#!/usr/bin/python -u # Copyright 2007-2008 Martin J. Bligh , Google Inc. # Released under the GPL v2 """ Run a control file through the server side engine """ import ast import datetime import getpass import logging import os import re import signal import socket import sys import traceback import time import urllib2 import common from autotest_lib.client.common_lib import control_data from autotest_lib.client.common_lib import error from autotest_lib.client.common_lib import global_config from autotest_lib.client.common_lib import utils from autotest_lib.client.common_lib.cros.graphite import autotest_es from autotest_lib.client.common_lib.cros.graphite import autotest_stats try: from autotest_lib.puppylab import results_mocker except ImportError: results_mocker = None _CONFIG = global_config.global_config require_atfork = _CONFIG.get_config_value( 'AUTOSERV', 'require_atfork_module', type=bool, default=True) # Number of seconds to wait before returning if testing mode is enabled TESTING_MODE_SLEEP_SECS = 1 try: import atfork atfork.monkeypatch_os_fork_functions() import atfork.stdlib_fixer # Fix the Python standard library for threading+fork safety with its # internal locks. http://code.google.com/p/python-atfork/ import warnings warnings.filterwarnings('ignore', 'logging module already imported') atfork.stdlib_fixer.fix_logging_module() except ImportError, e: from autotest_lib.client.common_lib import global_config if _CONFIG.get_config_value( 'AUTOSERV', 'require_atfork_module', type=bool, default=False): print >>sys.stderr, 'Please run utils/build_externals.py' print e sys.exit(1) from autotest_lib.server import frontend from autotest_lib.server import server_logging_config from autotest_lib.server import server_job, utils, autoserv_parser, autotest from autotest_lib.server import utils as server_utils from autotest_lib.site_utils import job_directories from autotest_lib.site_utils import job_overhead from autotest_lib.site_utils import lxc from autotest_lib.site_utils import lxc_utils from autotest_lib.client.common_lib import pidfile, logging_manager from autotest_lib.client.common_lib.cros.graphite import autotest_stats # Control segment to stage server-side package. STAGE_SERVER_SIDE_PACKAGE_CONTROL_FILE = server_job._control_segment_path( 'stage_server_side_package') # Command line to start servod in a moblab. START_SERVOD_CMD = 'sudo start servod BOARD=%s PORT=%s' STOP_SERVOD_CMD = 'sudo stop servod' def log_alarm(signum, frame): logging.error("Received SIGALARM. Ignoring and continuing on.") sys.exit(1) def _get_machines(parser): """Get a list of machine names from command line arg -m or a file. @param parser: Parser for the command line arguments. @return: A list of machine names from command line arg -m or the machines file specified in the command line arg -M. """ if parser.options.machines: machines = parser.options.machines.replace(',', ' ').strip().split() else: machines = [] machines_file = parser.options.machines_file if machines_file: machines = [] for m in open(machines_file, 'r').readlines(): # remove comments, spaces m = re.sub('#.*', '', m).strip() if m: machines.append(m) logging.debug('Read list of machines from file: %s', machines_file) logging.debug('Machines: %s', ','.join(machines)) if machines: for machine in machines: if not machine or re.search('\s', machine): parser.parser.error("Invalid machine: %s" % str(machine)) machines = list(set(machines)) machines.sort() return machines def _stage_ssp(parser): """Stage server-side package. This function calls a control segment to stage server-side package based on the job and autoserv command line option. The detail implementation could be different for each host type. Currently, only CrosHost has stage_server_side_package function defined. The script returns None if no server-side package is available. However, it may raise exception if it failed for reasons other than artifact (the server-side package) not found. @param parser: Command line arguments parser passed in the autoserv process. @return: url of the staged server-side package. Return None if server- side package is not found for the build. """ # If test_source_build is not specified, default to use server-side test # code from build specified in --image. namespace = {'machines': _get_machines(parser), 