def kill_workers(arguments):
status_in = 'ALL' if (len(arguments.status) == 1 and arguments.status[0]
== 'ALL') else arguments.status
computingSite_in = 'ALL' if (
len(arguments.sites) == 1
and arguments.sites[0] == 'ALL') else arguments.sites
computingElement_in = 'ALL' if (len(
arguments.ces) == 1 and arguments.ces[0] == 'ALL') else arguments.ces
submissionHost_in = 'ALL' if (len(arguments.submissionhosts) == 1
and arguments.submissionhosts[0]
== 'ALL') else arguments.submissionhosts
dbProxy = DBProxy()
retVal = dbProxy.kill_workers_by_query({
'status': status_in,
'computingSite': computingSite_in,
'computingElement': computingElement_in,
'submissionHost': submissionHost_in
})
if retVal is not None:
msg_temp = ('Sweeper will soon kill {n_workers} workers, with '
'status in {status_in}, '
'computingSite in {computingSite_in}, '
'computingElement in {computingElement_in}, '
'submissionHost in {submissionHost_in}')
print(
msg_temp.format(n_workers=retVal,
status_in=status_in,
computingSite_in=computingSite_in,
computingElement_in=computingElement_in,
submissionHost_in=submissionHost_in))
else:
mainLogger.critical('Failed to kill workers. See panda-db_proxy.log')
def qconf_purge(arguments):
queueName = arguments.queue
dbProxy = DBProxy()
retVal = dbProxy.purge_pq(queueName)
if retVal:
print('Purged {0} from harvester DB'.format(queueName))
else:
mainLogger.critical('Failed to purge {0} . See panda-db_proxy.log'.format(queueName))
class EventFeeder(AgentBase):
# constructor
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.queueConfigMapper = queue_config_mapper
self.communicator = communicator
self.pluginFactory = PluginFactory()
# main loop
def run(self):
lockedBy = 'eventfeeder-{0}'.format(self.ident)
while True:
mainLog = self.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
mainLog.debug('getting workers to feed events')
workSpecsPerQueue = self.dbProxy.get_workers_to_feed_events(harvester_config.eventfeeder.maxWorkers,
harvester_config.eventfeeder.lockInterval)
mainLog.debug('got {0} queues'.format(len(workSpecsPerQueue)))
# loop over all workers
for queueName, workSpecList in iteritems(workSpecsPerQueue):
tmpQueLog = self.make_logger(_logger, 'queue={0}'.format(queueName), method_name='run')
# check queue
if not self.queueConfigMapper.has_queue(queueName):
tmpQueLog.error('config not found')
continue
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
# get plugin
messenger = self.pluginFactory.get_plugin(queueConfig.messenger)
# loop over all workers
for workSpec in workSpecList:
tmpLog = core_utils.make_logger(_logger, 'workerID={0}'.format(workSpec.workerID),
method_name='run')
# get events
tmpLog.debug('get events')
tmpStat, events = self.communicator.get_event_ranges(workSpec.eventsRequestParams)
# failed
if tmpStat is False:
tmpLog.error('failed to get events with {0}'.format(events))
continue
tmpStat = messenger.feed_events(workSpec, events)
# failed
if tmpStat is False:
tmpLog.error('failed to feed events')
continue
# update worker
workSpec.eventsRequest = WorkSpec.EV_useEvents
workSpec.eventsRequestParams = None
workSpec.eventFeedTime = None
# update local database
tmpStat = self.dbProxy.update_worker(workSpec)
tmpLog.debug('done with {0}'.format(tmpStat))
tmpQueLog.debug('done')
mainLog.debug('done')
# check if being terminated
if self.terminated(harvester_config.eventfeeder.sleepTime):
mainLog.debug('terminated')
return
def __init__(self, single_mode=False, stop_event=None, daemon_mode=True):
# initialize database and config
self.singleMode = single_mode
self.stopEvent = stop_event
self.daemonMode = daemon_mode
from pandaharvester.harvestercore.communicator_pool import CommunicatorPool
self.communicatorPool = CommunicatorPool()
from pandaharvester.harvestercore.queue_config_mapper import QueueConfigMapper
self.queueConfigMapper = QueueConfigMapper()
from pandaharvester.harvestercore.db_proxy_pool import DBProxyPool as DBProxy
dbProxy = DBProxy()
dbProxy.make_tables(self.queueConfigMapper)
def __init__(self, single_mode=False, stop_event=None, daemon_mode=True):
# initialize database and config
self.singleMode = single_mode
self.stopEvent = stop_event
self.daemonMode = daemon_mode
from pandaharvester.harvestercore.communicator_pool import CommunicatorPool
self.communicatorPool = CommunicatorPool()
from pandaharvester.harvestercore.queue_config_mapper import QueueConfigMapper
self.queueConfigMapper = QueueConfigMapper()
from pandaharvester.harvestercore.db_proxy_pool import DBProxyPool as DBProxy
dbProxy = DBProxy()
dbProxy.make_tables(self.queueConfigMapper)
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queueConfigMapper = queue_config_mapper
self.dbProxy = DBProxy()
self.workerMaker = WorkerMaker()
self.workerAdjuster = WorkerAdjuster(queue_config_mapper)
self.pluginFactory = PluginFactory()
def HarvesterReport(self):
try:
from distutils.sysconfig import get_python_lib # pylint: disable=import-error
sys.path.append(get_python_lib()+'/pandacommon')
os.environ['PANDA_HOME']=os.environ['VIRTUAL_ENV']
from collections import defaultdict # pylint: disable=import-error
from pandaharvester.harvestercore.db_proxy_pool import DBProxyPool as DBProxy # pylint: disable=import-error
self.dbProxy = DBProxy()
workers = self.dbProxy.get_worker_stats_bulk(None)
rep = defaultdict(dict)
rtot = defaultdict(int)
for site, prodsourcelabels in workers.items():
for prodsourcelabel, resources in prodsourcelabels.items():
for resource, jobs in resources.items():
rep[f'{site}-{resource}'][prodsourcelabel or 'empty'] = jobs
for state, count in jobs.items():
rtot[state] += count
self.log(f"All Harvester jobs: {sum(rtot.values())} prodSourceLabel: submitted/running")
for k in sorted(rep.keys()):
log=f"{k:>28.28}:"
for psl, jobs in rep[k].items():
log += f"{psl:>10}: {jobs['submitted']}/{jobs['running']}"
self.log(log)
log = f"{'Totals':>28}: submitted: {rtot['submitted']} running: {rtot['running']}"
self.log(log+'\n\n')
except:
pass
def __init__(self, single_mode=False):
AgentBase.__init__(self, single_mode)
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
# get module and class names
moduleNames = self.get_list(harvester_config.credmanager.moduleName)
classNames = self.get_list(harvester_config.credmanager.className)
# file names of original certificates
if hasattr(harvester_config.credmanager, 'inCertFile'):
inCertFiles = self.get_list(
harvester_config.credmanager.inCertFile)
else:
inCertFiles = self.get_list(harvester_config.credmanager.certFile)
# file names of certificates to be generated
if hasattr(harvester_config.credmanager, 'outCertFile'):
outCertFiles = self.get_list(
harvester_config.credmanager.outCertFile)
else:
# use the file name of the certificate for panda connection as output name
outCertFiles = self.get_list(harvester_config.pandacon.cert_file)
# VOMS
vomses = self.get_list(harvester_config.credmanager.voms)
# get plugin
self.exeCores = []
for moduleName, className, inCertFile, outCertFile, voms in \
zip(moduleNames, classNames, inCertFiles, outCertFiles, vomses):
pluginPar = {}
pluginPar['module'] = moduleName
pluginPar['name'] = className
pluginPar['inCertFile'] = inCertFile
pluginPar['outCertFile'] = outCertFile
pluginPar['voms'] = voms
exeCore = self.pluginFactory.get_plugin(pluginPar)
self.exeCores.append(exeCore)
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.communicator = communicator
self.queueConfigMapper = queue_config_mapper
self._last_stats_update = None
self._last_metrics_update = None
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.db_proxy = DBProxy()
self.communicator = communicator
self.queueConfigMapper = queue_config_mapper
self.nodeName = socket.gethostname()
self.lastHeartbeat = None
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.communicator = communicator
self.nodeName = socket.gethostname()
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.db_proxy = DBProxy()
self.communicator = communicator
self.queueConfigMapper = queue_config_mapper
self.nodeName = socket.gethostname()
self.lastHeartbeat = None
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queueConfigMapper = queue_config_mapper
self.dbProxy = DBProxy()
self.pluginFactory = PluginFactory()
self.startTimestamp = time.time()
self.monitor_fifo = MonitorFIFO()
self.apfmon = Apfmon(self.queueConfigMapper)
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queueConfigMapper = queue_config_mapper
self.dbProxy = DBProxy()
self.workerMaker = WorkerMaker()
self.workerAdjuster = WorkerAdjuster(queue_config_mapper)
self.pluginFactory = PluginFactory()
self.monitor_fifo = MonitorFIFO()
self.apfmon = Apfmon(self.queueConfigMapper)
def __init__(self, queue_config_mapper):
self.queue_configMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
self.throttlerMap = dict()
self.apf_mon = Apfmon(self.queue_configMapper)
try:
self.maxNewWorkers = harvester_config.submitter.maxNewWorkers
except AttributeError:
self.maxNewWorkers = None
def __init__(self, queue_config_mapper):
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
self.throttlerMap = dict()
self.apf_mon = Apfmon(self.queueConfigMapper)
try:
self.maxNewWorkers = harvester_config.submitter.maxNewWorkers
except AttributeError:
self.maxNewWorkers = None
def __init__(self, **kwarg):
'''Set up DB connection and credentials'''
PluginBase.__init__(self, **kwarg)
self.dbproxy = DBProxy()
self.schedulerid = harvester_config.master.harvester_id
# Credential dictionary role: proxy file
self.certs = dict(zip([r.split('=')[1] for r in list(harvester_config.credmanager.voms)],
list(harvester_config.credmanager.outCertFile)))
self.cred_type = arc.initializeCredentialsType(arc.initializeCredentialsType.SkipCredentials)
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queue_config_mapper = queue_config_mapper
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
# plugin cores
self.exeCores = []
self.queue_exe_cores = []
# get plugin from harvester config
self.get_cores_from_harvester_config()
# update plugin cores from queue config
self.update_cores_from_queue_config()
def __init__(self, **kwarg):
'''Set up DB connection and credentials'''
PluginBase.__init__(self, **kwarg)
self.dbproxy = DBProxy()
self.schedulerid = harvester_config.master.harvester_id
# Credential dictionary role: proxy file
self.certs = dict(
zip([
r.split('=')[1]
for r in list(harvester_config.credmanager.voms)
], list(harvester_config.credmanager.outCertFile)))
self.cred_type = arc.initializeCredentialsType(
arc.initializeCredentialsType.SkipCredentials)
def __init__(self, pid_file, single_mode=False):
AgentBase.__init__(self, single_mode)
self.db_proxy = DBProxy()
if pid_file is not None:
self.pid_file = pid_file
else:
try:
self.pid_file = harvester_config.service_monitor.pidfile
except Exception:
self.pid_file = None
self.pid = self.get_master_pid()
self.master_process = psutil.Process(self.pid)
self.children = self.master_process.children(recursive=True)
self.cpu_count = multiprocessing.cpu_count()
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queueConfigMapper = queue_config_mapper
self.dbProxy = DBProxy()
self.pluginFactory = PluginFactory()
self.startTimestamp = time.time()
self.monitor_fifo = MonitorFIFO()
if self.monitor_fifo.enabled:
self.monitor_event_fifo = MonitorEventFIFO()
else:
self.monitor_event_fifo = None
self.apfmon = Apfmon(self.queueConfigMapper)
self.eventBasedMonCoreList = []
if getattr(harvester_config.monitor, 'eventBasedEnable', False):
for pluginConf in harvester_config.monitor.eventBasedPlugins:
pluginFactory = PluginFactory()
self.eventBasedMonCoreList.append(pluginFactory.get_plugin(pluginConf))
def __init__(self, pid_file, single_mode=False):
AgentBase.__init__(self, single_mode)
self.db_proxy = DBProxy()
if pid_file is not None:
self.pid_file = pid_file
else:
try:
self.pid_file = harvester_config.service_monitor.pidfile
except Exception:
self.pid_file = None
self.pid = self.get_master_pid()
self.master_process = psutil.Process(self.pid)
self.children = self.master_process.children(recursive=True)
self.cpu_count = multiprocessing.cpu_count()
class Submitter(AgentBase):
# constructor
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queueConfigMapper = queue_config_mapper
self.dbProxy = DBProxy()
self.workerMaker = WorkerMaker()
self.workerAdjuster = WorkerAdjuster(queue_config_mapper)
self.pluginFactory = PluginFactory()
# main loop
def run(self):
lockedBy = 'submitter-{0}'.format(self.ident)
while True:
mainLog = core_utils.make_logger(_logger,
'id={0}'.format(lockedBy),
method_name='run')
mainLog.debug('getting queues to submit workers')
# get queues associated to a site to submit workers
curWorkers, siteName, resMap = self.dbProxy.get_queues_to_submit(
harvester_config.submitter.nQueues,
harvester_config.submitter.lookupTime)
mainLog.debug('got {0} queues for site {1}'.format(
len(curWorkers), siteName))
# get commands
if siteName is not None:
comStr = '{0}:{1}'.format(CommandSpec.COM_setNWorkers,
siteName)
commandSpecs = self.dbProxy.get_commands_for_receiver(
'submitter', comStr)
mainLog.debug('got {0} commands'.format(len(commandSpecs)))
for commandSpec in commandSpecs:
newLimits = self.dbProxy.set_queue_limit(
siteName, commandSpec.params)
for tmpResource, tmpNewVal in iteritems(newLimits):
if tmpResource in resMap:
tmpQueueName = resMap[tmpResource]
if tmpQueueName in curWorkers:
curWorkers[tmpQueueName][
'nNewWorkers'] = tmpNewVal
# define number of new workers
if len(curWorkers) == 0:
nWorkersPerQueue = dict()
else:
nWorkersPerQueue = self.workerAdjuster.define_num_workers(
curWorkers, siteName)
if nWorkersPerQueue is None:
mainLog.error(
'WorkerAdjuster failed to define the number of workers')
elif len(nWorkersPerQueue) == 0:
pass
else:
# loop over all queues
for queueName, tmpVal in iteritems(nWorkersPerQueue):
tmpLog = core_utils.make_logger(
_logger,
'queue={0}'.format(queueName),
method_name='run')
tmpLog.debug('start')
nWorkers = tmpVal['nNewWorkers'] + tmpVal['nReady']
nReady = tmpVal['nReady']
# check queue
if not self.queueConfigMapper.has_queue(queueName):
tmpLog.error('config not found')
continue
# no new workers
if nWorkers == 0:
tmpLog.debug(
'skipped since no new worker is needed based on current stats'
)
continue
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
# actions based on mapping type
if queueConfig.mapType == WorkSpec.MT_NoJob:
# workers without jobs
jobChunks = []
for i in range(nWorkers):
jobChunks.append([])
elif queueConfig.mapType == WorkSpec.MT_OneToOne:
# one worker per one job
jobChunks = self.dbProxy.get_job_chunks_for_workers(
queueName, nWorkers, nReady, 1, None,
queueConfig.useJobLateBinding,
harvester_config.submitter.checkInterval,
harvester_config.submitter.lockInterval, lockedBy)
elif queueConfig.mapType == WorkSpec.MT_MultiJobs:
# one worker for multiple jobs
nJobsPerWorker = self.workerMaker.get_num_jobs_per_worker(
queueConfig, nWorkers)
tmpLog.debug(
'nJobsPerWorker={0}'.format(nJobsPerWorker))
jobChunks = self.dbProxy.get_job_chunks_for_workers(
queueName, nWorkers, nReady, nJobsPerWorker, None,
queueConfig.useJobLateBinding,
harvester_config.submitter.checkInterval,
harvester_config.submitter.lockInterval, lockedBy,
queueConfig.allowJobMixture)
elif queueConfig.mapType == WorkSpec.MT_MultiWorkers:
# multiple workers for one job
nWorkersPerJob = self.workerMaker.get_num_workers_per_job(
queueConfig, nWorkers)
jobChunks = self.dbProxy.get_job_chunks_for_workers(
queueName, nWorkers, nReady, None, nWorkersPerJob,
queueConfig.useJobLateBinding,
harvester_config.submitter.checkInterval,
harvester_config.submitter.lockInterval, lockedBy)
else:
tmpLog.error('unknown mapType={0}'.format(
queueConfig.mapType))
continue
tmpLog.debug('got {0} job chunks'.format(len(jobChunks)))
if len(jobChunks) == 0:
continue
# make workers
okChunks, ngChunks = self.workerMaker.make_workers(
jobChunks, queueConfig, nReady)
if len(ngChunks) == 0:
tmpLog.debug('successfully made {0} workers'.format(
len(okChunks)))
else:
tmpLog.debug(
'made {0} workers, while {1} workers failed'.
format(len(okChunks), len(ngChunks)))
timeNow = datetime.datetime.utcnow()
# NG
for ngJobs in ngChunks:
for jobSpec in ngJobs:
jobSpec.status = 'failed'
jobSpec.subStatus = 'failedtomake'
jobSpec.stateChangeTime = timeNow
jobSpec.lockedBy = None
jobSpec.trigger_propagation()
self.dbProxy.update_job(jobSpec, {
'lockedBy': lockedBy,
'subStatus': 'prepared'
})
# OK
pandaIDs = set()
workSpecList = []
if len(okChunks) > 0:
for workSpec, okJobs in okChunks:
# has job
if (queueConfig.useJobLateBinding and workSpec.workerID is None) \
or queueConfig.mapType == WorkSpec.MT_NoJob:
workSpec.hasJob = 0
else:
workSpec.hasJob = 1
if workSpec.nJobsToReFill in [None, 0]:
workSpec.set_jobspec_list(okJobs)
else:
# refill free slots during the worker is running
workSpec.set_jobspec_list(
okJobs[:workSpec.nJobsToReFill])
workSpec.nJobsToReFill = None
for jobSpec in okJobs[workSpec.
nJobsToReFill:]:
pandaIDs.add(jobSpec.PandaID)
# map type
workSpec.mapType = queueConfig.mapType
# queue name
workSpec.computingSite = queueConfig.queueName
# set access point
workSpec.accessPoint = queueConfig.messenger[
'accessPoint']
# events
if len(okJobs) > 0 and (
'eventService' in okJobs[0].jobParams
or 'cloneJob' in okJobs[0].jobParams):
workSpec.eventsRequest = WorkSpec.EV_useEvents
workSpecList.append(workSpec)
if len(workSpecList) > 0:
# get plugin for submitter
submitterCore = self.pluginFactory.get_plugin(
queueConfig.submitter)
if submitterCore is None:
# not found
tmpLog.error(
'submitter plugin for {0} not found'.format(
jobSpec.computingSite))
continue
# get plugin for messenger
messenger = self.pluginFactory.get_plugin(
queueConfig.messenger)
if messenger is None:
# not found
tmpLog.error(
'messenger plugin for {0} not found'.format(
jobSpec.computingSite))
continue
# setup access points
messenger.setup_access_points(workSpecList)
# feed jobs
for workSpec in workSpecList:
if workSpec.hasJob == 1:
tmpStat = messenger.feed_jobs(
workSpec, workSpec.get_jobspec_list())
if tmpStat is False:
tmpLog.error(
'failed to send jobs to workerID={0}'.
format(workSpec.workerID))
else:
tmpLog.debug(
'sent jobs to workerID={0} with {1}'.
format(workSpec.workerID, tmpStat))
# submit
tmpLog.debug('submitting {0} workers'.format(
len(workSpecList)))
workSpecList, tmpRetList, tmpStrList = self.submit_workers(
submitterCore, workSpecList)
for iWorker, (tmpRet, tmpStr) in enumerate(
zip(tmpRetList, tmpStrList)):
workSpec, jobList = okChunks[iWorker]
# use associated job list since it can be truncated for re-filling
jobList = workSpec.get_jobspec_list()
# set status
if not tmpRet:
# failed submission
tmpLog.error(
'failed to submit a workerID={0} with {1}'.
