| 1 |
slacapra |
1.1 |
from Scheduler import Scheduler
|
| 2 |
|
|
from crab_exceptions import *
|
| 3 |
|
|
from crab_logger import Logger
|
| 4 |
spiga |
1.38 |
from crab_util import getLocalDomain, uniqueTaskName
|
| 5 |
slacapra |
1.1 |
import common
|
| 6 |
spiga |
1.27 |
from PhEDExDatasvcInfo import PhEDExDatasvcInfo
|
| 7 |
slacapra |
1.1 |
|
| 8 |
|
|
import os,string
|
| 9 |
|
|
|
| 10 |
slacapra |
1.2 |
# Base class for all local scheduler
|
| 11 |
slacapra |
1.1 |
|
| 12 |
|
|
class SchedulerLocal(Scheduler) :
|
| 13 |
|
|
|
| 14 |
|
|
def configure(self, cfg_params):
|
| 15 |
|
|
|
| 16 |
spiga |
1.27 |
self.cfg_params = cfg_params
|
| 17 |
slacapra |
1.1 |
self.jobtypeName = cfg_params['CRAB.jobtype']
|
| 18 |
|
|
|
| 19 |
|
|
name=string.upper(self.name())
|
| 20 |
|
|
self.queue = cfg_params.get(name+'.queue',None)
|
| 21 |
|
|
|
| 22 |
|
|
self.res = cfg_params.get(name+'.resource',None)
|
| 23 |
|
|
|
| 24 |
|
|
if (cfg_params.has_key(self.name()+'.env_id')): self.environment_unique_identifier = cfg_params[self.name()+'.env_id']
|
| 25 |
|
|
|
| 26 |
slacapra |
1.3 |
self.return_data = int(cfg_params.get('USER.return_data',0))
|
| 27 |
spiga |
1.27 |
self.copy_data = int(cfg_params.get('USER.copy_data',0))
|
| 28 |
fanzago |
1.44 |
self.backup_copy = int(cfg_params.get('USER.backup_copy',0))
|
| 29 |
slacapra |
1.3 |
|
| 30 |
spiga |
1.37 |
self.check_RemoteDir = int(cfg_params.get('USER.check_user_remote_dir',0))
|
| 31 |
slacapra |
1.3 |
|
| 32 |
|
|
if ( self.return_data == 0 and self.copy_data == 0 ):
|
| 33 |
|
|
msg = 'Error: return_data = 0 and copy_data = 0 ==> your exe output will be lost\n'
|
| 34 |
|
|
msg = msg + 'Please modify return_data and copy_data value in your crab.cfg file\n'
|
| 35 |
|
|
raise CrabException(msg)
|
| 36 |
|
|
|
| 37 |
|
|
if ( self.return_data == 1 and self.copy_data == 1 ):
|
| 38 |
|
|
msg = 'Error: return_data and copy_data cannot be set both to 1\n'
|
| 39 |
|
|
msg = msg + 'Please modify return_data or copy_data value in your crab.cfg file\n'
|
| 40 |
|
|
raise CrabException(msg)
|
| 41 |
slacapra |
1.1 |
|
| 42 |
spiga |
1.27 |
self.publish_data = cfg_params.get("USER.publish_data",0)
|
| 43 |
|
|
if int(self.publish_data) == 1 and common.scheduler.name().upper() not in ['CAF']:
|
| 44 |
|
|
msg = "Error. User data publication not allowed while running on %s"%common.scheduler.name().upper()
|
| 45 |
|
|
raise CrabException(msg)
|
| 46 |
fanzago |
1.44 |
|
| 47 |
|
|
if ( int(self.copy_data) == 0 and int(self.backup_copy) == 1 ):
|
| 48 |
|
|
msg = 'Error: copy_data = 0 and backup_data = 1 ==> to use the backup_copy function, the copy_data value has to be = 1\n'
|
| 49 |
|
|
msg = msg + 'Please modify copy_data value in your crab.cfg file\n'
|
| 50 |
|
|
raise CrabException(msg)
|
| 51 |
|
|
|
| 52 |
|
|
if ( int(self.backup_copy) == 1 and int(self.publish_data) == 1 ):
|
| 53 |
|
|
msg = 'Warning: currently the publication is not supported with the backup copy. Work in progress....\n'
|
| 54 |
|
|
common.logger.message(msg)
|
| 55 |
|
|
raise CrabException(msg)
