| 1 |
slacapra |
1.1 |
from Scheduler import Scheduler
|
| 2 |
|
|
from crab_exceptions import *
|
| 3 |
|
|
from crab_logger import Logger
|
| 4 |
|
|
import common
|
| 5 |
|
|
|
| 6 |
|
|
import os,string
|
| 7 |
|
|
|
| 8 |
slacapra |
1.2 |
# Base class for all local scheduler
|
| 9 |
slacapra |
1.1 |
|
| 10 |
|
|
class SchedulerLocal(Scheduler) :
|
| 11 |
|
|
|
| 12 |
|
|
def configure(self, cfg_params):
|
| 13 |
|
|
|
| 14 |
|
|
self.jobtypeName = cfg_params['CRAB.jobtype']
|
| 15 |
|
|
|
| 16 |
|
|
name=string.upper(self.name())
|
| 17 |
|
|
self.queue = cfg_params.get(name+'.queue',None)
|
| 18 |
|
|
|
| 19 |
|
|
self.res = cfg_params.get(name+'.resource',None)
|
| 20 |
|
|
|
| 21 |
|
|
if (cfg_params.has_key(self.name()+'.env_id')): self.environment_unique_identifier = cfg_params[self.name()+'.env_id']
|
| 22 |
|
|
|
| 23 |
spiga |
1.14 |
self._taskId=str("_".join(common._db.queryTask('name').split('_')[:-1]))
|
| 24 |
slacapra |
1.1 |
|
| 25 |
slacapra |
1.3 |
self.return_data = int(cfg_params.get('USER.return_data',0))
|
| 26 |
|
|
|
| 27 |
fanzago |
1.23 |
## FEDE publication options
|
| 28 |
|
|
self.publish_data = cfg_params.get("USER.publish_data",0)
|
| 29 |
|
|
if int(self.publish_data) == 1 :
|
| 30 |
|
|
self.publish_data_name = cfg_params.get('USER.publish_data_name',None)
|
| 31 |
|
|
if not self.publish_data_name and int(self.publish_data) == 1:
|
| 32 |
|
|
msg = "Error. The [USER] section does not have 'publish_data_name'"
|
| 33 |
|
|
raise CrabException(msg)
|
| 34 |
|
|
self.SE = cfg_params.get("USER.storage_element", None)
|
| 35 |
|
|
if not self.SE:
|
| 36 |
|
|
msg = "Error. The [USER] section does not have 'storage_element'.\n"
|
| 37 |
|
|
msg = msg + "Please fill this field if you want to publish your data"
|
| 38 |
|
|
raise CrabException(msg)
|
| 39 |
slacapra |
1.3 |
self.copy_data = int(cfg_params.get("USER.copy_data",0))
|
| 40 |
|
|
if self.copy_data == 1:
|
| 41 |
fanzago |
1.24 |
self._copyCommand = cfg_params.get('USER.copycommand','rfcp')
|
| 42 |
|
|
common.logger.debug(3, "copyCommand set to "+ self._copyCommand)
|
| 43 |
slacapra |
1.3 |
self.SE_path= cfg_params.get('USER.storage_path',None)
|
| 44 |
|
|
if not self.SE_path:
|
| 45 |
fanzago |
1.23 |
# do not allow CASTOR_HOME if publish_data is enabled
|
| 46 |
|
|
if int(self.publish_data) == 0 and os.environ.has_key('CASTOR_HOME'):
|
| 47 |
slacapra |
1.3 |
self.SE_path=os.environ['CASTOR_HOME']
|
| 48 |
|
|
else:
|
| 49 |
|
|
msg='No USER.storage_path has been provided: cannot copy_output'
|
| 50 |
|
|
raise CrabException(msg)
|
| 51 |
|
|
pass
|
| 52 |
|
|
pass
|
| 53 |
fanzago |
1.23 |
### FEDE to improve the final / control
|
| 54 |
slacapra |
1.10 |
self.SE_path+='/'
|
| 55 |
slacapra |
1.3 |
|
| 56 |
|
|
if ( self.return_data == 0 and self.copy_data == 0 ):
|
| 57 |
|
|
msg = 'Error: return_data = 0 and copy_data = 0 ==> your exe output will be lost\n'
|
| 58 |
|
|
msg = msg + 'Please modify return_data and copy_data value in your crab.cfg file\n'
|
| 59 |
|
|
raise CrabException(msg)
|
| 60 |
|
|
|
| 61 |
|
|
if ( self.return_data == 1 and self.copy_data == 1 ):
|
| 62 |
|
|
msg = 'Error: return_data and copy_data cannot be set both to 1\n'
|
| 63 |
|
|
msg = msg + 'Please modify return_data or copy_data value in your crab.cfg file\n'
|
| 64 |
|
|
raise CrabException(msg)
|
| 65 |
slacapra |
1.1 |
|
| 66 |
|
|
## Get local domain name
|
| 67 |
|
|
import socket
|
| 68 |
|
|
tmp=socket.gethostname()
|
| 69 |
|
|
dot=string.find(tmp,'.')
