autoserv revision 6cc20de4cde89b9613d051dc947f68e425f342f3
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 if pid_file_manager: 60 pid_file_manager.close_file(1, signal.SIGTERM) 61 os.killpg(os.getpgrp(), signal.SIGKILL) 62 63 # Set signal handler 64 signal.signal(signal.SIGTERM, handle_sigterm) 65 66 # Ignore SIGTTOU's generated by output from forked children. 67 signal.signal(signal.SIGTTOU, signal.SIG_IGN) 68 69 # If we received a SIGALARM, let's be loud about it. 70 signal.signal(signal.SIGALRM, log_alarm) 71 72 # Server side tests that call shell scripts often depend on $USER being set 73 # but depending on how you launch your autotest scheduler it may not be set. 74 os.environ['USER'] = getpass.getuser() 75 76 if parser.options.machines: 77 machines = parser.options.machines.replace(',', ' ').strip().split() 78 else: 79 machines = [] 80 machines_file = parser.options.machines_file 81 label = parser.options.label 82 group_name = parser.options.group_name 83 user = parser.options.user 84 client = parser.options.client 85 server = parser.options.server 86 install_before = parser.options.install_before 87 install_after = parser.options.install_after 88 verify = parser.options.verify 89 repair = parser.options.repair 90 cleanup = parser.options.cleanup 91 provision = parser.options.provision 92 reset = parser.options.reset 93 no_tee = parser.options.no_tee 94 parse_job = parser.options.parse_job 95 execution_tag = parser.options.execution_tag 96 if not execution_tag: 97 execution_tag = parse_job 98 host_protection = parser.options.host_protection 99 ssh_user = parser.options.ssh_user 100 ssh_port = parser.options.ssh_port 101 ssh_pass = parser.options.ssh_pass 102 collect_crashinfo = parser.options.collect_crashinfo 103 control_filename = parser.options.control_filename 104 test_retry = parser.options.test_retry 105 verify_job_repo_url = parser.options.verify_job_repo_url 106 skip_crash_collection = parser.options.skip_crash_collection 107 ssh_verbosity = int(parser.options.ssh_verbosity) 108 ssh_options = parser.options.ssh_options 109 110 # can't be both a client and a server side test 111 if client and server: 112 parser.parser.error("Can not specify a test as both server and client!") 113 114 if provision and client: 115 parser.parser.error("Cannot specify provisioning and client!") 116 117 is_special_task = (verify or repair or cleanup or collect_crashinfo or 118 provision or reset) 119 if len(parser.args) < 1 and not is_special_task: 120 parser.parser.error("Missing argument: control file") 121 122 if ssh_verbosity > 0: 123 # ssh_verbosity is an integer between 0 and 3, inclusive 124 ssh_verbosity_flag = '-' + 'v' * ssh_verbosity 125 else: 126 ssh_verbosity_flag = '' 127 128 # We have a control file unless it's just a verify/repair/cleanup job 129 if len(parser.args) > 0: 130 control = parser.args[0] 131 else: 132 control = None 133 134 if machines_file: 135 machines = [] 136 for m in open(machines_file, 'r').readlines(): 137 # remove comments, spaces 138 m = re.sub('#.*', '', m).strip() 139 if m: 140 machines.append(m) 141 print "Read list of machines from file: %s" % machines_file 142 print ','.join(machines) 143 144 if machines: 145 for machine in machines: 146 if not machine or re.search('\s', machine): 147 parser.parser.error("Invalid machine: %s" % str(machine)) 148 machines = list(set(machines)) 149 machines.sort() 150 151 if group_name and len(machines) < 2: 152 parser.parser.error("-G %r may only be supplied with more than one machine." 153 % group_name) 154 155 kwargs = {'group_name': group_name, 'tag': execution_tag, 156 'disable_sysinfo': parser.options.disable_sysinfo} 157 if control_filename: 158 kwargs['control_filename'] = control_filename 159 job = server_job.server_job(control, parser.args[1:], results, label, 160 user, machines, client, parse_job, 161 ssh_user, ssh_port, ssh_pass, 162 ssh_verbosity_flag, ssh_options, 163 test_retry, **kwargs) 164 job.logging.start_logging() 165 job.init_parser() 166 167 # perform checks 168 job.precheck() 169 170 # run the job 171 exit_code = 0 172 try: 173 try: 174 if repair: 175 job.repair(host_protection) 176 elif verify: 177 job.verify() 178 elif provision: 179 job.provision(provision) 180 elif reset: 181 job.reset() 182 else: 183 job.run(cleanup, install_before, install_after, 184 verify_job_repo_url=verify_job_repo_url, 185 only_collect_crashinfo=collect_crashinfo, 186 skip_crash_collection=skip_crash_collection) 187 finally: 188 while job.hosts: 189 host = job.hosts.pop() 190 host.close() 191 except: 192 exit_code = 1 193 traceback.print_exc() 194 195 if pid_file_manager: 196 pid_file_manager.num_tests_failed = job.num_tests_failed 197 pid_file_manager.close_file(exit_code) 198 job.cleanup_parser() 199 200 sys.exit(exit_code) 201 202 203def main(): 204 # White list of tests with run time measurement enabled. 205 measure_run_time_tests_names = global_config.global_config.get_config_value( 206 'AUTOSERV', 'measure_run_time_tests', type=str) 207 if measure_run_time_tests_names: 208 measure_run_time_tests = [t.strip() for t in 209 measure_run_time_tests_names.split(',')] 210 else: 211 measure_run_time_tests = [] 212 # grab the parser 213 parser = autoserv_parser.autoserv_parser 214 parser.parse_args() 215 216 if len(sys.argv) == 1: 217 parser.parser.print_help() 218 sys.exit(1) 219 220 if parser.options.no_logging: 221 results = None 222 else: 223 results = parser.options.results 224 if not results: 225 results = 'results.' + time.strftime('%Y-%m-%d-%H.%M.%S') 226 results = os.path.abspath(results) 227 resultdir_exists = False 228 for filename in ('control.srv', 'status.log', '.autoserv_execute'): 229 if os.path.exists(os.path.join(results, filename)): 230 resultdir_exists = True 231 if not parser.options.use_existing_results and resultdir_exists: 232 error = "Error: results directory already exists: %s\n" % results 233 sys.stderr.write(error) 234 sys.exit(1) 235 236 # Now that we certified that there's no leftover results dir from 237 # previous jobs, lets create the result dir since the logging system 238 # needs to create the log file in there. 239 if not os.path.isdir(results): 240 os.makedirs(results) 241 242 logging_manager.configure_logging( 243 server_logging_config.ServerLoggingConfig(), results_dir=results, 244 use_console=not parser.options.no_tee, 245 verbose=parser.options.verbose, 246 no_console_prefix=parser.options.no_console_prefix) 247 if results: 248 logging.info("Results placed in %s" % results) 249 250 # wait until now to perform this check, so it get properly logged 251 if parser.options.use_existing_results and not resultdir_exists: 252 logging.error("No existing results directory found: %s", results) 253 sys.exit(1) 254 255 logging.debug('autoserv command was: %s', ' '.join(sys.argv)) 256 257 if parser.options.write_pidfile: 258 pid_file_manager = pidfile.PidFileManager(parser.options.pidfile_label, 259 results) 260 pid_file_manager.open_file() 261 else: 262 pid_file_manager = None 263 264 autotest.BaseAutotest.set_install_in_tmpdir( 265 parser.options.install_in_tmpdir) 266 267 timer = None 268 try: 269 # Take the first argument as control file name, get the test name from 270 # the control file. If the test name exists in the list of tests with 271 # run time measurement enabled, start a timer to begin measurement. 272 if (len(parser.args) > 0 and parser.args[0] != '' and 273 parser.options.machines): 274 try: 275 test_name = control_data.parse_control(parser.args[0], 276 raise_warnings=True).name 277 except control_data.ControlVariableException: 278 logging.debug('Failed to retrieve test name from control file.') 279 test_name = None 280 if test_name in measure_run_time_tests: 281 machines = parser.options.machines.replace(',', ' ' 282 ).strip().split() 283 afe = frontend.AFE() 284 board = server_utils.get_board_from_afe(machines[0], afe) 285 timer = stats.Timer('autoserv_run_time.%s.%s' % 286 (board, test_name)) 287 timer.start() 288 except control_data.ControlVariableException as e: 289 logging.error(str(e)) 290 exit_code = 0 291 try: 292 try: 293 run_autoserv(pid_file_manager, results, parser) 294 except SystemExit as e: 295 exit_code = e.code 296 if exit_code: 297 logging.exception(e) 298 except Exception as e: 299 # If we don't know what happened, we'll classify it as 300 # an 'abort' and return 1. 301 logging.exception(e) 302 exit_code = 1 303 finally: 304 if pid_file_manager: 305 pid_file_manager.close_file(exit_code) 306 if timer: 307 timer.stop() 308 sys.exit(exit_code) 309 310 311if __name__ == '__main__': 312 main() 313