def load_log(self, section):
# read these from config file
log_name = str(self.config[section]["log_name"])
extension = str(self.config[section]["extension"])
msg_types = str(self.config[section]["msg_types"])
max_days = float(self.config[section]["max_days"])
min_days = float(self.config[section]["min_days"])
max_mbytes = float(self.config[section]["max_mbytes"])
backupCount = 5
try:
backupCount = int(self.config[section]["backup_count"])
except:
pass # backup_count may not exist in all sections
compression = ''
try:
compression = str(self.config[section]["compression"])
except:
pass # compress may not exist in all sections
log_dir = "%s/%s" % (self.log_base_dir, log_name)
os.makedirs(log_dir)
logSupport.add_processlog_handler(log_name,
log_dir,
msg_types,
extension,
max_days,
min_days,
max_mbytes,
backupCount=backupCount,
compression=compression)
return logging.getLogger(log_name), log_dir
def load_log(self, section):
# read these from config file
log_name = str(self.config[section]["log_name"])
extension = str(self.config[section]["extension"])
msg_types = str(self.config[section]["msg_types"])
max_days = float(self.config[section]["max_days"])
min_days = float(self.config[section]["min_days"])
max_mbytes = float(self.config[section]["max_mbytes"])
backupCount = 5
try:
backupCount = int(self.config[section]["backup_count"])
except:
pass # backup_count may not exist in all sections
log_dir = "%s/%s" % (self.log_base_dir, log_name)
os.makedirs(log_dir)
logSupport.add_processlog_handler(
log_name,
log_dir,
msg_types,
extension,
max_days,
min_days,
max_mbytes,
backupCount=backupCount)
return logging.getLogger(log_name), log_dir
def main(work_dir, action):
startup_time=time.time()
glideinFrontendConfig.frontendConfig.frontend_descript_file = os.path.join(work_dir, glideinFrontendConfig.frontendConfig.frontend_descript_file)
frontendDescript = glideinFrontendConfig.FrontendDescript(work_dir)
# the log dir is shared between the frontend main and the groups, so use a subdir
logSupport.log_dir = os.path.join(frontendDescript.data['LogDir'], "frontend")
# Configure frontend process logging
process_logs = eval(frontendDescript.data['ProcessLogs'])
for plog in process_logs:
logSupport.add_processlog_handler("frontend", 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("frontend")
logSupport.log.info("Logging initialized")
logSupport.log.debug("Frontend startup time: %s" % str(startup_time))
try:
cleanup_environ()
# we use a dedicated config... ignore the system-wide
os.environ['CONDOR_CONFIG'] = frontendDescript.data['CondorConfig']
sleep_time = int(frontendDescript.data['LoopDelay'])
advertize_rate = int(frontendDescript.data['AdvertiseDelay'])
max_parallel_workers = int(frontendDescript.data['GroupParallelWorkers'])
restart_attempts = int(frontendDescript.data['RestartAttempts'])
restart_interval = int(frontendDescript.data['RestartInterval'])
groups = string.split(frontendDescript.data['Groups'], ',')
groups.sort()
glideinFrontendMonitorAggregator.monitorAggregatorConfig.config_frontend(os.path.join(work_dir, "monitor"), groups)
except:
logSupport.log.exception("Exception occurred configuring monitoring: ")
raise
glideinFrontendMonitoring.write_frontend_descript_xml(frontendDescript, os.path.join(work_dir, 'monitor/'))
logSupport.log.info("Enabled groups: %s" % groups)
# create lock file
pid_obj = glideinFrontendPidLib.FrontendPidSupport(work_dir)
# start
try:
pid_obj.register(action)
except glideinFrontendPidLib.pidSupport.AlreadyRunning, err:
pid_obj.load_registered()
logSupport.log.exception("Failed starting Frontend with action %s. Instance with pid %s is aready running for action %s. Exception during pid registration: %s" %
(action, pid_obj.mypid , str(pid_obj.action_type), err))
raise
def init_logs(name, log_dir, process_logs):
for plog in process_logs:
logSupport.add_processlog_handler(name, log_dir,
