| 1 |
spiga |
1.1 |
from Actor import *
|
| 2 |
|
|
from crab_util import *
|
| 3 |
|
|
import common
|
| 4 |
|
|
from ApmonIf import ApmonIf
|
| 5 |
|
|
import Statistic
|
| 6 |
|
|
import time
|
| 7 |
|
|
from ProgressBar import ProgressBar
|
| 8 |
|
|
from TerminalController import TerminalController
|
| 9 |
|
|
|
| 10 |
|
|
import xml.dom.minidom
|
| 11 |
|
|
import xml.dom.ext
|
| 12 |
|
|
import TaskDB
|
| 13 |
|
|
|
| 14 |
|
|
class StatusServer(Actor):
|
| 15 |
|
|
|
| 16 |
|
|
def __init__(self, cfg_params,):
|
| 17 |
|
|
self.cfg_params = cfg_params
|
| 18 |
mcinquil |
1.2 |
|
| 19 |
|
|
self.countSubmit = 0
|
| 20 |
|
|
self.countSubmitting = 0
|
| 21 |
|
|
self.countDone = 0
|
| 22 |
|
|
self.countReady = 0
|
| 23 |
|
|
self.countSched = 0
|
| 24 |
|
|
self.countRun = 0
|
| 25 |
|
|
self.countAbort = 0
|
| 26 |
|
|
self.countCancel = 0
|
| 27 |
|
|
self.countCleared = 0
|
| 28 |
|
|
self.countToTjob = 0
|
| 29 |
|
|
|
| 30 |
spiga |
1.3 |
try:
|
| 31 |
|
|
self.server_name = self.cfg_params['CRAB.server_name'] # gsiftp://pcpg01.cern.ch/data/SEDir/
|
| 32 |
|
|
except KeyError:
|
| 33 |
|
|
msg = 'No server selected ...'
|
| 34 |
|
|
msg = msg + 'Please specify a server in the crab cfg file'
|
| 35 |
|
|
raise CrabException(msg)
|
| 36 |
|
|
|
| 37 |
spiga |
1.1 |
return
|
| 38 |
mcinquil |
1.2 |
|
| 39 |
|
|
def translateStatus(self, status):
|
| 40 |
|
|
"""
|
| 41 |
|
|
simmetric as server
|
| 42 |
|
|
"""
|
| 43 |
|
|
|
| 44 |
mcinquil |
1.4 |
stateConverting = {'Running': 'R', 'Aborted': 'A', 'Done': 'D', 'Done (Failed)': 'D',\
|
| 45 |
mcinquil |
1.2 |
'Cleared': 'D', 'Cancelled': 'K', 'Killed': 'K'}
|
| 46 |
|
|
|
| 47 |
|
|
if status in stateConverting:
|
| 48 |
|
|
return stateConverting[status]
|
| 49 |
|
|
return None
|
| 50 |
|
|
|
| 51 |
|
|
|
| 52 |
spiga |
1.1 |
def run(self):
|
| 53 |
|
|
"""
|
| 54 |
|
|
The main method of the class: check the status of the task
|
| 55 |
|
|
"""
|
| 56 |
|
|
common.logger.debug(5, "status server::run() called")
|
| 57 |
|
|
start = time.time()
|
| 58 |
|
|
|
| 59 |
|
|
totalCreatedJobs = 0
|
| 60 |
|
|
flagSubmit = 1
|
| 61 |
|
|
for nj in range(common.jobDB.nJobs()):
|
| 62 |
|
|
if (common.jobDB.status(nj)!='S'):
|
| 63 |
|
|
totalCreatedJobs +=1
|
| 64 |
mcinquil |
1.2 |
# flagSubmit = 0
|
| 65 |
spiga |
1.1 |
|
| 66 |
|
|
if not flagSubmit:
|
| 67 |
|
|
common.logger.message("Not all jobs are submitted: before checking the status submit all the jobs.")
