleanup. 701 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP) 702 self._run_dispatcher() 703 self.mock_drone_manager.finish_process(_PidfileType.RESET) 704 self._run_dispatcher() 705 706 707 def test_job_abort_queued_synchronous(self): 708 self._initialize_test() 709 job = self._create_job(hosts=[1,2]) 710 job.synch_count = 2 711 job.save() 712 713 job.hostqueueentry_set.update(aborted=True) 714 self._run_dispatcher() 715 for host_queue_entry in job.hostqueueentry_set.all(): 716 self.assertEqual(host_queue_entry.status, 717 HqeStatus.ABORTED) 718 719 720 def test_no_pidfile_leaking(self): 721 self._initialize_test() 722 723 self.test_simple_job() 724 self.mock_drone_manager.refresh() 725 self.assertEquals(self.mock_drone_manager._pidfiles, {}) 726 727 self.test_job_abort_in_verify() 728 self.mock_drone_manager.refresh() 729 self.assertEquals(self.mock_drone_manager._pidfiles, {}) 730 731 self.test_job_abort() 732 self.mock_drone_manager.refresh() 733 self.assertEquals(self.mock_drone_manager._pidfiles, {}) 734 735 736 def _make_job_and_queue_entry(self): 737 job = self._create_job(hosts=[1]) 738 queue_entry = job.hostqueueentry_set.all()[0] 739 return job, queue_entry 740 741 742 def test_recover_running_no_process(self): 743 # recovery should re-execute a Running HQE if no process is found 744 _, queue_entry = self._make_job_and_queue_entry() 745 queue_entry.status = HqeStatus.RUNNING 746 queue_entry.execution_subdir = '1-myuser/host1' 747 queue_entry.save() 748 queue_entry.host.status = HostStatus.RUNNING 749 queue_entry.host.save() 750 751 self._initialize_test() 752 self._run_dispatcher() 753 self._finish_job(queue_entry) 754 755 756 def test_recover_verifying_hqe_no_special_task(self): 757 # recovery should move a Resetting HQE with no corresponding 758 # Verify or Reset SpecialTask back to Queued. 759 _, queue_entry = self._make_job_and_queue_entry() 760 queue_entry.status = HqeStatus.RESETTING 761 queue_entry.save() 762 763 # make some dummy SpecialTasks that shouldn't count 764 models.SpecialTask.objects.create( 765 host=queue_entry.host, 766 task=models.SpecialTask.Task.RESET, 767 requested_by=models.User.current_user()) 768 models.SpecialTask.objects.create( 769 host=queue_entry.host, 770 task=models.SpecialTask.Task.CLEANUP, 771 queue_entry=queue_entry, 772 is_complete=True, 773 requested_by=models.User.current_user()) 774 775 self._initialize_test() 776 self._check_statuses(queue_entry, HqeStatus.QUEUED) 777 778 779 def _test_recover_verifying_hqe_helper(self, task, pidfile_type): 780 _, queue_entry = self._make_job_and_queue_entry() 781 queue_entry.status = HqeStatus.VERIFYING 782 queue_entry.save() 783 784 special_task = models.SpecialTask.objects.create( 785 host=queue_entry.host, task=task, queue_entry=queue_entry) 786 787 self._initialize_test() 788 self._run_dispatcher() 789 self.mock_drone_manager.finish_process(pidfile_type) 790 self._run_dispatcher() 791 # don't bother checking the rest of the job execution, as long as the 792 # SpecialTask ran 793 794 795 def test_recover_verifying_hqe_with_cleanup(self): 796 # recover an HQE that was in pre-job cleanup 797 self._test_recover_verifying_hqe_helper(models.SpecialTask.Task.CLEANUP, 798 _PidfileType.CLEANUP) 799 800 801 def test_recover_verifying_hqe_with_verify(self): 802 # recover an HQE that was in pre-job verify 803 self._test_recover_verifying_hqe_helper(models.SpecialTask.Task.VERIFY, 