"""
Extract the state of the entry after doing work
@type entry: Entry
@param entry: Entry object
@type work_done: int
@param work_done: Work done info
"""
return_dict = entry.getState()
return_dict['work_done'] = work_done
return return_dict
############################################################
#
# S T A R T U P
#
############################################################
if __name__ == '__main__':
register_sighandler()
# Force integrity checks on all condor operations
gfl.set_condor_integrity_checks()
main(sys.argv[1], int(sys.argv[2]), int(sys.argv[3]), sys.argv[4],
sys.argv[5], sys.argv[6])
def compile_pickle_data(entry, work_done):
"""
Extract the state of the entry after doing work
@type entry: Entry
@param entry: Entry object
@type work_done: int
@param work_done: Work done info
"""
return_dict = entry.getState()
return_dict['work_done'] = work_done
return return_dict
############################################################
#
# S T A R T U P
#
############################################################
if __name__ == '__main__':
register_sighandler()
# Force integrity checks on all condor operations
gfl.set_condor_integrity_checks()
main(sys.argv[1], int(sys.argv[2]), int(sys.argv[3]),
sys.argv[4], sys.argv[5], sys.argv[6])
params['USE_MATCH_AUTH'] = ad_gc['GlideinParamUSE_MATCH_AUTH']
params['Report_Failed'] = 'NEVER'
# Now that we have everything submit the pilot!
logging.getLogger().setLevel(logging.DEBUG)
submitGlideins(entry_name,
"test.test",
int(nr_glideins),
idle_lifetime,
"test:test",
credentials,
client_web,
params,
status_sf,
log=logging.getLogger(),
factoryConfig=factory_config)
return 0
if __name__ == '__main__':
set_condor_integrity_checks()
try:
sys.exit(main())
except IOError as ioe:
if ioe.errno == 13: # Permission denied when accessing the credential
logging.error("Try to run the command as gfactory. Error: %s" %
ioe)
else:
raise
def main(startup_dir):
"""
Reads in the configuration file and starts up the factory
@type startup_dir: String
@param startup_dir: Path to glideinsubmit directory
"""
# Force integrity checks on all condor operations
glideFactoryLib.set_condor_integrity_checks()
glideFactoryInterface.factoryConfig.lock_dir = os.path.join(startup_dir,
"lock")
glideFactoryConfig.factoryConfig.glidein_descript_file = \
os.path.join(startup_dir,
glideFactoryConfig.factoryConfig.glidein_descript_file)
glideinDescript = glideFactoryConfig.GlideinDescript()
frontendDescript = glideFactoryConfig.FrontendDescript()
# set factory_collector at a global level, since we do not expect it to change
glideFactoryInterface.factoryConfig.factory_collector = glideinDescript.data['FactoryCollector']
# Setup the glideFactoryLib.factoryConfig so that we can process the
# globals classads
glideFactoryLib.factoryConfig.config_whoamI(
glideinDescript.data['FactoryName'],
glideinDescript.data['GlideinName'])
glideFactoryLib.factoryConfig.config_dirs(
startup_dir, glideinDescript.data['LogDir'],
glideinDescript.data['ClientLogBaseDir'],
glideinDescript.data['ClientProxiesBaseDir'])
# Set the Log directory
logSupport.log_dir = os.path.join(glideinDescript.data['LogDir'], "factory")
# Configure factory process logging
process_logs = eval(glideinDescript.data['ProcessLogs'])
for plog in process_logs:
if 'ADMIN' in plog['msg_types'].upper():
logSupport.add_processlog_handler("factoryadmin",
logSupport.log_dir,
"DEBUG,INFO,WARN,ERR",
plog['extension'],
int(float(plog['max_days'])),
int(float(plog['min_days'])),
int(float(plog['max_mbytes'])),
int(float(plog['backup_count'])),
plog['compression'])
else:
logSupport.add_processlog_handler("factory",
logSupport.log_dir,
plog['msg_types'],
plog['extension'],
int(float(plog['max_days'])),
int(float(plog['min_days'])),
int(float(plog['max_mbytes'])),
int(float(plog['backup_count'])),
plog['compression'])
logSupport.log = logging.getLogger("factory")
logSupport.log.info("Logging initialized")
if (glideinDescript.data['Entries'].strip() in ('', ',')):
# No entries are enabled. There is nothing to do. Just exit here.
log_msg = "No Entries are enabled. Exiting."
