def get_element_pids(work_dir, frontend_pid):
# get element pids
frontendDescript = glideinFrontendConfig.FrontendDescript(work_dir)
groups = sorted(string.split(frontendDescript.data['Groups'], ','))
element_pids = {}
for group in groups:
try:
element_pid, element_ppid = glideinFrontendPidLib.get_element_pid(
work_dir, group)
except RuntimeError as e:
print(e)
continue # report error and go to next group
if element_ppid != frontend_pid:
print("Group '%s' has an unexpected Parent PID: %s!=%s" %
(group, element_ppid, frontend_pid))
continue # report error and go to next group
element_pids[group] = element_pid
return element_pids
def __init__(self, frontend_workdir='/var/lib/gwms-frontend/vofrontend'):
self.frontend_workdir = frontend_workdir
glideinFrontendConfig.frontendConfig.frontend_descript_file = \
os.path.join(
self.frontend_workdir,
glideinFrontendConfig.frontendConfig.frontend_descript_file)
self.frontend_descript = glideinFrontendConfig.FrontendDescript(
self.frontend_workdir)
self.groups = self.frontend_descript.data['Groups'].split(',')
self.groups.sort()
self.group_descript = {}
self.group_params_descript = {}
self.group_sign_descript = {}
self.group_attrs_descript = {}
self.frontend_config = {'frontend': {}, 'group': {}}
self.frontend_config['frontend'] = {
'frontend_name': self.frontend_descript.data['FrontendName'],
'workdir': self.frontend_workdir,
'web_url': self.frontend_descript.data['WebURL'],
'monitoring_web_url':
self.frontend_descript.data['MonitoringWebURL'],
}
for g in self.groups:
self.group_descript[
g] = glideinFrontendConfig.ElementMergedDescript(
self.frontend_workdir, g)
self.group_params_descript[
g] = glideinFrontendConfig.ParamsDescript(
self.frontend_workdir, g)
self.group_sign_descript[
g] = glideinFrontendConfig.GroupSignatureDescript(
self.frontend_workdir, g)
self.group_attrs_descript[g] = glideinFrontendConfig.AttrsDescript(
self.frontend_workdir, g)
self.frontend_config['group'][g] = {
'workdir':
glideinFrontendConfig.get_group_dir(self.frontend_workdir, g),
'web_url':
os.path.join(self.frontend_descript.data['WebURL'],
'group_%s' % g),
'security_name':
self.group_descript[g].merged_data['SecurityName'],
'factory_collectors':
self.group_descript[g].merged_data['FactoryCollectors'],
'min_running':
self.group_descript[g].element_data['MinRunningPerEntry'],
'max_running':
self.group_descript[g].element_data['MaxRunningPerEntry'],
'fraction_running':
float(self.group_descript[g].
element_data['FracRunningPerEntry']),
'max_idle':
int(self.group_descript[g].element_data['MaxIdlePerEntry']),
'reserve_idle':
int(self.group_descript[g].element_data['ReserveIdlePerEntry']
),
#'idle_lifetime': int(self.group_descript[g].element_data['IdleLifetime']),
'max_vms_idle':
int(self.group_descript[g].element_data['MaxIdleVMsPerEntry']),
'curb_vms_idle':
int(self.group_descript[g].element_data['CurbIdleVMsPerEntry']
),
'total_max_glideins':
int(self.group_descript[g].element_data['MaxRunningTotal']),
'total_curb_glideins':
int(self.group_descript[g].element_data['CurbRunningTotal']),
'total_max_vms_idle':
int(self.group_descript[g].element_data['MaxIdleVMsTotal']),
'total_curb_vms_idle':
int(self.group_descript[g].element_data['CurbIdleVMsTotal']),
'fe_total_max_glideins':
int(self.group_descript[g].frontend_data['MaxRunningTotal']),
'fe_total_curb_glideins':
int(self.group_descript[g].frontend_data['CurbRunningTotal']),
'fe_total_max_vms_idle':
int(self.group_descript[g].frontend_data['MaxIdleVMsTotal']),
'fe_total_curb_vms_idle':
int(self.group_descript[g].frontend_data['CurbIdleVMsTotal']),
'global_total_max_glideins':
int(self.group_descript[g].
frontend_data['MaxRunningTotalGlobal']),
'global_total_curb_glideins':
int(self.group_descript[g].
frontend_data['CurbRunningTotalGlobal']),
'global_total_max_vms_idle':
int(self.group_descript[g].
frontend_data['MaxIdleVMsTotalGlobal']),
'global_total_curb_vms_idle':
int(self.group_descript[g].
