def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
self.log = core_utils.make_logger(baseLogger,
'aCT sweeper',
method_name='__init__')
self.actDB = aCTDBPanda(self.log)
def bootstrap():
logger = aCTLogger('aCTBootstrap')
log = logger()
dbpanda = aCTDBPanda(log)
if not dbpanda.createTables():
print('Failed to create Panda tables, see aCTBootstrap.log for details')
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
self.hostname = socket.getfqdn()
# Set up aCT DB connection
self.log = core_utils.make_logger(baseLogger, 'aCT submitter', method_name='__init__')
self.actDB = aCTDBPanda(self.log)
# 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)))
# Map of role to aCT proxyid
self.proxymap = {}
# Get proxy info
# TODO: better to send aCT the proxy file and let it handle it
for role, proxy in self.certs.items():
cred_type = arc.initializeCredentialsType(arc.initializeCredentialsType.SkipCredentials)
uc = arc.UserConfig(cred_type)
uc.ProxyPath(str(proxy))
cred = arc.Credential(uc)
dn = cred.GetIdentityName()
actp = aCTProxy(self.log)
attr = '/atlas/Role='+role
proxyid = actp.getProxyId(dn, attr)
if not proxyid:
raise Exception("Proxy with DN {0} and attribute {1} was not found in proxies table".format(dn, attr))
self.proxymap[role] = proxyid
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
# Set up aCT DB connection
self.log = core_utils.make_logger(baseLogger, 'aCT submitter', method_name='__init__')
self.conf = aCTConfigARC()
self.actDB = aCTDBPanda(self.log, self.conf.get(["db", "file"]))
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
# Set up aCT DB connection
self.log = core_utils.make_logger(baseLogger, 'aCT submitter', method_name='__init__')
self.actDB = aCTDBPanda(self.log)
# 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)))
# Map of role to aCT proxyid
self.proxymap = {}
# Get proxy info
# TODO: better to send aCT the proxy file and let it handle it
for role, proxy in self.certs.items():
cred_type = arc.initializeCredentialsType(arc.initializeCredentialsType.SkipCredentials)
uc = arc.UserConfig(cred_type)
uc.ProxyPath(str(proxy))
cred = arc.Credential(uc)
dn = cred.GetIdentityName()
self.log.info("Proxy {0} with DN {1} and role {2}".format(proxy, dn, role))
actp = aCTProxy(self.log)
attr = '/atlas/Role='+role
proxyid = actp.getProxyId(dn, attr)
if not proxyid:
raise Exception("Proxy with DN {0} and attribute {1} was not found in proxies table".format(dn, attr))
self.proxymap[role] = proxyid
def __init__(self, ceflavour=['ARC-CE']):
# Get agent name from /path/to/aCTAgent.py
self.name = os.path.basename(sys.argv[0])[:-3]
# logger
self.logger = aCTLogger.aCTLogger(self.name)
self.log = self.logger()
self.criticallogger = aCTLogger.aCTLogger('aCTCritical', arclog=False)
self.criticallog = self.criticallogger()
# config
self.conf = aCTConfig.aCTConfigAPP()
self.arcconf = aCTConfig.aCTConfigARC()
self.tmpdir = str(self.arcconf.get(['tmp', 'dir']))
# database
self.dbarc = aCTDBArc.aCTDBArc(self.log)
self.dbcondor = aCTDBCondor.aCTDBCondor(self.log)
self.dbpanda = aCTDBPanda.aCTDBPanda(self.log)
# APFMon
self.apfmon = aCTAPFMon.aCTAPFMon(self.conf)
# CRIC info
self.flavour = ceflavour
self.cricparser = aCTCRICParser.aCTCRICParser(self.log)
self.sites = {}
self.osmap = {}
self.sitesselect = ''
# start time for periodic restart
self.starttime = time.time()
self.log.info("Started %s", self.name)
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
# Set up aCT DB connection
self.log = core_utils.make_logger(baseLogger,
'aCT submitter',
