ViewVC Help

View File | Revision Log | Show Annotations | Root Listing

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

Comparing COMP/CRAB/python/SchedulerGlite.py (file contents):
Revision 1.35 by fanzago, Mon Mar 3 16:41:24 2008 UTC vs.
Revision 1.80 by farinafa, Thu Apr 29 10:23:06 2010 UTC

#	Line 1 | Line 1
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 *
11	>	import EdgLoggingInfo
12		import common
13	+	from WMCore.SiteScreening.BlackWhiteListParser import CEBlackWhiteListParser
14
15		import os, sys, time
16
#	Line 11 | Line 18 | class SchedulerGlite(SchedulerGrid):
18		def __init__(self, name="GLITE"):
19		SchedulerGrid.__init__(self,name)
20
21	+	self.OSBsize = 55000000
22	+
23		def configure(self,cfg_params):
24		SchedulerGrid.configure(self, cfg_params)
25	<	self.environment_unique_identifier = 'GLITE_WMS_JOBID'
25	>	self.environment_unique_identifier = '$GLITE_WMS_JOBID'
26	>
27	>	def realSchedParams(self,cfg_params):
28	>	"""
29	>	Return dictionary with specific parameters, to use
30	>	with real scheduler
31	>	"""
32	>	self.rb_param_file=''
33	>	if (not cfg_params.has_key('GRID.rb')):
34	>	cfg_params['GRID.rb']='CERN'
35	>	self.rb_param_file=common.scheduler.rb_configure(cfg_params.get("GRID.rb"))
36	>	self.wms_service=cfg_params.get("GRID.wms_service",'')
37	>	self.skipWMSAuth=cfg_params.get("GRID.skipwmsauth",1)
38	>	params = { 'service' : self.wms_service, \
39	>	'config' : self.rb_param_file, \
40	>	'skipWMSAuth' : self.skipWMSAuth
41	>	}
42	>	return  params
43	>
44
45		def rb_configure(self, RB):
46	+	url ='http://cmsdoc.cern.ch/cms/LCG/crab/config/'
47	+	from Downloader import Downloader
48	+	import httplib
49	+	common.logger.debug('Downloading config files for WMS: '+url)
50	+	## 25-Jun-2009 SL: patch to use Cream enabled WMS
51	+	if ( self.cfg_params.get('GRID.use_cream',None) ):
52	+	RB='CREAM'
53		if not RB: return None
20	–	glite_config = None
54		rb_param_file = None
55	+	configFileName = 'glite_wms_'+str(RB)+'.conf'
56
57	<	gliteConfig = GliteConfig(RB)
58	<	glite_config = gliteConfig.config()
57	>	results = Downloader(url)
58	>	try:
59	>	gliteConfig  = results.filePath(configFileName)
60	>	except httplib.HTTPException, ex:
61	>	raise CrabException( "Problem getting RB config file: %s, reason:"%(configFileName, ex) )
62
63	<	if (glite_config ):
64	<	rb_param_file = 'WMSconfig = '+glite_config+';\n'
63	>	if (gliteConfig ):
64	>	rb_param_file = gliteConfig
65		return rb_param_file
66
67	<	def sched_parameter(self):
67	>	def ce_list(self):
68		"""
69	<	Returns file with requirements and scheduler-specific parameters
69	>	Returns string with requirement CE related
70		"""
71	<	index = int(common.jobDB.nJobs()) - 1
72	<	job = common.job_list[index]
36	<	jbt = job.type()
37	<
38	<	lastDest=''
39	<	first = []
40	<	last  = []
41	<	for n in range(common.jobDB.nJobs()):
42	<	currDest=common.jobDB.destination(n)
43	<	if (currDest!=lastDest):
44	<	lastDest = currDest
45	<	first.append(n)
46	<	if n != 0:last.append(n-1)
47	<	if len(first)>len(last) :last.append(common.jobDB.nJobs())
48	<
71	>	ceParser = CEBlackWhiteListParser(self.EDG_ce_white_list,
72	>	self.EDG_ce_black_list, common.logger())
73		req = ''
74	<	req = req + jbt.getRequirements()
75	<
52	<
53	<	if self.EDG_requirements:
54	<	if (not req == ' '): req = req +  ' && '
55	<	req = req + self.EDG_requirements
56	<
74	>	ce_white_list = []
75	>	ce_black_list = []
76		if self.EDG_ce_white_list:
77	<	ce_white_list = self.EDG_ce_white_list
77	>	ce_white_list = ceParser.whiteList()
78		tmpCe=[]
79		concString = '&&'
80		for ce in ce_white_list:
81		tmpCe.append('RegExp("' + string.strip(ce) + '", other.GlueCEUniqueId)')