format(workSpec.workerID, tmpStr))
workSpec.set_status(WorkSpec.ST_missed)
jobList = []
elif queueConfig.useJobLateBinding and workSpec.hasJob == 1:
# directly go to running after feeding jobs for late biding
workSpec.set_status(WorkSpec.ST_running)
else:
# normal successful submission
workSpec.set_status(WorkSpec.ST_submitted)
workSpec.submitTime = timeNow
workSpec.modificationTime = timeNow
# prefetch events
if tmpRet and workSpec.hasJob == 1 and workSpec.eventsRequest == WorkSpec.EV_useEvents:
workSpec.eventsRequest = WorkSpec.EV_requestEvents
eventsRequestParams = dict()
for jobSpec in jobList:
eventsRequestParams[jobSpec.PandaID] = {
'pandaID': jobSpec.PandaID,
'taskID': jobSpec.taskID,
'jobsetID':
jobSpec.jobParams['jobsetID'],
'nRanges':
jobSpec.jobParams['coreCount'],
}
workSpec.eventsRequestParams = eventsRequestParams
# register worker
tmpStat = self.dbProxy.register_worker(
workSpec, jobList, lockedBy)
if jobList is not None:
for jobSpec in jobList:
pandaIDs.add(jobSpec.PandaID)
if tmpStat:
tmpStr = 'submitted a workerID={0} for PandaID={1} with batchID={2}'
tmpLog.info(
tmpStr.format(
workSpec.workerID,
jobSpec.PandaID,
workSpec.batchID))
else:
tmpStr = 'failed to register a worker for PandaID={0} with batchID={1}'
tmpLog.error(
tmpStr.format(
jobSpec.PandaID,
workSpec.batchID))
# release jobs
self.dbProxy.release_jobs(pandaIDs, lockedBy)
mainLog.debug('done')
# check if being terminated
if self.terminated(harvester_config.submitter.sleepTime):
mainLog.debug('terminated')
return
# wrapper for submitWorkers to skip ready workers
def submit_workers(self, submitter_core, workspec_list):
retList = []
strList = []
newSpecList = []
workersToSubmit = []
for workSpec in workspec_list:
if workSpec.status == WorkSpec.ST_ready:
newSpecList.append(workSpec)
retList.append(True)
strList.append('')
else:
workersToSubmit.append(workSpec)
tmpRetList = submitter_core.submit_workers(workersToSubmit)
for tmpRet, tmpStr in tmpRetList:
retList.append(tmpRet)
strList.append(tmpStr)
newSpecList += workersToSubmit
return newSpecList, retList, strList
if loggerName.split('.')[-1] in ['db_proxy']:
continue
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
msgStr = "plugin={0}".format(stagerCore.__class__.__name__)
tmpLog.debug(msgStr)
msgStr = "Initial queueConfig.stager = {}".format(initial_queueConfig_stager)
tmpLog.debug(msgStr)
msgStr = "Modified queueConfig.stager = {}".format(modified_queueConfig_stager)
tmpLog.debug(msgStr)
scope = 'panda'
proxy = DBProxy()
communicator = CommunicatorPool()
cacher = Cacher(communicator, single_mode=True)
cacher.run()
# check if db lock exits
locked = stagerCore.dbInterface.get_object_lock('dummy_id_for_out_0',lock_interval=120)
if not locked:
tmpLog.debug('DB Already locked by another thread')
# now unlock db
unlocked = stagerCore.dbInterface.release_object_lock('dummy_id_for_out_0')
if unlocked :
tmpLog.debug('unlocked db')
else:
class WorkerAdjuster(object):
# constructor
def __init__(self, queue_config_mapper):
self.queue_configMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
self.throttlerMap = dict()
self.apf_mon = Apfmon(self.queue_configMapper)
try:
self.maxNewWorkers = harvester_config.submitter.maxNewWorkers
except AttributeError:
self.maxNewWorkers = None
# define number of workers to submit based on various information
def define_num_workers(self, static_num_workers, site_name):
tmp_log = core_utils.make_logger(_logger,
'site={0}'.format(site_name),
method_name='define_num_workers')
tmp_log.debug('start')
tmp_log.debug('static_num_workers: {0}'.format(static_num_workers))
dyn_num_workers = copy.deepcopy(static_num_workers)
try:
# get queue status
queue_stat = self.dbProxy.get_cache("panda_queues.json", None)
if queue_stat is None:
queue_stat = dict()
else:
queue_stat = queue_stat.data
# get job statistics
job_stats = self.dbProxy.get_cache("job_statistics.json", None)
if job_stats is None:
job_stats = dict()
else:
job_stats = job_stats.data
# define num of new workers
for queue_name in static_num_workers:
# get queue
queue_config = self.queue_configMapper.get_queue(queue_name)
worker_limits_dict = self.dbProxy.get_worker_limits(queue_name)
max_workers = worker_limits_dict.get('maxWorkers', 0)
n_queue_limit = worker_limits_dict.get('nQueueLimitWorker', 0)
n_queue_limit_per_rt = worker_limits_dict[
'nQueueLimitWorkerPerRT']
n_queue_total, n_ready_total, n_running_total = 0, 0, 0
apf_msg = None
apf_data = None
for job_type, jt_values in iteritems(
static_num_workers[queue_name]):
for resource_type, tmp_val in iteritems(jt_values):
tmp_log.debug(
'Processing queue {0} job_type {1} resource_type {2} with static_num_workers {3}'
.format(queue_name, job_type, resource_type,
tmp_val))
# set 0 to num of new workers when the queue is disabled
if queue_name in queue_stat and queue_stat[queue_name][
'status'] in [
'offline', 'standby', 'maintenance'
]:
dyn_num_workers[queue_name][job_type][
resource_type]['nNewWorkers'] = 0
ret_msg = 'set n_new_workers=0 since status={0}'.format(
queue_stat[queue_name]['status'])
tmp_log.debug(ret_msg)
apf_msg = 'Not submitting workers since queue status = {0}'.format(
queue_stat[queue_name]['status'])
continue
# protection against not-up-to-date queue config
if queue_config is None:
dyn_num_workers[queue_name][job_type][
resource_type]['nNewWorkers'] = 0
ret_msg = 'set n_new_workers=0 due to missing queue_config'
tmp_log.debug(ret_msg)
apf_msg = 'Not submitting workers because of missing queue_config'
continue
# get throttler
if queue_name not in self.throttlerMap:
if hasattr(queue_config, 'throttler'):
throttler = self.pluginFactory.get_plugin(
queue_config.throttler)
else:
throttler = None
self.throttlerMap[queue_name] = throttler
# check throttler
throttler = self.throttlerMap[queue_name]
if throttler is not None:
to_throttle, tmp_msg = throttler.to_be_throttled(
queue_config)
if to_throttle:
dyn_num_workers[queue_name][job_type][
resource_type]['nNewWorkers'] = 0
ret_msg = 'set n_new_workers=0 by {0}:{1}'.format(
throttler.__class__.__name__, tmp_msg)
tmp_log.debug(ret_msg)
continue
# check stats
n_queue = tmp_val['nQueue']
n_ready = tmp_val['nReady']
n_running = tmp_val['nRunning']
if resource_type != 'ANY' and job_type != 'ANY' and job_type is not None:
n_queue_total += n_queue
n_ready_total += n_ready
n_running_total += n_running
if queue_config.runMode == 'slave':
n_new_workers_def = tmp_val['nNewWorkers']
if n_new_workers_def == 0:
dyn_num_workers[queue_name][job_type][
resource_type]['nNewWorkers'] = 0
ret_msg = 'set n_new_workers=0 by panda in slave mode'
tmp_log.debug(ret_msg)
continue
else:
n_new_workers_def = None
# define num of new workers based on static site config
n_new_workers = 0
if n_queue >= n_queue_limit_per_rt > 0:
# enough queued workers
ret_msg = 'No n_new_workers since n_queue({0})>=n_queue_limit_per_rt({1})'.format(
n_queue, n_queue_limit_per_rt)
tmp_log.debug(ret_msg)
pass
elif (n_queue + n_ready +
n_running) >= max_workers > 0:
# enough workers in the system
ret_msg = 'No n_new_workers since n_queue({0}) + n_ready({1}) + n_running({2}) '.format(
n_queue, n_ready, n_running)
ret_msg += '>= max_workers({0})'.format(
max_workers)
tmp_log.debug(ret_msg)
pass
else:
max_queued_workers = None
if n_queue_limit_per_rt > 0: # there is a limit set for the queue
max_queued_workers = n_queue_limit_per_rt
# Reset the maxQueueWorkers according to particular
if n_new_workers_def is not None: # don't surpass limits given centrally
maxQueuedWorkers_slave = n_new_workers_def + n_queue
if max_queued_workers is not None:
max_queued_workers = min(
maxQueuedWorkers_slave,
max_queued_workers)
else:
max_queued_workers = maxQueuedWorkers_slave
elif queue_config.mapType == 'NoJob': # for pull mode, limit to activated jobs
# limit the queue to the number of activated jobs to avoid empty pilots
try:
n_activated = max(
job_stats[queue_name]['activated'],
1) # avoid no activity queues
queue_limit = max_queued_workers
max_queued_workers = min(
n_activated, max_queued_workers)
tmp_log.debug(
'limiting max_queued_workers to min(n_activated={0}, queue_limit={1})'
.format(n_activated, queue_limit))
except KeyError:
tmp_log.warning(
'n_activated not defined, defaulting to configured queue limits'
)
pass
if max_queued_workers is None: # no value found, use default value
max_queued_workers = 1
# new workers
n_new_workers = max(max_queued_workers - n_queue,
0)
tmp_log.debug(
'setting n_new_workers to {0} in max_queued_workers calculation'
.format(n_new_workers))
if max_workers > 0:
n_new_workers = min(
n_new_workers,
max(
max_workers - n_queue - n_ready -
n_running, 0))
tmp_log.debug(
'setting n_new_workers to {0} to respect max_workers'
.format(n_new_workers))
if queue_config.maxNewWorkersPerCycle > 0:
n_new_workers = min(
n_new_workers,
queue_config.maxNewWorkersPerCycle)
tmp_log.debug(
'setting n_new_workers to {0} in order to respect maxNewWorkersPerCycle'
.format(n_new_workers))
if self.maxNewWorkers is not None and self.maxNewWorkers > 0:
n_new_workers = min(n_new_workers,
self.maxNewWorkers)
tmp_log.debug(
'setting n_new_workers to {0} in order to respect universal maxNewWorkers'
.format(n_new_workers))
dyn_num_workers[queue_name][job_type][resource_type][
'nNewWorkers'] = n_new_workers
# adjust n_new_workers for UCORE to let aggregations over RT respect nQueueLimitWorker and max_workers
if queue_config is None:
max_new_workers_per_cycle = 0
ret_msg = 'set max_new_workers_per_cycle=0 in UCORE aggregation due to missing queue_config'
tmp_log.debug(ret_msg)
else:
max_new_workers_per_cycle = queue_config.maxNewWorkersPerCycle
if len(dyn_num_workers[queue_name]) > 1:
total_new_workers_rts = 0
for _jt in dyn_num_workers[queue_name]:
for _rt in dyn_num_workers[queue_name][_jt]:
if _jt != 'ANY' and _rt != 'ANY':
total_new_workers_rts = total_new_workers_rts + dyn_num_workers[
queue_name][_jt][_rt]['nNewWorkers']
n_new_workers_max_agg = min(
max(n_queue_limit - n_queue_total, 0),
max(
max_workers - n_queue_total - n_ready_total -
n_running_total, 0))
if max_new_workers_per_cycle >= 0:
n_new_workers_max_agg = min(n_new_workers_max_agg,
max_new_workers_per_cycle)
if self.maxNewWorkers is not None and self.maxNewWorkers > 0:
n_new_workers_max_agg = min(n_new_workers_max_agg,
self.maxNewWorkers)
# exceeded max, to adjust
if total_new_workers_rts > n_new_workers_max_agg:
if n_new_workers_max_agg == 0:
for job_type in dyn_num_workers[queue_name]:
for resource_type in dyn_num_workers[
queue_name][job_type]:
dyn_num_workers[queue_name][job_type][
resource_type]['nNewWorkers'] = 0
tmp_log.debug(
'No n_new_workers since n_new_workers_max_agg=0 for UCORE'
)
else:
tmp_log.debug(
'n_new_workers_max_agg={0} for UCORE'.format(
n_new_workers_max_agg))
_d = dyn_num_workers[queue_name].copy()
del _d['ANY']
# TODO: needs to be recalculated
simple_rt_nw_list = []
for job_type in _d: # jt: job type
for resource_type in _d[
job_type]: # rt: resource type
simple_rt_nw_list.append([
(resource_type, job_type),
_d[job_type][resource_type].get(
'nNewWorkers', 0), 0
])
_countdown = n_new_workers_max_agg
for _rt_list in simple_rt_nw_list:
(resource_type,
job_type), n_new_workers_orig, _r = _rt_list
n_new_workers, remainder = divmod(
n_new_workers_orig * n_new_workers_max_agg,
total_new_workers_rts)
dyn_num_workers[queue_name][
job_type].setdefault(
resource_type, {
'nReady': 0,
'nRunning': 0,
'nQueue': 0,
'nNewWorkers': 0
})
dyn_num_workers[queue_name][job_type][
resource_type][
'nNewWorkers'] = n_new_workers
_rt_list[2] = remainder
_countdown -= n_new_workers
_s_list = sorted(simple_rt_nw_list,
key=(lambda x: x[1]))
sorted_rt_nw_list = sorted(_s_list,
key=(lambda x: x[2]),
reverse=True)
for (
resource_type, job_type
), n_new_workers_orig, remainder in sorted_rt_nw_list:
if _countdown <= 0:
break
dyn_num_workers[queue_name][job_type][
resource_type]['nNewWorkers'] += 1
_countdown -= 1
for job_type in dyn_num_workers[queue_name]:
for resource_type in dyn_num_workers[queue_name][
job_type]:
if job_type == 'ANY' or resource_type == 'ANY':
continue
n_new_workers = dyn_num_workers[queue_name][
job_type][resource_type]['nNewWorkers']
tmp_log.debug(
'setting n_new_workers to {0} of job_type {1} resource_type {2} in order to respect RT aggregations for UCORE'
.format(n_new_workers, job_type,
resource_type))
if not apf_msg:
apf_data = copy.deepcopy(dyn_num_workers[queue_name])
self.apf_mon.update_label(queue_name, apf_msg, apf_data)
# dump
tmp_log.debug('defined {0}'.format(str(dyn_num_workers)))
return dyn_num_workers
except Exception:
# dump error
err_msg = core_utils.dump_error_message(tmp_log)
return None
class ServiceMonitor(AgentBase):
# constructor
def __init__(self, pid_file, single_mode=False):
AgentBase.__init__(self, single_mode)
self.db_proxy = DBProxy()
if pid_file is not None:
self.pid_file = pid_file
else:
try:
self.pid_file = harvester_config.service_monitor.pidfile
except Exception:
self.pid_file = None
self.pid = self.get_master_pid()
self.master_process = psutil.Process(self.pid)
self.children = self.master_process.children(recursive=True)
self.cpu_count = multiprocessing.cpu_count()
def get_master_pid(self):
"""
Gets the master pid from the lock file
:return:
"""
try:
fh = open(self.pid_file, 'r')
pid = int(fh.readline())
fh.close()
except Exception:
_logger.error('Could not read pidfile "{0}"'.format(self.pid_file))
pid = None
return pid
def refresh_children_list(self, children):
children_refreshed = []
for child_current in children:
pid_current = child_current.pid
found = False
for child_stored in self.children:
pid_stored = child_stored.pid
if pid_stored == pid_current:
found = True
break
if found:
children_refreshed.append(child_stored)
else:
children_refreshed.append(child_current)
self.children = children_refreshed
return children_refreshed
def get_memory_n_cpu(self):
"""
sum memory of whole process tree starting from master pid
:return: rss in MiB
"""
try:
master_process = self.master_process
rss = master_process.memory_info()[0]
memory_pc = master_process.memory_percent()
cpu_pc = master_process.cpu_percent()
children = self.refresh_children_list(master_process.children(recursive=True))
for child in children:
rss += child.memory_info()[0]
memory_pc += child.memory_percent()
cpu_pc += child.cpu_percent()
# convert bytes to MiB
rss_mib = rss / float(2 ** 20)
# normalize cpu percentage by cpu count
cpu_pc = cpu_pc * 1.0 / self.cpu_count
except Exception:
_logger.error('Excepted with: {0}'.format(traceback.format_exc()))
rss_mib = None
memory_pc = None
cpu_pc = None
return rss_mib, memory_pc, cpu_pc
def volume_use(self, volume_name):
command = "df -Pkh /" + volume_name
used_amount = 0
tmp_array = command.split()
output = subprocess.Popen(tmp_array, stdout=subprocess.PIPE).communicate()[0].decode("utf-8")
for line in output.split('\n'):
if re.search(volume_name, line):
used_amount = re.search(r"(\d+)\%", line).group(1)
try:
used_amount_float = float(used_amount)
except ValueError:
used_amount_float = None
_logger.error('Could not convert used amount {0} to float for volume {1}'.format(used_amount, volume_name))
return used_amount_float
# main loop
def run(self):
while True:
_logger.debug('Running service monitor')
service_metrics = {}
# get memory usage
rss_mib, memory_pc, cpu_pc = self.get_memory_n_cpu()
service_metrics['rss_mib'] = rss_mib
service_metrics['memory_pc'] = memory_pc
service_metrics['cpu_pc'] = cpu_pc
_logger.debug('Memory usage: {0} MiB/{1}%, CPU usage: {2}'.format(rss_mib, memory_pc, cpu_pc))
# get volume usage
try:
volumes = harvester_config.service_monitor.disk_volumes.split(',')
except Exception:
volumes = []
for volume in volumes:
volume_use = self.volume_use(volume)
_logger.debug('Disk usage of {0}: {1} %'.format(volume, volume_use))
service_metrics['volume_{0}_pc'.format(volume)] = volume_use
service_metrics_spec = ServiceMetricSpec(service_metrics)
self.db_proxy.insert_service_metrics(service_metrics_spec)
# check if being terminated
try:
sleep_time = harvester_config.service_monitor.sleepTime
except Exception:
sleep_time = 300
if self.terminated(sleep_time, randomize=False):
return
class Submitter(AgentBase):
# constructor
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queueConfigMapper = queue_config_mapper
self.dbProxy = DBProxy()
self.workerMaker = WorkerMaker()
self.workerAdjuster = WorkerAdjuster(queue_config_mapper)
self.pluginFactory = PluginFactory()
self.monitor_fifo = MonitorFIFO()
self.apfmon = Apfmon(self.queueConfigMapper)
# main loop
def run(self):
lockedBy = 'submitter-{0}'.format(self.get_pid())
monitor_fifo = self.monitor_fifo
queueLockInterval = getattr(harvester_config.submitter, 'queueLockInterval',
harvester_config.submitter.lockInterval)
while True:
sw_main = core_utils.get_stopwatch()
mainLog = self.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
mainLog.debug('getting queues to submit workers')
# get queues associated to a site to submit workers
curWorkers, siteName, resMap = self.dbProxy.get_queues_to_submit(harvester_config.submitter.nQueues,
harvester_config.submitter.lookupTime,
harvester_config.submitter.lockInterval,
lockedBy, queueLockInterval)
submitted = False
if siteName is not None:
mainLog.debug('got {0} queues for site {1}'.format(len(curWorkers), siteName))
# get commands
comStr = '{0}:{1}'.format(CommandSpec.COM_setNWorkers, siteName)
commandSpecs = self.dbProxy.get_commands_for_receiver('submitter', comStr)
mainLog.debug('got {0} {1} commands'.format(commandSpecs, comStr))
for commandSpec in commandSpecs:
newLimits = self.dbProxy.set_queue_limit(siteName, commandSpec.params)
for tmpResource, tmpNewVal in iteritems(newLimits):
# if available, overwrite new worker value with the command from panda server
if tmpResource in resMap:
tmpQueueName = resMap[tmpResource]
if tmpQueueName in curWorkers:
curWorkers[tmpQueueName][tmpResource]['nNewWorkers'] = tmpNewVal
# define number of new workers
if len(curWorkers) == 0:
n_workers_per_queue_and_rt = dict()
else:
n_workers_per_queue_and_rt = self.workerAdjuster.define_num_workers(curWorkers, siteName)
if n_workers_per_queue_and_rt is None:
mainLog.error('WorkerAdjuster failed to define the number of workers')
elif len(n_workers_per_queue_and_rt) == 0:
pass
else:
# loop over all queues and resource types
for queueName in n_workers_per_queue_and_rt:
for resource_type, tmpVal in iteritems(n_workers_per_queue_and_rt[queueName]):
tmpLog = self.make_logger(_logger, 'id={0} queue={1} rtype={2}'.format(lockedBy,
queueName,
