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 |
52 |
|
glite_config = gliteConfig.config() |
53 |
|
|
54 |
|
if (glite_config ): |
55 |
< |
rb_param_file = 'WMSconfig = '+glite_config+';\n' |
55 |
> |
rb_param_file = glite_config |
56 |
|
return rb_param_file |
57 |
|
|
58 |
< |
def sched_parameter(self): |
58 |
> |
def ce_list(self): |
59 |
|
""" |
60 |
< |
Returns file with requirements and scheduler-specific parameters |
60 |
> |
Returns string with requirement CE related |
61 |
|
""" |
62 |
< |
index = int(common.jobDB.nJobs()) - 1 |
63 |
< |
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 |
< |
|
62 |
> |
ceParser = CEBlackWhiteListParser(self.EDG_ce_white_list, |
63 |
> |
self.EDG_ce_black_list, common.logger()) |
64 |
|
req = '' |
65 |
< |
req = req + jbt.getRequirements() |
66 |
< |
|
52 |
< |
|
53 |
< |
if self.EDG_requirements: |
54 |
< |
if (not req == ' '): req = req + ' && ' |
55 |
< |
req = req + self.EDG_requirements |
56 |
< |
|
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: |
72 |
|
tmpCe.append('RegExp("' + string.strip(ce) + '", other.GlueCEUniqueId)') |
63 |
– |
### MATTY' FIX: if more then one CE: && -> || |
64 |
– |
#print "list CE: " + str(tmpCe) |
73 |
|
if len(tmpCe) == 1: |
74 |
|
req += " && (" + concString.join(tmpCe) + ") " |
75 |
|
elif len(tmpCe) > 1: |
76 |
|
firstCE = 0 |
77 |
|
for reqTemp in tmpCe: |
70 |
– |
#print reqTemp |
78 |
|
if firstCE == 0: |
72 |
– |
#print "adding: "+str(" && ( (" + reqTemp + ") ") |
79 |
|
req += " && ( (" + reqTemp + ") " |
80 |
|
firstCE = 1 |
81 |
|
elif firstCE > 0: |
76 |
– |
#print "adding: "+str(" || (" + reqTemp + ") ") |
82 |
|
req += " || (" + reqTemp + ") " |
83 |
|
if firstCE > 0: |
84 |
|
req += ") " |
80 |
– |
## old code |
81 |
– |
# if len(tmpCe): req = req + " && (" + concString.join(tmpCe) + ") " |
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: |
91 |
|
tmpCe.append('(!RegExp("' + string.strip(ce) + '", other.GlueCEUniqueId))') |
92 |
< |
if len(tmpCe): req = req + " && (" + concString.join(tmpCe) + ") " |
92 |
> |
if len(tmpCe): req += " && (" + concString.join(tmpCe) + ") " |
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, retWL, retBL |
104 |
> |
|
105 |
> |
def se_list(self, dest): |
106 |
> |
""" |
107 |
> |
Returns string with requirement SE related |
108 |
> |
""" |
109 |
> |
hostList=self.findSites_(dest) |
110 |
> |
req='' |
111 |
> |
reqtmp=[] |
112 |
> |
concString = '||' |
113 |
> |
|
114 |
> |
for arg in hostList: |
115 |
> |
reqtmp.append(' Member("'+arg+'" , other.GlueCESEBindGroupSEUniqueID) ') |
116 |
> |
|
117 |
> |
if len(reqtmp): req += " && (" + concString.join(reqtmp) + ") " |
118 |
|
|
119 |
+ |
return req |
120 |
+ |
|
121 |
+ |
def jdlParam(self): |
122 |
+ |
""" |
123 |
+ |
Returns |
124 |
+ |
""" |
125 |
+ |
req='' |
126 |
+ |
if self.EDG_addJdlParam: |
127 |
+ |
if self.EDG_addJdlParam[-1] == '': self.EDG_addJdlParam= self.EDG_addJdlParam[:-1] |
128 |
+ |
for p in self.EDG_addJdlParam: |
129 |
+ |
req+=string.strip(p)+';\n' |
130 |
+ |
return req |
131 |
+ |
|
132 |
+ |
def specific_req(self): |
133 |
+ |
""" |
134 |
+ |
Returns string with specific requirements |
135 |
+ |
""" |
136 |
+ |
req='' |
137 |
|
if self.EDG_clock_time: |
138 |
|
if (not req == ' '): req = req + ' && ' |
139 |
|
req = req + 'other.GlueCEPolicyMaxWallClockTime>='+self.EDG_clock_time |
142 |
|