63	–	### MATTY' FIX: if more then one CE: && -> ||
64	–	#print "list CE: " + str(tmpCe)
82		if len(tmpCe) == 1:
83		req +=  " && (" + concString.join(tmpCe) + ") "
84		elif len(tmpCe) > 1:
85		firstCE = 0
86		for reqTemp in tmpCe:
70	–	#print reqTemp
87		if firstCE == 0:
72	–	#print "adding: "+str(" && ( (" + reqTemp + ") ")
88		req += " && ( (" + reqTemp + ") "
89		firstCE = 1
90		elif firstCE > 0:
76	–	#print "adding: "+str(" || (" + reqTemp + ") ")
91		req += " || (" + reqTemp + ") "
92		if firstCE > 0:
93		req += ") "
80	–	## old code
81	–	#            if len(tmpCe): req = req + " && (" + concString.join(tmpCe) + ") "
94
95		if self.EDG_ce_black_list:
96	<	ce_black_list = self.EDG_ce_black_list
96	>	ce_black_list = ceParser.blackList()
97		tmpCe=[]
98		concString = '&&'
99		for ce in ce_black_list:
100		tmpCe.append('(!RegExp("' + string.strip(ce) + '", other.GlueCEUniqueId))')
101	<	if len(tmpCe): req = req + " && (" + concString.join(tmpCe) + ") "
101	>	if len(tmpCe): req += " && (" + concString.join(tmpCe) + ") "
102	>
103	>	# requirement added to skip gliteCE
104	>	# not more needed
105	>	#       req += '&& (!RegExp("blah", other.GlueCEUniqueId))'
106	>	retWL = ','.join(ce_white_list)
107	>	retBL = ','.join(ce_black_list)
108	>	if not retWL:
109	>	retWL = None
110	>	if not retBL:
111	>	retBL = None
112	>
113	>	return req, retWL, retBL
114	>
115	>	def se_list(self, dest):
116	>	"""
117	>	Returns string with requirement SE related
118	>	"""
119	>	hostList=self.findSites_(dest)
120	>	req=''
121	>	reqtmp=[]
122	>	concString = '||'
123	>
124	>	for arg in hostList:
125	>	reqtmp.append(' Member("'+arg+'" , other.GlueCESEBindGroupSEUniqueID) ')
126	>
127	>	if len(reqtmp): req += " && (" + concString.join(reqtmp) + ") "
128	>
129	>	return req
130	>
131	>	def jdlParam(self):
132	>	"""
133	>	Returns
134	>	"""
135	>	req=''
136	>	if self.EDG_addJdlParam:
137	>	if self.EDG_addJdlParam[-1] == '': self.EDG_addJdlParam= self.EDG_addJdlParam[:-1]
138	>	for p in self.EDG_addJdlParam:
139	>	req+=string.strip(p)+';\n'
140	>	return req
141
142	+	def specific_req(self):
143	+	"""
144	+	Returns string with specific requirements
145	+	"""
146	+	req=''
147		if self.EDG_clock_time:
148		if (not req == ' '): req = req + ' && '
149		req = req + 'other.GlueCEPolicyMaxWallClockTime>='+self.EDG_clock_time
#	Line 96 | Line 152 | class SchedulerGlite(SchedulerGrid):
152		if (not req == ' '): req = req + ' && '
153		req = req + ' other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time
154
155	<	for i in range(len(first)): # Add loop DS
100	<	self.param='sched_param_'+str(i)+'.clad'
101	<	param_file = open(common.work_space.shareDir()+'/'+self.param, 'w')
102	<
103	<	itr4=self.findSites_(first[i])
104	<	reqSites=''
105	<	reqtmp=[]
106	<	concString = '||'
107	<
108	<	#############
109	<	# MC Changed matching syntax to avoid gang matching
110	<	#############
111	<	for arg in itr4:
112	<	reqtmp.append(' Member("'+arg+'" , other.GlueCESEBindGroupSEUniqueID) ')
113	<
114	<	if len(reqtmp): reqSites = reqSites + " && (" + concString.join(reqtmp) + ") "
115	<
116	<	# requirement added to skip gliteCE
117	<	reqSites = reqSites + '&& (!RegExp("blah", other.GlueCEUniqueId));\n'
118	<
119	<	param_file.write('Requirements = ' + req + reqSites )
120	<
121	<	if (self.rb_param_file):
122	<	param_file.write(self.rb_param_file)
123	<
124	<	if self.EDG_addJdlParam:
125	<	if self.EDG_addJdlParam[-1] == '': self.EDG_addJdlParam= self.EDG_addJdlParam[:-1]
126	<	for p in self.EDG_addJdlParam:
127	<	param_file.write(string.strip(p)+';\n')
128	<
129	<	param_file.close()
130	<
131	<	def wsSetupEnvironment(self):
132	<	"""
133	<	Returns part of a job script which does scheduler-specific work.