plog['msg_types'], plog['extension'],
int(float(plog['max_days'])),
int(float(plog['min_days'])),
int(float(plog['max_mbytes'])))
logSupport.log = logging.getLogger(name)
logSupport.log.info("Logging initialized for %s" % name)
def main(work_dir):
startup_time=time.time()
glideinFrontendConfig.frontendConfig.frontend_descript_file = os.path.join(work_dir, glideinFrontendConfig.frontendConfig.frontend_descript_file)
frontendDescript = glideinFrontendConfig.FrontendDescript(work_dir)
# the log dir is shared between the frontend main and the groups, so use a subdir
logSupport.log_dir = os.path.join(frontendDescript.data['LogDir'], "frontend")
# Configure frontend process logging
process_logs = eval(frontendDescript.data['ProcessLogs'])
for plog in process_logs:
logSupport.add_processlog_handler("frontend", logSupport.log_dir,
plog['msg_types'], plog['extension'],
int(float(plog['max_days'])),
int(float(plog['min_days'])),
int(float(plog['max_mbytes'])))
logSupport.log = logging.getLogger("frontend")
logSupport.log.info("Logging initialized")
logSupport.log.debug("Frontend startup time: %s" % str(startup_time))
try:
cleanup_environ()
# we use a dedicated config... ignore the system-wide
os.environ['CONDOR_CONFIG'] = frontendDescript.data['CondorConfig']
sleep_time = int(frontendDescript.data['LoopDelay'])
advertize_rate = int(frontendDescript.data['AdvertiseDelay'])
restart_attempts = int(frontendDescript.data['RestartAttempts'])
restart_interval = int(frontendDescript.data['RestartInterval'])
groups = string.split(frontendDescript.data['Groups'], ',')
groups.sort()
glideinFrontendMonitorAggregator.monitorAggregatorConfig.config_frontend(os.path.join(work_dir, "monitor"), groups)
except:
logSupport.log.exception("Exception occurred configuring monitoring: ")
raise
glideinFrontendMonitoring.write_frontend_descript_xml(frontendDescript, os.path.join(work_dir, 'monitor/'))
# create lock file
pid_obj = glideinFrontendPidLib.FrontendPidSupport(work_dir)
# start
pid_obj.register()
try:
try:
spawn(sleep_time, advertize_rate, work_dir,
frontendDescript, groups, restart_attempts, restart_interval)
except KeyboardInterrupt:
logSupport.log.info("Received signal...exit")
except:
logSupport.log.exception("Exception occurred trying to spawn: ")
finally:
pid_obj.relinquish()
def init_logs(name, log_dir, process_logs):
for plog in process_logs:
logSupport.add_processlog_handler(name, 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(name)
logSupport.log.info("Logging initialized for %s" % name)
def run(self):
import logging
from glideinwms.lib import logSupport
logger_name = self.logger_name or _default_logger_name
directory = self.directory or _default_directory
extension = self.extension or _default_extension
levels = self.levels or _default_levels
max_days = self.maxDays or _default_max_days
min_days = self.minDays or _default_min_days
max_size = self.maxDays or _default_max_size
logSupport.add_processlog_handler(logger_name, directory, levels,
extension, max_days, min_days,
max_size)
logSupport.log = logging.getLogger(self.logger_name)
_external_logger = logging.getLogger(self.logger_name)
logSupport.log.info("GlideTester logging initialized.")
def run(self):
import logging
from glideinwms.lib import logSupport
logger_name = self.logger_name or _default_logger_name
directory = self.directory or _default_directory
extension = self.extension or _default_extension
levels = self.levels or _default_levels
max_days = self.maxDays or _default_max_days
min_days = self.minDays or _default_min_days
max_size = self.maxDays or _default_max_size
logSupport.add_processlog_handler(logger_name, directory,
levels, extension,
max_days, min_days,
max_size
)
logSupport.log = logging.getLogger(self.logger_name)
_external_logger = logging.getLogger(self.logger_name)
logSupport.log.info("GlideTester logging initialized.")