resource_type),
method_name='run')
try:
tmpLog.debug('start')
tmpLog.debug('workers status: %s' % tmpVal)
nWorkers = tmpVal['nNewWorkers'] + tmpVal['nReady']
nReady = tmpVal['nReady']
# check queue
if not self.queueConfigMapper.has_queue(queueName):
tmpLog.error('config not found')
continue
# no new workers
if nWorkers == 0:
tmpLog.debug('skipped since no new worker is needed based on current stats')
continue
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
workerMakerCore = self.workerMaker.get_plugin(queueConfig)
# check if resource is ready
if hasattr(workerMakerCore, 'dynamicSizing') and workerMakerCore.dynamicSizing is True:
numReadyResources = self.workerMaker.num_ready_resources(queueConfig,
resource_type,
workerMakerCore)
tmpLog.debug('numReadyResources: %s' % numReadyResources)
if not numReadyResources:
if hasattr(workerMakerCore, 'staticWorkers'):
nQRWorkers = tmpVal['nQueue'] + tmpVal['nRunning']
tmpLog.debug('staticWorkers: %s, nQRWorkers(Queue+Running): %s' %
(workerMakerCore.staticWorkers, nQRWorkers))
if nQRWorkers >= workerMakerCore.staticWorkers:
tmpLog.debug('No left static workers, skip')
continue
else:
nWorkers = min(workerMakerCore.staticWorkers - nQRWorkers, nWorkers)
tmpLog.debug('staticWorkers: %s, nWorkers: %s' %
(workerMakerCore.staticWorkers, nWorkers))
else:
tmpLog.debug('skip since no resources are ready')
continue
else:
nWorkers = min(nWorkers, numReadyResources)
# post action of worker maker
if hasattr(workerMakerCore, 'skipOnFail') and workerMakerCore.skipOnFail is True:
skipOnFail = True
else:
skipOnFail = False
# actions based on mapping type
if queueConfig.mapType == WorkSpec.MT_NoJob:
# workers without jobs
jobChunks = []
for i in range(nWorkers):
jobChunks.append([])
elif queueConfig.mapType == WorkSpec.MT_OneToOne:
# one worker per one job
jobChunks = self.dbProxy.get_job_chunks_for_workers(
queueName,
nWorkers, nReady, 1, None,
queueConfig.useJobLateBinding,
harvester_config.submitter.checkInterval,
harvester_config.submitter.lockInterval,
lockedBy)
elif queueConfig.mapType == WorkSpec.MT_MultiJobs:
# one worker for multiple jobs
nJobsPerWorker = self.workerMaker.get_num_jobs_per_worker(queueConfig,
nWorkers,
resource_type,
maker=workerMakerCore)
tmpLog.debug('nJobsPerWorker={0}'.format(nJobsPerWorker))
jobChunks = self.dbProxy.get_job_chunks_for_workers(
queueName,
nWorkers, nReady, nJobsPerWorker, None,
queueConfig.useJobLateBinding,
harvester_config.submitter.checkInterval,
harvester_config.submitter.lockInterval,
lockedBy,
queueConfig.allowJobMixture)
elif queueConfig.mapType == WorkSpec.MT_MultiWorkers:
# multiple workers for one job
nWorkersPerJob = self.workerMaker.get_num_workers_per_job(queueConfig,
nWorkers,
resource_type,
maker=workerMakerCore)
maxWorkersPerJob = self.workerMaker.get_max_workers_per_job_in_total(
queueConfig, resource_type, maker=workerMakerCore)
maxWorkersPerJobPerCycle = self.workerMaker.get_max_workers_per_job_per_cycle(
queueConfig, resource_type, maker=workerMakerCore)
tmpLog.debug('nWorkersPerJob={0}'.format(nWorkersPerJob))
jobChunks = self.dbProxy.get_job_chunks_for_workers(
queueName,
nWorkers, nReady, None, nWorkersPerJob,
queueConfig.useJobLateBinding,
harvester_config.submitter.checkInterval,
harvester_config.submitter.lockInterval,
lockedBy, max_workers_per_job_in_total=maxWorkersPerJob,
max_workers_per_job_per_cycle=maxWorkersPerJobPerCycle)
else:
tmpLog.error('unknown mapType={0}'.format(queueConfig.mapType))
continue
tmpLog.debug('got {0} job chunks'.format(len(jobChunks)))
if len(jobChunks) == 0:
continue
# make workers
okChunks, ngChunks = self.workerMaker.make_workers(jobChunks, queueConfig,
nReady, resource_type,
maker=workerMakerCore)
if len(ngChunks) == 0:
tmpLog.debug('successfully made {0} workers'.format(len(okChunks)))
else:
tmpLog.debug('made {0} workers, while {1} workers failed'.format(len(okChunks),
len(ngChunks)))
timeNow = datetime.datetime.utcnow()
timeNow_timestamp = time.time()
pandaIDs = set()
# NG (=not good)
for ngJobs in ngChunks:
for jobSpec in ngJobs:
if skipOnFail:
# release jobs when workers are not made
pandaIDs.add(jobSpec.PandaID)
else:
jobSpec.status = 'failed'
jobSpec.subStatus = 'failed_to_make'
jobSpec.stateChangeTime = timeNow
jobSpec.lockedBy = None
errStr = 'failed to make a worker'
jobSpec.set_pilot_error(PilotErrors.ERR_SETUPFAILURE, errStr)
jobSpec.trigger_propagation()
self.dbProxy.update_job(jobSpec, {'lockedBy': lockedBy,
'subStatus': 'prepared'})
# OK
workSpecList = []
if len(okChunks) > 0:
for workSpec, okJobs in okChunks:
# has job
if (queueConfig.useJobLateBinding and workSpec.workerID is None) \
or queueConfig.mapType == WorkSpec.MT_NoJob:
workSpec.hasJob = 0
else:
workSpec.hasJob = 1
if workSpec.nJobsToReFill in [None, 0]:
workSpec.set_jobspec_list(okJobs)
else:
# refill free slots during the worker is running
workSpec.set_jobspec_list(okJobs[:workSpec.nJobsToReFill])
workSpec.nJobsToReFill = None
for jobSpec in okJobs[workSpec.nJobsToReFill:]:
pandaIDs.add(jobSpec.PandaID)
workSpec.set_num_jobs_with_list()
# map type
workSpec.mapType = queueConfig.mapType
# queue name
workSpec.computingSite = queueConfig.queueName
# set access point
workSpec.accessPoint = queueConfig.messenger['accessPoint']
# sync level
workSpec.syncLevel = queueConfig.get_synchronization_level()
# events
if len(okJobs) > 0 and \
('eventService' in okJobs[0].jobParams or
'cloneJob' in okJobs[0].jobParams):
workSpec.eventsRequest = WorkSpec.EV_useEvents
workSpecList.append(workSpec)
if len(workSpecList) > 0:
sw = core_utils.get_stopwatch()
# get plugin for submitter
submitterCore = self.pluginFactory.get_plugin(queueConfig.submitter)
if submitterCore is None:
# not found
tmpLog.error(
'submitter plugin for {0} not found'.format(jobSpec.computingSite))
continue
# get plugin for messenger
messenger = self.pluginFactory.get_plugin(queueConfig.messenger)
if messenger is None:
# not found
tmpLog.error(
'messenger plugin for {0} not found'.format(jobSpec.computingSite))
continue
# setup access points
messenger.setup_access_points(workSpecList)
# feed jobs
for workSpec in workSpecList:
if workSpec.hasJob == 1:
tmpStat = messenger.feed_jobs(workSpec, workSpec.get_jobspec_list())
if tmpStat is False:
tmpLog.error(
'failed to send jobs to workerID={0}'.format(workSpec.workerID))
else:
tmpLog.debug(
'sent jobs to workerID={0} with {1}'.format(workSpec.workerID,
tmpStat))
# insert workers
self.dbProxy.insert_workers(workSpecList, lockedBy)
# submit
sw.reset()
tmpLog.info('submitting {0} workers'.format(len(workSpecList)))
workSpecList, tmpRetList, tmpStrList = self.submit_workers(submitterCore,
workSpecList)
tmpLog.debug('done submitting {0} workers'.format(len(workSpecList))
+ sw.get_elapsed_time())
# collect successful jobs
okPandaIDs = set()
for iWorker, (tmpRet, tmpStr) in enumerate(zip(tmpRetList, tmpStrList)):
if tmpRet:
workSpec, jobList = okChunks[iWorker]
jobList = workSpec.get_jobspec_list()
if jobList is not None:
for jobSpec in jobList:
okPandaIDs.add(jobSpec.PandaID)
# loop over all workers
for iWorker, (tmpRet, tmpStr) in enumerate(zip(tmpRetList, tmpStrList)):
workSpec, jobList = okChunks[iWorker]
# set harvesterHost
workSpec.harvesterHost = socket.gethostname()
# use associated job list since it can be truncated for re-filling
jobList = workSpec.get_jobspec_list()
# set status
if not tmpRet:
# failed submission
errStr = 'failed to submit a workerID={0} with {1}'.format(
workSpec.workerID,
tmpStr)
tmpLog.error(errStr)
workSpec.set_status(WorkSpec.ST_missed)
workSpec.set_dialog_message(tmpStr)
workSpec.set_pilot_error(PilotErrors.ERR_SETUPFAILURE, errStr)
if jobList is not None:
# increment attempt number
newJobList = []
for jobSpec in jobList:
# skip if successful with another worker
if jobSpec.PandaID in okPandaIDs:
continue
if jobSpec.submissionAttempts is None:
jobSpec.submissionAttempts = 0
jobSpec.submissionAttempts += 1
# max attempt or permanent error
if tmpRet is False or \
jobSpec.submissionAttempts >= \
queueConfig.maxSubmissionAttempts:
newJobList.append(jobSpec)
else:
self.dbProxy.increment_submission_attempt(
jobSpec.PandaID,
jobSpec.submissionAttempts)
jobList = newJobList
elif queueConfig.useJobLateBinding and workSpec.hasJob == 1:
# directly go to running after feeding jobs for late biding
workSpec.set_status(WorkSpec.ST_running)
else:
# normal successful submission
workSpec.set_status(WorkSpec.ST_submitted)
workSpec.submitTime = timeNow
workSpec.modificationTime = timeNow
workSpec.checkTime = timeNow
if self.monitor_fifo.enabled:
workSpec.set_work_params({'lastCheckAt': timeNow_timestamp})
# prefetch events
if tmpRet and workSpec.hasJob == 1 and \
workSpec.eventsRequest == WorkSpec.EV_useEvents and \
queueConfig.prefetchEvents:
workSpec.eventsRequest = WorkSpec.EV_requestEvents
eventsRequestParams = dict()
for jobSpec in jobList:
eventsRequestParams[jobSpec.PandaID] = \
{'pandaID': jobSpec.PandaID,
'taskID': jobSpec.taskID,
'jobsetID': jobSpec.jobParams['jobsetID'],
'nRanges': max(int(math.ceil(workSpec.nCore / len(jobList))),
jobSpec.jobParams['coreCount']),
}
workSpec.eventsRequestParams = eventsRequestParams
# register worker
tmpStat = self.dbProxy.register_worker(workSpec, jobList, lockedBy)
if jobList is not None:
for jobSpec in jobList:
pandaIDs.add(jobSpec.PandaID)
if tmpStat:
if tmpRet:
tmpStr = \
'submitted a workerID={0} for PandaID={1} with batchID={2}'
tmpLog.info(tmpStr.format(workSpec.workerID,
jobSpec.PandaID,
workSpec.batchID))
else:
tmpStr = 'failed to submit a workerID={0} for PandaID={1}'
tmpLog.error(tmpStr.format(workSpec.workerID,
jobSpec.PandaID))
else:
tmpStr = \
'failed to register a worker for PandaID={0} with batchID={1}'
tmpLog.error(tmpStr.format(jobSpec.PandaID, workSpec.batchID))
# enqueue to monitor fifo
if self.monitor_fifo.enabled \
and queueConfig.mapType != WorkSpec.MT_MultiWorkers:
workSpecsToEnqueue = \
[[w] for w in workSpecList if w.status
in (WorkSpec.ST_submitted, WorkSpec.ST_running)]
check_delay = min(
getattr(harvester_config.monitor, 'eventBasedCheckInterval',
harvester_config.monitor.checkInterval),
getattr(harvester_config.monitor, 'fifoCheckInterval',
harvester_config.monitor.checkInterval))
monitor_fifo.put((queueName, workSpecsToEnqueue), time.time() + check_delay)
mainLog.debug('put workers to monitor FIFO')
submitted = True
# release jobs
self.dbProxy.release_jobs(pandaIDs, lockedBy)
tmpLog.info('done')
except Exception:
core_utils.dump_error_message(tmpLog)
# release the site
self.dbProxy.release_site(siteName, lockedBy)
if sw_main.get_elapsed_time_in_sec() > queueLockInterval:
mainLog.warning('a submitter cycle was longer than queueLockInterval {0} sec'.format(queueLockInterval)
+ sw_main.get_elapsed_time())
mainLog.debug('done')
# define sleep interval
if siteName is None:
sleepTime = harvester_config.submitter.sleepTime
else:
sleepTime = 0
if submitted and hasattr(harvester_config.submitter, 'minSubmissionInterval'):
interval = harvester_config.submitter.minSubmissionInterval
if interval > 0:
newTime = datetime.datetime.utcnow() + datetime.timedelta(seconds=interval)
self.dbProxy.update_panda_queue_attribute('submitTime', newTime, site_name=siteName)
# time the cycle
mainLog.debug('done a submitter cycle' + sw_main.get_elapsed_time())
# check if being terminated
if self.terminated(sleepTime):
mainLog.debug('terminated')
return
# wrapper for submitWorkers to skip ready workers
def submit_workers(self, submitter_core, workspec_list):
retList = []
strList = []
newSpecList = []
workersToSubmit = []
for workSpec in workspec_list:
if workSpec.status in [WorkSpec.ST_ready, WorkSpec.ST_running]:
newSpecList.append(workSpec)
retList.append(True)
strList.append('')
else:
workersToSubmit.append(workSpec)
tmpRetList = submitter_core.submit_workers(workersToSubmit)
# submit the workers to the monitoring
self.apfmon.create_workers(workersToSubmit)
for tmpRet, tmpStr in tmpRetList:
retList.append(tmpRet)
strList.append(tmpStr)
newSpecList += workersToSubmit
return newSpecList, retList, strList
import os
import sys
import json
from pandaharvester.harvestercore.db_proxy_pool import DBProxyPool as DBProxy
from pandaharvester.harvestermessenger import shared_file_messenger
workerID = int(sys.argv[1])
eventID = sys.argv[2]
status = sys.argv[3]
proxy = DBProxy()
workSpec = proxy.get_worker_with_id(workerID)
jobSpec = proxy.get_jobs_with_worker_id(workerID, None)[0]
accessPoint = workSpec.get_access_point()
try:
os.makedirs(accessPoint)
except:
pass
node = {}
node['eventRangeID'] = eventID
node['eventStatus'] = status
f = open(
os.path.join(accessPoint, shared_file_messenger.jsonEventsUpdateFileName),
'w')
json.dump([node], f)
f.close()
def __init__(self, **kwarg):
# logic type : AND: throttled if all rules are satisfied, OR: throttled if one rule is satisfied
self.logicType = 'OR'
PluginBase.__init__(self, **kwarg)
self.dbProxy = DBProxy()
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
msgStr = "plugin={0}".format(preparatorCore.__class__.__name__)
tmpLog.debug(msgStr)
msgStr = "Initial queueConfig.preparator = {}".format(
initial_queueConfig_preparator)
tmpLog.debug(msgStr)
msgStr = "Modified queueConfig.preparator = {}".format(
modified_queueConfig_preparator)
tmpLog.debug(msgStr)
scope = 'panda'
proxy = DBProxy()
communicator = CommunicatorPool()
cacher = Cacher(communicator, single_mode=True)
cacher.run()
tmpLog.debug("plugin={0}".format(preparatorCore.__class__.__name__))
tmpLog.debug("BasePath from preparator configuration: %s " %
preparatorCore.basePath)
# get all jobs in table in a preparing substate
tmpLog.debug('try to get all jobs in a preparing substate')
jobSpec_list = proxy.get_jobs_in_sub_status('preparing', 2000, None, None,
None, None, None, None)
tmpLog.debug('got {0} jobs'.format(len(jobSpec_list)))
# loop over all found jobs
if len(jobSpec_list) > 0:
import os
import sys
from pandaharvester.harvestercore.db_proxy_pool import DBProxyPool as DBProxy
from pandaharvester.harvestercore.communicator_pool import CommunicatorPool
from pandaharvester.harvestermessenger import shared_file_messenger
workerID = int(sys.argv[1])
proxy = DBProxy()
workSpec = proxy.get_worker_with_id(workerID)
jobSpec = proxy.get_jobs_with_worker_id(workerID, None)[0]
accessPoint = workSpec.get_access_point()
try:
os.makedirs(accessPoint)
except:
pass
node = {}
node['pandaID'] = jobSpec.PandaID
node['jobsetID'] = jobSpec.jobParams['jobsetID']
node['taskID'] = jobSpec.taskID
a = CommunicatorPool()
tmpStat, tmpVal = a.getEventRanges(node)
mess = shared_file_messenger.SharedFileMessenger()
mess.feed_events(workSpec, tmpVal)
import os
import sys
workerID = int(sys.argv[1])
from pandaharvester.harvestercore.db_proxy_pool import DBProxyPool as DBProxy
proxy = DBProxy()
workSpec = proxy.get_worker_with_id(workerID)
accessPoint = workSpec.get_access_point()
try:
os.makedirs(accessPoint)
except:
pass
from pandaharvester.harvestermessenger import shared_file_messenger
f = open(os.path.join(accessPoint, shared_file_messenger.jsonJobRequestFileName), 'w')
f.close()
except Exception:
pass
for loggerName, loggerObj in iteritems(logging.Logger.manager.loggerDict):
if loggerName.startswith('panda.log'):
if len(loggerObj.handlers) == 0:
continue
if loggerName.split('.')[-1] in ['db_proxy']:
continue
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
queueConfigMapper = QueueConfigMapper()
proxy = DBProxy()
proxy.make_tables(queueConfigMapper)
job = JobSpec()
job.PandaID = 1
job.modificationTime = datetime.datetime.now()
proxy.insert_jobs([job])
newJob = proxy.get_job(1)
a = CommunicatorPool()
a.get_jobs('siteName', 'nodeName', 'prodSourceLabel', 'computingElement', 1, {})
class ARCSubmitter(PluginBase):
'''Submitter for ARC CE'''
def __init__(self, **kwarg):
'''Set up DB connection and credentials'''
PluginBase.__init__(self, **kwarg)
self.dbproxy = DBProxy()
self.schedulerid = harvester_config.master.harvester_id
# Credential dictionary role: proxy file
self.certs = dict(zip([r.split('=')[1] for r in list(harvester_config.credmanager.voms)],
list(harvester_config.credmanager.outCertFile)))
self.cred_type = arc.initializeCredentialsType(arc.initializeCredentialsType.SkipCredentials)
def _run_submit(self, thr):
'''Run a thread to do the submission'''
try:
thr.start()
except:
pass
# Be careful to wait longer than submission timeout
thr.join(thr.userconfig.Timeout() + 60.0)
if thr.isAlive():
# abort due to timeout and try again
raise Exception("Submission timeout")
if thr.job is None:
raise Exception("Submission failed")
return thr.job
def _arc_submit(self, xrsl, arcces, userconfig, log):
'''Check the available CEs and submit'''
queuelist = []
for arcce in arcces:
(ce_endpoint, ce_queue) = arcce
aris = arc.URL(str(ce_endpoint))
ce_host = aris.Host()
if aris.Protocol() == 'https':
aris.ChangePath('/arex')
infoendpoints = [arc.Endpoint(aris.str(),
arc.Endpoint.COMPUTINGINFO,
'org.ogf.glue.emies.resourceinfo')]
else:
aris = 'ldap://'+aris.Host()+'/mds-vo-name=local,o=grid'
infoendpoints = [arc.Endpoint(aris,
arc.Endpoint.COMPUTINGINFO,
'org.nordugrid.ldapng')]
# retriever contains a list of CE endpoints
retriever = arc.ComputingServiceRetriever(userconfig, infoendpoints)
retriever.wait()
# targets is the list of queues
# parse target.ComputingService.ID for the CE hostname
# target.ComputingShare.Name is the queue name
targets = retriever.GetExecutionTargets()
# Filter only sites for this process
for target in targets:
if not target.ComputingService.ID:
log.info("Target {0} does not have ComputingService ID defined, skipping".format(target.ComputingService.Name))
continue
# If EMI-ES infoendpoint, force EMI-ES submission
if infoendpoints[0].InterfaceName == 'org.ogf.glue.emies.resourceinfo' \
and target.ComputingEndpoint.InterfaceName != 'org.ogf.glue.emies.activitycreation':
log.debug("Rejecting target interface {0} because not EMI-ES".format(target.ComputingEndpoint.InterfaceName))
continue
# Check for matching host and queue
targethost = re.sub(':arex$', '', re.sub('urn:ogf:ComputingService:', '', target.ComputingService.ID))
targetqueue = target.ComputingShare.Name
if targethost != ce_host:
log.debug('Rejecting target host {0} as it does not match {1}'.format(targethost, ce_host))
continue
if targetqueue != ce_queue:
log.debug('Rejecting target queue {0} as it does not match {1}'.format(targetqueue, ce_queue))
continue
queuelist.append(target)
log.debug("Adding target {0}:{1}".format(targethost, targetqueue))