|
| 56 |
spiga |
1.27 |
|
| 57 |
slacapra |
1.1 |
## is this ok?
|
| 58 |
spiga |
1.25 |
localDomainName = getLocalDomain(self)
|
| 59 |
slacapra |
1.16 |
if not cfg_params.has_key('EDG.se_white_list'):
|
| 60 |
|
|
cfg_params['EDG.se_white_list']=localDomainName
|
| 61 |
|
|
common.logger.message("Your domain name is "+str(localDomainName)+": only local dataset will be considered")
|
| 62 |
|
|
else:
|
| 63 |
|
|
common.logger.message("Your se_white_list is set to "+str(cfg_params['EDG.se_white_list'])+": only local dataset will be considered")
|
| 64 |
spiga |
1.40 |
self.debug_wrapper = int(cfg_params.get('USER.debug_wrapper',0))
|
| 65 |
mcinquil |
1.41 |
self.debugWrap=''
|
| 66 |
spiga |
1.40 |
if self.debug_wrapper==1: self.debugWrap='--debug'
|
| 67 |
fanzago |
1.44 |
if ( int(self.backup_copy) == 1 ):
|
| 68 |
|
|
self.debugWrap='--debug'
|
| 69 |
|
|
self.backup='--backup'
|
| 70 |
slacapra |
1.16 |
|
| 71 |
slacapra |
1.13 |
Scheduler.configure(self,cfg_params)
|
| 72 |
slacapra |
1.1 |
return
|
| 73 |
|
|
|
| 74 |
|
|
def userName(self):
|
| 75 |
|
|
""" return the user name """
|
| 76 |
slacapra |
1.5 |
import pwd,getpass
|
| 77 |
|
|
tmp=pwd.getpwnam(getpass.getuser())[4]
|
| 78 |
|
|
return "/CN="+tmp.strip()
|
| 79 |
slacapra |
1.1 |
|
| 80 |
|
|
def wsSetupEnvironment(self):
|
| 81 |
|
|
"""
|
| 82 |
|
|
Returns part of a job script which does scheduler-specific work.
|
| 83 |
|
|
"""
|
| 84 |
spiga |
1.38 |
taskId = uniqueTaskName(common._db.queryTask('name'))
|
| 85 |
slacapra |
1.1 |
if not self.environment_unique_identifier:
|
| 86 |
|
|
raise CrabException('environment_unique_identifier not set')
|
| 87 |
|
|
|
| 88 |
slacapra |
1.12 |
index = int(common._db.nJobs())
|
| 89 |
|
|
job = common.job_list[index-1]
|
| 90 |
|
|
jbt = job.type()
|
| 91 |
ewv |
1.20 |
# start with wrapper timing
|
| 92 |
spiga |
1.18 |
txt = 'export TIME_WRAP_INI=`date +%s` \n'
|
| 93 |
farinafa |
1.22 |
txt += 'export TIME_STAGEOUT=-2 \n\n'
|
| 94 |
spiga |
1.17 |
|
| 95 |
spiga |
1.19 |
txt += '# '+self.name()+' specific stuff\n'
|
| 96 |
slacapra |
1.1 |
txt += '# strip arguments\n'
|
| 97 |
|
|
txt += 'echo "strip arguments"\n'
|
| 98 |
|
|
txt += 'args=("$@")\n'
|
| 99 |
|
|
txt += 'nargs=$#\n'
|
| 100 |
|
|
txt += 'shift $nargs\n'
|
| 101 |
|
|
txt += "# job number (first parameter for job wrapper)\n"
|
| 102 |
ewv |
1.9 |
txt += "NJob=${args[0]}; export NJob\n"
|
| 103 |
slacapra |
1.1 |
|
| 104 |
slacapra |
1.12 |
txt += "out_files=out_files_${NJob}; export out_files\n"
|
| 105 |
|
|
txt += "echo $out_files\n"