'image': (parser.options.test_source_build or parser.options.image),} script_locals = {} execfile(STAGE_SERVER_SIDE_PACKAGE_CONTROL_FILE, namespace, script_locals) return script_locals['ssp_url'] def _run_with_ssp(job, container_name, job_id, results, parser, ssp_url): """Run the server job with server-side packaging. @param job: The server job object. @param container_name: Name of the container to run the test. @param job_id: ID of the test job. @param results: Folder to store results. This could be different from parser.options.results: parser.options.results can be set to None for results to be stored in a temp folder. results can be None for autoserv run requires no logging. @param parser: Command line parser that contains the options. @param ssp_url: url of the staged server-side package. """ bucket = lxc.ContainerBucket() control = (parser.args[0] if len(parser.args) > 0 and parser.args[0] != '' else None) try: test_container = bucket.setup_test(container_name, job_id, ssp_url, results, control=control) except Exception as e: job.record('FAIL', None, None, 'Failed to setup container for test: %s. Check logs in ' 'ssp_logs folder for more details.' % e) raise args = sys.argv[:] args.remove('--require-ssp') # --parent_job_id is only useful in autoserv running in host, not in # container. Include this argument will cause test to fail for builds before # CL 286265 was merged. if '--parent_job_id' in args: index = args.index('--parent_job_id') args.remove('--parent_job_id') # Remove the actual parent job id in command line arg. del args[index] # A dictionary of paths to replace in the command line. Key is the path to # be replaced with the one in value. paths_to_replace = {} # Replace the control file path with the one in container. if control: container_control_filename = os.path.join( lxc.CONTROL_TEMP_PATH, os.path.basename(control)) paths_to_replace[control] = container_control_filename # Update result directory with the one in container. if parser.options.results: container_result_dir = os.path.join(lxc.RESULT_DIR_FMT % job_id) paths_to_replace[parser.options.results] = container_result_dir # Update parse_job directory with the one in container. The assumption is # that the result folder to be parsed is always the same as the results_dir. if parser.options.parse_job: container_parse_dir = os.path.join(lxc.RESULT_DIR_FMT % job_id) paths_to_replace[parser.options.parse_job] = container_result_dir args = [paths_to_replace.get(arg, arg) for arg in args] # Apply --use-existing-results, results directory is aready created and # mounted in container. Apply this arg to avoid exception being raised. if not '--use-existing-results' in args: args.append('--use-existing-results') # Make sure autoserv running in container using a different pid file. if not '--pidfile-label' in args: args.extend(['--pidfile-label', 'container_autoserv']) cmd_line = ' '.join(["'%s'" % arg if ' ' in arg else arg for arg in args]) logging.info('Run command in container: %s', cmd_line) success = False try: test_container.attach_run(cmd_line) success = True except Exception as e: # If the test run inside container fails without generating any log, # write a message to status.log to help troubleshooting. debug_files = os.listdir(os.path.join(results, 'debug')) if not debug_files: job.record('FAIL', None, None, 'Failed to run test inside the container: %s. Check ' 'logs in ssp_logs folder for more details.' % e) raise finally: counter_key = '%s.%s' % (lxc.STATS_KEY, 'success' if success else 'fail') autotest_stats.Counter(counter_key).increment() # metadata is uploaded separately so it can use http to upload. metadata = {'drone': socket.gethostname(), 'job_id': job_id, 'success': success} autotest_es.post(use_http=True, type_str=lxc.CONTAINER_RUN_TEST_METADB_TYPE, metadata=metadata) test_container.destroy() def correct_results_folder_permission(results): """Make sure the results folder has the right permission settings. For tests running with server-side packaging, the results folder has the owner of root. This must be changed to the user running the autoserv process, so parsing job can access the results folder. TODO(dshi): crbug.com/459344 Remove this function when test container can be unprivileged container. @param results: Path to the results folder. """ if not