|
| 68 |
|
|
return
|
| 69 |
|
|
|
| 70 |
|
|
common.scheduler.checkProxy()
|
| 71 |
|
|
|
| 72 |
|
|
common.taskDB.load()
|
| 73 |
|
|
WorkDirName =os.path.basename(os.path.split(common.work_space.topDir())[0])
|
| 74 |
|
|
projectUniqName = 'crab_'+str(WorkDirName)+'_'+common.taskDB.dict('TasKUUID')
|
| 75 |
|
|
try:
|
| 76 |
|
|
common.logger.message ("Checking the status...\n")
|
| 77 |
spiga |
1.3 |
cmd = 'lcg-cp --vo cms gsiftp://' + str(self.server_name) + str(projectUniqName)+'/res/xmlReportFile.xml file://'+common.work_space.resDir()+'xmlReportFile.xml'
|
| 78 |
mcinquil |
1.2 |
common.logger.debug(6, cmd)
|
| 79 |
spiga |
1.1 |
os.system(cmd +' >& /dev/null')
|
| 80 |
|
|
|
| 81 |
|
|
except:
|
| 82 |
|
|
msg = ("task status not yet available")
|
| 83 |
|
|
raise CrabException(msg)
|
| 84 |
|
|
|
| 85 |
|
|
try:
|
| 86 |
|
|
file = open(common.work_space.resDir()+"xmlReportFile.xml", "r")
|
| 87 |
|
|
doc = xml.dom.minidom.parse(common.work_space.resDir()+ "xmlReportFile.xml" )
|
| 88 |
|
|
|
| 89 |
|
|
except:
|
| 90 |
|
|
msg = ("problems reading report file")
|
| 91 |
|
|
raise CrabException(msg)
|
| 92 |
|
|
|
| 93 |
mcinquil |
1.2 |
### <Job status='Submitted' job_exit='NULL' id='1' exe_exit='NULL'/>
|
| 94 |
|
|
|
| 95 |
spiga |
1.1 |
task = doc.childNodes[0].childNodes[1].getAttribute("taskName")
|
| 96 |
mcinquil |
1.2 |
self.countToTjob = int(doc.childNodes[0].childNodes[1].getAttribute("totJob") )
|
| 97 |
|
|
|
| 98 |
|
|
addTree = 3
|
| 99 |
|
|
|
| 100 |
|
|
common.jobDB.load()
|
| 101 |
|
|
|
| 102 |
|
|
if doc.childNodes[0].childNodes[3].getAttribute("id") == "all":
|
| 103 |
|
|
if doc.childNodes[0].childNodes[3].getAttribute("status") == "Submitted":
|
| 104 |
|
|
self.countSubmitting = common.jobDB.nJobs()
|
| 105 |
|
|
for nj in range(common.jobDB.nJobs()):
|
| 106 |
|
|
common.jobDB.setStatus(nj, 'S')
|
| 107 |
|
|
else:
|
| 108 |
|
|
self.countAbort = common.jobDB.nJobs()
|
| 109 |
|
|
for nj in range(common.jobDB.nJobs()):
|
| 110 |
|
|
common.jobDB.setStatus(nj, 'A')
|
| 111 |
|
|
self.countToTjob = common.jobDB.nJobs()
|
| 112 |
spiga |
1.1 |
else:
|
| 113 |
mcinquil |
1.2 |
printline = ''
|
| 114 |
|
|
printline+= "%-10s %-20s %-20s %-25s" % ('JOBID','STATUS','EXE_EXIT_CODE','JOB_EXIT_STATUS')
|
| 115 |
|
|
print printline
|
| 116 |
|
|
print '-------------------------------------------------------------------------------------'
|
| 117 |
|
|
|
| 118 |
|
|
for job in range( self.countToTjob ):
|
| 119 |
|
|
idJob = doc.childNodes[0].childNodes[job+addTree].getAttribute("id")
|
| 120 |
|
|
stato = doc.childNodes[0].childNodes[job+addTree].getAttribute("status")
|
| 121 |
|
|
exe_exit_code = doc.childNodes[0].childNodes[job+addTree].getAttribute("job_exit")
|
| 122 |
|
|
job_exit_status = doc.childNodes[0].childNodes[job+addTree].getAttribute("exe_exit")
|
| 123 |
mcinquil |
1.5 |
cleared = doc.childNodes[0].childNodes[job+addTree].getAttribute("cleared")
|
| 124 |
mcinquil |
1.2 |
jobDbStatus = self.translateStatus(stato)
|
| 125 |
spiga |
1.1 |
|
| 126 |
mcinquil |
1.2 |
if jobDbStatus != None:
|
| 127 |
|
|
common.logger.debug(5, '*** Updating jobdb for job %s ***' %idJob)
|
| 128 |
|
|
if common.jobDB.status( str(int(idJob)-1) ) != "Y":
|
| 129 |
mcinquil |
1.5 |
if jobDbStatus == 'D' and int(cleared) != 1:#exe_exit_code =='' and job_exit_status=='':
|
| 130 |
|
|
## 'Done' but not yet cleared (server side) still showing 'Running'
|
| 131 |
|
|
stato = 'Running'
|
| 132 |
|
|
jobDbStatus = 'R'
|
| 133 |
mcinquil |
1.2 |
common.jobDB.setStatus( str(int(idJob)-1), self.translateStatus(stato) )
|
| 134 |
|
|
else:
|
| 135 |
|
|
stato = "Cleared"
|
| 136 |
|
|
common.jobDB.setExitStatus( str(int(idJob)-1), job_exit_status )
|
| 137 |
mcinquil |
1.4 |
if stato != "Done" and stato != "Cleared" and stato != "Aborted" and stato != "Done (Failed)":
|
| 138 |
mcinquil |
1.2 |
print "%-10s %-20s" % (idJob,stato)
|
| 139 |
|
|
else:
|
| 140 |
|
|
print "%-10s %-20s %-20s %-25s" % (idJob,stato,exe_exit_code,job_exit_status)
|
| 141 |
|
|
|
| 142 |
|
|
if stato == 'Running':
|
| 143 |
|
|
self.countRun += 1
|
| 144 |
|
|
elif stato == 'Aborted':
|
| 145 |
|
|
self.countAbort += 1
|
| 146 |
|
|
elif stato == 'Done':
|
| 147 |
|
|
self.countDone += 1
|
| 148 |
|
|
elif stato == 'Cancelled':
|
| 149 |
|
|
self.countCancel += 1
|
| 150 |
|
|
elif stato == 'Submitted':
|
| 151 |
|
|
self.countSubmit += 1
|
| 152 |
|
|
elif stato == 'Submitting':
|
| 153 |
|
|
self.countSubmitting += 1
|
| 154 |
|
|
elif stato == 'Ready':
|
| 155 |
|
|
self.countReady += 1
|
| 156 |
|
|
elif stato == 'Scheduled':
|
| 157 |
|
|
self.countSched += 1
|
| 158 |
|
|
elif stato == 'Cleared':
|
| 159 |
|
|
self.countCleared += 1
|
| 160 |
|
|
|
| 161 |
|
|
addTree += 1
|
| 162 |
|
|
common.jobDB.save()
|
| 163 |
|
|
|
| 164 |
|
|
self.PrintReport_()
|
| 165 |
|
|
|
| 166 |
|
|
|
| 167 |
|
|
def PrintReport_(self) :
|
| 168 |
|
|
|
| 169 |
|
|
""" Report #jobs for each status """
|
| 170 |
|
|
|
| 171 |
|
|
|
| 172 |
|
|
print ''
|
| 173 |
|
|
print ">>>>>>>>> %i Total Jobs " % (self.countToTjob)
|
| 174 |
|
|
print ''
|
| 175 |
|
|
|
| 176 |
|
|
if (self.countReady != 0):
|
| 177 |
|
|
print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
|
| 178 |
|
|
if (self.countSubmitting != 0) :
|
| 179 |
|
|
print ">>>>>>>>> %i Jobs Submitting by the server" % (self.countSubmitting)
|
| 180 |
|
|
if (self.countSubmit != 0):
|
| 181 |
|
|
print ">>>>>>>>> %i Jobs Submitted" % (self.countSubmit)
|
| 182 |
|
|
if (self.countAbort != 0):
|
| 183 |
|
|
print ">>>>>>>>> %i Jobs Aborted" % (self.countAbort)
|
| 184 |
|
|
if (self.countSched != 0):
|
| 185 |
|
|
print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
|
| 186 |
|
|
if (self.countRun != 0):
|
| 187 |
|
|
print ">>>>>>>>> %i Jobs Running" % (self.countRun)
|
| 188 |
|
|
if (self.countCleared != 0):
|
| 189 |
|
|
print ">>>>>>>>> %i Jobs Cleared" % (self.countRun)
|
| 190 |
|
|
if (self.countDone != 0):
|
| 191 |
|
|
print ">>>>>>>>> %i Jobs Done" % (self.countDone)
|
| 192 |
|
|
print " Retrieve them with: crab.py -getoutput -continue"
|
| 193 |
|
|
print ''
|
| 194 |
|
|
pass
|
| 195 |
spiga |
1.1 |
|