|
| 106 |
|
|
txt += jbt.outList()
|
| 107 |
|
|
|
| 108 |
slacapra |
1.6 |
txt += 'SyncGridJobId=`echo '+self.environment_unique_identifier+'`\n'
|
| 109 |
|
|
txt += 'MonitorJobID=`echo ${NJob}_${SyncGridJobId}`\n'
|
| 110 |
spiga |
1.38 |
txt += 'MonitorID=`echo ' + taskId + '`\n'
|
| 111 |
slacapra |
1.1 |
|
| 112 |
|
|
txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
|
| 113 |
|
|
txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
|
| 114 |
|
|
txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
|
| 115 |
slacapra |
1.7 |
txt += 'echo "SyncCE='+self.name()+'.`hostname -d`" | tee -a $RUNTIME_AREA/$repo \n'
|
| 116 |
spiga |
1.27 |
|
| 117 |
fanzago |
1.23 |
txt += 'middleware='+self.name().upper()+' \n'
|
| 118 |
slacapra |
1.1 |
|
| 119 |
|
|
txt += 'dumpStatus $RUNTIME_AREA/$repo \n'
|
| 120 |
|
|
|
| 121 |
slacapra |
1.12 |
txt += 'InputSandBox=${args[3]}\n'
|
| 122 |
|
|
|
| 123 |
slacapra |
1.1 |
txt += '\n\n'
|
| 124 |
|
|
|
| 125 |
|
|
return txt
|
| 126 |
|
|
|
| 127 |
spiga |
1.27 |
def wsCopyOutput_comm(self, pool=None):
|
| 128 |
slacapra |
1.3 |
"""
|
| 129 |
|
|
Write a CopyResults part of a job script, e.g.
|
| 130 |
|
|
to copy produced output into a storage element.
|
| 131 |
|
|
"""
|
| 132 |
spiga |
1.37 |
index = int(common._db.nJobs())
|
| 133 |
|
|
job = common.job_list[index-1]
|
| 134 |
|
|
jbt = job.type()
|
| 135 |
slacapra |
1.8 |
txt = '\n'
|
| 136 |
spiga |
1.27 |
if int(self.copy_data) == 1:
|
| 137 |
slacapra |
1.3 |
|
| 138 |
spiga |
1.27 |
stageout = PhEDExDatasvcInfo(self.cfg_params)
|
| 139 |
|
|
endpoint, lfn, SE, SE_PATH, user = stageout.getEndpoint()
|
| 140 |
spiga |
1.37 |
if self.check_RemoteDir == 1 :
|
| 141 |
|
|
self.checkRemoteDir(endpoint,jbt.outList('list') )
|
| 142 |
slacapra |
1.3 |
|
| 143 |
spiga |
1.27 |
txt += '#\n'
|
| 144 |
|
|
txt += '# COPY OUTPUT FILE TO '+SE_PATH+ '\n'
|
| 145 |
|
|
txt += '#\n\n'
|
| 146 |
|
|
|
| 147 |
fanzago |
1.35 |
txt += 'export SE='+SE+'\n'
|
| 148 |
spiga |
1.27 |
txt += 'echo "SE = $SE"\n'
|
| 149 |
|
|
txt += 'export SE_PATH='+SE_PATH+'\n'
|
| 150 |
|
|
txt += 'echo "SE_PATH = $SE_PATH"\n'
|
| 151 |
|
|
txt += 'export LFNBaseName='+lfn+'\n'
|
| 152 |
spiga |
1.33 |
txt += 'echo "LFNBaseName = $LFNBaseName"\n'
|
| 153 |
spiga |
1.27 |
txt += 'export USER='+user+'\n'
|
| 154 |
spiga |
1.33 |
txt += 'echo "USER = $USER"\n'
|
| 155 |
spiga |
1.27 |
txt += 'export endpoint='+endpoint+'\n'
|
| 156 |
spiga |
1.33 |
txt += 'echo "endpoint = $endpoint"\n'
|
| 157 |
fanzago |
1.36 |
### Needed i.e. for caf
|