804 _PidfileType.VERIFY) 805 806 807 def test_recover_parsing(self): 808 self._initialize_test() 809 job, queue_entry = self._make_job_and_queue_entry() 810 job.run_verify = False 811 job.run_reset = False 812 job.reboot_after = model_attributes.RebootAfter.NEVER 813 job.save() 814 815 self._run_dispatcher() # launches job 816 self.mock_drone_manager.finish_process(_PidfileType.JOB) 817 self._run_dispatcher() # launches parsing 818 819 # now "restart" the scheduler 820 self._create_dispatcher() 821 self._initialize_test() 822 self._run_dispatcher() 823 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 824 self._run_dispatcher() 825 826 827 def test_recover_parsing__no_process_already_aborted(self): 828 _, queue_entry = self._make_job_and_queue_entry() 829 queue_entry.execution_subdir = 'host1' 830 queue_entry.status = HqeStatus.PARSING 831 queue_entry.aborted = True 832 queue_entry.save() 833 834 self._initialize_test() 835 self._run_dispatcher() 836 837 838 def test_job_scheduled_just_after_abort(self): 839 # test a pretty obscure corner case where a job is aborted while queued, 840 # another job is ready to run, and throttling is active. the post-abort 841 # cleanup must not be pre-empted by the second job. 842 # This test kind of doesn't make sense anymore after verify+cleanup 843 # were merged into reset. It should maybe just be removed. 844 job1, queue_entry1 = self._make_job_and_queue_entry() 845 queue_entry1.save() 846 job2, queue_entry2 = self._make_job_and_queue_entry() 847 job2.reboot_before = model_attributes.RebootBefore.IF_DIRTY 848 job2.save() 849 850 self.mock_drone_manager.process_capacity = 0 851 self._run_dispatcher() # schedule job1, but won't start verify 852 job1.hostqueueentry_set.update(aborted=True) 853 self.mock_drone_manager.process_capacity = 100 854 self._run_dispatcher() # reset must run here, not verify for job2 855 self._check_statuses(queue_entry1, HqeStatus.ABORTED, 856 HostStatus.RESETTING) 857 self.mock_drone_manager.finish_process(_PidfileType.RESET) 858 self._run_dispatcher() # now verify starts for job2 859 self._check_statuses(queue_entry2, HqeStatus.RUNNING, 860 HostStatus.RUNNING) 861 862 863 def test_reverify_interrupting_pre_job(self): 864 # ensure things behave sanely if a reverify is scheduled in the middle 865 # of pre-job actions 866 _, queue_entry = self._make_job_and_queue_entry() 867 868 self._run_dispatcher() # pre-job verify 869 self._create_reverify_request() 870 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 871 exit_status=256) 872 self._run_dispatcher() # repair 873 self.mock_drone_manager.finish_process(_PidfileType.REPAIR) 874 self._run_dispatcher() # reverify runs now 875 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 876 self._run_dispatcher() # pre-job verify 877 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 878 self._run_dispatcher() # and job runs... 879 self._check_statuses(queue_entry, HqeStatus.RUNNING, HostStatus.RUNNING) 880 self._finish_job(queue_entry) # reverify has been deleted 881 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 882 HostStatus.READY) 883 self._assert_nothing_is_running() 884 885 886 def test_reverify_while_job_running(self): 887 # once a job is running, a reverify must not be allowed to preempt 888 # Gathering 889 _, queue_entry = self._make_job_and_queue_entry() 890 self._run_pre_job_verify(queue_entry) 891 