| 1 |
+ |
""" |
| 2 |
+ |
CRAB interface to BossLite gLite Scheduler |
| 3 |
+ |
""" |
| 4 |
+ |
|
| 5 |
+ |
__revision__ = "$Id$" |
| 6 |
+ |
__version__ = "$Revision$" |
| 7 |
+ |
|
| 8 |
|
from SchedulerGrid import SchedulerGrid |
| 2 |
– |
from crab_logger import Logger |
| 9 |
|
from crab_exceptions import * |
| 10 |
|
from crab_util import * |
| 11 |
|
from GliteConfig import * |
| 12 |
+ |
import EdgLoggingInfo |
| 13 |
|
import common |
| 14 |
+ |
from WMCore.SiteScreening.BlackWhiteListParser import CEBlackWhiteListParser |
| 15 |
|
|
| 16 |
|
import os, sys, time |
| 17 |
|
|
| 19 |
|
def __init__(self, name="GLITE"): |
| 20 |
|
SchedulerGrid.__init__(self,name) |
| 21 |
|
|
| 22 |
+ |
self.OSBsize = 55000000 |
| 23 |
+ |
|
| 24 |
|
def configure(self,cfg_params): |
| 25 |
|
SchedulerGrid.configure(self, cfg_params) |
| 26 |
< |
self.environment_unique_identifier = 'GLITE_WMS_JOBID' |
| 26 |
> |
self.environment_unique_identifier = '$GLITE_WMS_JOBID' |
| 27 |
> |
|
| 28 |
> |
def realSchedParams(self,cfg_params): |
| 29 |
> |
""" |
| 30 |
> |
Return dictionary with specific parameters, to use |
| 31 |
> |
with real scheduler |
| 32 |
> |
""" |
| 33 |
> |
self.rb_param_file='' |
| 34 |
> |
if (not cfg_params.has_key('GRID.rb')): |
| 35 |
> |
cfg_params['GRID.rb']='CERN' |
| 36 |
> |
self.rb_param_file=common.scheduler.rb_configure(cfg_params.get("GRID.rb")) |
| 37 |
> |
self.wms_service=cfg_params.get("GRID.wms_service",'') |
| 38 |
> |
self.skipWMSAuth=cfg_params.get("GRID.skipwmsauth",1) |
| 39 |
> |
params = { 'service' : self.wms_service, \ |
| 40 |
> |
'config' : self.rb_param_file, \ |
| 41 |
> |
'skipWMSAuth' : self.skipWMSAuth |
| 42 |
> |
} |
| 43 |
> |
return params |
| 44 |
> |
|
| 45 |
|
|
| 46 |
|
def rb_configure(self, RB): |
| 47 |
|
if not RB: return None |
| 57 |
|
|
| 58 |
|
def ce_list(self): |
| 59 |
|
""" |
| 60 |
< |
Returns string with requirement CE related |
| 60 |
> |
Returns string with requirement CE related |
| 61 |
|
""" |
| 62 |
< |
req = '' |
| 62 |
> |
ceParser = CEBlackWhiteListParser(self.EDG_ce_white_list, |
| 63 |
> |
self.EDG_ce_black_list, common.logger) |
| 64 |
> |
req = '' |
| 65 |
> |
ce_white_list = [] |
| 66 |
> |
ce_black_list = [] |
| 67 |
|
if self.EDG_ce_white_list: |
| 68 |
< |
ce_white_list = self.EDG_ce_white_list |
| 68 |
> |
ce_white_list = ceParser.whiteList() |
| 69 |
|
tmpCe=[] |
| 70 |
|
concString = '&&' |
| 71 |
|
for ce in ce_white_list: |
| 84 |
|
req += ") " |
| 85 |
|
|
| 86 |
|
if self.EDG_ce_black_list: |
| 87 |
< |
ce_black_list = self.EDG_ce_black_list |
| 87 |
> |
ce_black_list = ceParser.blackList() |
| 88 |
|
tmpCe=[] |
| 89 |
|
concString = '&&' |
| 90 |
|
for ce in ce_black_list: |
| 93 |
|
|
| 94 |
|
# requirement added to skip gliteCE |
| 95 |
|
req += '&& (!RegExp("blah", other.GlueCEUniqueId))' |
| 96 |
+ |
retWL = ','.join(ce_white_list) |
| 97 |