134	<	"""
135	<	txt = ''
136	<	txt += '# strip arguments\n'
137	<	txt += 'echo "strip arguments"\n'
138	<	txt += 'args=("$@")\n'
139	<	txt += 'nargs=$#\n'
140	<	txt += 'shift $nargs\n'
141	<	txt += "# job number (first parameter for job wrapper)\n"
142	<	txt += "NJob=${args[0]}; export NJob\n"
143	<
144	<	txt += '# job identification to DashBoard \n'
145	<	#txt += 'MonitorJobID=`echo ${NJob}_$GLITE_WMS_JOBID`\n'
146	<	#txt += 'SyncGridJobId=`echo $GLITE_WMS_JOBID`\n'
147	<	#txt += 'MonitorID=`echo ' + self._taskId + '`\n'
148	<	txt += 'MonitorJobID=${NJob}_$GLITE_WMS_JOBID \n'
149	<	txt += 'SyncGridJobId=$GLITE_WMS_JOBID \n'
150	<	txt += 'MonitorID='+self._taskId+' \n'
151	<	txt += 'echo "MonitorJobID=$MonitorJobID" > $RUNTIME_AREA/$repo \n'
152	<	txt += 'echo "SyncGridJobId=$SyncGridJobId" >> $RUNTIME_AREA/$repo \n'
153	<	txt += 'echo "MonitorID=$MonitorID" >> $RUNTIME_AREA/$repo\n'
154	<	#txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
155	<	#txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
156	<	#txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
157	<
158	<	#txt += 'echo "middleware discovery: " \n'
159	<	txt += 'echo ">>> GridFlavour discovery: " \n'
160	<	txt += 'if [ $VO_CMS_SW_DIR ]; then \n'
161	<	txt += '    middleware=LCG\n'
162	<	#txt += '    echo "SyncCE=`glite-brokerinfo getCE`" | tee -a $RUNTIME_AREA/$repo \n'
163	<	txt += '    echo "SyncCE=`glite-brokerinfo getCE`" >> $RUNTIME_AREA/$repo \n'
164	<	#txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
165	<	txt += '    echo "GridFlavour=$middleware" | tee -a $RUNTIME_AREA/$repo \n'
166	<	#txt += '    echo ">>> middleware =$middleware" \n'
167	<	txt += 'elif [ $OSG_APP ]; then \n'
168	<	txt += '    middleware=OSG \n'
169	<	txt += '    if [ $OSG_JOB_CONTACT ]; then \n'
170	<	txt += '        SyncCE="$OSG_JOB_CONTACT"; \n'
171	<	#txt += '        echo "SyncCE=$SyncCE" | tee -a $RUNTIME_AREA/$repo ;\n'
172	<	txt += '        echo "SyncCE=$SyncCE" >> $RUNTIME_AREA/$repo ;\n'
173	<	txt += '    else\n'
174	<	txt += '        echo "not reporting SyncCE";\n'
175	<	txt += '    fi\n';
176	<	txt += '    echo "GridFlavour=$middleware" | tee -a $RUNTIME_AREA/$repo \n'
177	<	#txt += '    echo ">>> middleware =$middleware" \n'
178	<	txt += 'else \n'
179	<	txt += '    echo "ERROR ==> GridFlavour not identified" \n'
180	<	txt += '    job_exit_code=10030\n'
181	<	txt += '    func_exit \n'
182	<	#txt += '    echo "SET_CMS_ENV 10030 ==> middleware not identified" \n'
183	<	#txt += '    echo "JOB_EXIT_STATUS = 10030" \n'
184	<	#txt += '    echo "JobExitCode=10030" | tee -a $RUNTIME_AREA/$repo \n'
185	<	#txt += '    dumpStatus $RUNTIME_AREA/$repo \n'
186	<	#txt += '    exit 1 \n'
187	<	txt += 'fi \n'
188	<
189	<	txt += 'dumpStatus $RUNTIME_AREA/$repo \n'
190	<
191	<	txt += '\n\n'
192	<
193	<	txt += 'export VO='+self.VO+'\n'