# check if any queues are available, if not leave and try again next time
if not queuelist:
raise Exception("No free queues available")
log.debug("preparing submission")
jobdescs = arc.JobDescriptionList()
if not arc.JobDescription_Parse(str(xrsl), jobdescs):
raise Exception("Failed to prepare job description")
# Run the submission in a separate thread
thr = SubmitThr(queuelist, jobdescs, userconfig)
return self._run_submit(thr)
def _set_logdir(self, site):
date = time.strftime('%Y-%m-%d')
return os.path.join(date, site)
# submit workers
def submit_workers(self, workspec_list):
retlist = []
# Get queue info from DB
pandaqueues = self.dbproxy.get_cache("panda_queues.json", None)
if pandaqueues is None:
raise Exception("Failed to get panda queue info from database")
pandaqueues = pandaqueues.data
osmap = self.dbproxy.get_cache("ddmendpoints_objectstores.json", None)
if osmap is None:
raise Exception("Failed to get Object Store info from database")
osmap = osmap.data
for workspec in workspec_list:
arclog = arc_utils.ARCLogger(baselogger, workspec.workerID)
tmplog = arclog.log
# Assume for aCT that jobs are always pre-fetched (no late-binding)
for jobspec in workspec.get_jobspec_list():
tmplog.debug("JobSpec: {0}".format(jobspec.values_map()))
if jobspec.computingSite not in pandaqueues:
retlist.append((False, "No queue information for {0}".format(jobspec.computingSite)))
continue
# Get CEs from panda queue info
# List of (endpoint, queue) tuples
arcces = []
for endpoint in pandaqueues[jobspec.computingSite]['queues']:
ce_endpoint = endpoint['ce_endpoint']
if not re.search('://', ce_endpoint):
ce_endpoint = 'gsiftp://%s' % ce_endpoint
ce_queue = endpoint['ce_queue_name']
arcces.append((ce_endpoint, ce_queue))
if not arcces:
retlist.append((False, "No CEs defined for %{0}".format(jobspec.computingSite)))
continue
# Set true pilot or not
queueconfigmapper = QueueConfigMapper()
queueconfig = queueconfigmapper.get_queue(jobspec.computingSite)
pandaqueues[jobspec.computingSite]['truepilot'] = 'running' in queueconfig.noHeartbeat
# Set log URL for GTAG env in job description
logbaseurl = queueconfig.submitter.get('logBaseURL')
logsubdir = self._set_logdir(jobspec.computingSite)
logfileurl = '/'.join([logbaseurl, logsubdir, '%d.out' % jobspec.PandaID]) if logbaseurl else None
tmplog.debug("Converting to ARC XRSL format")
arcxrsl = ARCParser(jobspec.jobParams,
jobspec.computingSite,
pandaqueues[jobspec.computingSite],
logfileurl,
self.schedulerid,
osmap,
'/tmp', # tmpdir, TODO common tmp dir
None, #jobSpec.eventranges, # TODO event ranges
tmplog)
arcxrsl.parse()
xrsl = arcxrsl.getXrsl()
tmplog.debug("ARC xrsl: {0}".format(xrsl))
# Set the files to be downloaded at the end of the job
downloadfiles = 'gmlog/errors'
if 'logFile' in jobspec.jobParams:
downloadfiles += ';%s' %jobspec.jobParams['logFile'].replace('.tgz', '')
if not pandaqueues[jobspec.computingSite]['truepilot']:
downloadfiles += ';jobSmallFiles.tgz'
# Set certificate
userconfig = arc.UserConfig(self.cred_type)
proxyrole = ''
if jobspec.jobParams['prodSourceLabel'] == 'user':
userconfig.ProxyPath(str(self.certs['pilot']))
proxyrole = 'pilot'
else:
userconfig.ProxyPath(str(self.certs['production']))
proxyrole = 'production'
tmplog.debug("Submitting using {0} proxy at {1}".format(proxyrole, userconfig.ProxyPath()))
try:
tmplog.debug("Submission targets: {0}".format(arcces))
arcjob = self._arc_submit(xrsl, arcces, userconfig, tmplog)
tmplog.info("ARC CE job id {0}".format(arcjob.JobID))
arc_utils.arcjob2workspec(arcjob, workspec)
workspec.workAttributes['arcdownloadfiles'] = downloadfiles
workspec.workAttributes['proxyrole'] = proxyrole
workspec.workAttributes['logsubdir'] = logsubdir
workspec.batchID = arcjob.JobID
tmplog.debug(workspec.workAttributes)
result = (True, '')
except Exception as exc:
tmplog.error(traceback.format_exc())
result = (False, "Failed to submit ARC job: {0}".format(str(exc)))
retlist.append(result)
return retlist
class Sweeper(AgentBase):
# constructor
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
# main loop
def run(self):
lockedBy = 'sweeper-{0}'.format(self.ident)
while True:
mainLog = core_utils.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
mainLog.debug('try to get workers to kill')
# get workers to kill
workersToKill = self.dbProxy.get_workers_to_kill(harvester_config.sweeper.maxWorkers,
harvester_config.sweeper.checkInterval)
mainLog.debug('got {0} queues to kill workers'.format(len(workersToKill)))
# loop over all workers
for queueName, workSpecs in iteritems(workersToKill):
# get sweeper
if not self.queueConfigMapper.has_queue(queueName):
mainLog.error('queue config for {0} not found'.format(queueName))
continue
queueConfig = self.queueConfigMapper.get_queue(queueName)
sweeperCore = self.pluginFactory.get_plugin(queueConfig.sweeper)
for workSpec in workSpecs:
tmpLog = core_utils.make_logger(_logger, 'workerID={0}'.format(workSpec.workerID),
method_name='run')
tmpLog.debug('start killing')
tmpStat, tmpOut = sweeperCore.kill_worker(workSpec)
tmpLog.debug('done with status={0} diag={1}'.format(tmpStat, tmpOut))
mainLog.debug('done kill')
# timeout for missed
try:
keepMissed = harvester_config.sweeper.keepMissed
except:
keepMissed = 24
# get workers for cleanup
statusTimeoutMap = {'finished': harvester_config.sweeper.keepFinished,
'failed': harvester_config.sweeper.keepFailed,
'cancelled': harvester_config.sweeper.keepCancelled,
'missed': keepMissed
}
workersForCleanup = self.dbProxy.get_workers_for_cleanup(harvester_config.sweeper.maxWorkers,
statusTimeoutMap)
mainLog.debug('got {0} queues for workers cleanup'.format(len(workersForCleanup)))
for queueName, workSpecs in iteritems(workersForCleanup):
# get sweeper
if not self.queueConfigMapper.has_queue(queueName):
mainLog.error('queue config for {0} not found'.format(queueName))
continue
queueConfig = self.queueConfigMapper.get_queue(queueName)
sweeperCore = self.pluginFactory.get_plugin(queueConfig.sweeper)
for workSpec in workSpecs:
tmpLog = core_utils.make_logger(_logger, 'workerID={0}'.format(workSpec.workerID),
method_name='run')
tmpLog.debug('start cleanup')
tmpStat, tmpOut = sweeperCore.sweep_worker(workSpec)
tmpLog.debug('done with status={0} diag={1}'.format(tmpStat, tmpOut))
if tmpStat:
# delete from DB
self.dbProxy.delete_worker(workSpec.workerID)
mainLog.debug('done cleanup')
# check if being terminated
if self.terminated(harvester_config.sweeper.sleepTime):
mainLog.debug('terminated')
return
def __init__(self):
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
def __init__(self, **kwarg):
# logic type : AND: throttled if all rules are satisfied, OR: throttled if one rule is satisfied
self.logicType = 'OR'
PluginBase.__init__(self, **kwarg)
self.dbProxy = DBProxy()
if loggerName.split('.')[-1] in ['db_proxy']:
continue
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
msgStr = "plugin={0}".format(preparatorCore.__class__.__name__)
tmpLog.debug(msgStr)
msgStr = "Initial queueConfig.preparator = {}".format(initial_queueConfig_preparator)
tmpLog.debug(msgStr)
msgStr = "Modified queueConfig.preparator = {}".format(modified_queueConfig_preparator)
tmpLog.debug(msgStr)
scope = 'panda'
proxy = DBProxy()
communicator = CommunicatorPool()
cacher = Cacher(communicator, single_mode=True)
cacher.run()
tmpLog.debug("plugin={0}".format(preparatorCore.__class__.__name__))
tmpLog.debug("BasePath from preparator configuration: %s " % preparatorCore.basePath)
# get all jobs in table in a preparing substate
#tmpLog.debug('try to get all jobs in a preparing substate')
#jobSpec_list = proxy.get_jobs_in_sub_status('preparing',2000,None,None,None,None,None,None)
# get all jobs
if job_id > 0 :
tmpLog.debug('try to get job ID - {}'.format(job_id))
jobSpec_list = [proxy.get_job(job_id)]
else :
for loggerName, loggerObj in iteritems(logging.Logger.manager.loggerDict):
if loggerName.startswith('panda.log'):
if len(loggerObj.handlers) == 0:
continue
if loggerName.split('.')[-1] in ['db_proxy']:
continue
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
pp = pprint.PrettyPrinter(indent=4)
queueConfigMapper = QueueConfigMapper()
proxy = DBProxy()
sqlJ ="SELECT * FROM job_table"
resultsJobcur = proxy.execute(sqlJ)
resultsJob = resultsJobcur.fetchall()
proxy.commit()
sqlF ="SELECT * FROM file_table"
resultsFilescur = proxy.execute(sqlF)
resultsFiles = resultsFilescur.fetchall()
proxy.commit()
print "job_table - "
print resultsJob[0].keys()
if loggerName.split('.')[-1] in ['db_proxy']:
continue
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
msgStr = "plugin={0}".format(preparatorCore.__class__.__name__)
tmpLog.debug(msgStr)
msgStr = "Initial queueConfig.preparator = {}".format(initial_queueConfig_preparator)
tmpLog.debug(msgStr)
msgStr = "Modified queueConfig.preparator = {}".format(modified_queueConfig_preparator)
tmpLog.debug(msgStr)
scope = 'panda'
proxy = DBProxy()
communicator = CommunicatorPool()
cacher = Cacher(communicator, single_mode=True)
cacher.run()
tmpLog.debug("plugin={0}".format(preparatorCore.__class__.__name__))
tmpLog.debug("BasePath from preparator configuration: %s " % preparatorCore.basePath)
# get all jobs in table in a preparing substate
tmpLog.debug('try to get all jobs in a preparing substate')
jobSpec_list = proxy.get_jobs_in_sub_status('preparing',2000,None,None,None,None,None,None)
tmpLog.debug('got {0} jobs'.format(len(jobSpec_list)))
# loop over all found jobs
if len(jobSpec_list) > 0 :
for jobSpec in jobSpec_list:
tmpLog.debug(' PandaID = %d status = %s subStatus = %s lockedBy = %s' %
class Monitor(AgentBase):
# constructor
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queueConfigMapper = queue_config_mapper
self.dbProxy = DBProxy()
self.pluginFactory = PluginFactory()
self.startTimestamp = time.time()
self.monitor_fifo = MonitorFIFO()
self.apfmon = Apfmon(self.queueConfigMapper)
# main loop
def run(self):
lockedBy = 'monitor-{0}'.format(self.get_pid())
# init messengers
for queueConfig in self.queueConfigMapper.get_all_queues().values():
# just import for module initialization
self.pluginFactory.get_plugin(queueConfig.messenger)
# main
try:
fifoSleepTimeMilli = harvester_config.monitor.fifoSleepTimeMilli
except AttributeError:
fifoSleepTimeMilli = 5000
try:
fifoCheckDuration = harvester_config.monitor.fifoCheckDuration
except AttributeError:
fifoCheckDuration = 30
try:
fifoMaxWorkersPerChunk = harvester_config.monitor.fifoMaxWorkersPerChunk
except AttributeError:
fifoMaxWorkersPerChunk = 500
try:
fifoProtectiveDequeue = harvester_config.monitor.fifoProtectiveDequeue
except AttributeError:
fifoProtectiveDequeue = True
last_DB_cycle_timestamp = 0
monitor_fifo = self.monitor_fifo
sleepTime = (fifoSleepTimeMilli / 1000.0) \
if monitor_fifo.enabled else harvester_config.monitor.sleepTime
adjusted_sleepTime = sleepTime
if monitor_fifo.enabled:
monitor_fifo.restore()
while True:
sw_main = core_utils.get_stopwatch()
mainLog = self.make_logger(_logger,
'id={0}'.format(lockedBy),
method_name='run')
mainLog.debug('start a monitor cycle')
if time.time() >= last_DB_cycle_timestamp + harvester_config.monitor.sleepTime and \
not (monitor_fifo.enabled and self.singleMode):
# run with workers from DB
sw_db = core_utils.get_stopwatch()
mainLog.debug('starting run with DB')
mainLog.debug('getting workers to monitor')
workSpecsPerQueue = self.dbProxy.get_workers_to_update(
harvester_config.monitor.maxWorkers,
harvester_config.monitor.checkInterval,
harvester_config.monitor.lockInterval, lockedBy)
mainLog.debug('got {0} queues'.format(len(workSpecsPerQueue)))
# loop over all workers
for queueName, configIdWorkSpecs in iteritems(
workSpecsPerQueue):
for configID, workSpecsList in iteritems(
configIdWorkSpecs):
retVal = self.monitor_agent_core(lockedBy,
queueName,
workSpecsList,
config_id=configID)
if monitor_fifo.enabled and retVal is not None:
workSpecsToEnqueue, workSpecsToEnqueueToHead, timeNow_timestamp, fifoCheckInterval = retVal
if workSpecsToEnqueue:
mainLog.debug('putting workers to FIFO')
try:
score = fifoCheckInterval + timeNow_timestamp
monitor_fifo.put(
(queueName, workSpecsToEnqueue), score)
mainLog.info(
'put workers of {0} to FIFO with score {1}'
.format(queueName, score))
except Exception as errStr:
mainLog.error(
'failed to put object from FIFO: {0}'.
format(errStr))
if workSpecsToEnqueueToHead:
mainLog.debug('putting workers to FIFO head')
try:
score = fifoCheckInterval - timeNow_timestamp
monitor_fifo.put(
(queueName, workSpecsToEnqueueToHead),
score)
mainLog.info(
'put workers of {0} to FIFO with score {1}'
.format(queueName, score))
except Exception as errStr:
mainLog.error(
'failed to put object from FIFO head: {0}'
.format(errStr))
last_DB_cycle_timestamp = time.time()
if sw_db.get_elapsed_time_in_sec(
) > harvester_config.monitor.lockInterval:
mainLog.warning(
'a single DB cycle was longer than lockInterval ' +
sw_db.get_elapsed_time())
else:
mainLog.debug('done a DB cycle' + sw_db.get_elapsed_time())
mainLog.debug('ended run with DB')
elif monitor_fifo.enabled:
# run with workers from FIFO
sw = core_utils.get_stopwatch()
n_loops = 0
n_loops_hit = 0
last_fifo_cycle_timestamp = time.time()
to_break = False
obj_dequeued_id_list = []
obj_to_enqueue_dict = collections.defaultdict(
lambda: [[], 0, 0])
obj_to_enqueue_to_head_dict = collections.defaultdict(
lambda: [[], 0, 0])
remaining_obj_to_enqueue_dict = {}
remaining_obj_to_enqueue_to_head_dict = {}
n_chunk_peeked_stat, sum_overhead_time_stat = 0, 0.0
while time.time(
) < last_fifo_cycle_timestamp + fifoCheckDuration:
sw.reset()
n_loops += 1
retVal, overhead_time = monitor_fifo.to_check_workers()
if overhead_time is not None:
n_chunk_peeked_stat += 1
sum_overhead_time_stat += overhead_time
if retVal:
# check fifo size
fifo_size = monitor_fifo.size()
mainLog.debug('FIFO size is {0}'.format(fifo_size))
mainLog.debug('starting run with FIFO')
try:
obj_gotten = monitor_fifo.get(
timeout=1, protective=fifoProtectiveDequeue)
except Exception as errStr:
mainLog.error(
'failed to get object from FIFO: {0}'.format(
errStr))
else:
if obj_gotten is not None:
sw_fifo = core_utils.get_stopwatch()
if fifoProtectiveDequeue:
obj_dequeued_id_list.append(obj_gotten.id)
queueName, workSpecsList = obj_gotten.item
mainLog.debug(
'got a chunk of {0} workers of {1} from FIFO'
.format(len(workSpecsList), queueName) +
sw.get_elapsed_time())
sw.reset()
configID = None
for workSpecs in workSpecsList:
if configID is None and len(workSpecs) > 0:
configID = workSpecs[0].configID
for workSpec in workSpecs:
if workSpec.pandaid_list is None:
_jobspec_list = workSpec.get_jobspec_list(
)
if _jobspec_list is not None:
workSpec.pandaid_list = [
j.PandaID
for j in workSpec.
get_jobspec_list()
]
else:
workSpec.pandaid_list = []
workSpec.force_update(
'pandaid_list')
retVal = self.monitor_agent_core(
lockedBy,
queueName,
workSpecsList,
from_fifo=True,
config_id=configID)
if retVal is not None:
workSpecsToEnqueue, workSpecsToEnqueueToHead, timeNow_timestamp, fifoCheckInterval = retVal
try:
if len(obj_to_enqueue_dict[queueName]
[0]) + len(
workSpecsToEnqueue
) <= fifoMaxWorkersPerChunk:
obj_to_enqueue_dict[queueName][
0].extend(workSpecsToEnqueue)
obj_to_enqueue_dict[queueName][
1] = max(
obj_to_enqueue_dict[
queueName][1],
timeNow_timestamp)
obj_to_enqueue_dict[queueName][
2] = max(
obj_to_enqueue_dict[
queueName][2],
fifoCheckInterval)
else:
to_break = True
remaining_obj_to_enqueue_dict[
queueName] = [
workSpecsToEnqueue,
timeNow_timestamp,
fifoCheckInterval
]
except Exception as errStr:
mainLog.error(
'failed to gather workers for FIFO: {0}'
.format(errStr))
to_break = True
try:
if len(obj_to_enqueue_to_head_dict[
queueName][0]) + len(
workSpecsToEnqueueToHead
) <= fifoMaxWorkersPerChunk:
obj_to_enqueue_to_head_dict[
queueName][0].extend(
workSpecsToEnqueueToHead)
obj_to_enqueue_to_head_dict[
queueName][1] = max(
obj_to_enqueue_to_head_dict[
queueName][1],
timeNow_timestamp)
obj_to_enqueue_to_head_dict[
queueName][2] = max(
obj_to_enqueue_to_head_dict[
queueName][2],
fifoCheckInterval)
else:
to_break = True
remaining_obj_to_enqueue_to_head_dict[
queueName] = [
workSpecsToEnqueueToHead,
timeNow_timestamp,
fifoCheckInterval
]
except Exception as errStr:
mainLog.error(
'failed to gather workers for FIFO head: {0}'
.format(errStr))
to_break = True
mainLog.debug(
'checked {0} workers from FIFO'.format(
len(workSpecsList)) +
sw.get_elapsed_time())
else:
mainLog.debug(
'monitor_agent_core returned None. Skipped putting to FIFO'
)
if sw_fifo.get_elapsed_time_in_sec(
) > harvester_config.monitor.lockInterval:
mainLog.warning(
'a single FIFO cycle was longer than lockInterval '
+ sw_fifo.get_elapsed_time())
else:
mainLog.debug('done a FIFO cycle' +
sw_fifo.get_elapsed_time())
n_loops_hit += 1
if to_break:
break
else:
mainLog.debug('got nothing in FIFO')
else:
mainLog.debug(
'workers in FIFO too young to check. Skipped')
if self.singleMode:
break
if overhead_time is not None:
time.sleep(
max(-overhead_time * random.uniform(0.1, 1),
adjusted_sleepTime))
else:
time.sleep(
max(fifoCheckDuration * random.uniform(0.1, 1),
adjusted_sleepTime))
mainLog.debug(
'run {0} loops, including {1} FIFO cycles'.format(
n_loops, n_loops_hit))
# enqueue to fifo
sw.reset()
n_chunk_put = 0
mainLog.debug('putting worker chunks to FIFO')
for _dct in (obj_to_enqueue_dict,
remaining_obj_to_enqueue_dict):
for queueName, obj_to_enqueue in iteritems(_dct):
try:
workSpecsToEnqueue, timeNow_timestamp, fifoCheckInterval = obj_to_enqueue
if workSpecsToEnqueue:
score = fifoCheckInterval + timeNow_timestamp
monitor_fifo.put(
(queueName, workSpecsToEnqueue), score)
n_chunk_put += 1
mainLog.info(
'put a chunk of {0} workers of {1} to FIFO with score {2}'
.format(len(workSpecsToEnqueue), queueName,
score))
except Exception as errStr:
mainLog.error(
'failed to put object from FIFO: {0}'.format(
errStr))
mainLog.debug('putting worker chunks to FIFO head')
for _dct in (obj_to_enqueue_to_head_dict,
remaining_obj_to_enqueue_to_head_dict):
for queueName, obj_to_enqueue_to_head in iteritems(_dct):
try:
workSpecsToEnqueueToHead, timeNow_timestamp, fifoCheckInterval = obj_to_enqueue_to_head
if workSpecsToEnqueueToHead:
score = fifoCheckInterval + timeNow_timestamp - 2**32
monitor_fifo.put(
(queueName, workSpecsToEnqueueToHead),
score)
n_chunk_put += 1
mainLog.info(
'put a chunk of {0} workers of {1} to FIFO with score {2}'
.format(len(workSpecsToEnqueueToHead),
queueName, score))
except Exception as errStr:
mainLog.error(
'failed to put object from FIFO head: {0}'.