+ |
retBL = ','.join(ce_black_list) |
| 98 |
+ |
if not retWL: |
| 99 |
+ |
retWL = None |
| 100 |
+ |
if not retBL: |
| 101 |
+ |
retBL = None |
| 102 |
|
|
| 103 |
< |
return req,self.EDG_ce_white_list,self.EDG_ce_black_list |
| 103 |
> |
return req, retWL, retBL |
| 104 |
|
|
| 105 |
< |
def se_list(self, id, dest): |
| 105 |
> |
def se_list(self, dest): |
| 106 |
|
""" |
| 107 |
< |
Returns string with requirement SE related |
| 108 |
< |
""" |
| 109 |
< |
hostList=self.findSites_(id,dest) |
| 107 |
> |
Returns string with requirement SE related |
| 108 |
> |
""" |
| 109 |
> |
hostList=self.findSites_(dest) |
| 110 |
|
req='' |
| 111 |
|
reqtmp=[] |
| 112 |
|
concString = '||' |
| 120 |
|
|
| 121 |
|
def jdlParam(self): |
| 122 |
|
""" |
| 123 |
< |
Returns |
| 124 |
< |
""" |
| 123 |
> |
Returns |
| 124 |
> |
""" |
| 125 |
|
req='' |
| 126 |
|
if self.EDG_addJdlParam: |
| 127 |
< |
if self.EDG_addJdlParam[-1] == '': self.EDG_addJdlParam= self.EDG_addJdlParam[:-1] |
| 127 |
> |
if self.EDG_addJdlParam[-1] == '': self.EDG_addJdlParam= self.EDG_addJdlParam[:-1] |
| 128 |
|
for p in self.EDG_addJdlParam: |
| 129 |
< |
# param_file.write(string.strip(p)+';\n') |
| 92 |
< |
req+=string.strip(p)+';\n' ## BL--DS |
| 129 |
> |
req+=string.strip(p)+';\n' |
| 130 |
|
return req |
| 131 |
|
|
| 132 |
|
def specific_req(self): |
| 133 |
|
""" |
| 134 |
|
Returns string with specific requirements |
| 135 |
< |
""" |
| 135 |
> |
""" |
| 136 |
|
req='' |
| 137 |
|
if self.EDG_clock_time: |
| 138 |
|
if (not req == ' '): req = req + ' && ' |
| 141 |
|
if self.EDG_cpu_time: |
| 142 |
|
if (not req == ' '): req = req + ' && ' |
| 143 |
|
req = req + ' other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time |
| 107 |
– |
|
| 108 |
– |
return req |
| 109 |
– |
|
| 110 |
– |
def sched_fix_parameter(self): |
| 111 |
– |
""" |
| 112 |
– |
Returns string with requirements and scheduler-specific parameters |
| 113 |
– |
""" |
| 114 |
– |
index = int(common._db.nJobs()) |
| 115 |
– |
job = common.job_list[index-1] |
| 116 |
– |
jbt = job.type() |
| 117 |
– |
req = '' |
| 118 |
– |
req = req + jbt.getRequirements() |
| 144 |
|
|
| 145 |
< |
if self.EDG_requirements: |
| 121 |
< |
if (not req == ' '): req = req + ' && ' |
| 122 |
< |
req = req + self.EDG_requirements |
| 123 |
< |
|
| 124 |
< |
Task_Req={'jobType':req}## DS--BL |
| 125 |
< |
common._db.updateTask_(Task_Req) |
| 145 |
> |
return req |
| 146 |
|
|
| 147 |
|
def sched_parameter(self,i,task): |
| 148 |
|
""" |
| 149 |
|
Returns string with requirements and scheduler-specific parameters |
| 150 |
|
""" |
| 151 |
< |
dest= eval(task.jobs[i]['dlsDestination']) ## DS--BL |
| 151 |
> |
dest= task.jobs[i-1]['dlsDestination'] |
| 152 |
|
|
| 153 |
|
req='' |
| 154 |
|
req +=task['jobType'] |
| 155 |
|
|
| 156 |
|
sched_param='' |
| 157 |
< |
sched_param+='Requirements = ' + req +self.specific_req() + self.se_list(i,dest) +\ |