def main(work_dir, action):
startup_time = time.time()
glideinFrontendConfig.frontendConfig.frontend_descript_file = os.path.join(work_dir, glideinFrontendConfig.frontendConfig.frontend_descript_file)
frontendDescript = glideinFrontendConfig.FrontendDescript(work_dir)
# the log dir is shared between the frontend main and the groups, so use a subdir
logSupport.log_dir = os.path.join(frontendDescript.data['LogDir'],
"frontend")
# Configure frontend process logging
process_logs = eval(frontendDescript.data['ProcessLogs'])
for plog in process_logs:
logSupport.add_processlog_handler("frontend", 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("frontend")
logSupport.log.info("Logging initialized")
logSupport.log.debug("Frontend startup time: %s" % str(startup_time))
try:
cleanup_environ()
# we use a dedicated config... ignore the system-wide
os.environ['CONDOR_CONFIG'] = frontendDescript.data['CondorConfig']
sleep_time = int(frontendDescript.data['LoopDelay'])
advertize_rate = int(frontendDescript.data['AdvertiseDelay'])
max_parallel_workers = int(frontendDescript.data['GroupParallelWorkers'])
restart_attempts = int(frontendDescript.data['RestartAttempts'])
restart_interval = int(frontendDescript.data['RestartInterval'])
groups = sorted(frontendDescript.data['Groups'].split(','))
glideinFrontendMonitorAggregator.config_frontend(os.path.join(work_dir, "monitor"), groups)
except:
logSupport.log.exception("Exception occurred configuring monitoring: ")
raise
glideinFrontendMonitoring.write_frontend_descript_xml(
frontendDescript, os.path.join(work_dir, 'monitor/'))
logSupport.log.info("Enabled groups: %s" % groups)
# create lock file
pid_obj = glideinFrontendPidLib.FrontendPidSupport(work_dir)
# start
try:
pid_obj.register(action)
except glideinFrontendPidLib.pidSupport.AlreadyRunning as err:
pid_obj.load_registered()
logSupport.log.exception("Failed starting Frontend with action %s. Instance with pid %s is aready running for action %s. Exception during pid registration: %s" % (action, pid_obj.mypid, str(pid_obj.action_type), err))
raise
try:
try:
if action == "run":
spawn(sleep_time, advertize_rate, work_dir,
frontendDescript, groups, max_parallel_workers,
restart_interval, restart_attempts)
elif action in ('removeWait', 'removeIdle', 'removeAll', 'removeWaitExcess', 'removeIdleExcess', 'removeAllExcess'):
spawn_removal(work_dir, frontendDescript, groups,
max_parallel_workers, action)
else:
raise ValueError("Unknown action: %s" % action)
except KeyboardInterrupt:
logSupport.log.info("Received signal...exit")
except HUPException:
logSupport.log.info("Received SIGHUP, reload config")
pid_obj.relinquish()
os.execv( os.path.join(FRONTEND_DIR, "../creation/reconfig_frontend"), ['reconfig_frontend', '-sighupreload', '-xml', '/etc/gwms-frontend/frontend.xml'] )
except:
logSupport.log.exception("Exception occurred trying to spawn: ")
finally:
pid_obj.relinquish()
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 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 main(parent_pid, work_dir, group_name):
startup_time = time.time()
elementDescript = glideinFrontendConfig.ElementMergedDescript(work_dir, group_name)
# the log dir is shared between the frontend main and the groups, so use a subdir
logSupport.log_dir = os.path.join(elementDescript.frontend_data['LogDir'], "group_%s" % group_name)
# Configure frontend group process logging
process_logs = eval(elementDescript.frontend_data['ProcessLogs'])
for plog in process_logs:
logSupport.add_processlog_handler(group_name, logSupport.log_dir, plog['msg_types'], plog['extension'],
int(float(plog['max_days'])),
int(float(plog['min_days'])),
int(float(plog['max_mbytes'])))
logSupport.log = logging.getLogger(group_name)
logSupport.log.info("Logging initialized")
logSupport.log.debug("Frontend Element startup time: %s" % str(startup_time))
paramsDescript = glideinFrontendConfig.ParamsDescript(work_dir, group_name)
attrsDescript = glideinFrontendConfig.AttrsDescript(work_dir,group_name)
signatureDescript = glideinFrontendConfig.GroupSignatureDescript(work_dir, group_name)
#
# We decided we will not use the data from the stage area
# Leaving it commented in the code, in case we decide in the future
# it was a good validation of the Web server health
#
#stageArea=glideinFrontendConfig.MergeStageFiles(elementDescript.frontend_data['WebURL'],
# signatureDescript.signature_type,
# signatureDescript.frontend_descript_fname,signatureDescript.frontend_descript_signature,
# group_name,
# signatureDescript.group_descript_fname,signatureDescript.group_descript_signature)
# constsDescript=stageArea.get_constants()
#
attr_dict=attrsDescript.data
glideinFrontendMonitoring.monitoringConfig.monitor_dir = os.path.join(work_dir, "monitor/group_%s" % group_name)
glideinFrontendInterface.frontendConfig.advertise_use_tcp = (elementDescript.frontend_data['AdvertiseWithTCP'] in ('True', '1'))