format(errStr))
# release protective dequeued objects
if fifoProtectiveDequeue and len(obj_dequeued_id_list) > 0:
monitor_fifo.release(ids=obj_dequeued_id_list)
mainLog.debug(
'put {0} worker chunks into FIFO'.format(n_chunk_put) +
sw.get_elapsed_time())
# adjust adjusted_sleepTime
if n_chunk_peeked_stat > 0 and sum_overhead_time_stat > sleepTime:
speedup_factor = (sum_overhead_time_stat - sleepTime) / (
n_chunk_peeked_stat *
harvester_config.monitor.checkInterval)
speedup_factor = max(speedup_factor, 0)
adjusted_sleepTime = adjusted_sleepTime / (1. +
speedup_factor)
elif n_chunk_peeked_stat == 0 or sum_overhead_time_stat < 0:
adjusted_sleepTime = (sleepTime + adjusted_sleepTime) / 2
mainLog.debug('adjusted_sleepTime becomes {0:.3f} sec'.format(
adjusted_sleepTime))
# end run with fifo
mainLog.debug('ended run with FIFO')
# time the cycle
mainLog.debug('done a monitor cycle' + sw_main.get_elapsed_time())
# check if being terminated
if self.terminated(adjusted_sleepTime):
mainLog.debug('terminated')
return
# core of monitor agent to check workers in workSpecsList of queueName
def monitor_agent_core(self,
lockedBy,
queueName,
workSpecsList,
from_fifo=False,
config_id=None):
tmpQueLog = self.make_logger(_logger,
'id={0} queue={1}'.format(
lockedBy, queueName),
method_name='run')
# check queue
if not self.queueConfigMapper.has_queue(queueName, config_id):
tmpQueLog.error('config not found')
return None
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName, config_id)
try:
apfmon_status_updates = self.queueConfigMapper.queueConfig[
queueName].monitor['apfmon_status_updates']
except Exception:
apfmon_status_updates = False
tmpQueLog.debug(
'apfmon_status_updates: {0}'.format(apfmon_status_updates))
# get plugins
monCore = self.pluginFactory.get_plugin(queueConfig.monitor)
messenger = self.pluginFactory.get_plugin(queueConfig.messenger)
# workspec chunk of active workers
workSpecsToEnqueue_dict = {}
workSpecsToEnqueueToHead_dict = {}
timeNow_timestamp = time.time()
# get fifoCheckInterval for PQ and other fifo attributes
try:
fifoCheckInterval = monCore.fifoCheckInterval
except Exception:
if hasattr(harvester_config.monitor, 'fifoCheckInterval'):
fifoCheckInterval = harvester_config.monitor.fifoCheckInterval
else:
fifoCheckInterval = harvester_config.monitor.checkInterval
try:
forceEnqueueInterval = harvester_config.monitor.fifoForceEnqueueInterval
except AttributeError:
forceEnqueueInterval = 3600
try:
fifoMaxPreemptInterval = harvester_config.monitor.fifoMaxPreemptInterval
except AttributeError:
fifoMaxPreemptInterval = 60
# check workers
allWorkers = [item for sublist in workSpecsList for item in sublist]
tmpQueLog.debug('checking {0} workers'.format(len(allWorkers)))
tmpStat, tmpRetMap = self.check_workers(monCore, messenger, allWorkers,
queueConfig, tmpQueLog,
from_fifo)
if tmpStat:
# loop over all worker chunks
tmpQueLog.debug('update jobs and workers')
iWorker = 0
for workSpecs in workSpecsList:
jobSpecs = None
pandaIDsList = []
eventsToUpdateList = []
filesToStageOutList = dict()
isCheckedList = []
mapType = workSpecs[0].mapType
# loop over workSpecs
for workSpec in workSpecs:
tmpLog = self.make_logger(_logger,
'id={0} workerID={1}'.format(
lockedBy, workSpec.workerID),
method_name='run')
tmpOut = tmpRetMap[workSpec.workerID]
oldStatus = tmpOut['oldStatus']
newStatus = tmpOut['newStatus']
monStatus = tmpOut['monStatus']
diagMessage = tmpOut['diagMessage']
workAttributes = tmpOut['workAttributes']
eventsToUpdate = tmpOut['eventsToUpdate']
filesToStageOut = tmpOut['filesToStageOut']
eventsRequestParams = tmpOut['eventsRequestParams']
nJobsToReFill = tmpOut['nJobsToReFill']
pandaIDs = tmpOut['pandaIDs']
isChecked = tmpOut['isChecked']
tmpStr = 'newStatus={0} monitoredStatus={1} diag={2} '
tmpStr += 'postProcessed={3} files={4}'
tmpLog.debug(
tmpStr.format(newStatus, monStatus, diagMessage,
workSpec.is_post_processed(),
str(filesToStageOut)))
iWorker += 1
# check status
if newStatus not in WorkSpec.ST_LIST:
tmpLog.error('unknown status={0}'.format(newStatus))
return
# update worker
workSpec.set_status(newStatus)
workSpec.set_work_attributes(workAttributes)
workSpec.set_dialog_message(diagMessage)
if isChecked:
workSpec.checkTime = datetime.datetime.utcnow()
isCheckedList.append(isChecked)
if monStatus == WorkSpec.ST_failed:
if not workSpec.has_pilot_error():
workSpec.set_pilot_error(
PilotErrors.ERR_GENERALERROR, diagMessage)
elif monStatus == WorkSpec.ST_cancelled:
if not workSpec.has_pilot_error():
workSpec.set_pilot_error(PilotErrors.ERR_PANDAKILL,
diagMessage)
if monStatus in [
WorkSpec.ST_finished, WorkSpec.ST_failed,
WorkSpec.ST_cancelled
]:
workSpec.set_work_params({'finalMonStatus': monStatus})
# request events
if eventsRequestParams != {}:
workSpec.eventsRequest = WorkSpec.EV_requestEvents
workSpec.eventsRequestParams = eventsRequestParams
# jobs to refill
if nJobsToReFill is not None:
workSpec.nJobsToReFill = nJobsToReFill
# get associated jobs for the worker chunk
if workSpec.hasJob == 1 and jobSpecs is None:
jobSpecs = self.dbProxy.get_jobs_with_worker_id(
workSpec.workerID,
None,
only_running=True,
slim=True)
# pandaIDs for push
pandaIDsList.append(pandaIDs)
if len(eventsToUpdate) > 0:
eventsToUpdateList.append(eventsToUpdate)
if len(filesToStageOut) > 0:
filesToStageOutList[
workSpec.workerID] = filesToStageOut
# apfmon status update
if apfmon_status_updates and newStatus != oldStatus:
tmpQueLog.debug(
'apfmon_status_updates: {0} newStatus: {1} monStatus: {2} oldStatus: {3} workSpecStatus: {4}'
.format(apfmon_status_updates, newStatus,
monStatus, oldStatus, workSpec.status))
self.apfmon.update_worker(workSpec, monStatus)
# lock workers for fifo
if from_fifo:
# collect some attributes to be updated when workers are locked
worker_id_list = dict()
for workSpec, isChecked in zip(workSpecs, isCheckedList):
attrs = dict()
if isChecked:
attrs['checkTime'] = workSpec.checkTime
workSpec.force_not_update('checkTime')
if workSpec.has_updated_attributes():
attrs['lockedBy'] = lockedBy
workSpec.lockedBy = lockedBy
workSpec.force_not_update('lockedBy')
else:
attrs['lockedBy'] = None
worker_id_list[workSpec.workerID] = attrs
temRetLockWorker = self.dbProxy.lock_workers(
worker_id_list, harvester_config.monitor.lockInterval)
# skip if not locked
if not temRetLockWorker:
continue
# update jobs and workers
if jobSpecs is not None and len(jobSpecs) > 0:
tmpQueLog.debug(
'updating {0} jobs with {1} workers'.format(
len(jobSpecs), len(workSpecs)))
core_utils.update_job_attributes_with_workers(
mapType, jobSpecs, workSpecs, filesToStageOutList,
eventsToUpdateList)
# update local database
tmpRet = self.dbProxy.update_jobs_workers(
jobSpecs, workSpecs, lockedBy, pandaIDsList)
if not tmpRet:
for workSpec in workSpecs:
tmpLog = self.make_logger(_logger,
'id={0} workerID={1}'.format(
lockedBy,
workSpec.workerID),
method_name='run')
if from_fifo:
tmpLog.info(
'failed to update the DB. Maybe locked by other thread running with DB'
)
else:
if workSpec.status in [
WorkSpec.ST_finished, WorkSpec.ST_failed,
WorkSpec.ST_cancelled, WorkSpec.ST_missed
]:
tmpLog.info(
'worker already in final status. Skipped')
else:
tmpLog.error(
'failed to update the DB. lockInterval may be too short'
)
else:
if jobSpecs is not None:
for jobSpec in jobSpecs:
tmpLog = self.make_logger(
_logger,
'id={0} PandaID={1}'.format(
lockedBy, jobSpec.PandaID),
method_name='run')
tmpLog.debug(
'new status={0} subStatus={1} status_in_metadata={2}'
.format(
jobSpec.status, jobSpec.subStatus,
jobSpec.get_job_status_from_attributes()))
# send ACK to workers for events and files
if len(eventsToUpdateList) > 0 or len(filesToStageOutList) > 0:
for workSpec in workSpecs:
try:
messenger.acknowledge_events_files(workSpec)
except Exception:
core_utils.dump_error_message(tmpQueLog)
tmpQueLog.error(
'failed to send ACK to workerID={0}'.format(
workSpec.workerID))
# active workers for fifo
if self.monitor_fifo.enabled and workSpecs:
workSpec = workSpecs[0]
tmpOut = tmpRetMap[workSpec.workerID]
newStatus = tmpOut['newStatus']
monStatus = tmpOut['monStatus']
if newStatus in [WorkSpec.ST_submitted, WorkSpec.ST_running, WorkSpec.ST_idle] \
and workSpec.mapType != WorkSpec.MT_MultiWorkers \
and workSpec.workAttributes is not None:
timeNow = datetime.datetime.utcnow()
timeNow_timestamp = time.time()
# get lastCheckAt
_bool, lastCheckAt = workSpec.get_work_params(
'lastCheckAt')
try:
last_check_period = timeNow_timestamp - lastCheckAt
except TypeError:
last_check_period = forceEnqueueInterval + 1.0
# get lastForceEnqueueAt
_bool, lastForceEnqueueAt = workSpec.get_work_params(
'lastForceEnqueueAt')
if not (_bool and lastForceEnqueueAt is not None):
lastForceEnqueueAt = 0
# notification
intolerable_delay = max(
forceEnqueueInterval * 2,
harvester_config.monitor.checkInterval * 4)
if _bool and lastCheckAt is not None and last_check_period > harvester_config.monitor.checkInterval \
and timeNow_timestamp - harvester_config.monitor.checkInterval > self.startTimestamp:
if last_check_period > intolerable_delay:
tmpQueLog.error(
'last check period of workerID={0} is {1} sec, intolerably longer than monitor checkInterval. Will NOT enquque worker by force. Please check why monitor checks worker slowly'
.format(workSpec.workerID,
last_check_period))
else:
tmpQueLog.warning(
'last check period of workerID={0} is {1} sec, longer than monitor checkInterval'
.format(workSpec.workerID,
last_check_period))
# prepartion to enqueue fifo
if (from_fifo) \
or (not from_fifo
and timeNow_timestamp - harvester_config.monitor.sleepTime > self.startTimestamp
and last_check_period > forceEnqueueInterval
and last_check_period < intolerable_delay
and timeNow_timestamp - lastForceEnqueueAt > 86400 + forceEnqueueInterval):
if not from_fifo:
# in DB cycle
tmpQueLog.warning(
'last check period of workerID={0} is {1} sec, longer than monitor forceEnqueueInterval. Enqueue the worker by force'
.format(workSpec.workerID,
last_check_period))
workSpec.set_work_params(
{'lastForceEnqueueAt': timeNow_timestamp})
workSpec.set_work_params(
{'lastCheckAt': timeNow_timestamp})
workSpec.lockedBy = None
workSpec.force_update('lockedBy')
if monStatus in [
WorkSpec.ST_finished, WorkSpec.ST_failed,
WorkSpec.ST_cancelled
]:
# for post-processing
_bool, startFifoPreemptAt = workSpec.get_work_params(
'startFifoPreemptAt')
if not _bool or startFifoPreemptAt is None:
startFifoPreemptAt = timeNow_timestamp
workSpec.set_work_params({
'startFifoPreemptAt':
startFifoPreemptAt
})
tmpQueLog.debug(
'workerID={0} , startFifoPreemptAt: {1}'.
format(workSpec.workerID,
startFifoPreemptAt))
if timeNow_timestamp - startFifoPreemptAt < fifoMaxPreemptInterval:
workSpecsToEnqueueToHead_dict[
workSpec.workerID] = workSpecs
else:
workSpec.set_work_params({
'startFifoPreemptAt':
timeNow_timestamp
})
workSpec.modificationTime = timeNow
workSpec.force_update('modificationTime')
workSpecsToEnqueue_dict[
workSpec.workerID] = workSpecs
else:
workSpec.modificationTime = timeNow
workSpec.force_update('modificationTime')
workSpecsToEnqueue_dict[
workSpec.workerID] = workSpecs
else:
tmpQueLog.error('failed to check workers')
workSpecsToEnqueue = list(workSpecsToEnqueue_dict.values())
workSpecsToEnqueueToHead = list(workSpecsToEnqueueToHead_dict.values())
retVal = workSpecsToEnqueue, workSpecsToEnqueueToHead, timeNow_timestamp, fifoCheckInterval
tmpQueLog.debug('done')
return retVal
# wrapper for checkWorkers
def check_workers(self, mon_core, messenger, all_workers, queue_config,
tmp_log, from_fifo):
# check timeout value
try:
checkTimeout = mon_core.checkTimeout
except Exception:
try:
checkTimeout = harvester_config.monitor.checkTimeout
except Exception:
checkTimeout = None
try:
workerQueueTimeLimit = harvester_config.monitor.workerQueueTimeLimit
except AttributeError:
workerQueueTimeLimit = 172800
workersToCheck = []
thingsToPostProcess = []
retMap = dict()
for workSpec in all_workers:
eventsRequestParams = {}
eventsToUpdate = []
pandaIDs = []
workStatus = None
workAttributes = None
filesToStageOut = []
nJobsToReFill = None
if workSpec.has_work_params('finalMonStatus'):
# to post-process
_bool, finalMonStatus = workSpec.get_work_params(
'finalMonStatus')
_thing = (workSpec, (finalMonStatus, ''))
thingsToPostProcess.append(_thing)
else:
# job-level late binding
if workSpec.hasJob == 0 and workSpec.mapType != WorkSpec.MT_NoJob:
# check if job is requested
jobRequested = messenger.job_requested(workSpec)
if jobRequested:
# set ready when job is requested
workStatus = WorkSpec.ST_ready
else:
workStatus = workSpec.status
elif workSpec.nJobsToReFill in [0, None]:
# check if job is requested to refill free slots
jobRequested = messenger.job_requested(workSpec)
if jobRequested:
nJobsToReFill = jobRequested
workersToCheck.append(workSpec)
else:
workersToCheck.append(workSpec)
# add
retMap[workSpec.workerID] = {
'oldStatus': workSpec.status,
'newStatus': workStatus,
'monStatus': workStatus,
'workAttributes': workAttributes,
'filesToStageOut': filesToStageOut,
'eventsRequestParams': eventsRequestParams,
'eventsToUpdate': eventsToUpdate,
'diagMessage': '',
'pandaIDs': pandaIDs,
'nJobsToReFill': nJobsToReFill,
'isChecked': True
}
# check workers
tmp_log.debug('checking workers with plugin')
try:
if workersToCheck:
tmpStat, tmpOut = mon_core.check_workers(workersToCheck)
if not tmpStat:
tmp_log.error(
'failed to check workers with: {0}'.format(tmpOut))
workersToCheck = []
tmpOut = []
else:
tmp_log.debug('checked')
else:
tmp_log.debug('Nothing to be checked with plugin')
tmpOut = []
timeNow = datetime.datetime.utcnow()
for workSpec, (newStatus, diagMessage) in itertools.chain(
zip(workersToCheck, tmpOut), thingsToPostProcess):
workerID = workSpec.workerID
tmp_log.debug('Going to check workerID={0}'.format(workerID))
pandaIDs = []
if workerID in retMap:
# failed to check status
if newStatus is None:
tmp_log.warning(
'Failed to check workerID={0} with {1}'.format(
workerID, diagMessage))
retMap[workerID]['isChecked'] = False
# set status
if workSpec.checkTime is not None and checkTimeout is not None and \
timeNow - workSpec.checkTime > datetime.timedelta(seconds=checkTimeout):
# kill due to timeout
tmp_log.debug(
'kill workerID={0} due to consecutive check failures'
.format(workerID))
self.dbProxy.kill_worker(workSpec.workerID)
newStatus = WorkSpec.ST_cancelled
diagMessage = 'Killed by Harvester due to consecutive worker check failures. ' + diagMessage
workSpec.set_pilot_error(
PilotErrors.ERR_FAILEDBYSERVER, diagMessage)
else:
# use original status
newStatus = workSpec.status
# request kill
if messenger.kill_requested(workSpec):
tmp_log.debug(
'kill workerID={0} as requested'.format(workerID))
self.dbProxy.kill_worker(workSpec.workerID)
# stuck queuing for too long
if workSpec.status == WorkSpec.ST_submitted \
and timeNow > workSpec.submitTime + datetime.timedelta(seconds=workerQueueTimeLimit):
tmp_log.debug(
'kill workerID={0} due to queuing longer than {1} seconds'
.format(workerID, workerQueueTimeLimit))
self.dbProxy.kill_worker(workSpec.workerID)
diagMessage = 'Killed by Harvester due to worker queuing too long' + diagMessage
workSpec.set_pilot_error(
PilotErrors.ERR_FAILEDBYSERVER, diagMessage)
# expired heartbeat - only when requested in the configuration
try:
# check if the queue configuration requires checking for worker heartbeat
worker_heartbeat_limit = int(
queue_config.messenger['worker_heartbeat'])
except (AttributeError, KeyError):
worker_heartbeat_limit = None
tmp_log.debug(
'workerID={0} heartbeat limit is configured to {1}'.
format(workerID, worker_heartbeat_limit))
if worker_heartbeat_limit:
if messenger.is_alive(workSpec,
worker_heartbeat_limit):
tmp_log.debug(
'heartbeat for workerID={0} is valid'.format(
workerID))
else:
tmp_log.debug(
'heartbeat for workerID={0} expired: sending kill request'
.format(workerID))
self.dbProxy.kill_worker(workSpec.workerID)
diagMessage = 'Killed by Harvester due to worker heartbeat expired. ' + diagMessage
workSpec.set_pilot_error(
PilotErrors.ERR_FAILEDBYSERVER, diagMessage)
# get work attributes
workAttributes = messenger.get_work_attributes(workSpec)
retMap[workerID]['workAttributes'] = workAttributes
# get output files
filesToStageOut = messenger.get_files_to_stage_out(
workSpec)
retMap[workerID]['filesToStageOut'] = filesToStageOut
# get events to update
if workSpec.eventsRequest in [
WorkSpec.EV_useEvents, WorkSpec.EV_requestEvents
]:
eventsToUpdate = messenger.events_to_update(workSpec)
retMap[workerID]['eventsToUpdate'] = eventsToUpdate
# request events
if workSpec.eventsRequest == WorkSpec.EV_useEvents:
eventsRequestParams = messenger.events_requested(
workSpec)
retMap[workerID][
'eventsRequestParams'] = eventsRequestParams
# get PandaIDs for pull model
if workSpec.mapType == WorkSpec.MT_NoJob:
pandaIDs = messenger.get_panda_ids(workSpec)
retMap[workerID]['pandaIDs'] = pandaIDs
# keep original new status
retMap[workerID]['monStatus'] = newStatus
# set running or idle while there are events to update or files to stage out
if newStatus in [
WorkSpec.ST_finished, WorkSpec.ST_failed,
WorkSpec.ST_cancelled
]:
if len(retMap[workerID]['filesToStageOut']) > 0 or \
len(retMap[workerID]['eventsToUpdate']) > 0:
if workSpec.status == WorkSpec.ST_running:
newStatus = WorkSpec.ST_running
else:
newStatus = WorkSpec.ST_idle
elif not workSpec.is_post_processed():
if not queue_config.is_no_heartbeat_status(
newStatus):
# post processing unless heartbeat is suppressed
jobSpecs = self.dbProxy.get_jobs_with_worker_id(
workSpec.workerID,
None,
True,
only_running=True,
slim=True)
# post processing
messenger.post_processing(
workSpec, jobSpecs, workSpec.mapType)
workSpec.post_processed()
if workSpec.status == WorkSpec.ST_running:
newStatus = WorkSpec.ST_running
else:
newStatus = WorkSpec.ST_idle
# reset modification time to immediately trigger subsequent lookup
if not self.monitor_fifo.enabled:
workSpec.trigger_next_lookup()
retMap[workerID]['newStatus'] = newStatus
retMap[workerID]['diagMessage'] = diagMessage
else:
tmp_log.debug(
'workerID={0} not in retMap'.format(workerID))
return True, retMap
except Exception:
core_utils.dump_error_message(tmp_log)
return False, None
class Propagator(AgentBase):
# constructor
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.communicator = communicator
self.queueConfigMapper = queue_config_mapper
self._last_stats_update = None
self._last_metrics_update = None
# main loop
def run(self):
while True:
sw_main = core_utils.get_stopwatch()
mainLog = self.make_logger(_logger, 'id={0}'.format(self.get_pid()), method_name='run')
mainLog.debug('getting jobs to propagate')
sw = core_utils.get_stopwatch()
jobSpecs = self.dbProxy.get_jobs_to_propagate(harvester_config.propagator.maxJobs,
harvester_config.propagator.lockInterval,
harvester_config.propagator.updateInterval,
self.get_pid())
mainLog.debug('got {0} jobs {1}'.format(len(jobSpecs), sw.get_elapsed_time()))
# update jobs in central database
iJobs = 0
nJobs = harvester_config.propagator.nJobsInBulk
hbSuppressMap = dict()
while iJobs < len(jobSpecs):
jobList = jobSpecs[iJobs:iJobs + nJobs]
iJobs += nJobs
# collect jobs to update or check
jobListToSkip = []
jobListToUpdate = []