194	<	txt += 'if [ $middleware == LCG ]; then\n'
195	<	txt += '    CloseCEs=`glite-brokerinfo getCE`\n'
196	<	txt += '    echo "CloseCEs = $CloseCEs"\n'
197	<	txt += '    CE=`echo $CloseCEs | sed -e "s/:.*//"`\n'
198	<	txt += '    echo "CE = $CE"\n'
199	<	txt += 'elif [ $middleware == OSG ]; then \n'
200	<	txt += '    if [ $OSG_JOB_CONTACT ]; then \n'
201	<	txt += '        CE=`echo $OSG_JOB_CONTACT | /usr/bin/awk -F\/ \'{print $1}\'` \n'
202	<	txt += '    else \n'
203	<	txt += '        echo "ERROR ==> OSG mode in setting CE name from OSG_JOB_CONTACT" \n'
204	<	txt += '        job_exit_code=10099\n'
205	<	txt += '        func_exit \n'
206	<	#txt += '        echo "SET_CMS_ENV 10099 ==> OSG mode: ERROR in setting CE name from OSG_JOB_CONTACT" \n'
207	<	#txt += '        echo "JOB_EXIT_STATUS = 10099" \n'
208	<	#txt += '        echo "JobExitCode=10099" | tee -a $RUNTIME_AREA/$repo \n'
209	<	#txt += '        dumpStatus $RUNTIME_AREA/$repo \n'
210	<	#txt += '        exit 1 \n'
211	<	txt += '    fi \n'
212	<	txt += 'fi \n'
155	>	return req
156
157	<	return txt
157	>	def sched_parameter(self,i,task):
158	>	"""
159	>	Returns string with requirements and scheduler-specific parameters
160	>	"""
161	>	dest=  task.jobs[i-1]['dlsDestination']
162	>
163	>	req=''
164	>	req +=task['jobType']
165	>
166	>	sched_param=''
167	>	sched_param+='Requirements = ' + req +self.specific_req() + self.se_list(dest) +\
168	>	self.ce_list()[0] +';\n'
169	>	if self.EDG_addJdlParam: sched_param+=self.jdlParam()
170	>	sched_param+='MyProxyServer = "' + self.proxyServer + '";\n'
171	>	sched_param+='VirtualOrganisation = "' + self.VO + '";\n'
172	>	sched_param+='RetryCount = '+str(self.EDG_retry_count)+';\n'
173	>	sched_param+='DefaultNodeRetryCount = '+str(self.EDG_retry_count)+';\n'
174	>	sched_param+='ShallowRetryCount = '+str(self.EDG_shallow_retry_count)+';\n'
175	>	sched_param+='DefaultNodeShallowRetryCount = '+str(self.EDG_shallow_retry_count)+';\n'
176
177	<	def loggingInfo(self, id):
177	>	return sched_param
178	>
179	>	def decodeLogInfo(self, file):
180		"""
181	<	retrieve the logging info from logging and bookkeeping and return it
181	>	Parse logging info file and return main info
182		"""
183	<	self.checkProxy()
184	<	cmd = 'glite-job-logging-info -v 3 ' + id
185	<	cmd_out = runCommand(cmd)
223	<	return cmd_out
224	<
225	<	def queryDetailedStatus(self, id):
226	<	""" Query a detailed status of the job with id """
227	<	cmd = 'glite-job-status '+id
228	<	cmd_out = runCommand(cmd)
229	<	return cmd_out
183	>	loggingInfo = EdgLoggingInfo.EdgLoggingInfo()
184	>	reason = loggingInfo.decodeReason(file)
185	>	return reason
186
187	<	def findSites_(self, n):
187	>	def findSites_(self, sites):
188		itr4 =[]
233	–	sites = common.jobDB.destination(n)
189		if len(sites)>0 and sites[0]=="":
190		return itr4
191		if sites != [""]:
