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