| 158 |
|
|
if (pool) and (pool != 'None'):
|
| 159 |
spiga |
1.27 |
txt += 'export STAGE_SVCCLASS='+str(pool)+'\n'
|
| 160 |
|
|
|
| 161 |
|
|
txt += 'echo ">>> Copy output files from WN = `hostname` to $SE_PATH :"\n'
|
| 162 |
|
|
txt += 'export TIME_STAGEOUT_INI=`date +%s` \n'
|
| 163 |
|
|
txt += 'copy_exit_status=0\n'
|
| 164 |
fanzago |
1.44 |
#txt += 'echo "python cmscp.py --destination $endpoint --inputFileList $file_list --middleware $middleware '+self.debugWrap+' --lfn $LFNBaseName "\n'
|
| 165 |
|
|
#txt += 'python cmscp.py --destination $endpoint --inputFileList $file_list --middleware $middleware '+self.debugWrap+' --lfn $LFNBaseName \n'
|
| 166 |
|
|
txt += 'echo "python cmscp.py --destination $endpoint --inputFileList $file_list --middleware $middleware '+self.debugWrap+' --lfn $LFNBaseName '+self.backup+' "\n'
|
| 167 |
|
|
txt += 'python cmscp.py --destination $endpoint --inputFileList $file_list --middleware $middleware '+self.debugWrap+' --lfn $LFNBaseName '+self.backup+' \n'
|
| 168 |
fanzago |
1.43 |
|
| 169 |
spiga |
1.40 |
if self.debug_wrapper==1:
|
| 170 |
fanzago |
1.43 |
txt += 'if [ -f .SEinteraction.log ] ;then\n'
|
| 171 |
spiga |
1.27 |
txt += '########### details of SE interaction\n'
|
| 172 |
spiga |
1.39 |
txt += ' cat .SEinteraction.log\n'
|
| 173 |
|
|
txt += 'else\n'
|
| 174 |
|
|
txt += ' echo ".SEinteraction.log file not found"\n'
|
| 175 |
spiga |
1.40 |
txt += 'fi\n'
|
| 176 |
spiga |
1.45 |
txt += '##################################### \n'
|
| 177 |
spiga |
1.40 |
txt += 'if [ -f cmscpReport.sh ] ;then\n'
|
| 178 |
fanzago |
1.43 |
txt += ' cat cmscpReport.sh\n'
|
| 179 |
spiga |
1.40 |
txt += ' source cmscpReport.sh\n'
|
| 180 |
|
|
txt += 'else\n'
|
| 181 |
|
|
txt += ' echo "cmscpReport.sh file not found"\n'
|
| 182 |
|
|
txt += ' StageOutExitStatus=60307\n'
|
| 183 |
|
|
txt += 'fi\n'
|
| 184 |
spiga |
1.27 |
txt += 'if [ $StageOutExitStatus -ne 0 ]; then\n'
|
| 185 |
|
|
txt += ' echo "Problem copying file to $SE $SE_PATH"\n'
|
| 186 |
|
|
txt += ' copy_exit_status=$StageOutExitStatus \n'
|
| 187 |
spiga |
1.40 |
if not self.debug_wrapper==1:
|
| 188 |
spiga |
1.32 |
txt += ' ########### details of SE interaction\n'
|
| 189 |
spiga |
1.40 |
txt += ' if [ -f .SEinteraction.log ] ;then\n'
|
| 190 |
|
|
txt += ' cat .SEinteraction.log\n'
|
| 191 |
|
|
txt += ' else\n'
|
| 192 |
|
|
txt += ' echo ".SEinteraction.log file not found"\n'
|
| 193 |
|
|
txt += ' fi\n'
|
| 194 |
spiga |
1.45 |
txt += ' ##################################### \n'
|
| 195 |
spiga |
1.27 |