jobListToCheck = []
retList = []
for tmpJobSpec in jobList:
if tmpJobSpec.computingSite not in hbSuppressMap:
queueConfig = self.queueConfigMapper.get_queue(tmpJobSpec.computingSite,
tmpJobSpec.configID)
hbSuppressMap[tmpJobSpec.computingSite] = queueConfig.get_no_heartbeat_status()
# heartbeat is suppressed
if tmpJobSpec.get_status() in hbSuppressMap[tmpJobSpec.computingSite] and \
not tmpJobSpec.not_suppress_heartbeat():
# check running job to detect lost heartbeat
if tmpJobSpec.status == 'running':
jobListToCheck.append(tmpJobSpec)
else:
jobListToSkip.append(tmpJobSpec)
retList.append({'StatusCode': 0, 'command': None})
else:
jobListToUpdate.append(tmpJobSpec)
sw.reset()
retList += self.communicator.check_jobs(jobListToCheck)
mainLog.debug('check_jobs for {0} jobs {1}'.format(len(jobListToCheck), sw.get_elapsed_time()))
sw.reset()
retList += self.communicator.update_jobs(jobListToUpdate, self.get_pid())
mainLog.debug('update_jobs for {0} jobs took {1}'.format(len(jobListToUpdate),
sw.get_elapsed_time()))
# logging
for tmpJobSpec, tmpRet in zip(jobListToSkip+jobListToCheck+jobListToUpdate, retList):
if tmpRet['StatusCode'] == 0:
if tmpJobSpec in jobListToUpdate:
mainLog.debug('updated PandaID={0} status={1}'.format(tmpJobSpec.PandaID,
tmpJobSpec.status))
else:
mainLog.debug('skip updating PandaID={0} status={1}'.format(tmpJobSpec.PandaID,
tmpJobSpec.status))
# release job
tmpJobSpec.propagatorLock = None
if tmpJobSpec.is_final_status() and tmpJobSpec.status == tmpJobSpec.get_status():
# unset to disable further updating
tmpJobSpec.propagatorTime = None
tmpJobSpec.subStatus = 'done'
tmpJobSpec.modificationTime = datetime.datetime.utcnow()
elif tmpJobSpec.is_final_status() and not tmpJobSpec.all_events_done():
# trigger next propagation to update remaining events
tmpJobSpec.trigger_propagation()
else:
# check event availability
if tmpJobSpec.status == 'starting' and 'eventService' in tmpJobSpec.jobParams and \
tmpJobSpec.subStatus != 'submitted':
tmpEvStat, tmpEvRet = self.communicator.check_event_availability(tmpJobSpec)
if tmpEvStat:
if tmpEvRet is not None:
tmpJobSpec.nRemainingEvents = tmpEvRet
if tmpEvRet == 0:
mainLog.debug('kill PandaID={0} due to no event'.format(tmpJobSpec.PandaID))
tmpRet['command'] = 'tobekilled'
# got kill command
if 'command' in tmpRet and tmpRet['command'] in ['tobekilled']:
nWorkers = self.dbProxy.kill_workers_with_job(tmpJobSpec.PandaID)
if nWorkers == 0:
# no workers
tmpJobSpec.status = 'cancelled'
tmpJobSpec.subStatus = 'killed'
tmpJobSpec.set_pilot_error(PilotErrors.ERR_PANDAKILL,
PilotErrors.pilotError[PilotErrors.ERR_PANDAKILL])
tmpJobSpec.stateChangeTime = datetime.datetime.utcnow()
tmpJobSpec.trigger_propagation()
self.dbProxy.update_job(tmpJobSpec, {'propagatorLock': self.get_pid()})
else:
mainLog.error('failed to update PandaID={0} status={1}'.format(tmpJobSpec.PandaID,
tmpJobSpec.status))
mainLog.debug('getting workers to propagate')
sw.reset()
workSpecs = self.dbProxy.get_workers_to_propagate(harvester_config.propagator.maxWorkers,
harvester_config.propagator.updateInterval)
mainLog.debug('got {0} workers {1}'.format(len(workSpecs), sw.get_elapsed_time()))
# update workers in central database
sw.reset()
iWorkers = 0
nWorkers = harvester_config.propagator.nWorkersInBulk
while iWorkers < len(workSpecs):
workList = workSpecs[iWorkers:iWorkers + nWorkers]
iWorkers += nWorkers
retList, tmpErrStr = self.communicator.update_workers(workList)
# logging
if retList is None:
mainLog.error('failed to update workers with {0}'.format(tmpErrStr))
else:
for tmpWorkSpec, tmpRet in zip(workList, retList):
if tmpRet:
mainLog.debug('updated workerID={0} status={1}'.format(tmpWorkSpec.workerID,
tmpWorkSpec.status))
# update logs
for logFilePath, logOffset, logSize, logRemoteName in \
tmpWorkSpec.get_log_files_to_upload():
with open(logFilePath, 'rb') as logFileObj:
tmpStat, tmpErr = self.communicator.upload_file(logRemoteName, logFileObj,
logOffset, logSize)
if tmpStat:
tmpWorkSpec.update_log_files_to_upload(logFilePath, logOffset+logSize)
# disable further update
if tmpWorkSpec.is_final_status():
tmpWorkSpec.disable_propagation()
self.dbProxy.update_worker(tmpWorkSpec, {'workerID': tmpWorkSpec.workerID})
else:
mainLog.error('failed to update workerID={0} status={1}'.format(tmpWorkSpec.workerID,
tmpWorkSpec.status))
mainLog.debug('update_workers for {0} workers took {1}'.format(iWorkers,
sw.get_elapsed_time()))
mainLog.debug('getting commands')
commandSpecs = self.dbProxy.get_commands_for_receiver('propagator')
mainLog.debug('got {0} commands'.format(len(commandSpecs)))
for commandSpec in commandSpecs:
if commandSpec.command.startswith(CommandSpec.COM_reportWorkerStats):
# get worker stats
siteName = commandSpec.command.split(':')[-1]
workerStats = self.dbProxy.get_worker_stats(siteName)
if len(workerStats) == 0:
mainLog.error('failed to get worker stats for {0}'.format(siteName))
else:
# report worker stats
tmpRet, tmpStr = self.communicator.update_worker_stats(siteName, workerStats)
if tmpRet:
mainLog.debug('updated worker stats (command) for {0}'.format(siteName))
else:
mainLog.error('failed to update worker stats (command) for {0} err={1}'.format(siteName,
tmpStr))
if not self._last_stats_update or time.time() - self._last_stats_update > STATS_PERIOD:
# get active UPS queues. PanDA server needs to know about them and which harvester instance is taking
# care of them
active_ups_queues = self.queueConfigMapper.get_active_ups_queues()
# update worker stats for all sites
worker_stats_bulk = self.dbProxy.get_worker_stats_bulk(active_ups_queues)
if not worker_stats_bulk:
mainLog.error('failed to get worker stats in bulk')
else:
for site_name in worker_stats_bulk:
tmp_ret, tmp_str = self.communicator.update_worker_stats(site_name,
worker_stats_bulk[site_name])
if tmp_ret:
mainLog.debug('update of worker stats (bulk) for {0}'.format(site_name))
self._last_stats_update = time.time()
else:
mainLog.error('failed to update worker stats (bulk) for {0} err={1}'.format(site_name,
tmp_str))
if not self._last_metrics_update \
or datetime.datetime.utcnow() - self._last_metrics_update > datetime.timedelta(seconds=METRICS_PERIOD):
# get latest metrics from DB
service_metrics_list = self.dbProxy.get_service_metrics(self._last_metrics_update)
if not service_metrics_list:
mainLog.error('failed to get service metrics')
self._last_metrics_update = datetime.datetime.utcnow()
else:
tmp_ret, tmp_str = self.communicator.update_service_metrics(service_metrics_list)
if tmp_ret:
mainLog.debug('update of service metrics OK')
self._last_metrics_update = datetime.datetime.utcnow()
else:
mainLog.error('failed to update service metrics err={0}'.format(tmp_str))
# send dialog messages
mainLog.debug('getting dialog messages to propagate')
try:
maxDialogs = harvester_config.propagator.maxDialogs
except Exception:
maxDialogs = 50
diagSpecs = self.dbProxy.get_dialog_messages_to_send(maxDialogs,
harvester_config.propagator.lockInterval)
mainLog.debug('got {0} dialogs'.format(len(diagSpecs)))
if len(diagSpecs) > 0:
tmpStat, tmpStr = self.communicator.send_dialog_messages(diagSpecs)
if tmpStat:
diagIDs = [diagSpec.diagID for diagSpec in diagSpecs]
self.dbProxy.delete_dialog_messages(diagIDs)
mainLog.debug('sent {0} dialogs'.format(len(diagSpecs)))
else:
mainLog.error('failed to send dialogs err={0}'.format(tmpStr))
if sw_main.get_elapsed_time_in_sec() > harvester_config.propagator.lockInterval:
mainLog.warning('a single cycle was longer than lockInterval. done' + sw_main.get_elapsed_time())
else:
mainLog.debug('done' + sw_main.get_elapsed_time())
# check if being terminated
if self.terminated(harvester_config.propagator.sleepTime):
mainLog.debug('terminated')
return
class CredManager(AgentBase):
# constructor
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.queue_config_mapper = queue_config_mapper
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
# plugin cores
self.exeCores = []
self.queue_exe_cores = []
# get plugin from harvester config
self.get_cores_from_harvester_config()
# update plugin cores from queue config
self.update_cores_from_queue_config()
# get list
def get_list(self, data):
if isinstance(data, list):
return data
else:
return [data]
# get plugin cores from harvester config
def get_cores_from_harvester_config(self):
# get module and class names
if hasattr(harvester_config.credmanager, 'moduleName'):
moduleNames = self.get_list(
harvester_config.credmanager.moduleName)
else:
moduleNames = []
if hasattr(harvester_config.credmanager, 'className'):
classNames = self.get_list(harvester_config.credmanager.className)
else:
classNames = []
# file names of original certificates
if hasattr(harvester_config.credmanager, 'inCertFile'):
inCertFiles = self.get_list(
harvester_config.credmanager.inCertFile)
elif hasattr(harvester_config.credmanager, 'certFile'):
inCertFiles = self.get_list(harvester_config.credmanager.certFile)
else:
inCertFiles = []
# file names of certificates to be generated
if hasattr(harvester_config.credmanager, 'outCertFile'):
outCertFiles = self.get_list(
harvester_config.credmanager.outCertFile)
else:
# use the file name of the certificate for panda connection as output name
outCertFiles = self.get_list(harvester_config.pandacon.cert_file)
# VOMS
if hasattr(harvester_config.credmanager, 'voms'):
vomses = self.get_list(harvester_config.credmanager.voms)
else:
vomses = []
# direct and merged plugin configuration in json
if hasattr(harvester_config.credmanager, 'pluginConfigs'):
pluginConfigs = harvester_config.credmanager.pluginConfigs
else:
pluginConfigs = []
# from traditional attributes
for moduleName, className, inCertFile, outCertFile, voms in \
zip(moduleNames, classNames, inCertFiles, outCertFiles, vomses):
pluginPar = {}
pluginPar['module'] = moduleName
pluginPar['name'] = className
pluginPar['inCertFile'] = inCertFile
pluginPar['outCertFile'] = outCertFile
pluginPar['voms'] = voms
try:
exeCore = self.pluginFactory.get_plugin(pluginPar)
self.exeCores.append(exeCore)
except Exception:
_logger.error(
'failed to launch credmanager with traditional attributes for {0}'
.format(pluginPar))
core_utils.dump_error_message(_logger)
# from pluginConfigs
for pc in pluginConfigs:
try:
setup_maps = pc['configs']
for setup_name, setup_map in setup_maps.items():
try:
pluginPar = {}
pluginPar['module'] = pc['module']
pluginPar['name'] = pc['name']
pluginPar['setup_name'] = setup_name
pluginPar.update(setup_map)
exeCore = self.pluginFactory.get_plugin(pluginPar)
self.exeCores.append(exeCore)
except Exception:
_logger.error(
'failed to launch credmanager in pluginConfigs for {0}'
.format(pluginPar))
core_utils.dump_error_message(_logger)
except Exception:
_logger.error('failed to parse pluginConfigs {0}'.format(pc))
core_utils.dump_error_message(_logger)
# update plugin cores from queue config
def update_cores_from_queue_config(self):
self.queue_exe_cores = []
for queue_name, queue_config in self.queue_config_mapper.get_all_queues(
).items():
if queue_config.queueStatus == 'offline' \
or not hasattr(queue_config, 'credmanagers') \
or not isinstance(queue_config.credmanagers, list):
continue
for cm_setup in queue_config.credmanagers:
try:
pluginPar = {}
pluginPar['module'] = cm_setup['module']
pluginPar['name'] = cm_setup['name']
pluginPar['setup_name'] = queue_name
for k, v in cm_setup.items():
if k in ('module', 'name'):
pass
if isinstance(v, str) and '$' in v:
# replace placeholders
value = v
patts = re.findall('\$\{([a-zA-Z\d_.]+)\}', v)
for patt in patts:
tmp_ph = '${' + patt + '}'
tmp_val = None
if patt == 'harvesterID':
tmp_val = harvester_config.master.harvester_id
elif patt == 'queueName':
tmp_val = queue_name
elif patt.startswith('common.'):
# values from common blocks
attr = patt.replace('common.', '')
if hasattr(
queue_config, 'common'
) and attr in queue_config.common:
tmp_val = queue_config.common[attr]
if tmp_val is not None:
value = value.replace(tmp_ph, tmp_val)
# fill in
pluginPar[k] = value
else:
# fill in
pluginPar[k] = v
exe_core = self.pluginFactory.get_plugin(pluginPar)
self.queue_exe_cores.append(exe_core)
except Exception:
_logger.error(
'failed to launch about queue={0} for {1}'.format(
queue_name, pluginPar))
core_utils.dump_error_message(_logger)
# main loop
def run(self):
while True:
# update plugin cores from queue config
self.update_cores_from_queue_config()
# execute
self.execute() # this is the main run
# check if being terminated
if self.terminated(harvester_config.credmanager.sleepTime,
randomize=False):
return
# main
def execute(self):
# get lock
locked = self.dbProxy.get_process_lock(
'credmanager', self.get_pid(),
harvester_config.credmanager.sleepTime)
if not locked:
return
# loop over all plugins
for exeCore in itertools.chain(self.exeCores, self.queue_exe_cores):
# do nothing
if exeCore is None:
continue
# make logger
credmanager_name = ''
if hasattr(exeCore, 'setup_name'):
credmanager_name = exeCore.setup_name
else:
credmanager_name = '{0} {1}'.format(exeCore.inCertFile,
exeCore.outCertFile)
mainLog = self.make_logger(_logger,
'{0} {1}'.format(
exeCore.__class__.__name__,
credmanager_name),
method_name='execute')
try:
# check credential
mainLog.debug('check credential')
isValid = exeCore.check_credential()
if isValid:
mainLog.debug('valid')
elif not isValid:
# renew it if necessary
mainLog.debug('invalid')
mainLog.debug('renew credential')
tmpStat, tmpOut = exeCore.renew_credential()
if not tmpStat:
mainLog.error('failed : {0}'.format(tmpOut))
continue
except Exception:
core_utils.dump_error_message(mainLog)
mainLog.debug('done')
# monit main
def execute_monit(self):
self.update_cores_from_queue_config()
metrics = {}
# loop over all plugins
for exeCore in itertools.chain(self.exeCores, self.queue_exe_cores):
# do nothing
if exeCore is None:
continue
# make logger
if hasattr(exeCore, 'setup_name'):
credmanager_name = exeCore.setup_name
else:
credmanager_name = '{0} {1}'.format(exeCore.inCertFile,
exeCore.outCertFile)
subLog = self.make_logger(_logger,
'{0} {1}'.format(
exeCore.__class__.__name__,
credmanager_name),
method_name='execute_monit')
try:
# check credential
subLog.debug('check credential lifetime')
lifetime = exeCore.check_credential_lifetime()
if lifetime is not None:
metrics[exeCore.outCertFile] = lifetime
except Exception:
core_utils.dump_error_message(subLog)
subLog.debug('done')
return metrics
class WorkerAdjuster(object):
# constructor
def __init__(self, queue_config_mapper):
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
self.throttlerMap = dict()
self.apf_mon = Apfmon(self.queueConfigMapper)
try:
self.maxNewWorkers = harvester_config.submitter.maxNewWorkers
except AttributeError:
self.maxNewWorkers = None
# define number of workers to submit based on various information
def define_num_workers(self, static_num_workers, site_name):
tmpLog = core_utils.make_logger(_logger, 'site={0}'.format(site_name), method_name='define_num_workers')
tmpLog.debug('start')
tmpLog.debug('static_num_workers: {0}'.format(static_num_workers))
dyn_num_workers = copy.deepcopy(static_num_workers)
try:
# get queue status
queueStat = self.dbProxy.get_cache("panda_queues.json", None)
if queueStat is None:
queueStat = dict()
else:
queueStat = queueStat.data
# get job statistics
job_stats = self.dbProxy.get_cache("job_statistics.json", None)
if job_stats is None:
job_stats = dict()
else:
job_stats = job_stats.data
# define num of new workers
for queueName in static_num_workers:
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
workerLimits_dict = self.dbProxy.get_worker_limits(queueName)
maxWorkers = workerLimits_dict.get('maxWorkers', 0)
nQueueLimit = workerLimits_dict.get('nQueueLimitWorker', 0)
nQueueLimitPerRT = workerLimits_dict['nQueueLimitWorkerPerRT']
nQueue_total, nReady_total, nRunning_total = 0, 0, 0
apf_msg = None
apf_data = None
for resource_type, tmpVal in iteritems(static_num_workers[queueName]):
tmpLog.debug('Processing queue {0} resource {1} with static_num_workers {2}'.
format(queueName, resource_type, tmpVal))
# set 0 to num of new workers when the queue is disabled
if queueName in queueStat and queueStat[queueName]['status'] in ['offline', 'standby',
'maintenance']:
dyn_num_workers[queueName][resource_type]['nNewWorkers'] = 0
retMsg = 'set nNewWorkers=0 since status={0}'.format(queueStat[queueName]['status'])
tmpLog.debug(retMsg)
apf_msg = 'Not submitting workers since queue status = {0}'.format(queueStat[queueName]['status'])
continue
# protection against not-up-to-date queue config
if queueConfig is None:
dyn_num_workers[queueName][resource_type]['nNewWorkers'] = 0
retMsg = 'set nNewWorkers=0 due to missing queueConfig'
tmpLog.debug(retMsg)
apf_msg = 'Not submitting workers because of missing queueConfig'
continue
# get throttler
if queueName not in self.throttlerMap:
if hasattr(queueConfig, 'throttler'):
throttler = self.pluginFactory.get_plugin(queueConfig.throttler)
else:
throttler = None
self.throttlerMap[queueName] = throttler
# check throttler
throttler = self.throttlerMap[queueName]
if throttler is not None:
toThrottle, tmpMsg = throttler.to_be_throttled(queueConfig)
if toThrottle:
dyn_num_workers[queueName][resource_type]['nNewWorkers'] = 0
retMsg = 'set nNewWorkers=0 by {0}:{1}'.format(throttler.__class__.__name__, tmpMsg)
tmpLog.debug(retMsg)
continue
# check stats
nQueue = tmpVal['nQueue']
nReady = tmpVal['nReady']
nRunning = tmpVal['nRunning']
if resource_type != 'ANY':
nQueue_total += nQueue
nReady_total += nReady
nRunning_total += nRunning
if queueConfig.runMode == 'slave':
nNewWorkersDef = tmpVal['nNewWorkers']
if nNewWorkersDef == 0:
dyn_num_workers[queueName][resource_type]['nNewWorkers'] = 0
retMsg = 'set nNewWorkers=0 by panda in slave mode'
tmpLog.debug(retMsg)
continue
else:
nNewWorkersDef = None
# define num of new workers based on static site config
nNewWorkers = 0
if nQueue >= nQueueLimitPerRT > 0:
# enough queued workers
retMsg = 'No nNewWorkers since nQueue({0})>=nQueueLimitPerRT({1})'.format(nQueue, nQueueLimitPerRT)
tmpLog.debug(retMsg)
pass
elif (nQueue + nReady + nRunning) >= maxWorkers > 0:
# enough workers in the system
retMsg = 'No nNewWorkers since nQueue({0}) + nReady({1}) + nRunning({2}) '.format(nQueue,
nReady,
nRunning)
retMsg += '>= maxWorkers({0})'.format(maxWorkers)
tmpLog.debug(retMsg)
pass
else:
maxQueuedWorkers = None
if nQueueLimitPerRT > 0: # there is a limit set for the queue
maxQueuedWorkers = nQueueLimitPerRT
# Reset the maxQueueWorkers according to particular
if nNewWorkersDef is not None: # don't surpass limits given centrally
maxQueuedWorkers_slave = nNewWorkersDef + nQueue
if maxQueuedWorkers is not None:
maxQueuedWorkers = min(maxQueuedWorkers_slave, maxQueuedWorkers)
else:
maxQueuedWorkers = maxQueuedWorkers_slave
elif queueConfig.mapType == 'NoJob': # for pull mode, limit to activated jobs
# limit the queue to the number of activated jobs to avoid empty pilots
try:
n_activated = max(job_stats[queueName]['activated'], 1) # avoid no activity queues
queue_limit = maxQueuedWorkers
maxQueuedWorkers = min(n_activated, maxQueuedWorkers)
tmpLog.debug('limiting maxQueuedWorkers to min(n_activated={0}, queue_limit={1})'.