method_name='__init__')
self.conf = aCTConfigARC()
self.actDB = aCTDBPanda(self.log, self.conf.get(["db", "file"]))
# Get proxy info
# TODO: specify DN in conf instead
cred_type = arc.initializeCredentialsType(
arc.initializeCredentialsType.SkipCredentials)
uc = arc.UserConfig(cred_type)
uc.ProxyPath(str(self.conf.get(['voms', 'proxypath'])))
cred = arc.Credential(uc)
dn = cred.GetIdentityName()
self.log.info("Running under DN %s" % dn)
# Set up proxy map (prod/pilot roles)
self.proxymap = {}
actp = aCTProxy(self.log)
for role in self.conf.getList(['voms', 'roles', 'item']):
attr = '/atlas/Role=' + role
proxyid = actp.getProxyId(dn, attr)
if not proxyid:
raise Exception(
"Proxy with DN {0} and attribute {1} was not found in proxies table"
.format(dn, attr))
self.proxymap[role] = proxyid
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
# Set up aCT DB connection
self.log = core_utils.make_logger(baseLogger,
'aCT submitter',
method_name='__init__')
self.actDB = aCTDBPanda(self.log)
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
self.log = core_utils.make_logger(baseLogger, 'aCT sweeper', method_name='__init__')
try:
self.actDB = aCTDBPanda(self.log)
except Exception as e:
self.log.error('Could not connect to aCT database: {0}'.format(str(e)))
self.actDB = None
def __init__(self):
self.conf = aCTConfig.aCTConfigARC()
self.logger = aCTLogger.aCTLogger("aCTReport")
self.log = self.logger()
self.criticallogger = aCTLogger.aCTLogger('aCTCritical', arclog=False)
self.criticallog = self.criticallogger()
#self.db=aCTDB.aCTDB(None,self.conf.get(["db","file"]))
self.db = aCTDBArc.aCTDBArc(self.log, self.conf.get(["db", "file"]))
self.pandadb = aCTDBPanda.aCTDBPanda(self.log,
self.conf.get(["db", "file"]))
def main():
if len(sys.argv) != 2:
print("Usage: python aCTHeartbeatWatchdog.py timelimit")
sys.exit(1)
timelimit = int(sys.argv[1])
# logger
logger = aCTLogger('aCTHeartbeatWatchdog')
log = logger()
# database
dbarc = aCTDBArc(log)
dbpanda = aCTDBPanda(log)
# Query for running jobs with theartbeat longer than timelimit seconds ago
select = "sendhb=1 and " \
"pandastatus in ('sent', 'starting', 'running', 'transferring') and " \
"theartbeat != 0 and " + dbpanda.timeStampLessThan("theartbeat", timelimit)
columns = ['pandaid', 'pandastatus', 'proxyid', 'sitename', 'theartbeat']
jobs = dbpanda.getJobs(select, columns)
if jobs:
print(
'Found %d jobs with outdated heartbeat (older than %d seconds):\n'
% (len(jobs), timelimit))
print('\t'.join(
['pandaid', 'site', 'status', 'theartbeat', 'Panda response']))
# Panda server for each proxy
pandas = {}
for job in jobs:
proxyid = job['proxyid']
if proxyid not in pandas:
panda = aCTPanda(log, dbarc.getProxyPath(proxyid))
pandas[proxyid] = panda
response = pandas[proxyid].updateStatus(job['pandaid'],
job['pandastatus'])
print('\t'.join([
str(job['pandaid']), job['sitename'], job['pandastatus'],
str(job['theartbeat']),
str(response)
]))
# update heartbeat time in the DB
dbpanda.updateJob(
job['pandaid'],
{'theartbeat': dbpanda.getTimeStamp(time.time() + 1)})
def bootstrap_db():
'''Set up the DB tables'''
# TODO: setup only what is needed based on config and app
logger = aCTLogger('aCTBootstrap')
log = logger()
dbarc = aCTDBArc(log)
dbclient = ClientDB(log)
dbcondor = aCTDBCondor(log)
dbpanda = aCTDBPanda(log)
if not dbarc.createTables():
print('Error creating arc tables, see aCTBootstrap.log for details')
if not dbclient.createTables():
print('Error creating client tables, see aCTBootstrap.log for details')
if not dbcondor.createTables():
