ViewVC Help

View File | Revision Log | Show Annotations | Root Listing

root/cvsroot/COMP/CRAB/python/SchedulerRcondor.py

Comparing COMP/CRAB/python/SchedulerRcondor.py (file contents):
Revision 1.1 by belforte, Wed Jul 11 15:52:54 2012 UTC vs.
Revision 1.10 by belforte, Fri Aug 10 17:54:30 2012 UTC

#	Line 5 | Line 5 | Implements the vanilla (local) Remote Co
5		from SchedulerGrid  import SchedulerGrid
6		from crab_exceptions import CrabException
7		from crab_util import runCommand
8	+	#from WMCore.SiteScreening.BlackWhiteListParser import CEBlackWhiteListParser
9	+	from WMCore.SiteScreening.BlackWhiteListParser import SEBlackWhiteListParser
10	+	import Scram
11	+
12
13
14		import common
15		import os
16		import socket
17	+	import re
18
19		# FUTURE: for python 2.4 & 2.6
20		try:
#	Line 27 | Line 32 | class SchedulerRcondor(SchedulerGrid) :
32
33		def __init__(self):
34		SchedulerGrid.__init__(self,"RCONDOR")
35	+	self.rcondorHost   = os.getenv('RCONDOR_HOST')
36	+	if self.rcondorHost == None:
37	+	raise CrabException('FATAL ERROR: env.var RCONDOR_HOST not defined')
38		self.datasetPath   = None
39		self.selectNoInput = None
40	<	self.OSBsize = 100*1000*1000 # 100 MB
40	>	self.OSBsize = 50*1000*1000 # 50 MB
41
42		self.environment_unique_identifier = None
43	+
44		return
45
46
#	Line 45 | Line 54 | class SchedulerRcondor(SchedulerGrid) :
54		self.proxyValid=0
55		self.dontCheckProxy=int(cfg_params.get("GRID.dont_check_proxy",0))
56		self.space_token = cfg_params.get("USER.space_token",None)
57	<	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'
57	>	self.proxyServer= 'myproxy.cern.ch'
58		self.group = cfg_params.get("GRID.group", None)
59		self.role = cfg_params.get("GRID.role", None)
60		self.VO = cfg_params.get('GRID.virtual_organization','cms')
61	<
61	>
62		try:
63		tmp =  cfg_params['CMSSW.datasetpath']
64		if tmp.lower() == 'none':
#	Line 86 | Line 87 | class SchedulerRcondor(SchedulerGrid) :
87
88		def sched_parameter(self, i, task):
89		"""
90	<	Return scheduler-specific parameters
90	>	Return scheduler-specific parameters. Used at crab -submit time
91		"""
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'
92
93	<	return req
93	>	#SB paste from crab ScheduerGlidein
94	>
95	>	jobParams = ""
96	>
97	>	seDest = task.jobs[i-1]['dlsDestination']
98	>
99	>	if seDest == [''] :
100	>	seDest = self.blackWhiteListParser.expandList("T") # all of SiteDB
101	>
102	>	seString=self.blackWhiteListParser.cleanForBlackWhiteList(seDest)
103	>
104	>	jobParams += '+DESIRED_SEs = "'+seString+'"; '
105	>
106	>	scram = Scram.Scram(None)
107	>	cmsVersion = scram.getSWVersion()
108	>	scramArch  = scram.getArch()
109	>
110	>	cmsver=re.split('_', cmsVersion)
111	>	numericCmsVersion = "%s%.2d%.2d" %(cmsver[1], int(cmsver[2]), int(cmsver[3]))
112	>
113	>	jobParams += '+DESIRED_CMSVersion ="' +cmsVersion+'";'
114	>	jobParams += '+DESIRED_CMSVersionNr ="' +numericCmsVersion+'";'
115	>	jobParams += '+DESIRED_CMSScramArch ="' +scramArch+'";'
116	>
117	>	myschedName = self.rcondorHost
118	>	jobParams += '+Glidein_MonitorID = "https://'+ myschedName + '//$(Cluster).$(Process)"; '
119	>
120	>	if (self.EDG_clock_time):
121	>	jobParams += '+MaxWallTimeMins = '+self.EDG_clock_time+'; '
122	>	else:
123	>	jobParams += '+MaxWallTimeMins = %d; ' % (60*24)
124	>
125	>	common._db.updateTask_({'jobType':jobParams})
126	>
127	>
128	>	return jobParams
129
130
131		def realSchedParams(self, cfg_params):
132		"""
133		Return dictionary with specific parameters, to use with real scheduler
134	+	is called when scheduler is initialized in Boss, i.e. at each crab command
135		"""