| 158 |
< |
self.ce_list()[0] +';\n' ## BL--DS |
| 159 |
< |
if self.EDG_addJdlParam: sched_param+=self.jdlParam() ## BL--DS |
| 157 |
> |
sched_param+='Requirements = ' + req +self.specific_req() + self.se_list(dest) +\ |
| 158 |
> |
self.ce_list()[0] +';\n' |
| 159 |
> |
if self.EDG_addJdlParam: sched_param+=self.jdlParam() |
| 160 |
|
sched_param+='MyProxyServer = "' + self.proxyServer + '";\n' |
| 161 |
|
sched_param+='VirtualOrganisation = "' + self.VO + '";\n' |
| 162 |
|
sched_param+='RetryCount = '+str(self.EDG_retry_count)+';\n' |
| 163 |
|
sched_param+='ShallowRetryCount = '+str(self.EDG_shallow_retry_count)+';\n' |
| 164 |
|
|
| 165 |
|
return sched_param |
| 146 |
– |
|
| 147 |
– |
def wsSetupEnvironment(self): |
| 148 |
– |
""" |
| 149 |
– |
Returns part of a job script which does scheduler-specific work. |
| 150 |
– |
""" |
| 151 |
– |
txt = '' |
| 152 |
– |
txt += '# strip arguments\n' |
| 153 |
– |
txt += 'echo "strip arguments"\n' |
| 154 |
– |
txt += 'args=("$@")\n' |
| 155 |
– |
txt += 'nargs=$#\n' |
| 156 |
– |
txt += 'shift $nargs\n' |
| 157 |
– |
txt += "# job number (first parameter for job wrapper)\n" |
| 158 |
– |
txt += "NJob=${args[0]}; export NJob\n" |
| 159 |
– |
|
| 160 |
– |
txt += '# job identification to DashBoard \n' |
| 161 |
– |
#txt += 'MonitorJobID=`echo ${NJob}_$GLITE_WMS_JOBID`\n' |
| 162 |
– |
#txt += 'SyncGridJobId=`echo $GLITE_WMS_JOBID`\n' |
| 163 |
– |
#txt += 'MonitorID=`echo ' + self._taskId + '`\n' |
| 164 |
– |
txt += 'MonitorJobID=${NJob}_$GLITE_WMS_JOBID \n' |
| 165 |
– |
txt += 'SyncGridJobId=$GLITE_WMS_JOBID \n' |
| 166 |
– |
txt += 'MonitorID='+self._taskId+' \n' |
| 167 |
– |
txt += 'echo "MonitorJobID=$MonitorJobID" > $RUNTIME_AREA/$repo \n' |
| 168 |
– |
txt += 'echo "SyncGridJobId=$SyncGridJobId" >> $RUNTIME_AREA/$repo \n' |
| 169 |
– |
txt += 'echo "MonitorID=$MonitorID" >> $RUNTIME_AREA/$repo\n' |
| 170 |
– |
#txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n' |
| 171 |
– |
#txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n' |
| 172 |
– |
#txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n' |
| 173 |
– |
|
| 174 |
– |
#txt += 'echo "middleware discovery: " \n' |
| 175 |
– |
txt += 'echo ">>> GridFlavour discovery: " \n' |
| 176 |
– |
txt += 'if [ $VO_CMS_SW_DIR ]; then \n' |
| 177 |
– |
txt += ' middleware=LCG\n' |
| 178 |
– |
#txt += ' echo "SyncCE=`glite-brokerinfo getCE`" | tee -a $RUNTIME_AREA/$repo \n' |
| 179 |
– |
txt += ' echo "SyncCE=`glite-brokerinfo getCE`" >> $RUNTIME_AREA/$repo \n' |
| 180 |
– |
#txt += ' echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n' |
| 181 |
– |
txt += ' echo "GridFlavour=$middleware" | tee -a $RUNTIME_AREA/$repo \n' |
| 182 |
– |
#txt += ' echo ">>> middleware =$middleware" \n' |
| 183 |
– |
txt += 'elif [ $OSG_APP ]; then \n' |
| 184 |
– |
txt += ' middleware=OSG \n' |
| 185 |