print('Error creating condor tables, see aCTBootstrap.log for details')
if not dbpanda.createTables():
print('Error creating panda tables, see aCTBootstrap.log for details')
def __init__(self, args):
self.output = ""
self.harvester = args.harvester
self.outfile = args.web
self.actconfs = args.conffiles or [''] # empty string for default behaviour
self.logger=aCTLogger.aCTLogger("aCTReport")
self.actlog=self.logger()
self.actlog.logger.setLevel(logging.ERROR)
self.criticallogger = aCTLogger.aCTLogger('aCTCritical', arclog=False)
self.criticallog = self.criticallogger()
if self.outfile:
self.log('<META HTTP-EQUIV="refresh" CONTENT="60"><pre>')
self.log(time.asctime() + '\n')
self.db=aCTDBArc.aCTDBArc(self.actlog)
self.pandadb=aCTDBPanda.aCTDBPanda(self.actlog)
from datetime import datetime
from act.arc.aCTDBArc import aCTDBArc
from act.atlas.aCTDBPanda import aCTDBPanda
from act.common.aCTLogger import aCTLogger
from act.common.aCTConfig import aCTConfigARC
try:
service_id, webpage_url = sys.argv[1:3]
except:
print('Usage: kibana.py service_id webpage_url')
sys.exit(1)
logger = aCTLogger('kibana probe')
log = logger()
arcdb = aCTDBArc(log)
pandadb = aCTDBPanda(log)
config = aCTConfigARC()
def getARCJobs():
return str(arcdb.getNArcJobs('TRUE'))
def getARCSlots():
jobs = arcdb.getArcJobsInfo("state='Running'", ['RequestedSlots'])
slots = 0
for j in jobs:
slots += j['RequestedSlots']
return str(slots)
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
# Set up aCT DB connection
self.log = core_utils.make_logger(baseLogger, 'aCT submitter', method_name='__init__')
self.actDB = aCTDBPanda(self.log)
def report(actconfs):
actlogger = aCTLogger('aCTReport')
logger = actlogger()
rep = {}
rtot = {}
log = ''
states = [
"sent", "starting", "running", "slots", "tovalidate", "toresubmit",
"toclean", "finished", "done", "failed", "donefailed", "tobekilled",
"cancelled", "donecancelled"
]
for conf in actconfs:
if conf:
os.environ['ACTCONFIGARC'] = conf
db = aCTDBPanda(logger)
c = db.db.conn.cursor()
c.execute("select sitename, actpandastatus, corecount from pandajobs")
rows = c.fetchall()
for r in rows:
site, state = (str(r[0]), str(r[1]))
if r[2] is None:
corecount = 1
else:
corecount = int(r[2])
try:
rep[site][state] += 1
if state == "running":
rep[site]["slots"] += corecount
except:
try:
rep[site][state] = 1
if state == "running":
try:
rep[site]["slots"] += corecount
except:
rep[site]["slots"] = corecount
except:
rep[site] = {}
rep[site][state] = 1
if state == "running":
rep[site]["slots"] = corecount
try:
rtot[state] += 1
if state == "running":
rtot["slots"] += corecount
except:
rtot[state] = 1
if state == "running":
rtot["slots"] = corecount
log += f"All Panda jobs: {sum([v for k,v in rtot.items() if k != 'slots'])}\n"
log += f"{'':29} {' '.join([f'{s:>9}' for s in states])}\n"
for k in sorted(rep.keys()):
log += f"{k:>28.28}:"
for s in states:
try:
log += f'{rep[k][s]:>10}'
except KeyError:
log += f'{"-":>10}'
log += '\n'
log += f'{"Totals":>28}:'
for s in states:
try:
log += f'{rtot[s]:>10}'
except:
log += f'{"-":>10}'
log += '\n\n'
if len(actconfs) == 1:
log += HarvesterReport()
return log
def __init__(self, **kwarg):
BaseMessenger.__init__(self, **kwarg)
# Set up aCT DB connection
self.log = core_utils.make_logger(baseLogger, 'aCT messenger', method_name='__init__')
self.actDB = aCTDBPanda(self.log)
def __init__(self, **kwarg):
PluginBase.__init__(self, **kwarg)
self.log = core_utils.make_logger(baseLogger, 'aCT sweeper', method_name='__init__')
self.actDB = aCTDBPanda(self.log)