logSupport.log.error(log_msg)
sys.exit(1)
write_descript(glideinDescript,frontendDescript,os.path.join(startup_dir, 'monitor/'))
try:
os.chdir(startup_dir)
except:
logSupport.log.exception("Failed starting Factory. Unable to change to startup_dir: ")
raise
try:
if (is_file_old(glideinDescript.default_rsakey_fname,
int(glideinDescript.data['OldPubKeyGraceTime']))):
# First backup and load any existing key
logSupport.log.info("Backing up and loading old key")
glideinDescript.backup_and_load_old_key()
# Create a new key for this run
logSupport.log.info("Recreating and loading new key")
glideinDescript.load_pub_key(recreate=True)
else:
# Key is recent enough. Just reuse it.
logSupport.log.info("Key is recent enough, reusing for this run")
glideinDescript.load_pub_key(recreate=False)
logSupport.log.info("Loading old key")
glideinDescript.load_old_rsa_key()
except:
logSupport.log.exception("Failed starting Factory. Exception occurred loading factory keys: ")
raise
glideFactoryMonitorAggregator.glideFactoryMonitoring.monitoringConfig.my_name = "%s@%s" % (glideinDescript.data['GlideinName'],
glideinDescript.data['FactoryName'])
glideFactoryInterface.factoryConfig.advertise_use_tcp = (glideinDescript.data['AdvertiseWithTCP'] in ('True', '1'))
glideFactoryInterface.factoryConfig.advertise_use_multi = (glideinDescript.data['AdvertiseWithMultiple'] in ('True', '1'))
sleep_time = int(glideinDescript.data['LoopDelay'])
advertize_rate = int(glideinDescript.data['AdvertiseDelay'])
restart_attempts = int(glideinDescript.data['RestartAttempts'])
restart_interval = int(glideinDescript.data['RestartInterval'])
try:
glideinwms_dir = os.path.dirname(os.path.dirname(sys.argv[0]))
glideFactoryInterface.factoryConfig.glideinwms_version = glideinWMSVersion.GlideinWMSDistro(glideinwms_dir, 'checksum.factory').version()
except:
logSupport.log.exception("Non critical Factory error. Exception occurred while trying to retrieve the glideinwms version: ")
entries = glideinDescript.data['Entries'].split(',')
entries.sort()
glideFactoryMonitorAggregator.monitorAggregatorConfig.config_factory(
os.path.join(startup_dir, "monitor"), entries,
log = logSupport.log)
# create lock file
pid_obj = glideFactoryPidLib.FactoryPidSupport(startup_dir)
increase_process_limit()
# start
try:
pid_obj.register()
except glideFactoryPidLib.pidSupport.AlreadyRunning, err:
pid_obj.load_registered()
logSupport.log.exception("Failed starting Factory. Instance with pid %s is aready running. Exception during pid registration: %s" %
(pid_obj.mypid , err))
raise
def test_set_condor_integrity_checks(self):
set_condor_integrity_checks()
self.assertEqual(
os.environ['_CONDOR_SEC_DEFAULT_INTEGRITY'],
'REQUIRED')
def main(startup_dir):
"""
Reads in the configuration file and starts up the factory
@type startup_dir: String
@param startup_dir: Path to glideinsubmit directory
"""
# Force integrity checks on all condor operations
glideFactoryLib.set_condor_integrity_checks()
glideFactoryInterface.factoryConfig.lock_dir = os.path.join(startup_dir,
"lock")
glideFactoryConfig.factoryConfig.glidein_descript_file = \
os.path.join(startup_dir,
glideFactoryConfig.factoryConfig.glidein_descript_file)
glideinDescript = glideFactoryConfig.GlideinDescript()
frontendDescript = glideFactoryConfig.FrontendDescript()
# set factory_collector at a global level, since we do not expect it to change
glideFactoryInterface.factoryConfig.factory_collector = glideinDescript.data['FactoryCollector']
# Setup the glideFactoryLib.factoryConfig so that we can process the
# globals classads
glideFactoryLib.factoryConfig.config_whoamI(
glideinDescript.data['FactoryName'],
glideinDescript.data['GlideinName'])
glideFactoryLib.factoryConfig.config_dirs(
startup_dir, glideinDescript.data['LogDir'],
glideinDescript.data['ClientLogBaseDir'],
glideinDescript.data['ClientProxiesBaseDir'])
# Set the Log directory
logSupport.log_dir = os.path.join(glideinDescript.data['LogDir'], "factory")
# Configure factory process logging
process_logs = eval(glideinDescript.data['ProcessLogs'])
for plog in process_logs:
if 'ADMIN' in plog['msg_types'].upper():
logSupport.add_processlog_handler("factoryadmin",
logSupport.log_dir,
"DEBUG,INFO,WARN,ERR",
plog['extension'],
int(float(plog['max_days'])),
int(float(plog['min_days'])),
int(float(plog['max_mbytes'])),
int(float(plog['backup_count'])),
plog['compression'])
else:
logSupport.add_processlog_handler("factory",
logSupport.log_dir,
plog['msg_types'],
plog['extension'],
int(float(plog['max_days'])),
int(float(plog['min_days'])),
int(float(plog['max_mbytes'])),
int(float(plog['backup_count'])),
plog['compression'])
logSupport.log = logging.getLogger("factory")
logSupport.log.info("Logging initialized")
if (glideinDescript.data['Entries'].strip() in ('', ',')):
# No entries are enabled. There is nothing to do. Just exit here.
log_msg = "No Entries are enabled. Exiting."