192	<	##Addedd Daniele
238	<	replicas = self.blackWhiteListParser.checkBlackList(sites,n)
192	>	replicas = self.blackWhiteListParser.checkBlackList(sites)
193		if len(replicas)!=0:
194	<	replicas = self.blackWhiteListParser.checkWhiteList(replicas,n)
194	>	replicas = self.blackWhiteListParser.checkWhiteList(replicas)
195
196		itr4 = replicas
243	–	#####
197		return itr4
198
199	<	def tOut(self, list):
200	<	return 180
199	>	def delegateProxy(self):
200	>	self.boss().delegateProxy()
201	>	return
202	>
203	>	def wsExitFunc(self):
204	>	"""
205	>	"""
206	>	txt = '\n'
207	>
208	>	txt += '#\n'
209	>	txt += '# EXECUTE THIS FUNCTION BEFORE EXIT \n'
210	>	txt += '#\n\n'
211	>
212	>	txt += 'func_exit() { \n'
213	>	txt += self.wsExitFunc_common()
214	>	### specific Glite check for OSB
215	>	txt += '    tar zcvf ${out_files}.tgz  ${final_list}\n'
216	>	txt += '    tmp_size=`ls -gGrta ${out_files}.tgz | awk \'{ print $3 }\'`\n'
217	>	txt += '    rm ${out_files}.tgz\n'
218	>	txt += '    size=`expr $tmp_size`\n'
219	>	txt += '    echo "Total Output dimension: $size"\n'
220	>	txt += '    limit='+str(self.OSBsize) +' \n'
221	>	txt += '    echo "WARNING: output files size limit is set to: $limit"\n'
222	>	txt += '    if [ "$limit" -lt "$size" ]; then\n'
223	>	txt += '        exceed=1\n'
224	>	txt += '        job_exit_code=70000\n'
225	>	txt += '        echo "Output Sanbox too big. Produced output is lost "\n'
226	>	txt += '    else\n'
227	>	txt += '        exceed=0\n'
228	>	txt += '        echo "Total Output dimension $size is fine."\n'
229	>	txt += '    fi\n'
230	>
231	>	txt += '    echo "JOB_EXIT_STATUS = $job_exit_code"\n'
232	>	txt += '    echo "JobExitCode=$job_exit_code" >> $RUNTIME_AREA/$repo\n'
233	>	txt += '    dumpStatus $RUNTIME_AREA/$repo\n'
234	>	txt += '    if [ $exceed -ne 1 ]; then\n'
235	>	txt += '        tar zcvf ${out_files}.tgz  ${final_list}\n'
236	>	txt += '    else\n'
237	>	txt += '        tar zcvf ${out_files}.tgz CMSSW_${NJob}.stdout CMSSW_${NJob}.stderr\n'
238	>	txt += '    fi\n'
239	>	txt += '    python $RUNTIME_AREA/fillCrabFjr.py $RUNTIME_AREA/crab_fjr_$NJob.xml --errorcode $job_exit_code \n'
240	>	txt += '    exit $job_exit_code\n'
241	>
242	>	txt += '}\n'
243	>	return txt
244	>
245	>	def listMatch(self, dest, full):
246	>	matching='fast'
247	>
248	>	if self.boss().schedulerConfig['name'] == 'SchedulerGLite' :
249	>	taskId=common._db.getTask()
250	>	req=str(self.sched_parameter(1,taskId))
251	>	sites = self.boss().schedSession().matchResources(taskId, requirements=req)
252	>	else :
253	>	sites = SchedulerGrid.listMatch(self, dest, full)
254	>
255	>	if full == True: matching='full'
256	>	common.logger.debug("list of available site ( "+str(matching) +" matching ) : "+str(sites))
257	>
258	>	return sites

Diff Legend

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