| 5 |
|
from SchedulerGrid import SchedulerGrid |
| 6 |
|
from crab_exceptions import CrabException |
| 7 |
|
from crab_util import runCommand |
| 8 |
+ |
from WMCore.SiteScreening.BlackWhiteListParser import SEBlackWhiteListParser |
| 9 |
+ |
import Scram |
| 10 |
|
|
| 11 |
|
|
| 12 |
|
import common |
| 13 |
|
import os |
| 14 |
|
import socket |
| 15 |
+ |
import re |
| 16 |
|
|
| 17 |
|
# FUTURE: for python 2.4 & 2.6 |
| 18 |
|
try: |
| 30 |
|
|
| 31 |
|
def __init__(self): |
| 32 |
|
SchedulerGrid.__init__(self,"RCONDOR") |
| 33 |
+ |
self.rcondorHost = os.getenv('RCONDOR_HOST') |
| 34 |
+ |
if self.rcondorHost == None: |
| 35 |
+ |
raise CrabException('FATAL ERROR: env.var RCONDOR_HOST not defined') |
| 36 |
|
self.datasetPath = None |
| 37 |
|
self.selectNoInput = None |
| 38 |
< |
self.OSBsize = 100*1000*1000 # 100 MB |
| 38 |
> |
self.OSBsize = 50*1000*1000 # 50 MB |
| 39 |
|
|
| 40 |
|
self.environment_unique_identifier = None |
| 41 |
+ |
|
| 42 |
|
return |
| 43 |
|
|
| 44 |
|
|
| 52 |
|
self.proxyValid=0 |
| 53 |
|
self.dontCheckProxy=int(cfg_params.get("GRID.dont_check_proxy",0)) |
| 54 |
|
self.space_token = cfg_params.get("USER.space_token",None) |
| 55 |
< |
try: |
| 49 |
< |
self.proxyServer = Downloader("http://cmsdoc.cern.ch/cms/LCG/crab/config/").config("myproxy_server.conf") |
| 50 |
< |
self.proxyServer = self.proxyServer.strip() |
| 51 |
< |
if self.proxyServer is None: |
| 52 |
< |
raise CrabException("myproxy_server.conf retrieved but empty") |
| 53 |
< |
except Exception, e: |
| 54 |
< |
common.logger.info("Problem setting myproxy server endpoint: using myproxy.cern.ch") |
| 55 |
< |
common.logger.debug(e) |
| 56 |
< |
self.proxyServer= 'myproxy.cern.ch' |
| 55 |
> |
self.proxyServer= 'myproxy.cern.ch' |
| 56 |
|
self.group = cfg_params.get("GRID.group", None) |
| 57 |
|
self.role = cfg_params.get("GRID.role", None) |
| 58 |
|
self.VO = cfg_params.get('GRID.virtual_organization','cms') |
| 59 |
< |
|
| 59 |
> |
|
| 60 |
|
try: |
| 61 |
|
tmp = cfg_params['CMSSW.datasetpath'] |
| 62 |
|
if tmp.lower() == 'none': |
| 69 |
|
msg = "Error: datasetpath not defined " |
| 70 |
|
raise CrabException(msg) |
| 71 |
|
|
| 72 |
+ |
if cfg_params.get('GRID.ce_black_list', None) or \ |
| 73 |
+ |
cfg_params.get('GRID.ce_white_list', None) : |
| 74 |
+ |
msg="BEWARE: scheduler RGLIDEIN ignores CE black/white lists." |
| 75 |
+ |
msg+="\n Remove them from crab configuration to proceed." |
| 76 |
+ |
msg+="\n Use GRID.se_white_list and/or GRID.se_black_list instead" |
| 77 |
+ |
raise CrabException(msg) |
| 78 |
+ |
|
| 79 |
|
self.checkProxy() |
| 80 |
|
|
| 81 |
|
return |
| 92 |
|
|
| 93 |
|
def sched_parameter(self, i, task): |
| 94 |
|
""" |
| 95 |
< |
Return scheduler-specific parameters |
| 95 |
> |
Return scheduler-specific parameters. Used at crab -submit time |