results: return try: utils.run('sudo -n chown -R %s "%s"' % (os.getuid(), results)) utils.run('sudo -n chgrp -R %s "%s"' % (os.getgid(), results)) except error.CmdError as e: metadata = {'error': str(e), 'result_folder': results, 'drone': socket.gethostname()} autotest_es.post(use_http=True, type_str='correct_results_folder_failure', metadata=metadata) raise def _start_servod(machine): """Try to start servod in moblab if it's not already running or running with different board or port. @param machine: Name of the dut used for test. """ if not utils.is_moblab(): return logging.debug('Trying to start servod.') try: afe = frontend.AFE() board = server_utils.get_board_from_afe(machine, afe) hosts = afe.get_hosts(hostname=machine) servo_host = hosts[0].attributes.get('servo_host', None) servo_port = hosts[0].attributes.get('servo_port', 9999) if not servo_host in ['localhost', '127.0.0.1']: logging.warn('Starting servod is aborted. The dut\'s servo_host ' 'attribute is not set to localhost.') return except (urllib2.HTTPError, urllib2.URLError): # Ignore error if RPC failed to get board logging.error('Failed to get board name from AFE. Start servod is ' 'aborted') return try: pid = utils.run('pgrep servod').stdout cmd_line = utils.run('ps -fp %s' % pid).stdout if ('--board %s' % board in cmd_line and '--port %s' % servo_port in cmd_line): logging.debug('Servod is already running with given board and port.' ' There is no need to restart servod.') return logging.debug('Servod is running with different board or port. ' 'Stopping existing servod.') utils.run('sudo stop servod') except error.CmdError: # servod is not running. pass try: utils.run(START_SERVOD_CMD % (board, servo_port)) logging.debug('Servod is started') except error.CmdError as e: logging.error('Servod failed to be started, error: %s', e) def run_autoserv(pid_file_manager, results, parser, ssp_url, use_ssp): """Run server job with given options. @param pid_file_manager: PidFileManager used to monitor the autoserv process @param results: Folder to store results. @param parser: Parser for the command line arguments. @param ssp_url: Url to server-side package. @param use_ssp: Set to True to run with server-side packaging. """ if parser.options.warn_no_ssp: # Post a warning in the log. logging.warn('Autoserv is required to run with server-side packaging. ' 'However, no drone is found to support server-side ' 'packaging. The test will be executed in a drone without ' 'server-side packaging supported.') # send stdin to /dev/null dev_null = os.open(os.devnull, os.O_RDONLY) os.dup2(dev_null, sys.stdin.fileno()) os.close(dev_null) # Create separate process group os.setpgrp() # Container name is predefined so the container can be destroyed in # handle_sigterm. job_or_task_id = job_directories.get_job_id_or_task_id( parser.options.results) container_name = (lxc.TEST_CONTAINER_NAME_FMT % (job_or_task_id, time.time(), os.getpid())) # Implement SIGTERM handler def handle_sigterm(signum, frame): logging.debug('Received SIGTERM') if pid_file_manager: pid_file_manager.close_file(1, signal.SIGTERM) logging.debug('Finished writing to pid_file. Killing process.') # Update results folder's file permission. This needs to be done ASAP # before the parsing process tries to access the log. if use_ssp and results: correct_results_folder_permission(results) # TODO (sbasi) - remove the time.sleep when crbug.com/302815 is solved. # This sleep allows the pending output to be logged before the kill # signal is sent. time.sleep(.1) if use_ssp: logging.debug('Destroy container %s before aborting the autoserv ' 'process.', container_name) metadata = {'drone': socket.gethostname(), 'job_id': job_or_task_id, 'container_name': container_name, 'action': 'abort', 'success': True} try: bucket = lxc.ContainerBucket() container = bucket.get(container_name) if container: container.destroy() else: metadata['success'] = False metadata['error'] = 'container not found' logging.debug('Container %s is not found.', container_name) except: metadata['success'] = False metadata['error'] = 'Exception: %s' % sys.exc_info() # Handle any exception so the autoserv process can be aborted. logging.error('Failed to destroy container %s. Error: %s', container_name, sys.exc_info()) autotest_es.post(use_http=True, type_str=lxc.CONTAINER_RUN_TEST_METADB_TYPE, metadata=metadata) # Try