autoserv revision 49e21e6f66df4cca76824a9e10dec55c4e94cf0d
1#!/usr/bin/python -u 2# Copyright 2007-2008 Martin J. Bligh <mbligh@google.com>, Google Inc. 3# Released under the GPL v2 4 5""" 6Run a control file through the server side engine 7""" 8 9import sys, os, re, traceback, signal, time, logging, getpass 10 11import common 12 13from autotest_lib.client.common_lib import control_data 14from autotest_lib.client.common_lib import global_config 15require_atfork = global_config.global_config.get_config_value( 16 'AUTOSERV', 'require_atfork_module', type=bool, default=True) 17 18 19try: 20 import atfork 21 atfork.monkeypatch_os_fork_functions() 22 import atfork.stdlib_fixer 23 # Fix the Python standard library for threading+fork safety with its 24 # internal locks. http://code.google.com/p/python-atfork/ 25 import warnings 26 warnings.filterwarnings('ignore', 'logging module already imported') 27 atfork.stdlib_fixer.fix_logging_module() 28except ImportError, e: 29 from autotest_lib.client.common_lib import global_config 30 if global_config.global_config.get_config_value( 31 'AUTOSERV', 'require_atfork_module', type=bool, default=False): 32 print >>sys.stderr, 'Please run utils/build_externals.py' 33 print e 34 sys.exit(1) 35 36from autotest_lib.server import frontend 37from autotest_lib.server import server_logging_config 38from autotest_lib.server import server_job, utils, autoserv_parser, autotest 39from autotest_lib.server import utils as server_utils 40 41from autotest_lib.client.common_lib import pidfile, logging_manager 42from autotest_lib.site_utils.graphite import stats 43 44def log_alarm(signum, frame): 45 logging.error("Received SIGALARM. Ignoring and continuing on.") 46 sys.exit(1) 47 48def run_autoserv(pid_file_manager, results, parser): 49 # send stdin to /dev/null 50 dev_null = os.open(os.devnull, os.O_RDONLY) 51 os.dup2(dev_null, sys.stdin.fileno()) 52 os.close(dev_null) 53 54 # Create separate process group 55 os.setpgrp() 56 57 # Implement SIGTERM handler 58 def handle_sigterm(signum, frame): 59 logging.debug('Received SIGTERM') 60 if pid_file_manager: 61 pid_file_manager.close_file(1, signal.SIGTERM) 62 logging.debug('Finished writing to pid_file. Killing process.') 63 os.killpg(os.getpgrp(), signal.SIGKILL) 64 65 # Set signal handler 66 signal.signal(signal.SIGTERM, handle_sigterm) 67 68 # Ignore SIGTTOU's generated by output from forked children. 69 signal.signal(signal.SIGTTOU, signal.SIG_IGN) 70 71 # If we received a SIGALARM, let's be loud about it. 72 signal.signal(signal.SIGALRM, log_alarm) 73 74 # Server side tests that call shell scripts often depend on $USER being set 75 # but depending on how you launch your autotest scheduler it may not be set. 76 os.environ['USER'] = getpass.getuser() 77 78 if parser.options.machines: 79 machines = parser.options.machines.replace(',', ' ').strip().split() 80 else: 81 machines = [] 82 machines_file = parser.options.machines_file 83 label = parser.options.label 84 group_name = parser.options.group_name 85 user = parser.options.user 86 client = parser.options.client 87 server = parser.options.server 88 install_before = parser.options.install_before 89 install_after = parser.options.install_after 90 verify = parser.options.verify 91 repair = parser.options.repair 92 cleanup = parser.options.cleanup 93 provision = parser.options.provision 94 reset = parser.options.reset 95 no_tee = parser.options.no_tee 96 parse_job = parser.options.parse_job 97 execution_tag = parser.options.execution_tag 98 if not execution_tag: 99 execution_tag = parse_job 100 host_protection = parser.options.host_protection 101 ssh_user = parser.options.ssh_user 102 ssh_port = parser.options.ssh_port 103 ssh_pass = parser.options.ssh_pass 104 collect_crashinfo = parser.options.collect_crashinfo 105 control_filename = parser.options.control_filename 106 test_retry = parser.options.test_retry 107 verify_job_repo_url = parser.options.verify_job_repo_url 108 skip_crash_collection = parser.options.skip_crash_collection 109 ssh_verbosity = int(parser.options.ssh_verbosity) 110 ssh_options = parser.options.ssh_options 111 112 # can't be both a client and a server side test 113 if client and server: 114 parser.parser.error("Can not specify a test as both server and client!") 115 116 if provision and client: 117 parser.parser.error("Cannot specify provisioning and client!") 118 119 is_special_task = (verify or repair or cleanup or collect_crashinfo or 120 provision or reset) 121 if len(parser.args) < 1 and not is_special_task: 122 parser.parser.error("Missing argument: control file") 123 124 if ssh_verbosity > 0: 125 # ssh_verbosity is an integer between 0 and 3, inclusive 126 ssh_verbosity_flag = '-' + 'v' * ssh_verbosity 127 else: 128 ssh_verbosity_flag = '' 129 130 # We have a control file unless it's just a verify/repair/cleanup job 131 if len(parser.args) > 0: 132 control = parser.args[0] 133 else: 134 control = None 135 136 if machines_file: 137 machines = [] 138 for m in open(machines_file, 'r').readlines(): 139 # remove comments, spaces 140 m = re.sub('#.