| 96 |
|
""" |
| 91 |
– |
req = '' |
| 92 |
– |
if self.EDG_addJdlParam: |
| 93 |
– |
if self.EDG_addJdlParam[-1] == '': |
| 94 |
– |
self.EDG_addJdlParam = self.EDG_addJdlParam[:-1] |
| 95 |
– |
for p in self.EDG_addJdlParam: |
| 96 |
– |
req += p.strip()+';\n' |
| 97 |
|
|
| 98 |
< |
return req |
| 98 |
> |
#SB paste from crab ScheduerGlidein |
| 99 |
> |
|
| 100 |
> |
jobParams = "" |
| 101 |
> |
|
| 102 |
> |
seDest = task.jobs[i-1]['dlsDestination'] |
| 103 |
> |
|
| 104 |
> |
if seDest == [''] : |
| 105 |
> |
seDest = self.blackWhiteListParser.expandList("T") # all of SiteDB |
| 106 |
> |
|
| 107 |
> |
seString=self.blackWhiteListParser.cleanForBlackWhiteList(seDest) |
| 108 |
> |
|
| 109 |
> |
jobParams += '+DESIRED_SEs = "'+seString+'"; ' |
| 110 |
> |
|
| 111 |
> |
scram = Scram.Scram(None) |
| 112 |
> |
cmsVersion = scram.getSWVersion() |
| 113 |
> |
scramArch = scram.getArch() |
| 114 |
> |
|
| 115 |
> |
cmsver=re.split('_', cmsVersion) |
| 116 |
> |
numericCmsVersion = "%s%.2d%.2d" %(cmsver[1], int(cmsver[2]), int(cmsver[3])) |
| 117 |
> |
|
| 118 |
> |
jobParams += '+DESIRED_CMSVersion ="' +cmsVersion+'";' |
| 119 |
> |
jobParams += '+DESIRED_CMSVersionNr ="' +numericCmsVersion+'";' |
| 120 |
> |
jobParams += '+DESIRED_CMSScramArch ="' +scramArch+'";' |
| 121 |
> |
|
| 122 |
> |
myschedName = self.rcondorHost |
| 123 |
> |
jobParams += '+Glidein_MonitorID = "https://'+ myschedName + '//$(Cluster).$(Process)"; ' |
| 124 |
> |
|
| 125 |
> |
if (self.EDG_clock_time): |
| 126 |
> |
jobParams += '+MaxWallTimeMins = '+self.EDG_clock_time+'; ' |
| 127 |
> |
else: |
| 128 |
> |
jobParams += '+MaxWallTimeMins = %d; ' % (60*24) |
| 129 |
> |
|
| 130 |
> |
common._db.updateTask_({'jobType':jobParams}) |
| 131 |
> |
|
| 132 |
> |
|
| 133 |
> |
return jobParams |
| 134 |
|
|
| 135 |
|
|
| 136 |
|
def realSchedParams(self, cfg_params): |
| 137 |
|
""" |
| 138 |
|
Return dictionary with specific parameters, to use with real scheduler |
| 139 |
+ |
is called when scheduler is initialized in Boss, i.e. at each crab command |
| 140 |
|
""" |
| 141 |
+ |
#SB this method is used to pass directory names to Boss Scheduler |
| 142 |
+ |
# via params dictionary |
| 143 |
|
|
| 106 |
– |
tmpDir = os.path.join(common.work_space.shareDir(),'.condor_temp') |
| 107 |
– |
tmpDir = os.path.join(common.work_space.shareDir(),'.condor_temp') |
| 144 |
|
jobDir = common.work_space.jobDir() |
| 109 |
– |
|
| 145 |
|
taskDir=common.work_space.topDir().split('/')[-2] |
| 146 |
< |
rcondorDir = "/afs/cern.ch/user/b/belforte/w0/crabtest/rc/igor/" |
| 147 |
< |
rcondorDir ='%s/.rcondor/mount/' % os.getenv('HOME') |
| 113 |
< |
tmpDir = os.path.join(rcondorDir,taskDir) |
| 114 |
< |
tmpDir = os.path.join(tmpDir,'condor_temp') |
| 146 |
> |
shareDir = common.work_space.shareDir() |
