def __init__(self, site, cloud, nJobs):
"""Initialize class with parameters
"""
self.__site = site
self.__cloud = cloud
self.__nJobs = nJobs
taskBuffer.init(panda_config.dbhost, panda_config.dbpasswd, nDBConnection=1)
def run(inFile,v_onlyTA):
import cPickle as pickle
try:
# read Jobs from file
f = open(inFile)
jobs = pickle.load(f)
f.close()
except:
type, value, traceBack = sys.exc_info()
print("run() : %s %s" % (type,value))
return
# password
from config import panda_config
passwd = panda_config.dbpasswd
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# instantiate TB
from taskbuffer.TaskBuffer import taskBuffer
taskBuffer.init(panda_config.dbhost,panda_config.dbpasswd,nDBConnection=1)
# run Setupper
from dataservice.Setupper import Setupper
thr = Setupper(taskBuffer,jobs,onlyTA=v_onlyTA)
thr.start()
thr.join()
return
def main(backGround=False):
_logger.debug('starting ...')
# register signal handler
signal.signal(signal.SIGINT, catch_sig)
signal.signal(signal.SIGHUP, catch_sig)
signal.signal(signal.SIGTERM, catch_sig)
signal.signal(signal.SIGALRM, catch_sig)
signal.alarm(overallTimeout)
# forking
pid = os.fork()
if pid != 0:
# watch child process
os.wait()
time.sleep(1)
else:
# main loop
from taskbuffer.TaskBuffer import taskBuffer
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# instantiate TB
taskBuffer.init(panda_config.dbhost,
panda_config.dbpasswd,
nDBConnection=1)
# instantiate sitemapper
siteMapper = SiteMapper(taskBuffer)
# ActiveMQ params
clientid = 'PANDA-' + socket.getfqdn()
queue = '/queue/Consumer.test1.poc.pocMSG'
ssl_opts = {
'use_ssl': True,
'ssl_cert_file': '%s/hostcert.pem' % panda_config.certdir,
'ssl_key_file': '%s/hostkey.pem' % panda_config.certdir
}
# resolve multiple brokers
brokerList = socket.gethostbyname_ex('gridmsg007.cern.ch')[-1]
# set listener
for tmpBroker in brokerList:
try:
_logger.debug('setting listener on %s' % tmpBroker)
conn = stomp.Connection(host_and_ports=[(tmpBroker, 6162)],
**ssl_opts)
conn.set_listener(
'GenCallbackConsumer',
GenCallbackConsumer(conn, taskBuffer, siteMapper))
conn.start()
conn.connect(headers={'client-id': clientid})
conn.subscribe(destination=queue, ack='client-individual')
#,headers = {'selector':"cbtype='FileDoneMessage'"})
if not conn.is_connected():
_logger.error("connection failure to %s" % tmpBroker)
except:
errtype, errvalue = sys.exc_info()[:2]
_logger.error("failed to set listener on %s : %s %s" %
(tmpBroker, errtype, errvalue))
catch_sig(None, None)
def main(backGround=False):
_logger.debug('starting ...')