to correct the result file permission again after the # container is destroyed, as the container might have created some # new files in the result folder. if results: correct_results_folder_permission(results) os.killpg(os.getpgrp(), signal.SIGKILL) # Set signal handler signal.signal(signal.SIGTERM, handle_sigterm) # faulthandler is only needed to debug in the Lab and is not avaliable to # be imported in the chroot as part of VMTest, so Try-Except it. try: import faulthandler faulthandler.register(signal.SIGTERM, all_threads=True, chain=True) logging.debug('faulthandler registered on SIGTERM.') except ImportError: sys.exc_clear() # Ignore SIGTTOU's generated by output from forked children. signal.signal(signal.SIGTTOU, signal.SIG_IGN) # If we received a SIGALARM, let's be loud about it. signal.signal(signal.SIGALRM, log_alarm) # Server side tests that call shell scripts often depend on $USER being set # but depending on how you launch your autotest scheduler it may not be set. os.environ['USER'] = getpass.getuser() label = parser.options.label group_name = parser.options.group_name user = parser.options.user client = parser.options.client server = parser.options.server install_before = parser.options.install_before install_after = parser.options.install_after verify = parser.options.verify repair = parser.options.repair cleanup = parser.options.cleanup provision = parser.options.provision reset = parser.options.reset job_labels = parser.options.job_labels no_tee = parser.options.no_tee parse_job = parser.options.parse_job execution_tag = parser.options.execution_tag if not execution_tag: execution_tag = parse_job ssh_user = parser.options.ssh_user ssh_port = parser.options.ssh_port ssh_pass = parser.options.ssh_pass collect_crashinfo = parser.options.collect_crashinfo control_filename = parser.options.control_filename test_retry = parser.options.test_retry verify_job_repo_url = parser.options.verify_job_repo_url skip_crash_collection = parser.options.skip_crash_collection ssh_verbosity = int(parser.options.ssh_verbosity) ssh_options = parser.options.ssh_options no_use_packaging = parser.options.no_use_packaging host_attributes = parser.options.host_attributes in_lab = bool(parser.options.lab) # can't be both a client and a server side test if client and server: parser.parser.error("Can not specify a test as both server and client!") if provision and client: parser.parser.error("Cannot specify provisioning and client!") is_special_task = (verify or repair or cleanup or collect_crashinfo or provision or reset) if len(parser.args) < 1 and not is_special_task: parser.parser.error("Missing argument: control file") if ssh_verbosity > 0: # ssh_verbosity is an integer between 0 and 3, inclusive ssh_verbosity_flag = '-' + 'v' * ssh_verbosity else: ssh_verbosity_flag = '' # We have a control file unless it's just a verify/repair/cleanup job if len(parser.args) > 0: control = parser.args[0] else: control = None machines = _get_machines(parser) if group_name and len(machines) < 2: parser.parser.error('-G %r may only be supplied with more than one ' 'machine.' % group_name) kwargs = {'group_name': group_name, 'tag': execution_tag, 'disable_sysinfo': parser.options.disable_sysinfo} if parser.options.parent_job_id: kwargs['parent_job_id'] = int(parser.options.parent_job_id) if control_filename: kwargs['control_filename'] = control_filename if host_attributes: kwargs['host_attributes'] = host_attributes kwargs['in_lab'] = in_lab job = server_job.server_job(control, parser.args[1:], results, label, user, machines, client, parse_job, ssh_user, ssh_port, ssh_pass, ssh_verbosity_flag, ssh_options, test_retry, **kwargs) job.logging.start_logging() job.init_parser() # perform checks job.precheck() # run the job exit_code = 0 auto_start_servod = _CONFIG.get_config_value( 'AUTOSERV', 'auto_start_servod', type=bool, default=False) try: try: if repair: if auto_start_servod and len(machines) == 1: _start_servod(machines[0]) job.repair(job_labels) elif verify: job.verify(job_labels) elif provision: job.provision(job_labels) elif reset: job.reset(job_labels) elif cleanup: job.cleanup(job_labels) else: if auto_start_servod and len(machines) == 1: _start_servod(machines[0]) if use_ssp: try: _run_with_ssp(job, container_name, job_or_task_id, results, parser, ssp_url) finally: # Update the ownership of files in result folder. correct_results_folder_permission(results) else: job.run(install_before, install_after, verify_job_repo_url=verify_job_repo_url, only_collect_crashinfo=collect_crashinfo, skip_crash_collection=skip_crash_collection, job_labels=job_labels, use_packaging=(not no_use_packaging)) finally: while job.hosts: host = job.hosts.pop() host.close() except: exit_code = 1 traceback.print_exc() if pid_file_manager: pid_file_manager.num_tests_failed = job.num_tests_failed pid_file_manager.close_file(exit_code) job.cleanup_parser() sys.exit(exit_code) def record_autoserv(options, duration_secs): """Record autoserv end-to-end time in metadata db. @param options: parser options. @param duration_secs: How long autoserv has taken, in secs. """ # Get machine hostname machines = options.machines.replace( ',', ' ').strip().split() if options.machines else [] num_machines = len(machines) if num_machines > 1: # Skip the case where atomic group is used. return elif num_machines == 0: machines.append('hostless') # Determine the status that will be reported. s = job_overhead.STATUS task_mapping = { 'reset': s.RESETTING, 'verify': s.VERIFYING, 'provision': s.PROVISIONING, 'repair': s.REPAIRING, 'cleanup': s.CLEANING, 'collect_crashinfo': s.GATHERING} match = filter(lambda task: getattr(options, task, False) == True, task_mapping) status = task_mapping[match[0]] if match else s.RUNNING is_special_task = status not in [s.RUNNING, s.GATHERING] job_or_task_id = job_directories.get_job_id_or_task_id(options.results) job_overhead.record_state_duration( job_or_task_id, machines[0], status, duration_secs, is_special_task=is_special_task) def main(): start_time = datetime.datetime.now() # White list of tests with run time measurement enabled. measure_run_time_tests_names = _CONFIG.get_config_value( 'AUTOSERV', 'measure_run_time_tests', type=str) if measure_run_time_tests_names: measure_run_time_tests = [t.strip() for t in measure_run_time_tests_names.split(',')] else: measure_run_time_tests = [] # grab the parser parser = autoserv_parser.autoserv_parser parser.parse_args() if len(sys.argv) == 1: parser.parser.print_help() sys.exit(1) # If the job requires to run with server-side package, try to stage server- # side package first. If that fails with error that autotest server package # does not exist, fall back to run the job without using server-side # packaging. If option warn_no_ssp is specified, that means autoserv is # running in a drone does not support SSP, thus no need to stage server-side # package. ssp_url = None ssp_url_warning = False if (not parser.options.warn_no_ssp and parser.options.require_ssp): ssp_url = _stage_ssp(parser) # The build does not have autotest server package. Fall back to not # to use server-side package. Logging is postponed until logging being # set up. ssp_url_warning = not ssp_url if parser.options.no_logging: results = None else: results = parser.options.results if not results: results = 'results.' + time.strftime('%Y-%m-%d-%H.%M.%S') results = os.path.abspath(results) resultdir_exists = False for filename in ('control.srv', 'status.log', '.autoserv_execute'): if os.path.exists(os.path.join(results, filename)): resultdir_exists = True if not parser.options.use_existing_results and resultdir_exists: error = "Error: results directory already exists: %s\n" % results sys.stderr.write(error) sys.exit(1) # Now that we certified that there's no leftover results dir from # previous jobs, lets create the result dir since the logging system # needs to create the log file in there. if not os.path.isdir(results): os.makedirs(results) # Server-side packaging will only be used if it's required and the package # is available. If warn_no_ssp is specified, it means that autoserv is # running in a drone does not have SSP supported and a warning will be logs. # Therefore, it should not run with SSP. use_ssp = (not parser.options.warn_no_ssp and parser.options.require_ssp and ssp_url) if use_ssp: log_dir = os.path.join(results, 'ssp_logs') if results else None if log_dir and not os.path.exists(log_dir): os.makedirs(log_dir) else: log_dir = results logging_manager.configure_logging( server_logging_config.ServerLoggingConfig(), results_dir=log_dir, use_console=not