self._run_dispatcher() # job runs 892 self._create_reverify_request() 893 # make job end with a signal, so gathering will run 894 self.mock_drone_manager.finish_process(_PidfileType.JOB, 895 exit_status=271) 896 self._run_dispatcher() # gathering must start 897 self.mock_drone_manager.finish_process(_PidfileType.GATHER) 898 self._run_dispatcher() # parsing and cleanup 899 self._finish_parsing() 900 self._run_dispatcher() # now reverify runs 901 self._check_statuses(queue_entry, HqeStatus.FAILED, 902 HostStatus.VERIFYING) 903 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 904 self._run_dispatcher() 905 self._check_host_status(queue_entry.host, HostStatus.READY) 906 907 908 def test_reverify_while_host_pending(self): 909 # ensure that if a reverify is scheduled while a host is in Pending, it 910 # won't run until the host is actually free 911 job = self._create_job(hosts=[1,2]) 912 queue_entry = job.hostqueueentry_set.get(host__hostname='host1') 913 job.synch_count = 2 914 job.save() 915 916 host2 = self.hosts[1] 917 host2.locked = True 918 host2.save() 919 920 self._run_dispatcher() # verify host1 921 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 922 self._run_dispatcher() # host1 Pending 923 self._check_statuses(queue_entry, HqeStatus.PENDING, HostStatus.PENDING) 924 self._create_reverify_request() 925 self._run_dispatcher() # nothing should happen here 926 self._check_statuses(queue_entry, HqeStatus.PENDING, HostStatus.PENDING) 927 928 # now let the job run 929 host2.locked = False 930 host2.save() 931 self._run_dispatcher() # verify host2 932 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 933 self._run_dispatcher() # run job 934 self._finish_job(queue_entry) 935 # the reverify should now be running 936 self._check_statuses(queue_entry, HqeStatus.COMPLETED, 937 HostStatus.VERIFYING) 938 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 939 self._run_dispatcher() 940 self._check_host_status(queue_entry.host, HostStatus.READY) 941 942 943 def test_throttling(self): 944 job = self._create_job(hosts=[1,2,3]) 945 job.synch_count = 3 946 job.save() 947 948 queue_entries = list(job.hostqueueentry_set.all()) 949 def _check_hqe_statuses(*statuses): 950 for queue_entry, status in zip(queue_entries, statuses): 951 self._check_statuses(queue_entry, status) 952 953 self.mock_drone_manager.process_capacity = 2 954 self._run_dispatcher() # verify runs on 1 and 2 955 queue_entries = list(job.hostqueueentry_set.all()) 956 _check_hqe_statuses(HqeStatus.QUEUED, 957 HqeStatus.VERIFYING, HqeStatus.VERIFYING) 958 self.assertEquals(len(self.mock_drone_manager.running_pidfile_ids()), 2) 959 960 self.mock_drone_manager.finish_specific_process( 961 'hosts/host3/1-verify', drone_manager.AUTOSERV_PID_FILE) 962 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 963 self._run_dispatcher() # verify runs on 3 964 _check_hqe_statuses(HqeStatus.VERIFYING, HqeStatus.PENDING, 965 HqeStatus.PENDING) 966 967 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 968 self._run_dispatcher() # job won't run due to throttling 969 _check_hqe_statuses(HqeStatus.STARTING, HqeStatus.STARTING, 970 HqeStatus.STARTING) 971 self._assert_nothing_is_running() 972 973 self.mock_drone_manager.process_capacity = 3 974 self._run_dispatcher() # now job runs 975 _check_hqe_statuses(HqeStatus.RUNNING, HqeStatus.RUNNING, 976 HqeStatus.RUNNING) 977 978 self.mock_drone_manager.process_capacity = 2 