– |
txt += ' if [ $OSG_JOB_CONTACT ]; then \n' |
| 186 |
– |
txt += ' SyncCE="$OSG_JOB_CONTACT"; \n' |
| 187 |
– |
#txt += ' echo "SyncCE=$SyncCE" | tee -a $RUNTIME_AREA/$repo ;\n' |
| 188 |
– |
txt += ' echo "SyncCE=$SyncCE" >> $RUNTIME_AREA/$repo ;\n' |
| 189 |
– |
txt += ' else\n' |
| 190 |
– |
txt += ' echo "not reporting SyncCE";\n' |
| 191 |
– |
txt += ' fi\n'; |
| 192 |
– |
txt += ' echo "GridFlavour=$middleware" | tee -a $RUNTIME_AREA/$repo \n' |
| 193 |
– |
#txt += ' echo ">>> middleware =$middleware" \n' |
| 194 |
– |
txt += 'else \n' |
| 195 |
– |
txt += ' echo "ERROR ==> GridFlavour not identified" \n' |
| 196 |
– |
txt += ' job_exit_code=10030\n' |
| 197 |
– |
txt += ' func_exit \n' |
| 198 |
– |
#txt += ' echo "SET_CMS_ENV 10030 ==> middleware not identified" \n' |
| 199 |
– |
#txt += ' echo "JOB_EXIT_STATUS = 10030" \n' |
| 200 |
– |
#txt += ' echo "JobExitCode=10030" | tee -a $RUNTIME_AREA/$repo \n' |
| 201 |
– |
#txt += ' dumpStatus $RUNTIME_AREA/$repo \n' |
| 202 |
– |
#txt += ' exit 1 \n' |
| 203 |
– |
txt += 'fi \n' |
| 204 |
– |
|
| 205 |
– |
txt += 'dumpStatus $RUNTIME_AREA/$repo \n' |
| 206 |
– |
|
| 207 |
– |
txt += '\n\n' |
| 208 |
– |
|
| 209 |
– |
txt += 'export VO='+self.VO+'\n' |
| 210 |
– |
txt += 'if [ $middleware == LCG ]; then\n' |
| 211 |
– |
txt += ' CloseCEs=`glite-brokerinfo getCE`\n' |
| 212 |
– |
txt += ' echo "CloseCEs = $CloseCEs"\n' |
| 213 |
– |
txt += ' CE=`echo $CloseCEs | sed -e "s/:.*//"`\n' |
| 214 |
– |
txt += ' echo "CE = $CE"\n' |
| 215 |
– |
txt += 'elif [ $middleware == OSG ]; then \n' |
| 216 |
– |
txt += ' if [ $OSG_JOB_CONTACT ]; then \n' |
| 217 |
– |
txt += ' CE=`echo $OSG_JOB_CONTACT | /usr/bin/awk -F\/ \'{print $1}\'` \n' |
| 218 |
– |
txt += ' else \n' |
| 219 |
– |
txt += ' echo "ERROR ==> OSG mode in setting CE name from OSG_JOB_CONTACT" \n' |
| 220 |
– |
txt += ' job_exit_code=10099\n' |
| 221 |
– |
txt += ' func_exit \n' |
| 222 |
– |
#txt += ' echo "SET_CMS_ENV 10099 ==> OSG mode: ERROR in setting CE name from OSG_JOB_CONTACT" \n' |
| 223 |
– |
#txt += ' echo "JOB_EXIT_STATUS = 10099" \n' |
| 224 |
– |
#txt += ' echo "JobExitCode=10099" | tee -a $RUNTIME_AREA/$repo \n' |
| 225 |
– |
#txt += ' dumpStatus $RUNTIME_AREA/$repo \n' |
| 226 |
– |
#txt += ' exit 1 \n' |
| 227 |
– |
txt += ' fi \n' |
| 228 |
– |
txt += 'fi \n' |
| 166 |
|
|
| 167 |
< |
return txt |
| 231 |
< |
|
| 232 |
< |
def loggingInfo(self, id): |
| 167 |
> |
def decodeLogInfo(self, file): |
| 168 |
|
""" |
| 169 |
< |
retrieve the logging info from logging and bookkeeping and return it |
| 169 |
> |
Parse logging info file and return main info |
| 170 |
|
""" |
| 171 |
< |
self.checkProxy() |
| 172 |
< |
cmd = 'glite-job-logging-info -v 3 ' + id |
| 173 |
< |
cmd_out = runCommand(cmd) |
| 239 |
< |
return cmd_out |
| 240 |
< |
|
| 241 |
< |
def queryDetailedStatus(self, id): |
| 242 |
< |