parser.options.no_tee, verbose=parser.options.verbose, no_console_prefix=parser.options.no_console_prefix) if ssp_url_warning: logging.warn( 'Autoserv is required to run with server-side packaging. ' 'However, no server-side package can be found based on ' '`--image`, host attribute job_repo_url or host label of ' 'cros-version. The test will be executed without ' 'server-side packaging supported.') if results: logging.info("Results placed in %s" % results) # wait until now to perform this check, so it get properly logged if (parser.options.use_existing_results and not resultdir_exists and not utils.is_in_container()): logging.error("No existing results directory found: %s", results) sys.exit(1) logging.debug('autoserv is running in drone %s.', socket.gethostname()) logging.debug('autoserv command was: %s', ' '.join(sys.argv)) if parser.options.write_pidfile and results: pid_file_manager = pidfile.PidFileManager(parser.options.pidfile_label, results) pid_file_manager.open_file() else: pid_file_manager = None autotest.BaseAutotest.set_install_in_tmpdir( parser.options.install_in_tmpdir) timer = None try: # Take the first argument as control file name, get the test name from # the control file. If the test name exists in the list of tests with # run time measurement enabled, start a timer to begin measurement. if (len(parser.args) > 0 and parser.args[0] != '' and parser.options.machines): try: test_name = control_data.parse_control(parser.args[0], raise_warnings=True).name except control_data.ControlVariableException: logging.debug('Failed to retrieve test name from control file.') test_name = None if test_name in measure_run_time_tests: machines = parser.options.machines.replace(',', ' ' ).strip().split() try: afe = frontend.AFE() board = server_utils.get_board_from_afe(machines[0], afe) timer = autotest_stats.Timer('autoserv_run_time.%s.%s' % (board, test_name)) timer.start() except (urllib2.HTTPError, urllib2.URLError): # Ignore error if RPC failed to get board pass except control_data.ControlVariableException as e: logging.error(str(e)) exit_code = 0 # TODO(beeps): Extend this to cover different failure modes. # Testing exceptions are matched against labels sent to autoserv. Eg, # to allow only the hostless job to run, specify # testing_exceptions: test_suite in the shadow_config. To allow both # the hostless job and dummy_Pass to run, specify # testing_exceptions: test_suite,dummy_Pass. You can figure out # what label autoserv is invoked with by looking through the logs of a test # for the autoserv command's -l option. testing_exceptions = _CONFIG.get_config_value( 'AUTOSERV', 'testing_exceptions', type=list, default=[]) test_mode = _CONFIG.get_config_value( 'AUTOSERV', 'testing_mode', type=bool, default=False) test_mode = (results_mocker and test_mode and not any([ex in parser.options.label for ex in testing_exceptions])) is_task = (parser.options.verify or parser.options.repair or parser.options.provision or parser.options.reset or parser.options.cleanup or parser.options.collect_crashinfo) try: try: if test_mode: # The parser doesn't run on tasks anyway, so we can just return # happy signals without faking results. if not is_task: machine = parser.options.results.split('/')[-1] # TODO(beeps): The proper way to do this would be to # refactor job creation so we can invoke job.record # directly. To do that one needs to pipe the test_name # through run_autoserv and bail just before invoking # the server job. See the comment in # puppylab/results_mocker for more context. results_mocker.ResultsMocker( test_name if test_name else 'unknown-test', parser.options.results, machine ).mock_results() return else: run_autoserv(pid_file_manager, results, parser, ssp_url, use_ssp) except SystemExit as e: exit_code = e.code if exit_code: logging.exception(e) except Exception as e: # If we don't know what happened, we'll classify it as # an 'abort' and return 1. logging.exception(e) exit_code = 1 finally: if pid_file_manager: pid_file_manager.close_file(exit_code) if timer: timer.stop() # Record the autoserv duration time. Must be called # just before the system exits to ensure accuracy. duration_secs = (datetime.datetime.now() - start_time).total_seconds() record_autoserv(parser.options, duration_secs) sys.exit(exit_code) if __name__ == '__main__': main()