def updatePandaServer(job_state, xmlstr=None, spaceReport=False, log=None, jr=False, stdout_tail="", **kwargs):
""" Update the panda server with the latest job info """
from PandaServerClient import PandaServerClient
client = PandaServerClient(
pilot_version=DorE(kwargs, "version"),
pilot_version_tag=DorE(kwargs, "pilot_version_tag"),
pilot_initdir=DorE(kwargs, "pilot_initdir"),
jobSchedulerId=DorE(kwargs, "jobSchedulerId"),
pilotId=DorE(kwargs, "pilotId"),
updateServer=DorE(kwargs, "updateServerFlag"),
jobrec=DorE(kwargs, "jobrec"),
pshttpurl=DorE(kwargs, "pshttpurl"),
)
# update the panda server
return client.updatePandaServer(
job_state.job,
job_state.site,
DorE(kwargs, "workerNode"),
DorE(kwargs, "psport"),
xmlstr=xmlstr,
spaceReport=spaceReport,
log=log,
ra=job_state.recoveryAttempt,
jr=jr,
useCoPilot=DorE(kwargs, "useCoPilot"),
stdout_tail=stdout_tail,
)
def updatePandaServer(self, job, site, workerNode, port, xmlstr=None, spaceReport=False, log=None, ra=0, jr=False, schedulerID=None, pilotID=None, updateServer=True):
""" Update the PanDA server """
# create and instantiate the client object
from PandaServerClient import PandaServerClient
client = PandaServerClient(pilot_version=self.__pilot_version, \
pilot_version_tag=self.__pilot_version_tag, \
pilot_initdir=self.__pilot_initdir, \
jobSchedulerId=self.__jobSchedulerId, \
pilotId=self.__pilotId, \
updateServer=self.__updateServer, \
jobrec=True, \
pshttpurl=self.__pshttpurl)
# update the panda server
return client.updatePandaServer(job, site, workerNode, port, xmlstr=xmlstr, spaceReport=spaceReport, log=log, ra=ra, jr=jr)
def updatePandaServer(self, job, site, workerNode, port, xmlstr=None, spaceReport=False, log=None, ra=0, jr=False, schedulerID=None, pilotID=None, updateServer=True):
""" Update the PanDA server """
# create and instantiate the client object
from PandaServerClient import PandaServerClient
client = PandaServerClient(pilot_version=self.__pilot_version, \
pilot_version_tag=self.__pilot_version_tag, \
pilot_initdir=self.__pilot_initdir, \
jobSchedulerId=self.__jobSchedulerId, \
pilotId=self.__pilotId, \
updateServer=self.__updateServer, \
jobrec=True, \
pshttpurl=self.__pshttpurl)
# update the panda server
return client.updatePandaServer(job, site, workerNode, port, xmlstr=xmlstr, spaceReport=spaceReport, log=log, ra=ra, jr=jr)
def updatePandaServer(job_state, xmlstr=None, spaceReport=False,
log=None, jr=False, stdout_tail="", **kwargs):
""" Update the panda server with the latest job info """
from PandaServerClient import PandaServerClient
client = PandaServerClient(pilot_version=DorE(kwargs, 'version'),
pilot_version_tag=DorE(kwargs, 'pilot_version_tag'),
pilot_initdir=DorE(kwargs, 'pilot_initdir'),
jobSchedulerId=DorE(kwargs, 'jobSchedulerId'),
pilotId=DorE(kwargs, 'pilotId'),
updateServer=DorE(kwargs, 'updateServerFlag'),
jobrec=DorE(kwargs, 'jobrec'),
pshttpurl=DorE(kwargs, 'pshttpurl'))
# update the panda server
return client.updatePandaServer(job_state.job, job_state.site, DorE(kwargs, 'workerNode'),
DorE(kwargs, 'psport'), xmlstr=xmlstr,
spaceReport=spaceReport, log=log, ra=job_state.recoveryAttempt,
jr=jr, useCoPilot=DorE(kwargs, 'useCoPilot'),
stdout_tail=stdout_tail)
def __mk_gl_temp_dir(self):
"""Make the directory tree for glexec.