|
| 70 |
|
|
if (dot==-1):
|
| 71 |
|
|
msg='Unkown domain name. Cannot use local scheduler'
|
| 72 |
|
|
raise CrabException(msg)
|
| 73 |
|
|
localDomainName = string.split(tmp,'.',1)[-1]
|
| 74 |
slacapra |
1.11 |
#common.taskDB.setDict('localSite',localDomainName)
|
| 75 |
slacapra |
1.1 |
## is this ok?
|
| 76 |
slacapra |
1.16 |
if not cfg_params.has_key('EDG.se_white_list'):
|
| 77 |
|
|
cfg_params['EDG.se_white_list']=localDomainName
|
| 78 |
|
|
common.logger.message("Your domain name is "+str(localDomainName)+": only local dataset will be considered")
|
| 79 |
|
|
else:
|
| 80 |
|
|
common.logger.message("Your se_white_list is set to "+str(cfg_params['EDG.se_white_list'])+": only local dataset will be considered")
|
| 81 |
|
|
|
| 82 |
ewv |
1.20 |
|
| 83 |
|
|
|
| 84 |
slacapra |
1.1 |
|
| 85 |
slacapra |
1.13 |
Scheduler.configure(self,cfg_params)
|
| 86 |
slacapra |
1.1 |
return
|
| 87 |
|
|
|
| 88 |
|
|
def userName(self):
|
| 89 |
|
|
""" return the user name """
|
| 90 |
slacapra |
1.5 |
import pwd,getpass
|
| 91 |
|
|
tmp=pwd.getpwnam(getpass.getuser())[4]
|
| 92 |
|
|
return "/CN="+tmp.strip()
|
| 93 |
slacapra |
1.1 |
|
| 94 |
|
|
def wsSetupEnvironment(self):
|
| 95 |
|
|
"""
|
| 96 |
|
|
Returns part of a job script which does scheduler-specific work.
|
| 97 |
|
|
"""
|
| 98 |
|
|
if not self.environment_unique_identifier:
|
| 99 |
|
|
raise CrabException('environment_unique_identifier not set')
|
| 100 |
|
|
|
| 101 |
slacapra |
1.12 |
index = int(common._db.nJobs())
|
| 102 |
|
|
job = common.job_list[index-1]
|
| 103 |
|
|
jbt = job.type()
|
| 104 |
ewv |
1.20 |
# start with wrapper timing
|
| 105 |
spiga |
1.18 |
txt = 'export TIME_WRAP_INI=`date +%s` \n'
|
| 106 |
farinafa |
1.22 |
txt += 'export TIME_STAGEOUT=-2 \n\n'
|
| 107 |
spiga |
1.17 |
|
| 108 |
spiga |
1.19 |
txt += '# '+self.name()+' specific stuff\n'
|
| 109 |
slacapra |
1.1 |
txt += '# strip arguments\n'
|
| 110 |
|
|
txt += 'echo "strip arguments"\n'
|
| 111 |
|
|
txt += 'args=("$@")\n'
|
| 112 |
|
|
txt += 'nargs=$#\n'
|
| 113 |
|
|
txt += 'shift $nargs\n'
|
| 114 |
|
|
txt += "# job number (first parameter for job wrapper)\n"
|
| 115 |
ewv |
1.9 |
txt += "NJob=${args[0]}; export NJob\n"
|
| 116 |
slacapra |
1.1 |
|
| 117 |
slacapra |
1.12 |
txt += "out_files=out_files_${NJob}; export out_files\n"
|
| 118 |
|
|
txt += "echo $out_files\n"
|
| 119 |
|
|
txt += jbt.outList()
|
| 120 |
|
|
|
| 121 |
slacapra |
1.6 |
txt += 'SyncGridJobId=`echo '+self.environment_unique_identifier+'`\n'
|
| 122 |
|
|
txt += 'MonitorJobID=`echo ${NJob}_${SyncGridJobId}`\n'
|
| 123 |
slacapra |
1.1 |
txt += 'MonitorID=`echo ' + self._taskId + '`\n'
|
| 124 |
|
|
|
| 125 |
|
|
txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
|
| 126 |
|
|
txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
|
| 127 |
|
|
txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
|
| 128 |
slacapra |
1.7 |
txt += 'echo "SyncCE='+self.name()+'.`hostname -d`" | tee -a $RUNTIME_AREA/$repo \n'
|
| 129 |
fanzago |
1.23 |
###### FEDE
|
| 130 |
|
|
txt += 'middleware='+self.name().upper()+' \n'
|
| 131 |
slacapra |
1.1 |
|
| 132 |
|
|
txt += 'dumpStatus $RUNTIME_AREA/$repo \n'
|
| 133 |
|
|
|
| 134 |
slacapra |
1.12 |
txt += 'InputSandBox=${args[3]}\n'
|
| 135 |
|
|
|
| 136 |
slacapra |
1.1 |
txt += '\n\n'
|
| 137 |
|
|
|
| 138 |
|
|
return txt
|
| 139 |
|
|
|
| 140 |
slacapra |
1.3 |
def wsCopyOutput(self):
|
| 141 |
|
|
"""
|
| 142 |
|
|
Write a CopyResults part of a job script, e.g.