logSupport.log.error(log_msg)
sys.exit(1)
write_descript(glideinDescript, frontendDescript, os.path.join(startup_dir, 'monitor/'))
try:
os.chdir(startup_dir)
except:
logSupport.log.exception("Failed starting Factory. Unable to change to startup_dir: ")
raise
try:
if (is_file_old(glideinDescript.default_rsakey_fname,
int(glideinDescript.data['OldPubKeyGraceTime']))):
# First backup and load any existing key
logSupport.log.info("Backing up and loading old key")
glideinDescript.backup_and_load_old_key()
# Create a new key for this run
logSupport.log.info("Recreating and loading new key")
glideinDescript.load_pub_key(recreate=True)
else:
# Key is recent enough. Just reuse it.
logSupport.log.info("Key is recent enough, reusing for this run")
glideinDescript.load_pub_key(recreate=False)
logSupport.log.info("Loading old key")
glideinDescript.load_old_rsa_key()
except RSAError as e:
logSupport.log.exception("Failed starting Factory. Exception occurred loading factory keys: ")
key_fname = getattr(e, 'key_fname', None)
cwd = getattr(e, 'cwd', None)
if key_fname and cwd:
logSupport.log.error("Failed to load RSA key %s with current working direcotry %s", key_fname, cwd)
logSupport.log.error("If you think the rsa key might be corrupted, try to remove it, and then reconfigure the factory to recreate it")
raise
except IOError as ioe:
logSupport.log.exception("Failed starting Factory. Exception occurred loading factory keys: ")
if ioe.filename == 'rsa.key' and ioe.errno == 2:
logSupport.log.error("Missing rsa.key file. Please, reconfigure the factory to recreate it")
raise
except:
logSupport.log.exception("Failed starting Factory. Exception occurred loading factory keys: ")
raise
glideFactoryMonitorAggregator.glideFactoryMonitoring.monitoringConfig.my_name = "%s@%s" % (glideinDescript.data['GlideinName'],
glideinDescript.data['FactoryName'])
glideFactoryInterface.factoryConfig.advertise_use_tcp = (glideinDescript.data['AdvertiseWithTCP'] in ('True', '1'))
glideFactoryInterface.factoryConfig.advertise_use_multi = (glideinDescript.data['AdvertiseWithMultiple'] in ('True', '1'))
sleep_time = int(glideinDescript.data['LoopDelay'])
advertize_rate = int(glideinDescript.data['AdvertiseDelay'])
restart_attempts = int(glideinDescript.data['RestartAttempts'])
restart_interval = int(glideinDescript.data['RestartInterval'])
try:
glideinwms_dir = os.path.dirname(os.path.dirname(sys.argv[0]))
glideFactoryInterface.factoryConfig.glideinwms_version = glideinWMSVersion.GlideinWMSDistro(glideinwms_dir, 'checksum.factory').version()
except:
logSupport.log.exception("Non critical Factory error. Exception occurred while trying to retrieve the glideinwms version: ")
entries = sorted(glideinDescript.data['Entries'].split(','))
glideFactoryMonitorAggregator.monitorAggregatorConfig.config_factory(
os.path.join(startup_dir, "monitor"), entries,
log=logSupport.log
)
# create lock file
pid_obj = glideFactoryPidLib.FactoryPidSupport(startup_dir)
increase_process_limit()
# start
try:
pid_obj.register()
except glideFactoryPidLib.pidSupport.AlreadyRunning as err:
pid_obj.load_registered()
logSupport.log.exception("Failed starting Factory. Instance with pid %s is aready running. Exception during pid registration: %s" %
(pid_obj.mypid, err))
raise
try:
try:
spawn(sleep_time, advertize_rate, startup_dir, glideinDescript,
frontendDescript, entries, restart_attempts, restart_interval)
except KeyboardInterrupt as e:
raise e
except HUPException as e:
# inside spawn(), outermost try will catch HUPException,
# then the code within the finally will run
# which will terminate glideFactoryEntryGroup children processes
# and then the following 3 lines will be executed.
logSupport.log.info("Received SIGHUP, reload config uid = %d" % os.getuid())
# must empty the lock file so that when the thread returns from reconfig_glidein and
# begins from the beginning, it will not error out which will happen
# if the lock file is not empty
pid_obj.relinquish()
os.execv(os.path.join(FACTORY_DIR, "../creation/reconfig_glidein"),
['reconfig_glidein', '-update_scripts', 'no', '-sighupreload', '-xml', '/etc/gwms-factory/glideinWMS.xml'])
except:
logSupport.log.exception("Exception occurred spawning the factory: ")
finally:
pid_obj.relinquish()
def test_set_condor_integrity_checks(self):
set_condor_integrity_checks()
self.assertEqual(
os.environ['_CONDOR_SEC_DEFAULT_INTEGRITY'],
'REQUIRED')