glideinFrontendInterface.frontendConfig.advertise_use_multi = (elementDescript.frontend_data['AdvertiseWithMultiple'] in ('True', '1'))
try:
glideinwms_dir = os.path.dirname(os.path.dirname(sys.argv[0]))
glideinFrontendInterface.frontendConfig.glideinwms_version = glideinWMSVersion.GlideinWMSDistro(glideinwms_dir, 'checksum.frontend').version()
except:
logSupport.log.exception("Exception occurred while trying to retrieve the glideinwms version: ")
if len(elementDescript.merged_data['Proxies']) > 0:
if not glideinFrontendPlugins.proxy_plugins.has_key(elementDescript.merged_data['ProxySelectionPlugin']):
logSupport.log.warning("Invalid ProxySelectionPlugin '%s', supported plugins are %s" % (elementDescript.merged_data['ProxySelectionPlugin']), glideinFrontendPlugins.proxy_plugins.keys())
return 1
x509_proxy_plugin = glideinFrontendPlugins.proxy_plugins[elementDescript.merged_data['ProxySelectionPlugin']](os.path.join(work_dir, "group_%s" % group_name), glideinFrontendPlugins.createCredentialList(elementDescript))
else:
# no proxies, will try to use the factory one
x509_proxy_plugin = None
# set the condor configuration and GSI setup globally, so I don't need to worry about it later on
os.environ['CONDOR_CONFIG'] = elementDescript.frontend_data['CondorConfig']
os.environ['_CONDOR_CERTIFICATE_MAPFILE'] = elementDescript.element_data['MapFile']
os.environ['X509_USER_PROXY'] = elementDescript.frontend_data['ClassAdProxy']
# create lock file
pid_obj = glideinFrontendPidLib.ElementPidSupport(work_dir, group_name)
pid_obj.register(parent_pid)
try:
try:
logSupport.log.info("Starting up")
iterate(parent_pid, elementDescript, paramsDescript, attr_dict, signatureDescript, x509_proxy_plugin)
except KeyboardInterrupt:
logSupport.log.info("Received signal...exit")
except:
logSupport.log.exception("Unhandled exception, dying: ")
finally:
pid_obj.relinquish()
def main(work_dir, action):
startup_time = time.time()
glideinFrontendConfig.frontendConfig.frontend_descript_file = os.path.join(work_dir, glideinFrontendConfig.frontendConfig.frontend_descript_file)
frontendDescript = glideinFrontendConfig.FrontendDescript(work_dir)
# the log dir is shared between the frontend main and the groups, so use a subdir
logSupport.log_dir = os.path.join(frontendDescript.data['LogDir'],
"frontend")
# Configure frontend process logging
process_logs = eval(frontendDescript.data['ProcessLogs'])
for plog in process_logs:
logSupport.add_processlog_handler("frontend", 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("frontend")
logSupport.log.info("Logging initialized")
logSupport.log.debug("Frontend startup time: %s" % str(startup_time))
try:
cleanup_environ()
# we use a dedicated config... ignore the system-wide
os.environ['CONDOR_CONFIG'] = frontendDescript.data['CondorConfig']
sleep_time = int(frontendDescript.data['LoopDelay'])
advertize_rate = int(frontendDescript.data['AdvertiseDelay'])
max_parallel_workers = int(frontendDescript.data['GroupParallelWorkers'])
restart_attempts = int(frontendDescript.data['RestartAttempts'])
restart_interval = int(frontendDescript.data['RestartInterval'])
groups = sorted(frontendDescript.data['Groups'].split(','))
glideinFrontendMonitorAggregator.monitorAggregatorConfig.config_frontend(os.path.join(work_dir, "monitor"), groups)
except:
logSupport.log.exception("Exception occurred configuring monitoring: ")
raise
glideinFrontendMonitoring.write_frontend_descript_xml(
frontendDescript, os.path.join(work_dir, 'monitor/'))
logSupport.log.info("Enabled groups: %s" % groups)
# create lock file
pid_obj = glideinFrontendPidLib.FrontendPidSupport(work_dir)
# start
try:
pid_obj.register(action)
except glideinFrontendPidLib.pidSupport.AlreadyRunning as err:
pid_obj.load_registered()
logSupport.log.exception("Failed starting Frontend with action %s. Instance with pid %s is aready running for action %s. Exception during pid registration: %s" % (action, pid_obj.mypid, str(pid_obj.action_type), err))
raise
try:
try:
if action == "run":
spawn(sleep_time, advertize_rate, work_dir,
frontendDescript, groups, max_parallel_workers,
restart_interval, restart_attempts)
elif action in ('removeWait', 'removeIdle', 'removeAll', 'removeWaitExcess', 'removeIdleExcess', 'removeAllExcess'):
spawn_removal(work_dir, frontendDescript, groups,
max_parallel_workers, action)
else:
raise ValueError("Unknown action: %s" % action)
except KeyboardInterrupt:
logSupport.log.info("Received signal...exit")
except HUPException:
logSupport.log.info("Received SIGHUP, reload config")
pid_obj.relinquish()
os.execv( os.path.join(FRONTEND_DIR, "../creation/reconfig_frontend"), ['reconfig_frontend', '-sighupreload', '-xml', '/etc/gwms-frontend/frontend.xml'] )
except:
logSupport.log.exception("Exception occurred trying to spawn: ")
finally:
pid_obj.relinquish()