136	+	#SB this method is used to pass directory names to Boss Scheduler
137	+	# via params dictionary
138
106	–	tmpDir = os.path.join(common.work_space.shareDir(),'.condor_temp')
107	–	tmpDir = os.path.join(common.work_space.shareDir(),'.condor_temp')
139		jobDir = common.work_space.jobDir()
109	–
140		taskDir=common.work_space.topDir().split('/')[-2]
141	<	rcondorDir = "/afs/cern.ch/user/b/belforte/w0/crabtest/rc/igor/"
142	<	rcondorDir ='%s/.rcondor/mount/' % os.getenv('HOME')
113	<	tmpDir = os.path.join(rcondorDir,taskDir)
114	<	tmpDir = os.path.join(tmpDir,'condor_temp')
141	>	shareDir = common.work_space.shareDir()
142	>	#SBtmpDir = common.work_space.tmpDir()
143
144	<	params = {'tmpDir':tmpDir,
145	<	'jobDir':jobDir}
144	>	params = {'rcondorHost':self.rcondorHost,
145	>	'shareDir':shareDir,
146	>	#SB'tmpDir':tmpDir,
147	>	'jobDir':jobDir,
148	>	'taskDir':taskDir}
149	>
150		return params
151
152
#	Line 159 | Line 191 | class SchedulerRcondor(SchedulerGrid) :
191		txt += 'func_exit() { \n'
192		txt += self.wsExitFunc_common()
193
194	<	txt += '    cp  ${out_files}.tgz $_CONDOR_SCRATCH_DIR/\n'
195	<	txt += '    cp  crab_fjr_$NJob.xml $_CONDOR_SCRATCH_DIR/\n'
194	>	txt += '    tar zcvf ${out_files}.tgz  ${final_list}\n'
195	>	txt += '    tmp_size=`ls -gGrta ${out_files}.tgz | awk \'{ print $3 }\'`\n'
196	>	txt += '    rm ${out_files}.tgz\n'
197	>	txt += '    size=`expr $tmp_size`\n'
198	>	txt += '    echo "Total Output dimension: $size"\n'
199	>	txt += '    limit='+str(self.OSBsize) +' \n'
200	>	txt += '    echo "WARNING: output files size limit is set to: $limit"\n'
201	>	txt += '    if [ "$limit" -lt "$size" ]; then\n'
202	>	txt += '        exceed=1\n'
203	>	txt += '        job_exit_code=70000\n'
204	>	txt += '        echo "Output Sanbox too big. Produced output is lost "\n'
205	>	txt += '    else\n'
206	>	txt += '        exceed=0\n'
207	>	txt += '        echo "Total Output dimension $size is fine."\n'
208	>	txt += '    fi\n'
209	>
210	>	txt += '    echo "JOB_EXIT_STATUS = $job_exit_code"\n'
211	>	txt += '    echo "JobExitCode=$job_exit_code" >> $RUNTIME_AREA/$repo\n'
212	>	txt += '    dumpStatus $RUNTIME_AREA/$repo\n'
213	>	txt += '    if [ $exceed -ne 1 ]; then\n'
214	>	txt += '        tar zcvf ${out_files}.tgz  ${final_list}\n'
215	>	txt += '    else\n'
216	>	txt += '        tar zcvf ${out_files}.tgz CMSSW_${NJob}.stdout CMSSW_${NJob}.stderr\n'
217	>	txt += '    fi\n'
218	>	txt += '    python $RUNTIME_AREA/fillCrabFjr.py $RUNTIME_AREA/crab_fjr_$NJob.xml --errorcode $job_exit_code \n'
219
220		txt += '    exit $job_exit_code\n'
221		txt += '}\n'
#	Line 179 | Line 234 | class SchedulerRcondor(SchedulerGrid) :
234		common._db.updateTask_(taskReq)
235
236
182	–	# presa di brutto da SchedulerGrid.py
183	–	"""
184	–
185	–	def wsSetupEnvironment(self):
186	–
187	–	#Returns part of a job script which does scheduler-specific work.
188	–
189	–	taskId =common._db.queryTask('name')
190	–	index = int(common._db.nJobs())
191	–	job = common.job_list[index-1]
192	–	jbt = job.type()
193	–	if not self.environment_unique_identifier:
194	–	try :
195	–	self.environment_unique_identifier = self.envUniqueID()
196	–	except :
197	–	raise CrabException('environment_unique_identifier not set')
198	–
199	–	# start with wrapper timing
200	–	txt  = 'export TIME_WRAP_INI=`date +%s` \n'
201	–	txt += 'export TIME_STAGEOUT=-2 \n\n'
202	–	txt += '# '+self.name()+' specific stuff\n'
203	–	txt += '# strip arguments\n'
204	–	txt += 'echo "strip arguments"\n'
205	–	txt += 'args=("$@")\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	–	"""

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines