server/autoserv

   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 """
   6 Run an control file through the server side engine
   7 """
   8
   9 import sys, os, re, traceback, signal, time, logging, getpass
  10
  11 import common
  12
  13 from autotest_lib.client.common_lib.global_config import global_config
  14 require_atfork = global_config.get_config_value(
  15         'AUTOSERV', 'require_atfork_module', type=bool, default=True)
  16
  17 try:
  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()
  26 except 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
  34 from autotest_lib.server import server_logging_config
  35 from autotest_lib.server import server_job, utils, autoserv_parser, autotest
  36 from autotest_lib.client.common_lib import pidfile, logging_manager
  37
  38 def run_autoserv(pid_file_manager, results, parser):
  39     # send stdin to /dev/null
  40     dev_null = os.open(os.devnull, os.O_RDONLY)
  41     os.dup2(dev_null, sys.stdin.fileno())
  42     os.close(dev_null)
  43
  44     # Create separate process group
  45     os.setpgrp()
  46
  47     # Implement SIGTERM handler
  48     def handle_sigterm(signum, frame):
  49         if pid_file_manager:
  50             pid_file_manager.close_file(1, signal.SIGTERM)
  51         os.killpg(os.getpgrp(), signal.SIGKILL)
  52
  53     # Set signal handler
  54     signal.signal(signal.SIGTERM, handle_sigterm)
  55
  56     # Server side tests that call shell scripts often depend on $USER being set
  57     # but depending on how you launch your autotest scheduler it may not be set.
  58     os.environ['USER'] = getpass.getuser()
  59
  60     if parser.options.machines:
  61         machines = parser.options.machines.replace(',', ' ').strip().split()
  62     else:
  63         machines = []
  64     machines_file = parser.options.machines_file
  65     label = parser.options.label
  66     group_name = parser.options.group_name
  67     user = parser.options.user
  68     client = parser.options.client
  69     server = parser.options.server
  70     install_before = parser.options.install_before
  71     install_after = parser.options.install_after
  72     verify = parser.options.verify
  73     repair = parser.options.repair
  74     cleanup = parser.options.cleanup
  75     no_tee = parser.options.no_tee
  76     parse_job = parser.options.parse_job
  77     execution_tag = parser.options.execution_tag
  78     if not execution_tag:
  79         execution_tag = parse_job
  80     host_protection = parser.options.host_protection
  81     ssh_user = parser.options.ssh_user
  82     ssh_port = parser.options.ssh_port
  83     ssh_pass = parser.options.ssh_pass
  84     collect_crashinfo = parser.options.collect_crashinfo
  85     control_filename = parser.options.control_filename
  86
  87     # can't be both a client and a server side test
  88     if client and server:
  89         parser.parser.error("Can not specify a test as both server and client!")
  90
  91     if len(parser.args) < 1 and not (verify or repair or cleanup
  92                                      or collect_crashinfo):
  93         parser.parser.error("Missing argument: control file")
  94
  95     # We have a control file unless it's just a verify/repair/cleanup job
  96     if len(parser.args) > 0:
  97         control = parser.args[0]
  98     else:
  99         control = None
 100
 101     if machines_file:
 102         machines = []
 103         for m in open(machines_file, 'r').readlines():
 104             # remove comments, spaces
 105             m = re.sub('#.*', '', m).strip()
 106             if m:
 107                 machines.append(m)
 108         print "Read list of machines from file: %s" % machines_file
 109         print ','.join(machines)
 110
 111     if machines:
 112         for machine in machines:
 113             if not machine or re.search('\s', machine):
 114                 parser.parser.error("Invalid machine: %s" % str(machine))
 115         machines = list(set(machines))
 116         machines.sort()
 117
 118     if group_name and len(machines) < 2:
 119         parser.parser.error("-G %r may only be supplied with more than one machine."
 120                % group_name)
 121
 122     kwargs = {'group_name': group_name, 'tag': execution_tag}
 123     if control_filename:
 124         kwargs['control_filename'] = control_filename
 125     job = server_job.server_job(control, parser.args[1:], results, label,
 126                                 user, machines, client, parse_job,
 127                                 ssh_user, ssh_port, ssh_pass, **kwargs)
 128     job.logging.start_logging()
 129     job.init_parser()
 130
 131     # perform checks
 132     job.precheck()
 133
 134     # run the job
 135     exit_code = 0
 136     try:
 137         try:
 138             if repair:
 139                 job.repair(host_protection)
 140             elif verify:
 141                 job.verify()
 142             else:
 143                 job.run(cleanup, install_before, install_after,
 144                         only_collect_crashinfo=collect_crashinfo)
 145         finally:
 146             while job.hosts:
 147                 host = job.hosts.pop()
 148                 host.close()
 149     except:
 150         exit_code = 1
 151         traceback.print_exc()
 152
 153     if pid_file_manager:
 154         pid_file_manager.num_tests_failed = job.num_tests_failed
 155         pid_file_manager.close_file(exit_code)
 156     job.cleanup_parser()
 157
 158     sys.exit(exit_code)
 159
 160
 161 def main():
 162     # grab the parser
 163     parser = autoserv_parser.autoserv_parser
 164     parser.parse_args()
 165
 166     if len(sys.argv) == 1:
 167         parser.parser.print_help()
 168         sys.exit(1)
 169
 170     if parser.options.no_logging:
 171         results = None
 172     else:
 173         results = parser.options.results
 174         if not results:
 175             results = 'results.' + time.strftime('%Y-%m-%d-%H.%M.%S')
 176         results  = os.path.abspath(results)
 177         resultdir_exists = False
 178         for filename in ('control.srv', 'status.log', '.autoserv_execute'):
 179             if os.path.exists(os.path.join(results, filename)):
 180                 resultdir_exists = True
 181         if not parser.options.use_existing_results and resultdir_exists:
 182             error = "Error: results directory already exists: %s\n" % results
 183             sys.stderr.write(error)
 184             sys.exit(1)
 185
 186         # Now that we certified that there's no leftover results dir from
 187         # previous jobs, lets create the result dir since the logging system
 188         # needs to create the log file in there.
 189         if not os.path.isdir(results):
 190             os.makedirs(results)
 191
 192     logging_manager.configure_logging(
 193             server_logging_config.ServerLoggingConfig(), results_dir=results,
 194             use_console=not parser.options.no_tee,
 195             verbose=parser.options.verbose,
 196             no_console_prefix=parser.options.no_console_prefix)
 197     if results:
 198         logging.info("Results placed in %s" % results)
 199
 200         # wait until now to perform this check, so it get properly logged
 201         if parser.options.use_existing_results and not resultdir_exists:
 202             logging.error("No existing results directory found: %s", results)
 203             sys.exit(1)
 204
 205
 206     if parser.options.write_pidfile:
 207         pid_file_manager = pidfile.PidFileManager(parser.options.pidfile_label,
 208                                                   results)
 209         pid_file_manager.open_file()
 210     else:
 211         pid_file_manager = None
 212
 213     autotest.BaseAutotest.set_install_in_tmpdir(
 214         parser.options.install_in_tmpdir)
 215
 216     exit_code = 0
 217     try:
 218         try:
 219             run_autoserv(pid_file_manager, results, parser)
 220         except SystemExit, e:
 221             exit_code = e.code
 222         except:
 223             traceback.print_exc()
 224             # If we don't know what happened, we'll classify it as
 225             # an 'abort' and return 1.
 226             exit_code = 1
 227     finally:
 228         if pid_file_manager:
 229             pid_file_manager.close_file(exit_code)
 230     sys.exit(exit_code)
 231
 232
 233 if __name__ == '__main__':
 234     main()