format(n_activated, queue_limit))
except KeyError:
tmpLog.warning('n_activated not defined, defaulting to configured queue limits')
pass
if maxQueuedWorkers is None: # no value found, use default value
maxQueuedWorkers = 1
# new workers
nNewWorkers = max(maxQueuedWorkers - nQueue, 0)
tmpLog.debug('setting nNewWorkers to {0} in maxQueuedWorkers calculation'
.format(nNewWorkers))
if maxWorkers > 0:
nNewWorkers = min(nNewWorkers, max(maxWorkers - nQueue - nReady - nRunning, 0))
tmpLog.debug('setting nNewWorkers to {0} to respect maxWorkers'
.format(nNewWorkers))
if queueConfig.maxNewWorkersPerCycle > 0:
nNewWorkers = min(nNewWorkers, queueConfig.maxNewWorkersPerCycle)
tmpLog.debug('setting nNewWorkers to {0} in order to respect maxNewWorkersPerCycle'
.format(nNewWorkers))
if self.maxNewWorkers is not None and self.maxNewWorkers > 0:
nNewWorkers = min(nNewWorkers, self.maxNewWorkers)
tmpLog.debug('setting nNewWorkers to {0} in order to respect universal maxNewWorkers'
.format(nNewWorkers))
dyn_num_workers[queueName][resource_type]['nNewWorkers'] = nNewWorkers
# adjust nNewWorkers for UCORE to let aggregations over RT respect nQueueLimitWorker and maxWorkers
if queueConfig is None:
maxNewWorkersPerCycle = 0
retMsg = 'set maxNewWorkersPerCycle=0 in UCORE aggregation due to missing queueConfig'
tmpLog.debug(retMsg)
else:
maxNewWorkersPerCycle = queueConfig.maxNewWorkersPerCycle
if len(dyn_num_workers[queueName]) > 1:
total_new_workers_rts = sum( dyn_num_workers[queueName][_rt]['nNewWorkers']
if _rt != 'ANY' else 0
for _rt in dyn_num_workers[queueName] )
nNewWorkers_max_agg = min(
max(nQueueLimit - nQueue_total, 0),
max(maxWorkers - nQueue_total - nReady_total - nRunning_total, 0),
)
if maxNewWorkersPerCycle >= 0:
nNewWorkers_max_agg = min(nNewWorkers_max_agg, maxNewWorkersPerCycle)
if self.maxNewWorkers is not None and self.maxNewWorkers > 0:
nNewWorkers_max_agg = min(nNewWorkers_max_agg, self.maxNewWorkers)
# exceeded max, to adjust
if total_new_workers_rts > nNewWorkers_max_agg:
if nNewWorkers_max_agg == 0:
for resource_type in dyn_num_workers[queueName]:
dyn_num_workers[queueName][resource_type]['nNewWorkers'] = 0
tmpLog.debug('No nNewWorkers since nNewWorkers_max_agg=0 for UCORE')
else:
tmpLog.debug('nNewWorkers_max_agg={0} for UCORE'.format(nNewWorkers_max_agg))
_d = dyn_num_workers[queueName].copy()
del _d['ANY']
simple_rt_nw_list = [ [_rt, _d[_rt].get('nNewWorkers', 0), 0] for _rt in _d ]
_countdown = nNewWorkers_max_agg
for _rt_list in simple_rt_nw_list:
resource_type, nNewWorkers_orig, _r = _rt_list
nNewWorkers, remainder = divmod(nNewWorkers_orig*nNewWorkers_max_agg, total_new_workers_rts)
dyn_num_workers[queueName][resource_type]['nNewWorkers'] = nNewWorkers
_rt_list[2] = remainder
_countdown -= nNewWorkers
_s_list = sorted(simple_rt_nw_list, key=(lambda x: x[1]))
sorted_rt_nw_list = sorted(_s_list, key=(lambda x: x[2]), reverse=True)
for resource_type, nNewWorkers_orig, remainder in sorted_rt_nw_list:
if _countdown <= 0:
break
dyn_num_workers[queueName][resource_type]['nNewWorkers'] += 1
_countdown -= 1
for resource_type in dyn_num_workers[queueName]:
if resource_type == 'ANY':
continue
nNewWorkers = dyn_num_workers[queueName][resource_type]['nNewWorkers']
tmpLog.debug('setting nNewWorkers to {0} of type {1} in order to respect RT aggregations for UCORE'
.format(nNewWorkers, resource_type))
if not apf_msg:
apf_data = copy.deepcopy(dyn_num_workers[queueName])
self.apf_mon.update_label(queueName, apf_msg, apf_data)
# dump
tmpLog.debug('defined {0}'.format(str(dyn_num_workers)))
return dyn_num_workers
except Exception:
# dump error
errMsg = core_utils.dump_error_message(tmpLog)
return None
os.remove(harvester_config.db.database_filename)
except Exception:
pass
for loggerName, loggerObj in iteritems(logging.Logger.manager.loggerDict):
if loggerName.startswith('panda.log'):
if len(loggerObj.handlers) == 0:
continue
if loggerName.split('.')[-1] in ['db_proxy']:
continue
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
queueConfigMapper = QueueConfigMapper()
proxy = DBProxy()
proxy.make_tables(queueConfigMapper)
job = JobSpec()
job.PandaID = 1
job.modificationTime = datetime.datetime.now()
proxy.insert_jobs([job])
newJob = proxy.get_job(1)
a = CommunicatorPool()
a.get_jobs('siteName', 'nodeName', 'prodSourceLabel', 'computingElement', 1,
{})
class CommandManager(AgentBase):
# constructor
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.db_proxy = DBProxy()
self.communicator = communicator
self.queueConfigMapper = queue_config_mapper
self.nodeName = socket.gethostname()
self.lastHeartbeat = None
# set single mode
def set_single_mode(self, single_mode):
self.singleMode = single_mode
def convert_to_command_specs(self, commands):
"""
Generates a list of CommandSpec objects
"""
command_specs = []
for command in commands:
command_spec = CommandSpec()
command_spec.convert_command_json(command)
for comStr, receiver in iteritems(CommandSpec.receiver_map):
if command_spec.command.startswith(comStr):
command_spec.receiver = receiver
break
if command_spec.receiver is not None:
command_specs.append(command_spec)
return command_specs
def run(self):
"""
main
"""
main_log = self.make_logger(_logger, 'id={0}'.format(self.get_pid()), method_name='run')
bulk_size = harvester_config.commandmanager.commands_bulk_size
locked = self.db_proxy.get_process_lock('commandmanager', self.get_pid(),
harvester_config.commandmanager.sleepTime)
if locked:
# send command list to be received
siteNames = set()
commandList = []
for queueName, queueConfig in iteritems(self.queueConfigMapper.get_active_queues()):
if queueConfig is None or queueConfig.runMode != 'slave':
continue
# one command for all queues in one site
if queueConfig.siteName not in siteNames:
commandItem = {'command': CommandSpec.COM_reportWorkerStats,
'computingSite': queueConfig.siteName,
'resourceType': queueConfig.resourceType
}
commandList.append(commandItem)
siteNames.add(queueConfig.siteName)
# one command for each queue
commandItem = {'command': CommandSpec.COM_setNWorkers,
'computingSite': queueConfig.siteName,
'resourceType': queueConfig.resourceType
}
commandList.append(commandItem)
data = {'startTime': datetime.datetime.utcnow(),
'sw_version': panda_pkg_info.release_version,
'commit_stamp': commit_timestamp.timestamp}
if len(commandList) > 0:
main_log.debug('sending command list to receive')
data['commands'] = commandList
self.communicator.is_alive(data)
# main loop
while True:
# get lock
locked = self.db_proxy.get_process_lock('commandmanager', self.get_pid(),
harvester_config.commandmanager.sleepTime)
if locked or self.singleMode:
main_log.debug('polling commands loop')
# send heartbeat
if self.lastHeartbeat is None \
or self.lastHeartbeat < datetime.datetime.utcnow() - datetime.timedelta(minutes=10):
self.lastHeartbeat = datetime.datetime.utcnow()
self.communicator.is_alive({})
continuous_loop = True # as long as there are commands, retrieve them
while continuous_loop:
# get commands from panda server for this harvester instance
commands = self.communicator.get_commands(bulk_size)
main_log.debug('got {0} commands (bulk size: {1})'.format(len(commands), bulk_size))
command_specs = self.convert_to_command_specs(commands)
# cache commands in internal DB
self.db_proxy.store_commands(command_specs)
main_log.debug('cached {0} commands in internal DB'.format(len(command_specs)))
# retrieve processed commands from harvester cache
command_ids_ack = self.db_proxy.get_commands_ack()
for shard in core_utils.create_shards(command_ids_ack, bulk_size):
# post acknowledgements to panda server
self.communicator.ack_commands(shard)
main_log.debug('acknowledged {0} commands to panda server'.format(len(shard)))
# clean acknowledged commands
self.db_proxy.clean_commands_by_id(shard)
# clean commands that have been processed and do not need acknowledgement
self.db_proxy.clean_processed_commands()
# if we didn't collect the full bulk, give panda server a break
if len(commands) < bulk_size:
continuous_loop = False
# check if being terminated
if self.terminated(harvester_config.commandmanager.sleepTime, randomize=False):
main_log.debug('terminated')
return
class EventFeeder(AgentBase):
# constructor
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.queueConfigMapper = queue_config_mapper
self.communicator = communicator
self.pluginFactory = PluginFactory()
# main loop
def run(self):
lockedBy = 'eventfeeder-{0}'.format(self.get_pid())
while True:
mainLog = self.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
mainLog.debug('getting workers to feed events')
workSpecsPerQueue = self.dbProxy.get_workers_to_feed_events(harvester_config.eventfeeder.maxWorkers,
harvester_config.eventfeeder.lockInterval,
lockedBy)
mainLog.debug('got {0} queues'.format(len(workSpecsPerQueue)))
# loop over all workers
for queueName, workSpecList in iteritems(workSpecsPerQueue):
tmpQueLog = self.make_logger(_logger, 'queue={0}'.format(queueName), method_name='run')
# check queue
if not self.queueConfigMapper.has_queue(queueName):
tmpQueLog.error('config not found')
continue
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
if hasattr(queueConfig, 'scatteredEvents') and queueConfig.scatteredEvents:
scattered = True
else:
scattered = False
# get plugin
messenger = self.pluginFactory.get_plugin(queueConfig.messenger)
# loop over all workers
for workSpec in workSpecList:
tmpLog = core_utils.make_logger(_logger, 'workerID={0}'.format(workSpec.workerID),
method_name='run')
# lock worker again
lockedFlag = self.dbProxy.lock_worker_again_to_feed_events(workSpec.workerID, lockedBy)
if not lockedFlag:
tmpLog.debug('skipped since locked by another')
continue
# get events
tmpLog.debug('get events')
tmpStat, events = self.communicator.get_event_ranges(workSpec.eventsRequestParams,
scattered)
# failed
if tmpStat is False:
tmpLog.error('failed to get events with {0}'.format(events))
continue
# lock worker again
lockedFlag = self.dbProxy.lock_worker_again_to_feed_events(workSpec.workerID, lockedBy)
if not lockedFlag:
tmpLog.debug('skipped before feeding since locked by another')
continue
tmpStat = messenger.feed_events(workSpec, events)
# failed
if tmpStat is False:
tmpLog.error('failed to feed events')
continue
# dump
for pandaID, eventList in iteritems(events):
try:
nRanges = workSpec.eventsRequestParams[pandaID]['nRanges']
except Exception:
nRanges = None
tmpLog.debug('got {0} events for PandaID={1} while getting {2} events'.format(len(eventList),
pandaID,
nRanges))
# disable multi workers
if workSpec.mapType == WorkSpec.MT_MultiWorkers:
if len(eventList) == 0 or (nRanges is not None and len(eventList) < nRanges):
tmpStat = self.dbProxy.disable_multi_workers(pandaID)
if tmpStat == 1:
tmpStr = 'disabled MultiWorkers for PandaID={0}'.format(pandaID)
tmpLog.debug(tmpStr)
# update worker
workSpec.eventsRequest = WorkSpec.EV_useEvents
workSpec.eventsRequestParams = None
workSpec.eventFeedTime = None
workSpec.eventFeedLock = None
# update local database
tmpStat = self.dbProxy.update_worker(workSpec, {'eventFeedLock': lockedBy})
tmpLog.debug('done with {0}'.format(tmpStat))
tmpQueLog.debug('done')
mainLog.debug('done')
# check if being terminated
if self.terminated(harvester_config.eventfeeder.sleepTime):
mainLog.debug('terminated')
return
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
for loggerName, loggerObj in iteritems(logging.Logger.manager.loggerDict):
if loggerName.startswith('panda.log'):
if len(loggerObj.handlers) == 0:
continue
if loggerName.split('.')[-1] in ['db_proxy']:
continue
stdoutHandler = logging.StreamHandler(sys.stdout)
stdoutHandler.setFormatter(loggerObj.handlers[0].formatter)
loggerObj.addHandler(stdoutHandler)
pp = pprint.PrettyPrinter(indent=4)
queueConfigMapper = QueueConfigMapper()
proxy = DBProxy()
sqlJ = "SELECT * FROM job_table"
resultsJobcur = proxy.execute(sqlJ)
resultsJob = resultsJobcur.fetchall()
proxy.commit()
sqlF = "SELECT * FROM file_table"
resultsFilescur = proxy.execute(sqlF)
resultsFiles = resultsFilescur.fetchall()
proxy.commit()
print "job_table - "
print resultsJob[0].keys()
class Sweeper(AgentBase):
# constructor
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
# main loop
def run(self):
lockedBy = 'sweeper-{0}'.format(self.get_pid())
while True:
sw_main = core_utils.get_stopwatch()
mainLog = self.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
# killing stage
sw_kill = core_utils.get_stopwatch()
mainLog.debug('try to get workers to kill')
# get workers to kill
workersToKill = self.dbProxy.get_workers_to_kill(harvester_config.sweeper.maxWorkers,
harvester_config.sweeper.checkInterval)
mainLog.debug('got {0} queues to kill workers'.format(len(workersToKill)))
# loop over all workers
sw = core_utils.get_stopwatch()
for queueName, configIdWorkSpecList in iteritems(workersToKill):
for configID, workspec_list in iteritems(configIdWorkSpecList):
# get sweeper
if not self.queueConfigMapper.has_queue(queueName, configID):
mainLog.error('queue config for {0}/{1} not found'.format(queueName, configID))
continue
queueConfig = self.queueConfigMapper.get_queue(queueName, configID)
sweeperCore = self.pluginFactory.get_plugin(queueConfig.sweeper)
sw.reset()
n_workers = len(workspec_list)
try:
# try bulk method
tmpLog = self.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
tmpLog.debug('start killing')
tmpList = sweeperCore.kill_workers(workspec_list)
except AttributeError:
# fall back to single-worker method
for workspec in workspec_list:
tmpLog = self.make_logger(_logger, 'workerID={0}'.format(workspec.workerID),
method_name='run')
try:
tmpLog.debug('start killing one worker')
tmpStat, tmpOut = sweeperCore.kill_worker(workspec)
tmpLog.debug('done killing with status={0} diag={1}'.format(tmpStat, tmpOut))
except Exception:
core_utils.dump_error_message(tmpLog)
except Exception:
core_utils.dump_error_message(mainLog)
else:
# bulk method
n_killed = 0
for workspec, (tmpStat, tmpOut) in zip(workspec_list, tmpList):
tmpLog.debug('done killing workerID={0} with status={1} diag={2}'.format(
workspec.workerID, tmpStat, tmpOut))
if tmpStat:
n_killed += 1
tmpLog.debug('killed {0}/{1} workers'.format(n_killed, n_workers))
mainLog.debug('done killing {0} workers'.format(n_workers) + sw.get_elapsed_time())
mainLog.debug('done all killing' + sw_kill.get_elapsed_time())
# cleanup stage
sw_cleanup = core_utils.get_stopwatch()
# timeout for missed
try:
keepMissed = harvester_config.sweeper.keepMissed
except Exception:
keepMissed = 24
try:
keepPending = harvester_config.sweeper.keepPending
except Exception:
keepPending = 24
# get workers for cleanup
statusTimeoutMap = {'finished': harvester_config.sweeper.keepFinished,
'failed': harvester_config.sweeper.keepFailed,
'cancelled': harvester_config.sweeper.keepCancelled,
'missed': keepMissed,
'pending': keepPending
}
workersForCleanup = self.dbProxy.get_workers_for_cleanup(harvester_config.sweeper.maxWorkers,
statusTimeoutMap)
mainLog.debug('got {0} queues for workers cleanup'.format(len(workersForCleanup)))
sw = core_utils.get_stopwatch()
for queueName, configIdWorkSpecList in iteritems(workersForCleanup):
for configID, workspec_list in iteritems(configIdWorkSpecList):
# get sweeper
if not self.queueConfigMapper.has_queue(queueName, configID):
mainLog.error('queue config for {0}/{1} not found'.format(queueName, configID))
continue
queueConfig = self.queueConfigMapper.get_queue(queueName, configID)
sweeperCore = self.pluginFactory.get_plugin(queueConfig.sweeper)
messenger = self.pluginFactory.get_plugin(queueConfig.messenger)
sw.reset()
n_workers = len(workspec_list)
# make sure workers to clean up are all terminated
mainLog.debug('making sure workers to clean up are all terminated')
try:
# try bulk method
tmpList = sweeperCore.kill_workers(workspec_list)
except AttributeError:
# fall back to single-worker method
for workspec in workspec_list:
tmpLog = self.make_logger(_logger, 'workerID={0}'.format(workspec.workerID),
method_name='run')
try:
tmpStat, tmpOut = sweeperCore.kill_worker(workspec)
except Exception:
core_utils.dump_error_message(tmpLog)
except Exception:
core_utils.dump_error_message(mainLog)
mainLog.debug('made sure workers to clean up are all terminated')
# start cleanup
for workspec in workspec_list:
tmpLog = self.make_logger(_logger, 'workerID={0}'.format(workspec.workerID),
method_name='run')
try:
tmpLog.debug('start cleaning up one worker')
# sweep worker
tmpStat, tmpOut = sweeperCore.sweep_worker(workspec)
tmpLog.debug('swept_worker with status={0} diag={1}'.format(tmpStat, tmpOut))
tmpLog.debug('start messenger cleanup')
mc_tmpStat, mc_tmpOut = messenger.clean_up(workspec)
tmpLog.debug('messenger cleaned up with status={0} diag={1}'.format(mc_tmpStat, mc_tmpOut))
if tmpStat:
self.dbProxy.delete_worker(workspec.workerID)
except Exception:
core_utils.dump_error_message(tmpLog)
mainLog.debug('done cleaning up {0} workers'.format(n_workers) + sw.get_elapsed_time())
mainLog.debug('done all cleanup' + sw_cleanup.get_elapsed_time())
# old-job-deletion stage
sw_delete = core_utils.get_stopwatch()
mainLog.debug('delete old jobs')
jobTimeout = max(statusTimeoutMap.values()) + 1
self.dbProxy.delete_old_jobs(jobTimeout)
# delete orphaned job info
self.dbProxy.delete_orphaned_job_info()
mainLog.debug('done deletion of old jobs' + sw_delete.get_elapsed_time())
# time the cycle
mainLog.debug('done a sweeper cycle' + sw_main.get_elapsed_time())
# check if being terminated
if self.terminated(harvester_config.sweeper.sleepTime):
mainLog.debug('terminated')
return
class WorkerMaker(object):
# constructor
def __init__(self):
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
# get plugin
def get_plugin(self, queue_config):
return self.pluginFactory.get_plugin(queue_config.workerMaker)
# make workers
def make_workers(self, jobchunk_list, queue_config, n_ready, resource_type, maker=None):
tmpLog = core_utils.make_logger(_logger, 'queue={0} rtype={1}'.format(queue_config.queueName, resource_type),
method_name='make_workers')
tmpLog.debug('start')
try:
# get plugin
if maker is None:
maker = self.pluginFactory.get_plugin(queue_config.workerMaker)
if maker is None:
# not found
tmpLog.error('plugin for {0} not found'.format(queue_config.queueName))
return [], jobchunk_list
# get ready workers
readyWorkers = self.dbProxy.get_ready_workers(queue_config.queueName, n_ready)
# loop over all chunks
okChunks = []
ngChunks = []
for iChunk, jobChunk in enumerate(jobchunk_list):
# make a worker
if iChunk >= n_ready:
workSpec = maker.make_worker(jobChunk, queue_config, resource_type)
else:
# use ready worker
if iChunk < len(readyWorkers):
workSpec = readyWorkers[iChunk]
else:
workSpec = None
# failed
if workSpec is None:
ngChunks.append(jobChunk)
continue
# set workerID
if workSpec.workerID is None:
workSpec.workerID = self.dbProxy.get_next_seq_number('SEQ_workerID')
workSpec.configID = queue_config.configID
workSpec.isNew = True
okChunks.append((workSpec, jobChunk))
# dump
tmpLog.debug('made {0} workers while {1} chunks failed'.format(len(okChunks),
len(ngChunks)))
return okChunks, ngChunks
except Exception:
# dump error
core_utils.dump_error_message(tmpLog)
return [], jobchunk_list
# get number of jobs per worker
def get_num_jobs_per_worker(self, queue_config, n_workers, resource_type, maker=None):
# get plugin
if maker is None:
maker = self.pluginFactory.get_plugin(queue_config.workerMaker)
return maker.get_num_jobs_per_worker(n_workers)
# get number of workers per job
def get_num_workers_per_job(self, queue_config, n_workers, resource_type, maker=None):
# get plugin
if maker is None:
maker = self.pluginFactory.get_plugin(queue_config.workerMaker)
return maker.get_num_workers_per_job(n_workers)
# check number of ready resources
def num_ready_resources(self, queue_config, resource_type, maker=None):
# get plugin
if maker is None:
maker = self.pluginFactory.get_plugin(queue_config.workerMaker)
return maker.num_ready_resources()
# get upper limit on the cumulative total of workers per job
def get_max_workers_per_job_in_total(self, queue_config, resource_type, maker=None):
# get plugin
if maker is None:
maker = self.pluginFactory.get_plugin(queue_config.workerMaker)
return maker.get_max_workers_per_job_in_total()
# get upper limit on the number of new workers per job in a cycle
def get_max_workers_per_job_per_cycle(self, queue_config, resource_type, maker=None):
# get plugin
if maker is None:
maker = self.pluginFactory.get_plugin(queue_config.workerMaker)
return maker.get_max_workers_per_job_per_cycle()
def __init__(self, queue_config_mapper):
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
self.dbProxy = DBProxy()
self.throttlerMap = dict()
class SimpleThrottler(PluginBase):
# constructor
def __init__(self, **kwarg):
# logic type : AND: throttled if all rules are satisfied, OR: throttled if one rule is satisfied
self.logicType = 'OR'
PluginBase.__init__(self, **kwarg)
self.dbProxy = DBProxy()
# check if to be throttled
def to_be_throttled(self, queue_config):
tmpLog = self.make_logger(baseLogger,
'computingSite={0}'.format(
queue_config.queueName),
method_name='to_be_throttled')
tmpLog.debug('start')
# set default return vale
if self.logicType == 'OR':
retVal = False, "no rule was satisfied"
else:
retVal = True, "all rules were satisfied"
# loop over all rules
criteriaList = []
maxMissedList = []
timeNow = datetime.datetime.utcnow()
for rule in self.rulesForMissed:
# convert rule to criteria
if rule['level'] == 'site':
criteria = dict()
criteria['siteName'] = queue_config.siteName
criteria['timeLimit'] = timeNow - datetime.timedelta(
minutes=rule['timeWindow'])
criteriaList.append(criteria)
maxMissedList.append(rule['maxMissed'])
elif rule['level'] == 'pq':
criteria = dict()
criteria['computingSite'] = queue_config.queueName
criteria['timeLimit'] = timeNow - datetime.timedelta(
minutes=rule['timeWindow'])
criteriaList.append(criteria)
maxMissedList.append(rule['maxMissed'])
elif rule['level'] == 'ce':
elmName = 'computingElements'
if elmName not in queue_config.submitter:
tmpLog.debug(
'skipped since {0} is undefined in submitter config'.
