| 1 |
slacapra |
1.1 |
from Actor import *
|
| 2 |
|
|
import common, crab_util
|
| 3 |
|
|
import string, os
|
| 4 |
|
|
|
| 5 |
|
|
class Status(Actor):
|
| 6 |
|
|
def __init__(self, cfg_params, nj_list):
|
| 7 |
|
|
self.cfg_params = cfg_params
|
| 8 |
|
|
self.nj_list = nj_list
|
| 9 |
|
|
self.countDone = 0
|
| 10 |
|
|
self.countReady = 0
|
| 11 |
|
|
self.countSched = 0
|
| 12 |
|
|
self.countRun = 0
|
| 13 |
|
|
self.countCleared = 0
|
| 14 |
|
|
self.countToTjob = 0
|
| 15 |
|
|
|
| 16 |
|
|
Status = crab_util.importName('edg_wl_userinterface_common_LbWrapper', 'Status')
|
| 17 |
|
|
# Bypass edg-job-status interfacing directly to C++ API
|
| 18 |
|
|
# Job attribute vector to retrieve status without edg-job-status
|
| 19 |
|
|
self.level = 0
|
| 20 |
|
|
# Instance of the Status class provided by LB API
|
| 21 |
|
|
self.jobStat = Status()
|
| 22 |
|
|
|
| 23 |
|
|
self.states = [ "Acl", "cancelReason", "cancelling","ce_node","children", \
|
| 24 |
|
|
"children_hist","children_num","children_states","condorId","condor_jdl", \
|
| 25 |
|
|
"cpuTime","destination", "done_code","exit_code","expectFrom", \
|
| 26 |
|
|
"expectUpdate","globusId","jdl","jobId","jobtype", \
|
| 27 |
|
|
"lastUpdateTime","localId","location", "matched_jdl","network_server", \
|
| 28 |
|
|
"owner","parent_job", "reason","resubmitted","rsl","seed","stateEnterTime","stateEnterTimes","subjob_failed", \
|
| 29 |
|
|
"user tags" , "status" , "status_code","hierarchy"]
|
| 30 |
|
|
self.hstates = {}
|
| 31 |
|
|
|
| 32 |
|
|
|
| 33 |
|
|
return
|
| 34 |
|
|
|
| 35 |
|
|
def run(self):
|
| 36 |
|
|
"""
|
| 37 |
|
|
The main method of the class.
|
| 38 |
|
|
"""
|
| 39 |
|
|
common.logger.debug(5, "Status::run() called")
|
| 40 |
|
|
|
| 41 |
|
|
common.jobDB.load()
|
| 42 |
|
|
for nj in self.nj_list:
|
| 43 |
|
|
st = common.jobDB.status(nj)
|
| 44 |
|
|
self.countToTjob = self.countToTjob + 1
|
| 45 |
|
|
jid = common.jobDB.jobId(nj)
|
| 46 |
|
|
if st == 'S':
|
| 47 |
|
|
result = self.getJobStatus_(jid, 'status')
|
| 48 |
|
|
self.processResult_(nj, result)
|
| 49 |
|
|
print 'Job %03d:'%(nj+1),jid,result
|
| 50 |
|
|
pass
|
| 51 |
|
|
else:
|
| 52 |
|
|
print 'Job %03d:'%(nj+1),jid,crabJobStatusToString(st)
|
| 53 |
|
|
pass
|
| 54 |
|
|
pass
|
| 55 |
|
|
|
| 56 |
slacapra |
1.2 |
common.jobDB.save()
|
| 57 |
|
|
|
| 58 |
slacapra |
1.1 |
self.Report_()
|
| 59 |
|
|
pass
|
| 60 |
|
|
|
| 61 |
|
|
|
| 62 |
|
|
def processResult_(self, nj, result):
|
| 63 |
|
|
#######################################################################################
|
| 64 |
|
|
self.hstates['destination'] = self.hstates['destination'].strip()
|
| 65 |
|
|
destination = self.hstates['destination'].split(":")[0]
|
| 66 |
|
|
self.hstates['jobId'] = self.hstates['jobId'].strip()
|
| 67 |
|
|
ID3 = self.hstates['jobId'].split("/")[3]
|
| 68 |
|
|
brokTmp = self.hstates['jobId'].split("/")[2]
|
| 69 |
|
|
broker = brokTmp.split(":")[0]
|
| 70 |
|
|
self.hstates['destination'] = self.hstates['destination'].strip()
|
| 71 |
|
|
destination = self.hstates['destination'].split(":")[0]
|
| 72 |
|
|
resFlag = 0
|
| 73 |
|
|
#######################################################################################
|
| 74 |
|
|
try:
|
| 75 |
|
|
if result == 'Done':
|
| 76 |
|
|
self.countDone = self.countDone + 1
|
| 77 |
|
|
exCode = self.hstates['exit_code']
|
| 78 |
|
|
#statistic.notify('checkstatus',resFlag,exCode,dataset,owner,destination,broker,ID3,ID1,NJC)
|
| 79 |
|
|
elif result == 'Ready':
|
| 80 |
|
|
self.countReady = self.countReady + 1
|
| 81 |
|
|
#statistic.notify('checkstatus',resFlag,'-----',dataset,owner,destination,broker,ID3,ID1,NJC)
|
| 82 |
|
|
elif result == 'Scheduled':
|
| 83 |
|
|
self.countSched = self.countSched + 1