|
| 143 |
|
|
to copy produced output into a storage element.
|
| 144 |
|
|
"""
|
| 145 |
slacapra |
1.8 |
txt = '\n'
|
| 146 |
|
|
if not self.copy_data: return txt
|
| 147 |
slacapra |
1.3 |
|
| 148 |
|
|
|
| 149 |
|
|
txt += '#\n'
|
| 150 |
fanzago |
1.23 |
txt += '# COPY OUTPUT FILE TO '+self.SE_path+ '\n'
|
| 151 |
slacapra |
1.3 |
txt += '#\n\n'
|
| 152 |
|
|
|
| 153 |
|
|
txt += 'export SE_PATH='+self.SE_path+'\n'
|
| 154 |
|
|
|
| 155 |
fanzago |
1.24 |
txt += 'export CP_CMD='+self._copyCommand+'\n'
|
| 156 |
|
|
common.logger.debug(3, "Wrapper script CP_CMD set to "+ self._copyCommand)
|
| 157 |
slacapra |
1.3 |
|
| 158 |
fanzago |
1.24 |
txt += 'echo ">>> Copy output files from WN = `hostname` to PATH = $SE_PATH using $CP_CMD :"\n'
|
| 159 |
slacapra |
1.3 |
|
| 160 |
fanzago |
1.24 |
txt += 'if [ $job_exit_code -eq 60302 ]; then\n'
|
| 161 |
|
|
txt += ' echo "--> No output file to copy to $SE"\n'
|
| 162 |
|
|
txt += ' copy_exit_status=$job_exit_code\n'
|
| 163 |
|
|
txt += ' echo "COPY_EXIT_STATUS = $copy_exit_status"\n'
|
| 164 |
|
|
txt += 'else\n'
|
| 165 |
|
|
txt += ' for out_file in $file_list ; do\n'
|
| 166 |
slacapra |
1.3 |
txt += ' echo "Trying to copy output file to $SE_PATH"\n'
|
| 167 |
fanzago |
1.24 |
txt += ' $CP_CMD $SOFTWARE_DIR/$out_file ${SE_PATH}/$out_file\n'
|
| 168 |
|
|
txt += ' copy_exit_status=$?\n'
|
| 169 |
|
|
txt += ' echo "COPY_EXIT_STATUS = $copy_exit_status"\n'
|
| 170 |
|
|
txt += ' echo "STAGE_OUT = $copy_exit_status"\n'
|
| 171 |
|
|
txt += ' if [ $copy_exit_status -ne 0 ]; then\n'
|
| 172 |
|
|
txt += ' echo "Problem copying $out_file to $SE $SE_PATH"\n'
|
| 173 |
|
|
txt += ' echo "StageOutExitStatus = $copy_exit_status " | tee -a $RUNTIME_AREA/$repo\n'
|
| 174 |
slacapra |
1.3 |
txt += ' else\n'
|
| 175 |
fanzago |
1.24 |
txt += ' echo "StageOutSE = $SE" | tee -a $RUNTIME_AREA/$repo\n'
|
| 176 |
|
|
txt += ' echo "StageOutCatalog = " | tee -a $RUNTIME_AREA/$repo\n'
|
| 177 |
slacapra |
1.3 |
txt += ' echo "output copied into $SE/$SE_PATH directory"\n'
|
| 178 |
fanzago |
1.24 |
txt += ' echo "StageOutExitStatus = 0" | tee -a $RUNTIME_AREA/$repo\n'
|
| 179 |
slacapra |
1.3 |
txt += ' fi\n'
|
| 180 |
fanzago |
1.24 |
txt += ' done\n'
|
| 181 |
slacapra |
1.3 |
txt += 'fi\n'
|
| 182 |
fanzago |
1.24 |
txt += 'exit_status=$copy_exit_status\n'
|
| 183 |
fanzago |
1.23 |
txt += 'export TIME_STAGEOUT_END=`date +%s` \n'