979 self.mock_drone_manager.finish_process(_PidfileType.JOB, 980 exit_status=271) 981 self._run_dispatcher() # gathering won't run due to throttling 982 _check_hqe_statuses(HqeStatus.GATHERING, HqeStatus.GATHERING, 983 HqeStatus.GATHERING) 984 self._assert_nothing_is_running() 985 986 self.mock_drone_manager.process_capacity = 3 987 self._run_dispatcher() # now gathering runs 988 989 self.mock_drone_manager.process_capacity = 0 990 self.mock_drone_manager.finish_process(_PidfileType.GATHER) 991 self._run_dispatcher() # parsing runs despite throttling 992 _check_hqe_statuses(HqeStatus.PARSING, HqeStatus.PARSING, 993 HqeStatus.PARSING) 994 995 996 def test_abort_starting_while_throttling(self): 997 self._initialize_test() 998 job = self._create_job(hosts=[1,2], synchronous=True) 999 queue_entry = job.hostqueueentry_set.all()[0] 1000 job.run_verify = False 1001 job.run_reset = False 1002 job.reboot_after = model_attributes.RebootAfter.NEVER 1003 job.save() 1004 1005 self.mock_drone_manager.process_capacity = 0 1006 self._run_dispatcher() # go to starting, but don't start job 1007 self._check_statuses(queue_entry, HqeStatus.STARTING, 1008 HostStatus.PENDING) 1009 1010 job.hostqueueentry_set.update(aborted=True) 1011 self._run_dispatcher() 1012 self._check_statuses(queue_entry, HqeStatus.GATHERING, 1013 HostStatus.RUNNING) 1014 1015 self.mock_drone_manager.process_capacity = 5 1016 self._run_dispatcher() 1017 self._check_statuses(queue_entry, HqeStatus.ABORTED, 1018 HostStatus.CLEANING) 1019 1020 1021 def test_simple_metahost_assignment(self): 1022 job = self._create_job(metahosts=[1]) 1023 self._run_dispatcher() 1024 entry = job.hostqueueentry_set.all()[0] 1025 self.assertEquals(entry.host.hostname, 'host1') 1026 self._check_statuses(entry, HqeStatus.VERIFYING, HostStatus.VERIFYING) 1027 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 1028 self._run_dispatcher() 1029 self._check_statuses(entry, HqeStatus.RUNNING, HostStatus.RUNNING) 1030 # rest of job proceeds normally 1031 1032 1033 def test_metahost_fail_verify(self): 1034 self.hosts[1].labels.add(self.labels[0]) # put label1 also on host2 1035 job = self._create_job(metahosts=[1]) 1036 self._run_dispatcher() # assigned to host1 1037 self.mock_drone_manager.finish_process(_PidfileType.VERIFY, 1038 exit_status=256) 1039 self._run_dispatcher() # host1 failed, gets reassigned to host2 1040 entry = job.hostqueueentry_set.all()[0] 1041 self.assertEquals(entry.host.hostname, 'host2') 1042 self._check_statuses(entry, HqeStatus.VERIFYING, HostStatus.VERIFYING) 1043 self._check_host_status(self.hosts[0], HostStatus.REPAIRING) 1044 1045 self.mock_drone_manager.finish_process(_PidfileType.VERIFY) 1046 self._run_dispatcher() 1047 self._check_statuses(entry, HqeStatus.RUNNING, HostStatus.RUNNING) 1048 1049 1050 def test_hostless_job(self): 1051 job = self._create_job(hostless=True) 1052 entry = job.hostqueueentry_set.all()[0] 1053 1054 self._run_dispatcher() 1055 self._check_entry_status(entry, HqeStatus.RUNNING) 1056 1057 self.mock_drone_manager.finish_process(_PidfileType.JOB) 1058 self._run_dispatcher() 1059 self._check_entry_status(entry, HqeStatus.PARSING) 1060 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1061 self._run_dispatcher() 1062 self._check_entry_status(entry, HqeStatus.COMPLETED) 1063 1064 1065 def test_pre_job_keyvals(self): 1066 job = self._create_job(hosts=[1]) 1067 job.run_verify = False 