See usage example in:
http://wiki.nikhef.nl/grid/GLExec_TransientPilotJobs
"""
pUtil.tolog('sys path is %s' % sys.path)
pUtil.tolog("folder is : %s" % self.__mkgltempdir_path)
cmd = '%s -t 777 `pwd`' % self.__mkgltempdir_path
attempts = 0
while attempts < 3:
stdout, stderr, status = execute(cmd)
pUtil.tolog('cmd: %s' % cmd)
pUtil.tolog('output: %s' % stdout)
pUtil.tolog('error: %s' % stderr)
pUtil.tolog('status: %s' % status)
if not (status or stderr):
self.__target_path = stdout.rstrip('\n')
os.environ['GLEXEC_TARGET_DIR'] = self.__target_path
os.environ['GLEXEC_TARGET_PROXY'] = os.path.join(self.__target_path, 'user_proxy')
pUtil.tolog("gltmpdir created and added to env: %s" % self.__target_path)
pUtil.tolog("now adding sandbox to sys.path")
sys.path.append(self.__target_path)
pUtil.tolog("New sys.path is %s " %sys.path)
return 0
else:
pUtil.tolog('error! gltmpdir has failed')
attempts += 1
#raise GlexecException("mkgltempdir failed: %s" % stderr)
pUtil.tolog("mkgltempdir failed: %s" % stderr)
if attempts == 3:
pUtil.tolog('sys path is %s' % sys.path)
pUtil.tolog('os environ is %s' % os.environ)
ec = 1226
env = Configuration.Configuration()
pUtil.tolog("Updating PanDA server for the failed job (error code %d)" % (ec))
env['job'].result[0] = 'failed'
env['job'].currentState = env['job'].result[0]
env['job'].result[2] = ec
env['pilotErrorDiag'] = "gLExec related failure - %s" %stderr
env['job'].pilotErrorDiag = env['pilotErrorDiag']
from pilot import getProperNodeName
if 'https://' not in env['pshttpurl']:
env['pshttpurl'] = 'https://' + env['pshttpurl']
import Node#, Site
env['workerNode'] = Node.Node()
env['workerNode'].setNodeName(getProperNodeName(os.uname()[1]))
env['job'].workdir = os.getcwd()
env['thisSite'].workdir = os.getcwd()
from PandaServerClient import PandaServerClient
strXML = pUtil.getMetadata(env['thisSite'].workdir, env['job'].jobId)
client = PandaServerClient(pilot_version = env['version'], pilot_version_tag = env['pilot_version_tag'],
pilot_initdir = env['pilot_initdir'], jobSchedulerId = env['jobSchedulerId'],
pilotId = env['pilotId'], updateServer = env['updateServerFlag'],
jobrec = env['jobrec'], pshttpurl = env['pshttpurl'])
client.updatePandaServer(env['job'], env['thisSite'], env['workerNode'], env['psport'],
log = env['pilotErrorDiag'], useCoPilot = env['useCoPilot'], xmlstr = strXML)
return 1
else:
pUtil.tolog('[Trial %s] Sleeping for 10 secs and retrying' % attempts)
time.sleep(10)
def __mk_gl_temp_dir(self):
"""Make the directory tree for glexec.
See usage example in:
http://wiki.nikhef.nl/grid/GLExec_TransientPilotJobs
"""
pUtil.tolog('sys path is %s' % sys.path)
pUtil.tolog("folder is : %s" % self.__mkgltempdir_path)
cmd = '%s -t 777 `pwd`' % self.__mkgltempdir_path
attempts = 0
while attempts < 3:
stdout, stderr, status = execute(cmd)
pUtil.tolog('cmd: %s' % cmd)
pUtil.tolog('output: %s' % stdout)
pUtil.tolog('error: %s' % stderr)
pUtil.tolog('status: %s' % status)
if not (status or stderr):
self.__target_path = stdout.rstrip('\n')
os.environ['GLEXEC_TARGET_DIR'] = self.__target_path
os.environ['GLEXEC_TARGET_PROXY'] = os.path.join(self.__target_path, 'user_proxy')
pUtil.tolog("gltmpdir created and added to env: %s" % self.__target_path)
pUtil.tolog("now adding sandbox to sys.path")
sys.path.append(self.__target_path)
pUtil.tolog("New sys.path is %s " %sys.path)
return 0
else:
pUtil.tolog('error! gltmpdir has failed')
attempts += 1
#raise GlexecException("mkgltempdir failed: %s" % stderr)
pUtil.tolog("mkgltempdir failed: %s" % stderr)
if attempts == 3:
pUtil.tolog('sys path is %s' % sys.path)
pUtil.tolog('os environ is %s' % os.environ)
ec = 1226
env = Configuration.Configuration()
pUtil.tolog("Updating PanDA server for the failed job (error code %d)" % (ec))
env['job'].result[0] = 'failed'
env['job'].currentState = env['job'].result[0]
env['job'].result[2] = ec
env['pilotErrorDiag'] = "gLExec related failure - %s" %stderr
env['job'].pilotErrorDiag = env['pilotErrorDiag']
from pilot import getProperNodeName
if 'https://' not in env['pshttpurl']:
env['pshttpurl'] = 'https://' + env['pshttpurl']
import Node#, Site
env['workerNode'] = Node.Node()
env['workerNode'].setNodeName(getProperNodeName(os.uname()[1]))
env['job'].workdir = os.getcwd()
env['thisSite'].workdir = os.getcwd()
from PandaServerClient import PandaServerClient
strXML = pUtil.getMetadata(env['thisSite'].workdir, env['job'].jobId)
client = PandaServerClient(pilot_version = env['version'], pilot_version_tag = env['pilot_version_tag'],
pilot_initdir = env['pilot_initdir'], jobSchedulerId = env['jobSchedulerId'],
pilotId = env['pilotId'], updateServer = env['updateServerFlag'],
jobrec = env['jobrec'], pshttpurl = env['pshttpurl'])
client.updatePandaServer(env['job'], env['thisSite'], env['workerNode'], env['psport'],
log = env['pilotErrorDiag'], useCoPilot = env['useCoPilot'], xmlstr = strXML)
return 1
else:
pUtil.tolog('[Trial %s] Sleeping for 10 secs and retrying' % attempts)
time.sleep(10)