|
| 184 |
|
|
txt += 'let "TIME_STAGEOUT = TIME_STAGEOUT_END - TIME_STAGEOUT_INI" \n'
|
| 185 |
slacapra |
1.3 |
|
| 186 |
|
|
return txt
|
| 187 |
spiga |
1.21 |
|
| 188 |
|
|
|
| 189 |
|
|
def wsExitFunc_comm(self):
|
| 190 |
|
|
"""
|
| 191 |
|
|
"""
|
| 192 |
|
|
txt = ''
|
| 193 |
|
|
txt += ' if [ $PYTHONPATH ]; then \n'
|
| 194 |
|
|
txt += ' if [ ! -s $RUNTIME_AREA/fillCrabFjr.py ]; then \n'
|
| 195 |
|
|
txt += ' echo "WARNING: it is not possible to create crab_fjr.xml to final report" \n'
|
| 196 |
|
|
txt += ' else \n'
|
| 197 |
|
|
txt += ' python $RUNTIME_AREA/fillCrabFjr.py $RUNTIME_AREA/crab_fjr_$NJob.xml --errorcode $job_exit_code $executable_exit_status \n'
|
| 198 |
|
|
txt += ' fi\n'
|
| 199 |
|
|
txt += ' fi\n'
|
| 200 |
|
|
txt += ' cd $RUNTIME_AREA \n'
|
| 201 |
|
|
txt += ' for file in $filesToCheck ; do\n'
|
| 202 |
|
|
txt += ' if [ -e $file ]; then\n'
|
| 203 |
|
|
txt += ' echo "tarring file $file in $out_files"\n'
|
| 204 |
|
|
txt += ' else\n'
|
| 205 |
|
|
txt += ' echo "WARNING: output file $file not found!"\n'
|
| 206 |
|
|
txt += ' fi\n'
|
| 207 |
|
|
txt += ' done\n'
|
| 208 |
|
|
txt += ' export TIME_WRAP_END=`date +%s`\n'
|
| 209 |
|
|
txt += ' let "TIME_WRAP = TIME_WRAP_END - TIME_WRAP_END_INI" \n'
|
| 210 |
|
|
txt += ' if [ $PYTHONPATH ]; then \n'
|
| 211 |
|
|
txt += ' if [ ! -s $RUNTIME_AREA/fillCrabFjr.py ]; then \n'
|
| 212 |
|
|
txt += ' echo "WARNING: it is not possible to create crab_fjr.xml to final report" \n'
|
| 213 |
|
|
txt += ' else \n'
|
| 214 |
|
|
# call timing FJR filling
|
| 215 |
|
|
txt += ' python $RUNTIME_AREA/fillCrabFjr.py $RUNTIME_AREA/crab_fjr_$NJob.xml --timing $TIME_WRAP $TIME_EXE $TIME_STAGEOUT \n'
|
| 216 |
|
|
txt += ' echo "CrabWrapperTime=$TIME_WRAP" >> $RUNTIME_AREA/$repo \n'
|
| 217 |
|
|
txt += ' if [ $TIME_STAGEOUT -lt 0 ]; then \n'
|
| 218 |
|
|
txt += ' export TIME_STAGEOUT=NULL \n'
|
| 219 |
|
|
txt += ' fi\n'
|
| 220 |
|
|
txt += ' echo "CrabStageoutTime=$TIME_STAGEOUT" >> $RUNTIME_AREA/$repo \n'
|
| 221 |
|
|
txt += ' fi\n'
|
| 222 |
|
|
txt += ' fi\n'
|
| 223 |
|
|
txt += ' dumpStatus $RUNTIME_AREA/$repo \n\n'
|
| 224 |
|
|
txt += ' echo "JOB_EXIT_STATUS = $job_exit_code"\n'
|
| 225 |
|
|
txt += ' echo "JobExitCode=$job_exit_code" >> $RUNTIME_AREA/$repo\n'
|
| 226 |
|
|
txt += ' dumpStatus $RUNTIME_AREA/$repo\n'
|
| 227 |
|
|
|
| 228 |
|
|
return txt
|
| 229 |
|
|
|