# register signal handler
signal.signal(signal.SIGINT, catch_sig)
signal.signal(signal.SIGHUP, catch_sig)
signal.signal(signal.SIGTERM,catch_sig)
signal.signal(signal.SIGALRM,catch_sig)
signal.alarm(overallTimeout)
# forking
pid = os.fork()
if pid != 0:
# watch child process
os.wait()
time.sleep(1)
else:
# main loop
from taskbuffer.TaskBuffer import taskBuffer
# check certificate
certName = '/data/atlpan/pandasv1_usercert.pem'
_logger.debug('checking certificate {0}'.format(certName))
certOK,certMsg = DataServiceUtils.checkCertificate(certName)
if not certOK:
_logger.error('bad certificate : {0}'.format(certMsg))
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# instantiate TB
taskBuffer.init(panda_config.dbhost,panda_config.dbpasswd,nDBConnection=1)
# instantiate sitemapper
siteMapper = SiteMapper(taskBuffer)
# ActiveMQ params
queue = '/queue/Consumer.PANDA.atlas.ddm.siteservices'
ssl_opts = {'use_ssl' : True,
'ssl_cert_file' : certName,
'ssl_key_file' : '/data/atlpan/pandasv1_userkey.pem'}
# resolve multiple brokers
brokerList = socket.gethostbyname_ex('atlasddm-mb.cern.ch')[-1]
# set listener
for tmpBroker in brokerList:
try:
clientid = 'PANDA-' + socket.getfqdn() + '-' + tmpBroker
_logger.debug('setting listener %s' % clientid)
conn = stomp.Connection(host_and_ports = [(tmpBroker, 6162)], **ssl_opts)
conn.set_listener('FileCallbackListener', FileCallbackListener(conn,taskBuffer,siteMapper))
conn.start()
conn.connect(headers = {'client-id': clientid})
conn.subscribe(destination=queue, ack='client-individual')
#,headers = {'selector':"cbtype='FileDoneMessage'"})
if not conn.is_connected():
_logger.error("connection failure to %s" % tmpBroker)
_logger.debug('listener %s is up and running' % clientid)
except:
errtype,errvalue = sys.exc_info()[:2]
_logger.error("failed to set listener on %s : %s %s" % (tmpBroker,errtype,errvalue))
catch_sig(None,None)
def main(backGround=False):
_logger.debug('starting ...')
# register signal handler
signal.signal(signal.SIGINT, catch_sig)
signal.signal(signal.SIGHUP, catch_sig)
signal.signal(signal.SIGTERM,catch_sig)
signal.signal(signal.SIGALRM,catch_sig)
signal.alarm(overallTimeout)
# forking
pid = os.fork()
if pid != 0:
# watch child process
os.wait()
time.sleep(1)
else:
# main loop
from taskbuffer.TaskBuffer import taskBuffer
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# instantiate TB
taskBuffer.init(panda_config.dbhost,panda_config.dbpasswd,nDBConnection=1)
# instantiate sitemapper
siteMapper = SiteMapper(taskBuffer)
# ActiveMQ params
clientid = 'PANDA-' + socket.getfqdn()
queue = '/queue/Consumer.test1.poc.pocMSG'
ssl_opts = {'use_ssl' : True,
'ssl_cert_file' : '%s/hostcert.pem'%panda_config.certdir,
'ssl_key_file' : '%s/hostkey.pem'%panda_config.certdir}
# resolve multiple brokers
brokerList = socket.gethostbyname_ex('gridmsg007.cern.ch')[-1]
# set listener
for tmpBroker in brokerList:
try:
_logger.debug('setting listener on %s' % tmpBroker)
conn = stomp.Connection(host_and_ports = [(tmpBroker, 6162)], **ssl_opts)
conn.set_listener('GenCallbackConsumer', GenCallbackConsumer(conn,taskBuffer,siteMapper))
conn.start()
conn.connect(headers = {'client-id': clientid})
conn.subscribe(destination=queue, ack='client-individual')
#,headers = {'selector':"cbtype='FileDoneMessage'"})
if not conn.is_connected():
_logger.error("connection failure to %s" % tmpBroker)
except:
errtype,errvalue = sys.exc_info()[:2]
_logger.error("failed to set listener on %s : %s %s" % (tmpBroker,errtype,errvalue))
catch_sig(None,None)
def __init__(self):
threading.Thread.__init__(self)
taskBuffer.init(panda_config.dbhost, panda_config.dbpasswd, nDBConnection=1)
if hasattr(panda_config, 'NWS_URL'):
self.NWS_URL = panda_config.NWS_URL
else:
self.NWS_URL = 'http://atlas-adc-netmetrics-lb.cern.ch/metrics/latest.json'
_logger.debug('Getting NWS dump...')