txt += ' job_exit_code=$StageOutExitStatus\n'
|
| 196 |
|
|
txt += 'fi\n'
|
| 197 |
|
|
txt += 'export TIME_STAGEOUT_END=`date +%s` \n'
|
| 198 |
|
|
txt += 'let "TIME_STAGEOUT = TIME_STAGEOUT_END - TIME_STAGEOUT_INI" \n'
|
| 199 |
|
|
else:
|
| 200 |
|
|
txt += 'export TIME_STAGEOUT=-1 \n'
|
| 201 |
slacapra |
1.3 |
return txt
|
| 202 |
spiga |
1.21 |
|
| 203 |
|
|
def wsExitFunc_comm(self):
|
| 204 |
|
|
"""
|
| 205 |
|
|
"""
|
| 206 |
|
|
txt = ''
|
| 207 |
|
|
txt += ' if [ $PYTHONPATH ]; then \n'
|
| 208 |
|
|
txt += ' if [ ! -s $RUNTIME_AREA/fillCrabFjr.py ]; then \n'
|
| 209 |
|
|
txt += ' echo "WARNING: it is not possible to create crab_fjr.xml to final report" \n'
|
| 210 |
|
|
txt += ' else \n'
|
| 211 |
|
|
txt += ' python $RUNTIME_AREA/fillCrabFjr.py $RUNTIME_AREA/crab_fjr_$NJob.xml --errorcode $job_exit_code $executable_exit_status \n'
|
| 212 |
|
|
txt += ' fi\n'
|
| 213 |
|
|
txt += ' fi\n'
|
| 214 |
|
|
txt += ' cd $RUNTIME_AREA \n'
|
| 215 |
|
|
txt += ' for file in $filesToCheck ; do\n'
|
| 216 |
|
|
txt += ' if [ -e $file ]; then\n'
|
| 217 |
|
|
txt += ' echo "tarring file $file in $out_files"\n'
|
| 218 |
|
|
txt += ' else\n'
|
| 219 |
|
|
txt += ' echo "WARNING: output file $file not found!"\n'
|
| 220 |
|
|
txt += ' fi\n'
|
| 221 |
|
|
txt += ' done\n'
|
| 222 |
|
|
txt += ' export TIME_WRAP_END=`date +%s`\n'
|
| 223 |
|
|
txt += ' let "TIME_WRAP = TIME_WRAP_END - TIME_WRAP_END_INI" \n'
|
| 224 |
|
|
txt += ' if [ $PYTHONPATH ]; then \n'
|
| 225 |
|
|
txt += ' if [ ! -s $RUNTIME_AREA/fillCrabFjr.py ]; then \n'
|
| 226 |
|
|
txt += ' echo "WARNING: it is not possible to create crab_fjr.xml to final report" \n'
|
| 227 |
|
|
txt += ' else \n'
|
| 228 |
|
|
# call timing FJR filling
|
| 229 |
|
|
txt += ' python $RUNTIME_AREA/fillCrabFjr.py $RUNTIME_AREA/crab_fjr_$NJob.xml --timing $TIME_WRAP $TIME_EXE $TIME_STAGEOUT \n'
|
| 230 |
|
|
txt += ' echo "CrabWrapperTime=$TIME_WRAP" >> $RUNTIME_AREA/$repo \n'
|
| 231 |
|
|
txt += ' if [ $TIME_STAGEOUT -lt 0 ]; then \n'
|
| 232 |
|
|
txt += ' export TIME_STAGEOUT=NULL \n'
|
| 233 |
|
|
txt += ' fi\n'
|
| 234 |
|
|
txt += ' echo "CrabStageoutTime=$TIME_STAGEOUT" >> $RUNTIME_AREA/$repo \n'
|
| 235 |
|
|
txt += ' fi\n'
|
| 236 |
|
|
txt += ' fi\n'
|
| 237 |
|
|
txt += ' dumpStatus $RUNTIME_AREA/$repo \n\n'
|
| 238 |
|
|
txt += ' echo "JOB_EXIT_STATUS = $job_exit_code"\n'
|
| 239 |
|
|
txt += ' echo "JobExitCode=$job_exit_code" >> $RUNTIME_AREA/$repo\n'
|
| 240 |
|
|
txt += ' dumpStatus $RUNTIME_AREA/$repo\n'
|
| 241 |
|
|
|
| 242 |
|
|
return txt
|
| 243 |
|
|
|