""" Query a detailed status of the job with id """ |
| 243 |
< |
cmd = 'glite-job-status '+id |
| 244 |
< |
cmd_out = runCommand(cmd) |
| 245 |
< |
return cmd_out |
| 171 |
> |
loggingInfo = EdgLoggingInfo.EdgLoggingInfo() |
| 172 |
> |
reason = loggingInfo.decodeReason(file) |
| 173 |
> |
return reason |
| 174 |
|
|
| 175 |
< |
def findSites_(self, n, sites): |
| 175 |
> |
def findSites_(self, sites): |
| 176 |
|
itr4 =[] |
| 177 |
|
if len(sites)>0 and sites[0]=="": |
| 178 |
|
return itr4 |
| 179 |
|
if sites != [""]: |
| 180 |
< |
##Addedd Daniele |
| 253 |
< |
replicas = self.blackWhiteListParser.checkBlackList(sites,n) |
| 180 |
> |
replicas = self.blackWhiteListParser.checkBlackList(sites) |
| 181 |
|
if len(replicas)!=0: |
| 182 |
< |
replicas = self.blackWhiteListParser.checkWhiteList(replicas,n) |
| 183 |
< |
|
| 182 |
> |
replicas = self.blackWhiteListParser.checkWhiteList(replicas) |
| 183 |
> |
|
| 184 |
|
itr4 = replicas |
| 258 |
– |
##### |
| 185 |
|
return itr4 |
| 186 |
|
|
| 187 |
|
|
| 188 |
+ |
def wsExitFunc(self): |
| 189 |
+ |
""" |
| 190 |
+ |
""" |
| 191 |
+ |
txt = '\n' |
| 192 |
|
|
| 193 |
+ |
txt += '#\n' |
| 194 |
+ |
txt += '# EXECUTE THIS FUNCTION BEFORE EXIT \n' |
| 195 |
+ |
txt += '#\n\n' |
| 196 |
+ |
|
| 197 |
+ |
txt += 'func_exit() { \n' |
| 198 |
+ |
txt += self.wsExitFunc_common() |
| 199 |
+ |
### specific Glite check for OSB |
| 200 |
+ |
txt += ' tar zcvf ${out_files}.tgz ${final_list}\n' |
| 201 |
+ |
txt += ' tmp_size=`ls -gGrta ${out_files}.tgz | awk \'{ print $3 }\'`\n' |
| 202 |
+ |
txt += ' rm ${out_files}.tgz\n' |
| 203 |
+ |
txt += ' size=`expr $tmp_size`\n' |
| 204 |
+ |
txt += ' echo "Total Output dimension: $size"\n' |
| 205 |
+ |
txt += ' limit='+str(self.OSBsize) +' \n' |
| 206 |
+ |
txt += ' echo "WARNING: output files size limit is set to: $limit"\n' |
| 207 |
+ |
txt += ' if [ "$limit" -lt "$size" ]; then\n' |
| 208 |
+ |
txt += ' exceed=1\n' |
| 209 |
+ |
txt += ' job_exit_code=70000\n' |
| 210 |
+ |
txt += ' echo "Output Sanbox too big. Produced output is lost "\n' |
| 211 |
+ |
txt += ' else\n' |
| 212 |
+ |
txt += ' exceed=0\n' |
| 213 |
+ |
txt += ' echo "Total Output dimension $size is fine."\n' |
| 214 |
+ |
txt += ' fi\n' |
| 215 |
+ |
|
| 216 |
+ |
txt += ' echo "JOB_EXIT_STATUS = $job_exit_code"\n' |
| 217 |
+ |
txt += ' echo "JobExitCode=$job_exit_code" >> $RUNTIME_AREA/$repo\n' |
| 218 |
+ |
txt += ' dumpStatus $RUNTIME_AREA/$repo\n' |
| 219 |
+ |
txt += ' if [ $exceed -ne 1 ]; then\n' |
| 220 |
+ |
txt += ' tar zcvf ${out_files}.tgz ${final_list}\n' |
| 221 |
+ |
txt += ' else\n' |
| 222 |
+ |
txt += ' tar zcvf ${out_files}.tgz CMSSW_${NJob}.stdout CMSSW_${NJob}.stderr\n' |
| 223 |
+ |
txt += ' fi\n' |
| 224 |
+ |
txt += ' exit $job_exit_code\n' |
| 225 |
|
|
| 226 |
< |
def tOut(self, list): |
| 227 |
< |
return 180 |
| 226 |
> |
txt += '}\n' |
| 227 |
> |
return txt |