| 147 |
> |
#SBtmpDir = common.work_space.tmpDir() |
| 148 |
|
|
| 149 |
< |
params = {'tmpDir':tmpDir, |
| 150 |
< |
'jobDir':jobDir} |
| 149 |
> |
params = {'rcondorHost':self.rcondorHost, |
| 150 |
> |
'shareDir':shareDir, |
| 151 |
> |
#SB'tmpDir':tmpDir, |
| 152 |
> |
'jobDir':jobDir, |
| 153 |
> |
'taskDir':taskDir} |
| 154 |
> |
|
| 155 |
|
return params |
| 156 |
|
|
| 157 |
|
|
| 196 |
|
txt += 'func_exit() { \n' |
| 197 |
|
txt += self.wsExitFunc_common() |
| 198 |
|
|
| 199 |
< |
txt += ' cp ${out_files}.tgz $_CONDOR_SCRATCH_DIR/\n' |
| 200 |
< |
txt += ' cp crab_fjr_$NJob.xml $_CONDOR_SCRATCH_DIR/\n' |
| 199 |
> |
txt += ' tar zcvf ${out_files}.tgz ${final_list}\n' |
| 200 |
> |
txt += ' tmp_size=`ls -gGrta ${out_files}.tgz | awk \'{ print $3 }\'`\n' |
| 201 |
> |
txt += ' rm ${out_files}.tgz\n' |
| 202 |
> |
txt += ' size=`expr $tmp_size`\n' |
| 203 |
> |
txt += ' echo "Total Output dimension: $size"\n' |
| 204 |
> |
txt += ' limit='+str(self.OSBsize) +' \n' |
| 205 |
> |
txt += ' echo "WARNING: output files size limit is set to: $limit"\n' |
| 206 |
> |
txt += ' if [ "$limit" -lt "$size" ]; then\n' |
| 207 |
> |
txt += ' exceed=1\n' |
| 208 |
> |
txt += ' job_exit_code=70000\n' |
| 209 |
> |
txt += ' echo "Output Sanbox too big. Produced output is lost "\n' |
| 210 |
> |
txt += ' else\n' |
| 211 |
> |
txt += ' exceed=0\n' |
| 212 |
> |
txt += ' echo "Total Output dimension $size is fine."\n' |
| 213 |
> |
txt += ' fi\n' |
| 214 |
> |
|
| 215 |
> |
txt += ' echo "JOB_EXIT_STATUS = $job_exit_code"\n' |
| 216 |
> |
txt += ' echo "JobExitCode=$job_exit_code" >> $RUNTIME_AREA/$repo\n' |
| 217 |
> |
txt += ' dumpStatus $RUNTIME_AREA/$repo\n' |
| 218 |
> |
txt += ' if [ $exceed -ne 1 ]; then\n' |
| 219 |
> |
txt += ' tar zcvf ${out_files}.tgz ${final_list}\n' |
| 220 |
> |
txt += ' else\n' |
| 221 |
> |
txt += ' tar zcvf ${out_files}.tgz CMSSW_${NJob}.stdout CMSSW_${NJob}.stderr\n' |
| 222 |
> |
txt += ' fi\n' |
| 223 |
> |
txt += ' python $RUNTIME_AREA/fillCrabFjr.py $RUNTIME_AREA/crab_fjr_$NJob.xml --errorcode $job_exit_code \n' |
| 224 |
|
|
| 225 |
|
txt += ' exit $job_exit_code\n' |
| 226 |
|
txt += '}\n' |
| 239 |
|
common._db.updateTask_(taskReq) |
| 240 |
|
|
| 241 |
|
|
| 182 |
– |
# presa di brutto da SchedulerGrid.py |
| 183 |
– |
""" |
| 184 |
– |
|
| 185 |
– |
def wsSetupEnvironment(self): |
| 186 |
– |
|
| 187 |
– |
#Returns part of a job script which does scheduler-specific work. |
| 188 |
– |
|
| 189 |
– |
taskId =common._db.queryTask('name') |
| 190 |
– |
index = int(common._db.nJobs()) |
| 191 |
– |
job = common.job_list[index-1] |
| 192 |
– |
jbt = job.type() |
| 193 |
– |
if not self.environment_unique_identifier: |
| 194 |
– |
try : |
| 195 |
– |
self.environment_unique_identifier = self.envUniqueID() |
| 196 |