self.nws_dump = aux.get_dump(self.NWS_URL)
_logger.debug('Done')
if hasattr(panda_config, 'AGIS_URL_CM'):
self.AGIS_URL_CM = panda_config.AGIS_URL_CM
else:
self.AGIS_URL_CM = 'http://atlas-agis-api.cern.ch/request/site/query/list_links/?json'
_logger.debug('Getting AGIS cost matrix dump...')
self.agis_cm_dump = aux.get_dump(self.AGIS_URL_CM)
_logger.debug('Done')
def __init__(self):
threading.Thread.__init__(self)
taskBuffer.init(panda_config.dbhost, panda_config.dbpasswd, nDBConnection=1)
if hasattr(panda_config,'NWS_URL'):
self.NWS_URL = panda_config.NWS_URL
else:
self.NWS_URL = 'http://atlas-adc-netmetrics-lb.cern.ch/metrics/latest.json'
_logger.debug('Getting NWS dump...')
self.nws_dump = aux.get_dump(self.NWS_URL)
_logger.debug('Done')
if hasattr(panda_config, 'AGIS_URL_CM'):
self.AGIS_URL_CM = panda_config.AGIS_URL_CM
else:
self.AGIS_URL_CM = 'http://atlas-agis-api.cern.ch/request/site/query/list_links/?json'
_logger.debug('Getting AGIS cost matrix dump...')
self.agis_cm_dump = aux.get_dump(self.AGIS_URL_CM)
_logger.debug('Done')
pid = items[1]
# start time
timeM = re.search('(\S+\s+\d+ \d+:\d+:\d+ \d+)', line)
startTime = datetime.datetime(
*time.strptime(timeM.group(1), '%b %d %H:%M:%S %Y')[:6])
# kill old process
if startTime < timeLimit:
tmpLog.debug("old process : %s %s" % (pid, startTime))
tmpLog.debug(line)
commands.getoutput('kill -9 %s' % pid)
except:
type, value, traceBack = sys.exc_info()
tmpLog.error("kill process : %s %s" % (type, value))
# instantiate TB
taskBuffer.init(panda_config.dbhost, panda_config.dbpasswd, nDBConnection=1)
# instantiate sitemapper
aSiteMapper = SiteMapper(taskBuffer)
# delete
tmpLog.debug("Del session")
status, retSel = taskBuffer.querySQLS(
"SELECT MAX(PandaID) FROM ATLAS_PANDA.jobsDefined4", {})
if retSel != None:
try:
maxID = retSel[0][0]
tmpLog.debug("maxID : %s" % maxID)
if maxID != None:
varMap = {}
varMap[':maxID'] = maxID
- jobs that don't belong to any task (usually HammerCloud test jobs)
- analysis tasks, since there has not been any unification yet
"""
import datetime
import sys
from elasticsearch import Elasticsearch
from pandalogger.PandaLogger import PandaLogger
_logger = PandaLogger().getLogger('frontier_retagging')
from config import panda_config
from taskbuffer.TaskBuffer import taskBuffer
taskBuffer.init(panda_config.dbhost, panda_config.dbpasswd, nDBConnection=1)
def get_frontier_failure_count_by_task():
"""
retrieve failure count by task from Elastic Search
"""
es_host = 'atlas-kibana.mwt2.org'
es_port = 9200
es_index = 'frontier'
# es_index = 'frontier-%d-%02d' % (ct.year, ct.month)
# prepare time window for query
n_hours = 1
ct = datetime.datetime.utcnow()
st = ct - datetime.timedelta(hours=n_hours)
current_time = ct.strftime('%Y%m%dT%H%M%S.%f')[:-3] + 'Z'
"""
import datetime
import traceback
import types
# config file
from config import panda_config
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# initialzie TaskBuffer
from taskbuffer.TaskBuffer import taskBuffer
taskBuffer.init(panda_config.dbhost, panda_config.dbpasswd,
panda_config.nDBConnection, True)
# initialize JobDispatcher
from jobdispatcher.JobDispatcher import jobDispatcher
if panda_config.nDBConnection != 0:
jobDispatcher.init(taskBuffer)
# initialize DataService
from dataservice.DataService import dataService
if panda_config.nDBConnection != 0:
dataService.init(taskBuffer)
# initialize UserIF
from userinterface.UserIF import userIF
if panda_config.nDBConnection != 0:
userIF.init(taskBuffer)
def main(backGround=False):
_logger.debug('starting ...')