|
| 84 |
|
|
#statistic.notify('checkstatus',resFlag,'-----',dataset,owner,destination,broker,ID3,ID1,NJC)
|
| 85 |
|
|
elif result == 'Running':
|
| 86 |
|
|
self.countRun = self.countRun + 1
|
| 87 |
|
|
#statistic.notify('checkstatus',resFlag,'-----',dataset,owner,destination,broker,ID3,ID1,NJC)
|
| 88 |
|
|
elif result == 'Aborted':
|
| 89 |
slacapra |
1.2 |
common.jobDB.setStatus(nj, 'A')
|
| 90 |
slacapra |
1.1 |
#job.saveJobStatus()
|
| 91 |
|
|
#statistic.notify('checkstatus',resFlag,'abort',dataset,owner,destination,broker,ID3,ID1,NJC)
|
| 92 |
|
|
pass
|
| 93 |
|
|
elif result == 'Cancelled':
|
| 94 |
|
|
#job.setStatus('K')
|
| 95 |
|
|
#job.saveJobStatus()
|
| 96 |
|
|
#statistic.notify('checkstatus',resFlag,'cancel',dataset,owner,destination,broker,ID3,ID1,NJC)
|
| 97 |
|
|
pass
|
| 98 |
|
|
elif result == 'Cleared':
|
| 99 |
|
|
#job.setStatus('P')
|
| 100 |
|
|
exCode = self.hstates['exit_code']
|
| 101 |
|
|
#statistic.notify('checkstatus',resFlag,exCode,dataset,owner,destination,broker,ID3,ID1,NJC)
|
| 102 |
|
|
#job.saveJobStatus()
|
| 103 |
|
|
self.countCleared = self.countCleared + 1
|
| 104 |
|
|
except UnboundLocalError:
|
| 105 |
|
|
common.logger.message('ERROR: UnboundLocalError with ')
|
| 106 |
|
|
|
| 107 |
|
|
def Report_(self) :
|
| 108 |
|
|
|
| 109 |
|
|
""" Report #jobs for each status """
|
| 110 |
|
|
|
| 111 |
|
|
#job_stat = common.job_list.loadStatus()
|
| 112 |
|
|
|
| 113 |
|
|
print ''
|
| 114 |
|
|
print ">>>>>>>>> %i Total Jobs " % (self.countToTjob)
|
| 115 |
|
|
|
| 116 |
|
|
if (self.countReady != 0):
|
| 117 |
|
|
print ''
|
| 118 |
|
|
print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
|
| 119 |
|
|
if (self.countSched != 0):
|
| 120 |
|
|
print ''
|
| 121 |
|
|
print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
|
| 122 |
|
|
if (self.countRun != 0):
|
| 123 |
|
|
print ''
|
| 124 |
|
|
print ">>>>>>>>> %i Jobs Running" % (self.countRun)
|
| 125 |
|
|
if (self.countCleared != 0):
|
| 126 |
|
|
print ''
|
| 127 |
|
|
print ">>>>>>>>> %i Jobs Retrieved (=Cleared)" % (self.countCleared)
|
| 128 |
|
|
print " You can resubmit them specifying JOB numbers: crab.py -resubmit JOB_number (or range of JOB) -continue"
|
| 129 |
|
|
print " (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"
|
| 130 |
|
|
# if job_stat[6] or job_stat[7]:
|
| 131 |
|
|
# print ''
|
| 132 |
|
|
# print ">>>>>>>>> %i Jobs aborted or killed(=cancelled by user)" % (job_stat[6] + job_stat[7])
|
| 133 |
|
|
# print " Resubmit them with: crab.py -resubmit -continue to resubmit all"
|
| 134 |
|
|
# print " or specifying JOB numbers (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"
|
| 135 |
|
|
# print " "
|
| 136 |
|
|
if (self.countDone != 0):
|
| 137 |
|
|
print ">>>>>>>>> %i Jobs Done" % (self.countDone)
|
| 138 |
|
|
print " Retrieve them with: crab.py -getoutput -continue to retrieve all"
|
| 139 |
|
|
print " or specifying JOB numbers (i.e -getoutput 1-3 => 1 and 2 and 3 or -getoutput 1,3 => 1 and 3)"
|
| 140 |
|
|
print('\n')
|
| 141 |
|
|
pass
|
| 142 |
|
|
|
| 143 |
|
|
|
| 144 |
|
|
def getJobStatus_(self, sid, attr):
|
| 145 |
|
|
result = ''
|
| 146 |
|
|
st = 0
|
| 147 |
|
|
self.jobStat.getStatus(sid, self.level)
|
| 148 |
|
|
(err, apiMsg) = self.jobStat.get_error()
|
| 149 |
|
|
if err:
|
| 150 |
|
|
common.logger.message(apiMsg)
|
| 151 |
|
|
return None
|
| 152 |
|
|
else:
|
| 153 |
|
|
for i in range(len(self.states)):
|
| 154 |
|
|
#print "states = ", self.states
|
| 155 |
|
|
# Fill an hash table with all information retrieved from LB API
|
| 156 |
|
|
self.hstates[ self.states[i] ] = self.jobStat.loadStatus(st)[i]
|
| 157 |
|
|
result = self.jobStat.loadStatus(st)[ self.states.index(attr) ]
|
| 158 |
|
|
return result
|