– |
except : |
| 197 |
– |
raise CrabException('environment_unique_identifier not set') |
| 198 |
– |
|
| 199 |
– |
# start with wrapper timing |
| 200 |
– |
txt = 'export TIME_WRAP_INI=`date +%s` \n' |
| 201 |
– |
txt += 'export TIME_STAGEOUT=-2 \n\n' |
| 202 |
– |
txt += '# '+self.name()+' specific stuff\n' |
| 203 |
– |
txt += '# strip arguments\n' |
| 204 |
– |
txt += 'echo "strip arguments"\n' |
| 205 |
– |
txt += 'args=("$@")\n' |
| 206 |
– |
txt += 'nargs=$#\n' |
| 207 |
– |
txt += 'shift $nargs\n' |
| 208 |
– |
txt += "# job number (first parameter for job wrapper)\n" |
| 209 |
– |
txt += "NJob=${args[0]}; export NJob\n" |
| 210 |
– |
txt += "NResub=${args[1]}; export NResub\n" |
| 211 |
– |
txt += "NRand=`getRandSeed`; export NRand\n" |
| 212 |
– |
# append random code |
| 213 |
– |
txt += 'OutUniqueID=_$NRand\n' |
| 214 |
– |
txt += 'OutUniqueID=_$NResub$OutUniqueID\n' |
| 215 |
– |
txt += 'OutUniqueID=$NJob$OutUniqueID; export OutUniqueID\n' |
| 216 |
– |
txt += 'CRAB_UNIQUE_JOB_ID=%s_${OutUniqueID}; export CRAB_UNIQUE_JOB_ID\n' % taskId |
| 217 |
– |
txt += 'echo env var CRAB_UNIQUE_JOB_ID set to: ${CRAB_UNIQUE_JOB_ID}\n' |
| 218 |
– |
# if we want to prepend |
| 219 |
– |
#txt += 'OutUniqueID=_$NResub\n' |
| 220 |
– |
#txt += 'OutUniqueID=_$NJob$OutUniqueID\n' |
| 221 |
– |
#txt += 'OutUniqueID=$NRand$OutUniqueID; export OutUniqueID\n' |
| 222 |
– |
|
| 223 |
– |
txt += "out_files=out_files_${NJob}; export out_files\n" |
| 224 |
– |
txt += "echo $out_files\n" |
| 225 |
– |
txt += jbt.outList() |
| 226 |
– |
# txt += 'if [ $JobRunCount ] && [ `expr $JobRunCount - 1` -gt 0 ] && [ $Glidein_MonitorID ]; then \n' |
| 227 |
– |
txt += 'if [ $Glidein_MonitorID ]; then \n' |
| 228 |
– |
# txt += ' attempt=`expr $JobRunCount - 1` \n' |
| 229 |
– |
# txt += ' MonitorJobID=${NJob}_${Glidein_MonitorID}__${attempt}\n' |
| 230 |
– |
# txt += ' SyncGridJobId=${Glidein_MonitorID}__${attempt}\n' |
| 231 |
– |
txt += ' MonitorJobID=${NJob}_${Glidein_MonitorID}\n' |
| 232 |
– |
txt += ' SyncGridJobId=${Glidein_MonitorID}\n' |
| 233 |
– |
txt += 'else \n' |
| 234 |
– |
txt += ' MonitorJobID=${NJob}_'+self.environment_unique_identifier+'\n' |
| 235 |
– |
txt += ' SyncGridJobId='+self.environment_unique_identifier+'\n' |
| 236 |
– |
txt += 'fi\n' |
| 237 |
– |
txt += 'MonitorID='+taskId+'\n' |
| 238 |
– |
txt += 'echo "MonitorJobID=$MonitorJobID" > $RUNTIME_AREA/$repo \n' |
| 239 |
– |
txt += 'echo "SyncGridJobId=$SyncGridJobId" >> $RUNTIME_AREA/$repo \n' |
| 240 |
– |
txt += 'echo "MonitorID=$MonitorID" >> $RUNTIME_AREA/$repo\n' |
| 241 |
– |
|
| 242 |
– |
txt += 'echo ">>> GridFlavour discovery: " \n' |
| 243 |
– |
txt += 'if [ $OSG_GRID ]; then \n' |
| 244 |
– |
txt += ' middleware=OSG \n' |
| 245 |
– |
txt += ' if [ $OSG_JOB_CONTACT ]; then \n' |
| 246 |
– |
txt += ' SyncCE="$OSG_JOB_CONTACT"; \n' |
| 247 |
– |
txt += ' echo "SyncCE=$SyncCE" >> $RUNTIME_AREA/$repo ;\n' |
| 248 |
– |
txt += ' else\n' |
| 249 |
– |
txt += ' echo "not reporting SyncCE";\n' |
| 250 |
– |
txt += ' fi\n'; |
| 251 |
– |
txt += ' echo "GridFlavour=$middleware" | tee -a $RUNTIME_AREA/$repo \n' |
| 252 |
– |
txt += ' echo "source OSG GRID setup script" \n' |
| 253 |
– |
txt += ' source $OSG_GRID/setup.sh \n' |
| 254 |
– |
txt += 'elif [ $NORDUGRID_CE ]; then \n' # We look for $NORDUGRID_CE before $VO_CMS_SW_DIR, |
| 255 |
– |
txt += ' middleware=ARC \n' # because the latter is defined for ARC too |
| 256 |
– |
txt += ' echo "SyncCE=${NORDUGRID_CE}:2811/nordugrid-GE-${QUEUE:-queue}" >> $RUNTIME_AREA/$repo \n' |
| 257 |
– |
txt += ' echo "GridFlavour=$middleware" | tee -a $RUNTIME_AREA/$repo \n' |
| 258 |
– |
txt += 'elif [ $VO_CMS_SW_DIR ]; then \n' |
| 259 |
– |
txt += ' middleware=LCG \n' |
| 260 |
– |
txt += ' if [ $GLIDEIN_Gatekeeper ]; then \n' |
| 261 |
– |
txt += ' echo "SyncCE=`echo $GLIDEIN_Gatekeeper | sed -e s/:2119//`" >> $RUNTIME_AREA/$repo \n' |
| 262 |
– |
txt += ' else \n' |
| 263 |
– |
txt += ' echo "SyncCE=`glite-brokerinfo getCE`" >> $RUNTIME_AREA/$repo \n' |
| 264 |
– |
txt += ' fi \n' |
| 265 |
– |
txt += ' echo "GridFlavour=$middleware" | tee -a $RUNTIME_AREA/$repo \n' |
| 266 |
– |
txt += 'else \n' |
| 267 |
– |
txt += ' echo "ERROR ==> GridFlavour not identified" \n' |
| 268 |
– |
txt += ' job_exit_code=10030 \n' |
| 269 |
– |
txt += ' func_exit \n' |
| 270 |
– |
txt += 'fi \n' |
| 271 |
– |
|
| 272 |
– |
txt += 'dumpStatus $RUNTIME_AREA/$repo \n' |
| 273 |
– |
txt += '\n\n' |
| 274 |
– |
|
| 275 |
– |
|
| 276 |
– |
txt += 'export VO='+self.VO+'\n' |
| 277 |
– |
txt += 'if [ $middleware == LCG ]; then\n' |
| 278 |
– |
txt += ' if [ $GLIDEIN_Gatekeeper ]; then\n' |
| 279 |
– |
txt += ' CloseCEs=$GLIDEIN_Gatekeeper \n' |
| 280 |
– |
txt += ' else\n' |
| 281 |
– |
txt += ' CloseCEs=`glite-brokerinfo getCE`\n' |
| 282 |
– |
txt += ' fi\n' |
| 283 |
– |
txt += ' echo "CloseCEs = $CloseCEs"\n' |
| 284 |
– |
txt += ' CE=`echo $CloseCEs | sed -e "s/:.*//"`\n' |
| 285 |
– |
txt += ' echo "CE = $CE"\n' |
| 286 |
– |
txt += 'elif [ $middleware == OSG ]; then \n' |
| 287 |
– |
txt += ' if [ $OSG_JOB_CONTACT ]; then \n' |
| 288 |
– |
txt += ' CE=`echo $OSG_JOB_CONTACT | /usr/bin/awk -F\/ \'{print $1}\'` \n' |
| 289 |
– |
txt += ' else \n' |
| 290 |
– |
txt += ' echo "ERROR ==> OSG mode in setting CE name from OSG_JOB_CONTACT" \n' |
| 291 |
– |
txt += ' job_exit_code=10099\n' |
| 292 |
– |
txt += ' func_exit\n' |
| 293 |
– |
txt += ' fi \n' |
| 294 |
– |
txt += 'elif [ $middleware == ARC ]; then \n' |
| 295 |
– |
txt += ' echo "CE = $NORDUGRID_CE"\n' |
| 296 |
– |
txt += 'fi \n' |
| 297 |
– |
|
| 298 |
– |
return txt |
| 299 |
– |
""" |