format(elmName))
continue
for ce in queue_config.submitter[elmName]:
criteria = dict()
criteria['computingElement'] = ce
criteria['timeLimit'] = timeNow - datetime.timedelta(
minutes=rule['timeWindow'])
criteriaList.append(criteria)
maxMissedList.append(rule['maxMissed'])
# loop over all criteria
for criteria, maxMissed in zip(criteriaList, maxMissedList):
nMissed = self.dbProxy.get_num_missed_workers(
queue_config.queueName, criteria)
if nMissed > maxMissed:
if self.logicType == 'OR':
tmpMsg = 'logic={0} and '.format(self.logicType)
tmpMsg += 'nMissed={0} > maxMissed={1} for {2}'.format(
nMissed, maxMissed, str(criteria))
retVal = True, tmpMsg
break
else:
if self.logicType == 'AND':
tmpMsg = 'logic={0} and '.format(self.logicType)
tmpMsg += 'nMissed={0} <= maxMissed={1} for {2}'.format(
nMissed, maxMissed, str(criteria))
retVal = False, tmpMsg
break
tmpLog.debug('ret={0} : {1}'.format(*retVal))
return retVal
class JobFetcher(AgentBase):
# constructor
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.communicator = communicator
self.nodeName = socket.gethostname()
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
# main loop
def run(self):
while True:
mainLog = self.make_logger(_logger,
'id={0}'.format(self.get_pid()),
method_name='run')
mainLog.debug('getting number of jobs to be fetched')
# get number of jobs to be fetched
nJobsPerQueue = self.dbProxy.get_num_jobs_to_fetch(
harvester_config.jobfetcher.nQueues,
harvester_config.jobfetcher.lookupTime)
mainLog.debug('got {0} queues'.format(len(nJobsPerQueue)))
# loop over all queues
for queueName, nJobs in iteritems(nJobsPerQueue):
# check queue
if not self.queueConfigMapper.has_queue(queueName):
continue
tmpLog = self.make_logger(_logger,
'queueName={0}'.format(queueName),
method_name='run')
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
# upper limit
if nJobs > harvester_config.jobfetcher.maxJobs:
nJobs = harvester_config.jobfetcher.maxJobs
# get jobs
default_prodSourceLabel = queueConfig.get_source_label()
pdpm = getattr(queueConfig,
'prodSourceLabelRandomWeightsPermille', {})
choice_list = core_utils.make_choice_list(
pdpm=pdpm, default=default_prodSourceLabel)
prodSourceLabel = random.choice(choice_list)
tmpLog.debug('getting {0} jobs for prodSourceLabel {1}'.format(
nJobs, prodSourceLabel))
sw = core_utils.get_stopwatch()
siteName = queueConfig.siteName
jobs, errStr = self.communicator.get_jobs(
siteName, self.nodeName, prodSourceLabel, self.nodeName,
nJobs, queueConfig.getJobCriteria)
tmpLog.info('got {0} jobs with {1} {2}'.format(
len(jobs), errStr, sw.get_elapsed_time()))
# convert to JobSpec
if len(jobs) > 0:
# get extractor plugin
if hasattr(queueConfig, 'extractor'):
extractorCore = self.pluginFactory.get_plugin(
queueConfig.extractor)
else:
extractorCore = None
jobSpecs = []
fileStatMap = dict()
sw_startconvert = core_utils.get_stopwatch()
for job in jobs:
timeNow = datetime.datetime.utcnow()
jobSpec = JobSpec()
jobSpec.convert_job_json(job)
jobSpec.computingSite = queueName
jobSpec.status = 'starting'
jobSpec.subStatus = 'fetched'
jobSpec.creationTime = timeNow
jobSpec.stateChangeTime = timeNow
jobSpec.configID = queueConfig.configID
jobSpec.set_one_attribute(
'schedulerID', 'harvester-{0}'.format(
harvester_config.master.harvester_id))
if queueConfig.zipPerMB is not None and jobSpec.zipPerMB is None:
jobSpec.zipPerMB = queueConfig.zipPerMB
fileGroupDictList = [
jobSpec.get_input_file_attributes()
]
if extractorCore is not None:
fileGroupDictList.append(
extractorCore.get_aux_inputs(jobSpec))
for fileGroupDict in fileGroupDictList:
for tmpLFN, fileAttrs in iteritems(fileGroupDict):
# check file status
if tmpLFN not in fileStatMap:
fileStatMap[
tmpLFN] = self.dbProxy.get_file_status(
tmpLFN, 'input',
queueConfig.ddmEndpointIn,
'starting')
# make file spec
fileSpec = FileSpec()
fileSpec.PandaID = jobSpec.PandaID
fileSpec.taskID = jobSpec.taskID
fileSpec.lfn = tmpLFN
fileSpec.endpoint = queueConfig.ddmEndpointIn
fileSpec.scope = fileAttrs['scope']
# set preparing to skip stage-in if the file is (being) taken care of by another job
if 'ready' in fileStatMap[tmpLFN] or 'preparing' in fileStatMap[tmpLFN] \
or 'to_prepare' in fileStatMap[tmpLFN]:
fileSpec.status = 'preparing'
else:
fileSpec.status = 'to_prepare'
if fileSpec.status not in fileStatMap[tmpLFN]:
fileStatMap[tmpLFN][fileSpec.status] = 0
fileStatMap[tmpLFN][fileSpec.status] += 1
if 'INTERNAL_FileType' in fileAttrs:
fileSpec.fileType = fileAttrs[
'INTERNAL_FileType']
jobSpec.auxInput = JobSpec.AUX_hasAuxInput
else:
fileSpec.fileType = 'input'
if 'INTERNAL_URL' in fileAttrs:
fileSpec.url = fileAttrs['INTERNAL_URL']
jobSpec.add_in_file(fileSpec)
jobSpec.trigger_propagation()
jobSpecs.append(jobSpec)
# insert to DB
tmpLog.debug("Converting of {0} jobs {1}".format(
len(jobs), sw_startconvert.get_elapsed_time()))
sw_insertdb = core_utils.get_stopwatch()
self.dbProxy.insert_jobs(jobSpecs)
tmpLog.debug('Insert of {0} jobs {1}'.format(
len(jobSpecs), sw_insertdb.get_elapsed_time()))
mainLog.debug('done')
# check if being terminated
if self.terminated(harvester_config.jobfetcher.sleepTime):
mainLog.debug('terminated')
return
class EventFeeder(AgentBase):
# constructor
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.queueConfigMapper = queue_config_mapper
self.communicator = communicator
self.pluginFactory = PluginFactory()
# main loop
def run(self):
lockedBy = 'eventfeeder-{0}'.format(self.get_pid())
while True:
mainLog = self.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
mainLog.debug('getting workers to feed events')
workSpecsPerQueue = self.dbProxy.get_workers_to_feed_events(harvester_config.eventfeeder.maxWorkers,
harvester_config.eventfeeder.lockInterval,
lockedBy)
mainLog.debug('got {0} queues'.format(len(workSpecsPerQueue)))
# loop over all workers
for queueName, workSpecList in iteritems(workSpecsPerQueue):
tmpQueLog = self.make_logger(_logger, 'queue={0}'.format(queueName), method_name='run')
# check queue
if not self.queueConfigMapper.has_queue(queueName):
tmpQueLog.error('config not found')
continue
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
if hasattr(queueConfig, 'scatteredEvents') and queueConfig.scatteredEvents:
scattered = True
else:
scattered = False
# get plugin
messenger = self.pluginFactory.get_plugin(queueConfig.messenger)
# loop over all workers
for workSpec in workSpecList:
tmpLog = core_utils.make_logger(_logger, 'workerID={0}'.format(workSpec.workerID),
method_name='run')
# lock worker again
lockedFlag = self.dbProxy.lock_worker_again_to_feed_events(workSpec.workerID, lockedBy)
if not lockedFlag:
tmpLog.debug('skipped since locked by another')
continue
# get events
tmpLog.debug('get events')
tmpStat, events = self.communicator.get_event_ranges(workSpec.eventsRequestParams,
scattered,
workSpec.get_access_point())
# failed
if tmpStat is False:
tmpLog.error('failed to get events with {0}'.format(events))
continue
# lock worker again
lockedFlag = self.dbProxy.lock_worker_again_to_feed_events(workSpec.workerID, lockedBy)
if not lockedFlag:
tmpLog.debug('skipped before feeding since locked by another')
continue
tmpStat = messenger.feed_events(workSpec, events)
# failed
if tmpStat is False:
tmpLog.error('failed to feed events')
continue
# dump
for pandaID, eventList in iteritems(events):
try:
nRanges = workSpec.eventsRequestParams[pandaID]['nRanges']
except Exception:
nRanges = None
tmpLog.debug('got {0} events for PandaID={1} while getting {2} events'.format(len(eventList),
pandaID,
nRanges))
# disable multi workers
if workSpec.mapType == WorkSpec.MT_MultiWorkers:
if len(eventList) == 0 or (nRanges is not None and len(eventList) < nRanges):
tmpStat = self.dbProxy.disable_multi_workers(pandaID)
if tmpStat == 1:
tmpStr = 'disabled MultiWorkers for PandaID={0}'.format(pandaID)
tmpLog.debug(tmpStr)
# update worker
workSpec.eventsRequest = WorkSpec.EV_useEvents
workSpec.eventsRequestParams = None
workSpec.eventFeedTime = None
workSpec.eventFeedLock = None
# update local database
tmpStat = self.dbProxy.update_worker(workSpec, {'eventFeedLock': lockedBy})
tmpLog.debug('done with {0}'.format(tmpStat))
tmpQueLog.debug('done')
mainLog.debug('done')
# check if being terminated
if self.terminated(harvester_config.eventfeeder.sleepTime):
mainLog.debug('terminated')
return
class SimpleThrottler(PluginBase):
# constructor
def __init__(self, **kwarg):
# logic type : AND: throttled if all rules are satisfied, OR: throttled if one rule is satisfied
self.logicType = 'OR'
PluginBase.__init__(self, **kwarg)
self.dbProxy = DBProxy()
# check if to be throttled
def to_be_throttled(self, queue_config):
tmpLog = self.make_logger(baseLogger, 'computingSite={0}'.format(queue_config.queueName),
method_name='to_be_throttled')
tmpLog.debug('start')
# set default return vale
if self.logicType == 'OR':
retVal = False, "no rule was satisfied"
else:
retVal = True, "all rules were satisfied"
# loop over all rules
criteriaList = []
maxMissedList = []
timeNow = datetime.datetime.utcnow()
for rule in self.rulesForMissed:
# convert rule to criteria
if rule['level'] == 'site':
criteria = dict()
criteria['siteName'] = queue_config.siteName
criteria['timeLimit'] = timeNow - datetime.timedelta(minutes=rule['timeWindow'])
criteriaList.append(criteria)
maxMissedList.append(rule['maxMissed'])
elif rule['level'] == 'pq':
criteria = dict()
criteria['computingSite'] = queue_config.queueName
criteria['timeLimit'] = timeNow - datetime.timedelta(minutes=rule['timeWindow'])
criteriaList.append(criteria)
maxMissedList.append(rule['maxMissed'])
elif rule['level'] == 'ce':
elmName = 'computingElements'
if elmName not in queue_config.submitter:
tmpLog.debug('skipped since {0} is undefined in submitter config'.format(elmName))
continue
for ce in queue_config.submitter[elmName]:
criteria = dict()
criteria['computingElement'] = ce
criteria['timeLimit'] = timeNow - datetime.timedelta(minutes=rule['timeWindow'])
criteriaList.append(criteria)
maxMissedList.append(rule['maxMissed'])
# loop over all criteria
for criteria, maxMissed in zip(criteriaList, maxMissedList):
nMissed = self.dbProxy.get_num_missed_workers(queue_config.queueName, criteria)
if nMissed > maxMissed:
if self.logicType == 'OR':
tmpMsg = 'logic={0} and '.format(self.logicType)
tmpMsg += 'nMissed={0} > maxMissed={1} for {2}'.format(nMissed, maxMissed, str(criteria))
retVal = True, tmpMsg
break
else:
if self.logicType == 'AND':
tmpMsg = 'logic={0} and '.format(self.logicType)
tmpMsg += 'nMissed={0} <= maxMissed={1} for {2}'.format(nMissed, maxMissed, str(criteria))
retVal = False, tmpMsg
break
tmpLog.debug('ret={0} : {1}'.format(*retVal))
return retVal
class JobFetcher(AgentBase):
# constructor
def __init__(self, communicator, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.communicator = communicator
self.nodeName = socket.gethostname()
self.queueConfigMapper = queue_config_mapper
# main loop
def run(self):
while True:
mainLog = self.make_logger(_logger, 'id={0}'.format(self.get_pid()), method_name='run')
mainLog.debug('getting number of jobs to be fetched')
# get number of jobs to be fetched
nJobsPerQueue = self.dbProxy.get_num_jobs_to_fetch(harvester_config.jobfetcher.nQueues,
harvester_config.jobfetcher.lookupTime)
mainLog.debug('got {0} queues'.format(len(nJobsPerQueue)))
# loop over all queues
for queueName, nJobs in iteritems(nJobsPerQueue):
# check queue
if not self.queueConfigMapper.has_queue(queueName):
continue
tmpLog = self.make_logger(_logger, 'queueName={0}'.format(queueName),
method_name='run')
# get queue
queueConfig = self.queueConfigMapper.get_queue(queueName)
# upper limit
if nJobs > harvester_config.jobfetcher.maxJobs:
nJobs = harvester_config.jobfetcher.maxJobs
# get jobs
tmpLog.debug('getting {0} jobs'.format(nJobs))
sw = core_utils.get_stopwatch()
siteName = queueConfig.siteName
jobs, errStr = self.communicator.get_jobs(siteName, self.nodeName,
queueConfig.get_source_label(),
self.nodeName, nJobs,
queueConfig.getJobCriteria)
tmpLog.info('got {0} jobs with {1} {2}'.format(len(jobs), errStr, sw.get_elapsed_time()))
# convert to JobSpec
if len(jobs) > 0:
jobSpecs = []
fileStatMap = dict()
sw_startconvert = core_utils.get_stopwatch()
for job in jobs:
timeNow = datetime.datetime.utcnow()
jobSpec = JobSpec()
jobSpec.convert_job_json(job)
jobSpec.computingSite = queueName
jobSpec.status = 'starting'
jobSpec.subStatus = 'fetched'
jobSpec.creationTime = timeNow
jobSpec.stateChangeTime = timeNow
jobSpec.configID = queueConfig.configID
jobSpec.set_one_attribute('schedulerID',
'harvester-{0}'.format(harvester_config.master.harvester_id))
if queueConfig.zipPerMB is not None and jobSpec.zipPerMB is None:
jobSpec.zipPerMB = queueConfig.zipPerMB
for tmpLFN, fileAttrs in iteritems(jobSpec.get_input_file_attributes()):
# check file status
if tmpLFN not in fileStatMap:
fileStatMap[tmpLFN] = self.dbProxy.get_file_status(tmpLFN, 'input',
queueConfig.ddmEndpointIn,
'starting')
# make file spec
fileSpec = FileSpec()
fileSpec.PandaID = jobSpec.PandaID
fileSpec.taskID = jobSpec.taskID
fileSpec.lfn = tmpLFN
fileSpec.endpoint = queueConfig.ddmEndpointIn
fileSpec.scope = fileAttrs['scope']
# set preparing to skip stage-in if the file is (being) taken care of by another job
if 'ready' in fileStatMap[tmpLFN] or 'preparing' in fileStatMap[tmpLFN] \
or 'to_prepare' in fileStatMap[tmpLFN]:
fileSpec.status = 'preparing'
else:
fileSpec.status = 'to_prepare'
if fileSpec.status not in fileStatMap[tmpLFN]:
fileStatMap[tmpLFN][fileSpec.status] = 0
fileStatMap[tmpLFN][fileSpec.status] += 1
fileSpec.fileType = 'input'
jobSpec.add_in_file(fileSpec)
jobSpec.trigger_propagation()
jobSpecs.append(jobSpec)
# insert to DB
tmpLog.debug("Converting of {0} jobs {1}".format(len(jobs),sw_startconvert.get_elapsed_time()))
sw_insertdb =core_utils.get_stopwatch()
self.dbProxy.insert_jobs(jobSpecs)
tmpLog.debug('Insert of {0} jobs {1}'.format(len(jobSpecs), sw_insertdb.get_elapsed_time()))
mainLog.debug('done')
# check if being terminated
if self.terminated(harvester_config.jobfetcher.sleepTime):
mainLog.debug('terminated')
return
class Stager(AgentBase):
# constructor
def __init__(self, queue_config_mapper, single_mode=False):
AgentBase.__init__(self, single_mode)
self.dbProxy = DBProxy()
self.queueConfigMapper = queue_config_mapper
self.pluginFactory = PluginFactory()
# main loop
def run(self):
lockedBy = 'stager-{0}'.format(self.get_pid())
while True:
sw = core_utils.get_stopwatch()
mainLog = self.make_logger(_logger, 'id={0}'.format(lockedBy), method_name='run')
mainLog.debug('try to get jobs to check')
# get jobs to check preparation
try:
maxFilesPerJob = harvester_config.stager.maxFilesPerJobToCheck
except Exception:
maxFilesPerJob = None
jobsToCheck = self.dbProxy.get_jobs_for_stage_out(harvester_config.stager.maxJobsToCheck,
harvester_config.stager.checkInterval,
harvester_config.stager.lockInterval,
lockedBy, 'transferring',
JobSpec.HO_hasTransfer,
max_files_per_job=maxFilesPerJob)
mainLog.debug('got {0} jobs to check'.format(len(jobsToCheck)))
# loop over all jobs
for jobSpec in jobsToCheck:
tmpLog = self.make_logger(_logger, 'PandaID={0}'.format(jobSpec.PandaID),
method_name='run')
try:
tmpLog.debug('start checking')
# configID
configID = jobSpec.configID
if not core_utils.dynamic_plugin_change():
configID = None
# get queue
if not self.queueConfigMapper.has_queue(jobSpec.computingSite, configID):
tmpLog.error('queue config for {0}/{1} not found'.format(jobSpec.computingSite,
configID))
continue
queueConfig = self.queueConfigMapper.get_queue(jobSpec.computingSite, configID)
# get plugin
stagerCore = self.pluginFactory.get_plugin(queueConfig.stager)
if stagerCore is None:
# not found
tmpLog.error('plugin for {0} not found'.format(jobSpec.computingSite))
continue
# lock job again
lockedAgain = self.dbProxy.lock_job_again(jobSpec.PandaID, 'stagerTime', 'stagerLock', lockedBy)
if not lockedAgain:
tmpLog.debug('skip since locked by another thread')
continue
tmpStat, tmpStr = stagerCore.check_status(jobSpec)
# check result
if tmpStat is True:
# succeeded
newSubStatus = self.dbProxy.update_job_for_stage_out(jobSpec, True, lockedBy)
tmpLog.debug('succeeded new subStatus={0}'.format(newSubStatus))
elif tmpStat is False:
# fatal error
tmpLog.debug('fatal error when checking status with {0}'.format(tmpStr))
# update job
for fileSpec in jobSpec.outFiles:
if fileSpec.status != 'finished':
fileSpec.status = 'failed'
errStr = 'stage-out failed with {0}'.format(tmpStr)
jobSpec.set_pilot_error(PilotErrors.ERR_STAGEOUTFAILED, errStr)
jobSpec.trigger_propagation()
newSubStatus = self.dbProxy.update_job_for_stage_out(jobSpec, True, lockedBy)
tmpLog.debug('updated new subStatus={0}'.format(newSubStatus))
else:
# on-going
tmpLog.debug('try to check later since {0}'.format(tmpStr))
except Exception:
core_utils.dump_error_message(tmpLog)
# get jobs to trigger stage-out
try:
maxFilesPerJob = harvester_config.stager.maxFilesPerJobToTrigger
except Exception:
maxFilesPerJob = None
jobsToTrigger = self.dbProxy.get_jobs_for_stage_out(harvester_config.stager.maxJobsToTrigger,
harvester_config.stager.triggerInterval,
harvester_config.stager.lockInterval,
lockedBy, 'to_transfer',
JobSpec.HO_hasOutput,
JobSpec.HO_hasZipOutput,
max_files_per_job=maxFilesPerJob)
mainLog.debug('got {0} jobs to trigger'.format(len(jobsToTrigger)))
# loop over all jobs
for jobSpec in jobsToTrigger:
tmpLog = self.make_logger(_logger, 'PandaID={0}'.format(jobSpec.PandaID),
method_name='run')
try:
tmpLog.debug('try to trigger stage-out')
# configID
configID = jobSpec.configID
if not core_utils.dynamic_plugin_change():
configID = None
# get queue
if not self.queueConfigMapper.has_queue(jobSpec.computingSite, configID):
tmpLog.error('queue config for {0}/{1} not found'.format(jobSpec.computingSite,
configID))
continue
queueConfig = self.queueConfigMapper.get_queue(jobSpec.computingSite, configID)
# get plugin
stagerCore = self.pluginFactory.get_plugin(queueConfig.stager)
if stagerCore is None:
# not found
tmpLog.error('plugin for {0} not found'.format(jobSpec.computingSite))
continue
# lock job again
lockedAgain = self.dbProxy.lock_job_again(jobSpec.PandaID, 'stagerTime', 'stagerLock', lockedBy)
if not lockedAgain:
tmpLog.debug('skip since locked by another thread')
continue
# trigger stage-out
tmpStat, tmpStr = stagerCore.trigger_stage_out(jobSpec)
# check result
if tmpStat is True:
# succeeded
jobSpec.all_files_triggered_to_stage_out()
newSubStatus = self.dbProxy.update_job_for_stage_out(jobSpec, True, lockedBy)
tmpLog.debug('triggered new subStatus={0}'.format(newSubStatus))
elif tmpStat is False:
# fatal error
tmpLog.debug('fatal error to trigger with {0}'.format(tmpStr))
# update job
for fileSpec in jobSpec.outFiles:
if fileSpec.status != 'finished':
fileSpec.status = 'failed'
errStr = 'stage-out failed with {0}'.format(tmpStr)
jobSpec.set_pilot_error(PilotErrors.ERR_STAGEOUTFAILED, errStr)
jobSpec.trigger_propagation()
newSubStatus = self.dbProxy.update_job_for_stage_out(jobSpec, True, lockedBy)
tmpLog.debug('updated new subStatus={0}'.format(newSubStatus))
else:
# temporary error
tmpLog.debug('try to trigger later since {0}'.format(tmpStr))
except Exception:
core_utils.dump_error_message(tmpLog)
# get jobs to zip output
try:
maxFilesPerJob = harvester_config.stager.maxFilesPerJobToZip
except Exception:
maxFilesPerJob = None
try:
zipInterval = harvester_config.stager.zipInterval
except Exception:
zipInterval = harvester_config.stager.triggerInterval
jobsToZip = self.dbProxy.get_jobs_for_stage_out(harvester_config.stager.maxJobsToZip,
zipInterval,
harvester_config.stager.lockInterval,
lockedBy, 'to_transfer',
JobSpec.HO_hasZipOutput,
JobSpec.HO_hasOutput,
max_files_per_job=maxFilesPerJob)
mainLog.debug('got {0} jobs to zip'.format(len(jobsToZip)))
# loop over all jobs
for jobSpec in jobsToZip:
tmpLog = self.make_logger(_logger, 'PandaID={0}'.format(jobSpec.PandaID),
method_name='run')
try:
tmpLog.debug('try to zip output')
# configID
configID = jobSpec.configID
if not core_utils.dynamic_plugin_change():
configID = None
# get queue
if not self.queueConfigMapper.has_queue(jobSpec.computingSite, configID):
tmpLog.error('queue config for {0}/{1} not found'.format(jobSpec.computingSite,
configID))
continue
queueConfig = self.queueConfigMapper.get_queue(jobSpec.computingSite, configID)
# get plugin
stagerCore = self.pluginFactory.get_plugin(queueConfig.stager)
if stagerCore is None:
# not found
tmpLog.error('plugin for {0} not found'.format(jobSpec.computingSite))
continue
# lock job again
lockedAgain = self.dbProxy.lock_job_again(jobSpec.PandaID, 'stagerTime', 'stagerLock', lockedBy)
if not lockedAgain:
tmpLog.debug('skip since locked by another thread')
continue
# trigger preparation
tmpStat, tmpStr = stagerCore.zip_output(jobSpec)
# succeeded
if tmpStat is True:
# update job
jobSpec.all_files_zipped()
newSubStatus = self.dbProxy.update_job_for_stage_out(jobSpec, False, lockedBy)
tmpLog.debug('zipped new subStatus={0}'.format(newSubStatus))
else:
# failed
tmpLog.debug('failed to zip with {0}'.format(tmpStr))
except Exception:
core_utils.dump_error_message(tmpLog)
mainLog.debug('done' + sw.get_elapsed_time())
# check if being terminated
if self.terminated(harvester_config.stager.sleepTime):
mainLog.debug('terminated')
return
