autoserv revision f594c5ee0a769d00b0f0e8a6ebe1c6e19cf53417
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.global_config import global_config 14require_atfork = global_config.get_config_value( 15 'AUTOSERV', 'require_atfork_module', type=bool, default=True) 16 17try: 18 import atfork 19 atfork.monkeypatch_os_fork_functions() 20 import atfork.stdlib_fixer 21 # Fix the Python standard library for threading+fork safety with its 22 # internal locks. http://code.google.com/p/python-atfork/ 23 import warnings 24 warnings.filterwarnings('ignore', 'logging module already imported') 25 atfork.stdlib_fixer.fix_logging_module() 26except ImportError, e: 27 from autotest_lib.client.common_lib import global_config 28 if global_config.global_config.get_config_value( 29 'AUTOSERV', 'require_atfork_module', type=bool, default=False): 30 print >>sys.stderr, 'Please run utils/build_externals.py' 31 print e 32 sys.exit(1) 33 34from autotest_lib.server import server_logging_config 35from autotest_lib.server import server_job, utils, autoserv_parser, autotest 36from autotest_lib.client.common_lib import pidfile, logging_manager 37 38def log_alarm(signum, frame): 39 logging.error("Received SIGALARM. Ignoring and continuing on.") 40 sys.exit(1) 41 42def run_autoserv(pid_file_manager, results, parser): 43 # send stdin to /dev/null 44 dev_null = os.open(os.devnull, os.O_RDONLY) 45 os.dup2(dev_null, sys.stdin.fileno()) 46 os.close(dev_null) 47 48 # Create separate process group 49 os.setpgrp() 50 51 # Implement SIGTERM handler 52 def handle_sigterm(signum, frame): 53 if pid_file_manager: 54 pid_file_manager.close_file(1, signal.SIGTERM) 55 os.killpg(os.getpgrp(), signal.SIGKILL) 56 57 # Set signal handler 58 signal.signal(signal.SIGTERM, handle_sigterm) 59 60 # Ignore SIGTTOU's generated by output from forked children. 61 signal.signal(signal.SIGTTOU, signal.SIG_IGN) 62 63 # If we received a SIGALARM, let's be loud about it. 64 signal.signal(signal.SIGALRM, log_alarm) 65 66 # Server side tests that call shell scripts often depend on $USER being set 67 # but depending on how you launch your autotest scheduler it may not be set. 68 os.environ['USER'] = getpass.getuser() 69 70 if parser.options.machines: 71 machines = parser.options.machines.replace(',', ' ').strip().split() 72 else: 73 machines = [] 74 machines_file = parser.options.machines_file 75 label = parser.options.label 76 group_name = parser.options.group_name 77 user = parser.options.user 78 client = parser.options.client 79 server = parser.options.server 80 install_before = parser.options.install_before 81 install_after = parser.options.install_after 82 verify = parser.options.verify 83 repair = parser.options.repair 84 cleanup = parser.options.cleanup 85 provision = parser.options.provision 86 no_tee = parser.options.no_tee 87 parse_job = parser.options.parse_job 88 execution_tag = parser.options.execution_tag 89 if not execution_tag: 90 execution_tag = parse_job 91 host_protection = parser.options.host_protection 92 ssh_user = parser.options.ssh_user 93 ssh_port = parser.options.ssh_port 94 ssh_pass = parser.options.ssh_pass 95 collect_crashinfo = parser.options.collect_crashinfo 96 control_filename = parser.options.control_filename 97 test_retry = parser.options.test_retry 98 verify_job_repo_url = parser.options.verify_job_repo_url 99 skip_crash_collection = parser.options.skip_crash_collection 100 101 # can't be both a client and a server side test 102 if client and server: 103 parser.parser.error("Can not specify a test as both server and client!") 104 105 if provision and client: 106 parser.parser.error("Cannot specify provisioning and client!") 107 108 is_special_task = (verify or repair or cleanup or collect_crashinfo or 109 provision) 110 if len(parser.args) < 1 and not is_special_task: 111 parser.parser.error("Missing argument: control file") 112 113 # We have a control file unless it's just a verify/repair/cleanup job 114 if len(parser.args) > 0: 115 control = parser.args[0] 116 else: 117 control = None 118 119 if machines_file: 120 machines = [] 121 for m in open(machines_file, 'r').readlines(): 122 # remove comments, spaces 123 m = re.sub('#.*', '', m).strip() 124 if m: 125 machines.append(m) 126 print "Read list of machines from file: %s" % machines_file 127 print ','.join(machines) 128 129 if machines: 130 for machine in machines: 131 if not machine or re.search('\s', machine): 132 parser.parser.error("Invalid machine: %s" % str(machine)) 133 machines = list(set(machines)) 134 machines.sort() 135 136 if group_name and len(machines) < 2: 137 parser.parser.error("-G %r may only be supplied with more than one machine." 138 % group_name) 139 140 kwargs = {'group_name': group_name, 'tag': execution_tag} 141 if control_filename: 142 kwargs['control_filename'] = control_filename 143 job = server_job.server_job(control, parser.args[1:], results, label, 144 user, machines, client, parse_job, 145 ssh_user, ssh_port, ssh_pass, test_retry, 146 **kwargs) 147 job.logging.start_logging() 148 job.init_parser() 149 150 # perform checks 151 job.precheck() 152 153 # run the job 154 exit_code = 0 155 try: 156 try: 157 if repair: 158 job.repair(host_protection) 159 elif verify: 160 job.verify() 161 elif provision: 162 job.provision(provision) 163 else: 164 job.run(cleanup, install_before, install_after, 165 verify_job_repo_url=verify_job_repo_url, 166 only_collect_crashinfo=collect_crashinfo, 167 skip_crash_collection=skip_crash_collection) 168 finally: 169 while job.hosts: 170 host = job.hosts.pop() 171 host.close() 172 except: 173 exit_code = 1 174 traceback.print_exc() 175 176 if pid_file_manager: 177 pid_file_manager.num_tests_failed = job.num_tests_failed 178 pid_file_manager.close_file(exit_code) 179 job.cleanup_parser() 180 181 sys.exit(exit_code) 182 183 184def main(): 185 # grab the parser 186 parser = autoserv_parser.autoserv_parser 187 parser.parse_args() 188 189 if len(sys.argv) == 1: 190 parser.parser.print_help() 191 sys.exit(1) 192 193 if parser.options.no_logging: 194 results = None 195 else: 196 results = parser.options.results 197 if not results: 198 results = 'results.' + time.strftime('%Y-%m-%d-%H.%M.%S') 199 results = os.path.abspath(results) 200 resultdir_exists = False 201 for filename in ('control.srv', 'status.log', '.autoserv_execute'): 202 if os.path.exists(os.path.join(results, filename)): 203 resultdir_exists = True 204 if not parser.options.use_existing_results and resultdir_exists: 205 error = "Error: results directory already exists: %s\n" % results 206 sys.stderr.write(error) 207 sys.exit(1) 208 209 # Now that we certified that there's no leftover results dir from 210 # previous jobs, lets create the result dir since the logging system 211 # needs to create the log file in there. 212 if not os.path.isdir(results): 213 os.makedirs(results) 214 215 logging_manager.configure_logging( 216 server_logging_config.ServerLoggingConfig(), results_dir=results, 217 use_console=not parser.options.no_tee, 218 verbose=parser.options.verbose, 219 no_console_prefix=parser.options.no_console_prefix) 220 if results: 221 logging.info("Results placed in %s" % results) 222 223 # wait until now to perform this check, so it get properly logged 224 if parser.options.use_existing_results and not resultdir_exists: 225 logging.error("No existing results directory found: %s", results) 226 sys.exit(1) 227 228 229 if parser.options.write_pidfile: 230 pid_file_manager = pidfile.PidFileManager(parser.options.pidfile_label, 231 results) 232 pid_file_manager.open_file() 233 else: 234 pid_file_manager = None 235 236 autotest.BaseAutotest.set_install_in_tmpdir( 237 parser.options.install_in_tmpdir) 238 239 exit_code = 0 240 try: 241 try: 242 run_autoserv(pid_file_manager, results, parser) 243 except SystemExit, e: 244 exit_code = e.code 245 except: 246 traceback.print_exc() 247 # If we don't know what happened, we'll classify it as 248 # an 'abort' and return 1. 249 exit_code = 1 250 finally: 251 if pid_file_manager: 252 pid_file_manager.close_file(exit_code) 253 sys.exit(exit_code) 254 255 256if __name__ == '__main__': 257 main() 258