1068 job.run_reset = False 1069 job.reboot_before = model_attributes.RebootBefore.NEVER 1070 job.save() 1071 models.JobKeyval.objects.create(job=job, key='mykey', value='myvalue') 1072 1073 self._run_dispatcher() 1074 self._finish_job(job.hostqueueentry_set.all()[0]) 1075 1076 attached_files = self.mock_drone_manager.attached_files( 1077 '1-autotest_system/host1') 1078 job_keyval_path = '1-autotest_system/host1/keyval' 1079 self.assert_(job_keyval_path in attached_files, attached_files) 1080 keyval_contents = attached_files[job_keyval_path] 1081 keyval_dict = dict(line.strip().split('=', 1) 1082 for line in keyval_contents.splitlines()) 1083 self.assert_('job_queued' in keyval_dict, keyval_dict) 1084 self.assertEquals(keyval_dict['mykey'], 'myvalue') 1085 1086 1087 # This tests the scheduler functions with archiving step disabled 1088 class SchedulerFunctionalTestNoArchiving(SchedulerFunctionalTest): 1089 def _set_global_config_values(self): 1090 super(SchedulerFunctionalTestNoArchiving, self 1091 )._set_global_config_values() 1092 self.mock_config.set_config_value('SCHEDULER', 'enable_archiving', 1093 False) 1094 1095 1096 def _finish_parsing(self): 1097 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1098 self._run_dispatcher() 1099 1100 1101 def _run_post_job_cleanup_failure_up_to_repair(self, queue_entry, 1102 include_verify=True): 1103 if include_verify: 1104 self._run_pre_job_verify(queue_entry) 1105 self._run_dispatcher() # job 1106 self.mock_drone_manager.finish_process(_PidfileType.JOB) 1107 self._run_dispatcher() # parsing + cleanup 1108 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1109 self.mock_drone_manager.finish_process(_PidfileType.CLEANUP, 1110 exit_status=256) 1111 self._run_dispatcher() # repair, HQE unaffected 1112 return queue_entry 1113 1114 1115 def test_hostless_job(self): 1116 job = self._create_job(hostless=True) 1117 entry = job.hostqueueentry_set.all()[0] 1118 1119 self._run_dispatcher() 1120 self._check_entry_status(entry, HqeStatus.RUNNING) 1121 1122 self.mock_drone_manager.finish_process(_PidfileType.JOB) 1123 self._run_dispatcher() 1124 self._check_entry_status(entry, HqeStatus.PARSING) 1125 self.mock_drone_manager.finish_process(_PidfileType.PARSE) 1126 self._run_dispatcher() 1127 self._check_entry_status(entry, HqeStatus.COMPLETED) 1128 1129 def test_synchronous_with_reset(self): 1130 # For crbug/621257. 1131 job = self._create_job(hosts=[1, 2]) 1132 job.synch_count = 2 1133 job.reboot_before = model_attributes.RebootBefore.ALWAYS 1134 job.save() 1135 1136 hqe1 = job.hostqueueentry_set.get(host__hostname='host1') 1137 hqe2 = job.hostqueueentry_set.get(host__hostname='host2') 1138 1139 self._run_dispatcher() 1140 1141 self._check_statuses(hqe1, HqeStatus.RESETTING, HostStatus.RESETTING) 1142 self._check_statuses(hqe2, HqeStatus.RESETTING, HostStatus.RESETTING) 1143 1144 self.mock_drone_manager.finish_active_process_on_host(1) 1145 self._run_dispatcher() 1146 1147 self._check_statuses(hqe1, HqeStatus.PENDING, HostStatus.PENDING) 1148 self._check_statuses(hqe2, HqeStatus.RESETTING, HostStatus.RESETTING) 1149 1150 self.mock_drone_manager.finish_active_process_on_host(2) 1151 self._run_dispatcher() 1152 1153 self._check_statuses(hqe1, HqeStatus.RUNNING, HostStatus.RUNNING) 1154 self._check_statuses(hqe2, HqeStatus.RUNNING, HostStatus.RUNNING) 1155 1156 1157 if __name__ == '__main__': 1158 unittest.main() 1159