*', '', m).strip() 141 if m: 142 machines.append(m) 143 print "Read list of machines from file: %s" % machines_file 144 print ','.join(machines) 145 146 if machines: 147 for machine in machines: 148 if not machine or re.search('\s', machine): 149 parser.parser.error("Invalid machine: %s" % str(machine)) 150 machines = list(set(machines)) 151 machines.sort() 152 153 if group_name and len(machines) < 2: 154 parser.parser.error("-G %r may only be supplied with more than one machine." 155 % group_name) 156 157 kwargs = {'group_name': group_name, 'tag': execution_tag, 158 'disable_sysinfo': parser.options.disable_sysinfo} 159 if control_filename: 160 kwargs['control_filename'] = control_filename 161 job = server_job.server_job(control, parser.args[1:], results, label, 162 user, machines, client, parse_job, 163 ssh_user, ssh_port, ssh_pass, 164 ssh_verbosity_flag, ssh_options, 165 test_retry, **kwargs) 166 job.logging.start_logging() 167 job.init_parser() 168 169 # perform checks 170 job.precheck() 171 172 # run the job 173 exit_code = 0 174 try: 175 try: 176 if repair: 177 job.repair(host_protection) 178 elif verify: 179 job.verify() 180 elif provision: 181 job.provision(provision) 182 elif reset: 183 job.reset() 184 else: 185 job.run(cleanup, install_before, install_after, 186 verify_job_repo_url=verify_job_repo_url, 187 only_collect_crashinfo=collect_crashinfo, 188 skip_crash_collection=skip_crash_collection) 189 finally: 190 while job.hosts: 191 host = job.hosts.pop() 192 host.close() 193 except: 194 exit_code = 1 195 traceback.print_exc() 196 197 if pid_file_manager: 198 pid_file_manager.num_tests_failed = job.num_tests_failed 199 pid_file_manager.close_file(exit_code) 200 job.cleanup_parser() 201 202 sys.exit(exit_code) 203 204 205def main(): 206 # White list of tests with run time measurement enabled. 207 measure_run_time_tests_names = global_config.global_config.get_config_value( 208 'AUTOSERV', 'measure_run_time_tests', type=str) 209 if measure_run_time_tests_names: 210 measure_run_time_tests = [t.strip() for t in 211 measure_run_time_tests_names.split(',')] 212 else: 213 measure_run_time_tests = [] 214 # grab the parser 215 parser = autoserv_parser.autoserv_parser 216 parser.parse_args() 217 218 if len(sys.argv) == 1: 219 parser.parser.print_help() 220 sys.exit(1) 221 222 if parser.options.no_logging: 223 results = None 224 else: 225 results = parser.options.results 226 if not results: 227 results = 'results.' + time.strftime('%Y-%m-%d-%H.%M.%S') 228 results = os.path.abspath(results) 229 resultdir_exists = False 230 for filename in ('control.srv', 'status.log', '.autoserv_execute'): 231 if os.path.exists(os.path.join(results, filename)): 232 resultdir_exists = True 233 if not parser.options.use_existing_results and resultdir_exists: 234 error = "Error: results directory already exists: %s\n" % results 235 sys.stderr.write(error) 236 sys.exit(1) 237 238 # Now that we certified that there's no leftover results dir from 239 # previous jobs, lets create the result dir since the logging system 240 # needs to create the log file in there. 241 if not os.path.isdir(results): 242 os.makedirs(results) 243 244 logging_manager.configure_logging( 245 server_logging_config.ServerLoggingConfig(), results_dir=results, 246 use_console=not parser.options.no_tee, 247 verbose=parser.options.verbose, 248 no_console_prefix=parser.options.no_console_prefix) 249 if results: 250 logging.info("Results placed in %s" % results) 251 252 # wait until now to perform this check, so it get properly logged 253 if parser.options.use_existing_results and not resultdir_exists: 254 logging.error("No existing results directory found: %s", results) 255 sys.exit(1) 256 257 logging.debug('autoserv command was: %s', ' '.join(sys.argv)) 258 259 if parser.options.write_pidfile: 260 pid_file_manager = pidfile.PidFileManager(parser.options.pidfile_label, 261 results) 262 pid_file_manager.open_file() 263 else: 264 pid_file_manager = None 265 266 autotest.BaseAutotest.set_install_in_tmpdir( 267 parser.options.install_in_tmpdir) 268 269 timer = None 270 try: 271 # Take the first argument as control file name, get the test name from 272 # the control file. If the test name exists in the list of tests with 273 # run time measurement enabled, start a timer to begin measurement. 274 if (len(parser.args) > 0 and parser.args[0] != '' and 275 parser.options.machines): 276 try: 277 test_name = control_data.parse_control(parser.args[0], 278 raise_warnings=True).name 279 except control_data.ControlVariableException: 280 logging.debug('Failed to retrieve test name from control file.') 281 test_name = None 282 if test_name in measure_run_time_tests: 283 machines = parser.options.machines.replace(',', ' ' 284 ).strip().split() 285 afe = frontend.AFE() 286 board = server_utils.get_board_from_afe(machines[0], afe) 287 timer = stats.Timer('autoserv_run_time.%s.%s' % 288 (board, test_name)) 289 timer.start() 290 except control_data.ControlVariableException as e: 291 logging.error(str(e)) 292 exit_code = 0 293 try: 294 try: 295 run_autoserv(pid_file_manager, results, parser) 296 except SystemExit as e: 297 exit_code = e.code 298 if exit_code: 299 logging.exception(e) 300 except Exception as e: 301 # If we don't know what happened, we'll classify it as 302 # an 'abort' and return 1. 303 logging.exception(e) 304 exit_code = 1 305 finally: 306 if pid_file_manager: 307 pid_file_manager.close_file(exit_code) 308 if timer: 309 timer.stop() 310 sys.exit(exit_code) 311 312 313if __name__ == '__main__': 314 main() 315