# register signal handler
signal.signal(signal.SIGINT, catch_sig)
signal.signal(signal.SIGHUP, catch_sig)
signal.signal(signal.SIGTERM, catch_sig)
signal.signal(signal.SIGALRM, catch_sig)
signal.alarm(overallTimeout)
# forking
pid = os.fork()
if pid != 0:
# watch child process
os.wait()
time.sleep(1)
else:
# main loop
from taskbuffer.TaskBuffer import taskBuffer
# check certificate
certName = '%s/pandasv1_usercert.pem' % panda_config.certdir
keyName = '%s/pandasv1_userkey.pem' % panda_config.certdir
_logger.debug('checking certificate {0}'.format(certName))
certOK, certMsg = DataServiceUtils.checkCertificate(certName)
if not certOK:
_logger.error('bad certificate : {0}'.format(certMsg))
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# instantiate TB
taskBuffer.init(panda_config.dbhost,
panda_config.dbpasswd,
nDBConnection=1)
# instantiate sitemapper
siteMapper = SiteMapper(taskBuffer)
# ActiveMQ params
queue = '/queue/Consumer.PANDA.atlas.ddm.siteservices'
ssl_opts = {
'use_ssl': True,
'ssl_version': ssl.PROTOCOL_TLSv1,
'ssl_cert_file': certName,
'ssl_key_file': keyName
}
# resolve multiple brokers
brokerList = socket.gethostbyname_ex('atlas-mb.cern.ch')[-1]
# set listener
connList = []
for tmpBroker in brokerList:
try:
clientid = 'PANDA-' + socket.getfqdn() + '-' + tmpBroker
subscription_id = 'panda-server-consumer-' + socket.getfqdn()
_logger.debug('setting listener %s' % clientid)
conn = stomp.Connection(host_and_ports=[(tmpBroker, 61023)],
**ssl_opts)
connList.append(conn)
except:
errtype, errvalue = sys.exc_info()[:2]
_logger.error("failed to connect to %s : %s %s" %
(tmpBroker, errtype, errvalue))
catch_sig(None, None)
while True:
for conn in connList:
try:
if not conn.is_connected():
conn.set_listener(
'FileCallbackListener',
FileCallbackListener(conn, taskBuffer, siteMapper,
subscription_id))
conn.start()
conn.connect(headers={'client-id': clientid})
conn.subscribe(destination=queue,
id=subscription_id,
ack='client-individual')
_logger.debug('listener %s is up and running' %
clientid)
except:
errtype, errvalue = sys.exc_info()[:2]
_logger.error("failed to set listener on %s : %s %s" %
(tmpBroker, errtype, errvalue))
catch_sig(None, None)
time.sleep(5)
def main(backGround=False):
_logger.debug('starting ...')