frontend_data['CurbIdleVMsTotalGlobal']),
'max_matchmakers':
int(self.group_descript[g].element_data['MaxMatchmakers']),
'proxies':
self.group_descript[g].merged_data['Proxies'],
'proxy_selection_plugin':
self.group_descript[g].merged_data['ProxySelectionPlugin'],
'condor_config':
self.group_descript[g].frontend_data['CondorConfig'],
'condor_mapfile':
self.group_descript[g].element_data['MapFile'],
'classad_proxy':
self.group_descript[g].frontend_data['ClassAdProxy'],
'schedds':
self.group_descript[g].merged_data['JobSchedds'],
'sign_descript': {
'frontend_descript_fname':
self.group_sign_descript[g].frontend_descript_fname,
'group_descript_fname':
self.group_sign_descript[g].group_descript_fname,
'signature_type':
self.group_sign_descript[g].signature_type,
'frontend_descript_signature':
self.group_sign_descript[g].frontend_descript_signature,
'group_descript_signature':
self.group_sign_descript[g].group_descript_signature,
}
}
self.get_proxy_descript_data(g)
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 get_downtime_fd(work_dir):
frontendDescript = glideinFrontendConfig.FrontendDescript(work_dir)
fd = glideinFrontendDowntimeLib.DowntimeFile(
os.path.join(work_dir, frontendDescript.data['DowntimesFile']))
return fd
def __init__(self, frontend_workdir="/var/lib/gwms-frontend/vofrontend"):
self.frontend_workdir = frontend_workdir
glideinFrontendConfig.frontendConfig.frontend_descript_file = os.path.join(
self.frontend_workdir,
glideinFrontendConfig.frontendConfig.frontend_descript_file)
self.frontend_descript = glideinFrontendConfig.FrontendDescript(
self.frontend_workdir)
self.groups = self.frontend_descript.data["Groups"].split(",")
self.groups.sort()
self.group_descript = {}
self.group_params_descript = {}
self.group_sign_descript = {}
self.group_attrs_descript = {}
self.frontend_config = {"frontend": {}, "group": {}}
self.frontend_config["frontend"] = {
"frontend_name": self.frontend_descript.data["FrontendName"],
"workdir": self.frontend_workdir,
"web_url": self.frontend_descript.data["WebURL"],
"monitoring_web_url":
self.frontend_descript.data["MonitoringWebURL"],
}
for g in self.groups:
self.group_descript[
g] = glideinFrontendConfig.ElementMergedDescript(
self.frontend_workdir, g)
self.group_params_descript[
g] = glideinFrontendConfig.ParamsDescript(
self.frontend_workdir, g)
self.group_sign_descript[
g] = glideinFrontendConfig.GroupSignatureDescript(
self.frontend_workdir, g)
self.group_attrs_descript[g] = glideinFrontendConfig.AttrsDescript(
self.frontend_workdir, g)
self.frontend_config["group"][g] = {
"workdir":
glideinFrontendConfig.get_group_dir(self.frontend_workdir, g),
"web_url":
os.path.join(self.frontend_descript.data["WebURL"],
f"group_{g}"),
"security_name":
self.group_descript[g].merged_data["SecurityName"],
"factory_collectors":
self.group_descript[g].merged_data["FactoryCollectors"],
"min_running":
self.group_descript[g].element_data["MinRunningPerEntry"],
"max_running":
self.group_descript[g].element_data["MaxRunningPerEntry"],
"fraction_running":
float(self.group_descript[g].
element_data["FracRunningPerEntry"]),
"max_idle":
int(self.group_descript[g].element_data["MaxIdlePerEntry"]),
"reserve_idle":
int(self.group_descript[g].element_data["ReserveIdlePerEntry"]
),
#'idle_lifetime': int(self.group_descript[g].element_data['IdleLifetime']),
"max_vms_idle":
int(self.group_descript[g].element_data["MaxIdleVMsPerEntry"]),
"curb_vms_idle":
int(self.group_descript[g].element_data["CurbIdleVMsPerEntry"]
),
"total_max_glideins":
int(self.group_descript[g].element_data["MaxRunningTotal"]),
"total_curb_glideins":
int(self.group_descript[g].element_data["CurbRunningTotal"]),
"total_max_vms_idle":
int(self.group_descript[g].element_data["MaxIdleVMsTotal"]),
"total_curb_vms_idle":
int(self.group_descript[g].element_data["CurbIdleVMsTotal"]),
"fe_total_max_glideins":
int(self.group_descript[g].frontend_data["MaxRunningTotal"]),
"fe_total_curb_glideins":
int(self.group_descript[g].frontend_data["CurbRunningTotal"]),
"fe_total_max_vms_idle":
int(self.group_descript[g].frontend_data["MaxIdleVMsTotal"]),
"fe_total_curb_vms_idle":
int(self.group_descript[g].frontend_data["CurbIdleVMsTotal"]),
"global_total_max_glideins":
int(self.group_descript[g].
frontend_data["MaxRunningTotalGlobal"]),
"global_total_curb_glideins":
int(self.group_descript[g].
frontend_data["CurbRunningTotalGlobal"]),
"global_total_max_vms_idle":
int(self.group_descript[g].
frontend_data["MaxIdleVMsTotalGlobal"]),
"global_total_curb_vms_idle":
int(self.group_descript[g].
frontend_data["CurbIdleVMsTotalGlobal"]),
"max_matchmakers":
int(self.group_descript[g].element_data["MaxMatchmakers"]),
"proxies":
self.group_descript[g].merged_data["Proxies"],
"proxy_selection_plugin":
self.group_descript[g].merged_data["ProxySelectionPlugin"],
"condor_config":
self.group_descript[g].frontend_data["CondorConfig"],
"condor_mapfile":
self.group_descript[g].element_data["MapFile"],
"classad_proxy":
self.group_descript[g].frontend_data["ClassAdProxy"],
"schedds":
self.group_descript[g].merged_data["JobSchedds"],
"sign_descript": {
"frontend_descript_fname":
self.group_sign_descript[g].frontend_descript_fname,
"group_descript_fname":
self.group_sign_descript[g].group_descript_fname,
"signature_type":
self.group_sign_descript[g].signature_type,
"frontend_descript_signature":
self.group_sign_descript[g].frontend_descript_signature,
"group_descript_signature":
self.group_sign_descript[g].group_descript_signature,
},
}
self.get_proxy_descript_data(g)