if (not req == ' '): req = req + ' && ' |
143 |
|
req = req + ' other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time |
144 |
|
|
145 |
< |
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' |
145 |
> |
return req |
146 |
|
|
147 |
< |
return txt |
147 |
> |
def sched_parameter(self,i,task): |
148 |
> |
""" |
149 |
> |
Returns string with requirements and scheduler-specific parameters |
150 |
> |
""" |
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(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 |
166 |
|
|
167 |
< |
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) |
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 |
171 |
> |
loggingInfo = EdgLoggingInfo.EdgLoggingInfo() |
172 |
> |
reason = loggingInfo.decodeReason(file) |
173 |
> |
return reason |
174 |
|
|
175 |
< |
def findSites_(self, n): |
175 |
> |
def findSites_(self, sites): |
176 |
|
itr4 =[] |
233 |
– |
sites = common.jobDB.destination(n) |
177 |
|
if len(sites)>0 and sites[0]=="": |
178 |
|
return itr4 |
179 |
|
if sites != [""]: |
180 |
< |
##Addedd Daniele |
238 |
< |
replicas = self.blackWhiteListParser.checkBlackList(sites,n) |
180 |
> |
replicas = self.blackWhiteListParser.checkBlackList(sites) |
181 |
|
if len(replicas)!=0: |
182 |
< |
replicas = self.blackWhiteListParser.checkWhiteList(replicas,n) |
182 |
> |
replicas = self.blackWhiteListParser.checkWhiteList(replicas) |
183 |
|
|
184 |
|
itr4 = replicas |
243 |
– |
##### |
185 |
|
return itr4 |
186 |
|
|
187 |
< |
def tOut(self, list): |
188 |
< |
return 180 |
187 |
> |
def delegateProxy(self): |
188 |
> |
self.boss().delegateProxy() |
189 |
> |
return |
190 |
> |
|
191 |
> |
def wsExitFunc(self): |
192 |
> |
""" |
193 |
> |
""" |
194 |
> |
txt = '\n' |
195 |
> |
|
196 |
> |
txt += '#\n' |
197 |
> |
txt += '# EXECUTE THIS FUNCTION BEFORE EXIT \n' |
198 |
> |
txt += '#\n\n' |
199 |
> |
|
200 |
> |
txt += 'func_exit() { \n' |
201 |
> |
txt += self.wsExitFunc_common() |
202 |
> |
### specific Glite check for OSB |
203 |
> |
txt += ' tar zcvf ${out_files}.tgz ${final_list}\n' |
204 |
> |
txt += ' tmp_size=`ls -gGrta ${out_files}.tgz | awk \'{ print $3 }\'`\n' |
205 |
> |
txt += ' rm ${out_files}.tgz\n' |
206 |
> |
txt += ' size=`expr $tmp_size`\n' |
207 |
> |
txt += ' echo "Total Output dimension: $size"\n' |
208 |
> |
txt += ' limit='+str(self.OSBsize) +' \n' |
209 |
> |
txt += ' echo "WARNING: output files size limit is set to: $limit"\n' |
210 |
> |
txt += ' if [ "$limit" -lt "$size" ]; then\n' |
211 |
> |
txt += ' exceed=1\n' |
212 |
> |
txt += ' job_exit_code=70000\n' |
213 |
> |
txt += ' echo "Output Sanbox too big. Produced output is lost "\n' |
214 |
> |
txt += ' else\n' |
215 |
> |
txt += ' exceed=0\n' |
216 |
> |
txt += ' echo "Total Output dimension $size is fine."\n' |
217 |
> |
txt += ' fi\n' |
218 |
> |
|
219 |
> |
txt += ' echo "JOB_EXIT_STATUS = $job_exit_code"\n' |
220 |
> |
txt += ' echo "JobExitCode=$job_exit_code" >> $RUNTIME_AREA/$repo\n' |
221 |
> |
txt += ' dumpStatus $RUNTIME_AREA/$repo\n' |
222 |
> |
txt += ' if [ $exceed -ne 1 ]; then\n' |
223 |
> |
txt += ' tar zcvf ${out_files}.tgz ${final_list}\n' |
224 |
> |
txt += ' else\n' |
225 |
> |
txt += ' tar zcvf ${out_files}.tgz CMSSW_${NJob}.stdout CMSSW_${NJob}.stderr\n' |
226 |
> |
txt += ' fi\n' |
227 |
> |
txt += ' exit $job_exit_code\n' |
228 |
> |
|
229 |
> |
txt += '}\n' |
230 |
> |
return txt |