# register signal handler
signal.signal(signal.SIGINT, catch_sig)
signal.signal(signal.SIGHUP, catch_sig)
signal.signal(signal.SIGTERM,catch_sig)
signal.signal(signal.SIGALRM,catch_sig)
signal.alarm(overallTimeout)
# forking
pid = os.fork()
if pid != 0:
# watch child process
os.wait()
time.sleep(1)
else:
# main loop
from taskbuffer.TaskBuffer import taskBuffer
# check certificate
certName = '/data/atlpan/pandasv1_usercert.pem'
#certName = '/etc/grid-security/hostcert.pem'
_logger.debug('checking certificate {0}'.format(certName))
certOK,certMsg = DataServiceUtils.checkCertificate(certName)
if not certOK:
_logger.error('bad certificate : {0}'.format(certMsg))
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# instantiate TB
taskBuffer.init(panda_config.dbhost,panda_config.dbpasswd,nDBConnection=1)
# instantiate sitemapper
siteMapper = SiteMapper(taskBuffer)
# ActiveMQ params
queue = '/queue/Consumer.panda.rucio.events'
ssl_opts = {'use_ssl' : True,
'ssl_version' : ssl.PROTOCOL_TLSv1,
'ssl_cert_file' : certName,
'ssl_key_file' : '/data/atlpan/pandasv1_userkey.pem'}
# resolve multiple brokers
brokerList = socket.gethostbyname_ex('atlas-mb.cern.ch')[-1]
# set listener
connList = []
for tmpBroker in brokerList:
try:
clientid = 'PANDA-' + socket.getfqdn() + '-' + tmpBroker
subscription_id = 'panda-server-consumer'
_logger.debug('setting listener %s to broker %s' % (clientid, tmpBroker))
conn = stomp.Connection(host_and_ports = [(tmpBroker, 61023)], **ssl_opts)
connList.append(conn)
except:
errtype,errvalue = sys.exc_info()[:2]
_logger.error("failed to connect to %s : %s %s" % (tmpBroker,errtype,errvalue))
catch_sig(None,None)
while True:
for conn in connList:
try:
if not conn.is_connected():
conn.set_listener('DatasetCallbackListener', DatasetCallbackListener(conn,taskBuffer,siteMapper,
subscription_id))
conn.start()
conn.connect(headers = {'client-id': clientid})
conn.subscribe(destination=queue, id=subscription_id, ack='auto')
_logger.debug('listener %s is up and running' % clientid)
except:
errtype,errvalue = sys.exc_info()[:2]
_logger.error("failed to set listener on %s : %s %s" % (tmpBroker,errtype,errvalue))
catch_sig(None,None)
time.sleep(5)
entry point
"""
import datetime
# config file
from config import panda_config
# initialize cx_Oracle using dummy connection
from taskbuffer.Initializer import initializer
initializer.init()
# initialzie TaskBuffer
from taskbuffer.TaskBuffer import taskBuffer
taskBuffer.init(panda_config.dbhost,panda_config.dbpasswd,panda_config.nDBConnection,True)
# initialize JobDispatcher
from jobdispatcher.JobDispatcher import jobDispatcher
if panda_config.nDBConnection != 0:
jobDispatcher.init(taskBuffer)
# initialize DataService
from dataservice.DataService import dataService
if panda_config.nDBConnection != 0:
dataService.init(taskBuffer)
# initialize UserIF
from userinterface.UserIF import userIF
if panda_config.nDBConnection != 0:
userIF.init(taskBuffer)
job.homepackage='JobTransforms-11-00-01-01'
job.transformation='share/rome.g4sim.standard.trf'
job.jobParameters='%s %s 1 2 14268' % (lfnI,lfnO)
jobs.append(job)
self.taskbuffer.storeJobs(jobs,None)
time.sleep(self.interval)
from taskbuffer.TaskBuffer import taskBuffer
from jobdispatcher.JobDispatcher import jobDispatcher
from userinterface.UserIF import userIF
import getpass
passwd = getpass.getpass()
taskBuffer.init('adbpro.usatlas.bnl.gov',passwd,nDBConnection=3)
jobDispatcher.init(taskBuffer)
userIF.init(taskBuffer)
jobDefID = int(time.time()) % 10000
thr1 = TestThread(taskBuffer,4,jobDefID,"myhost")
thr2 = TestThread(taskBuffer,3,jobDefID+1,"testsite")
thr1.start()
#thr2.start()
from jobdispatcher.JobDispatcher import getJob,updateJob
from userinterface.UserIF import submitJobs,getJobStatus,queryPandaIDs
