ViewVC Help

View File | Revision Log | Show Annotations | Root Listing

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

Comparing COMP/CRAB/python/SchedulerEdg.py (file contents):
Revision 1.31 by slacapra, Tue Dec 6 13:18:48 2005 UTC vs.
Revision 1.80 by slacapra, Wed Sep 20 17:29:52 2006 UTC

#	Line 2 | Line 2 | from Scheduler import Scheduler
2		from crab_logger import Logger
3		from crab_exceptions import *
4		from crab_util import *
5	+	from EdgConfig import *
6		import common
7
8		import os, sys, time
#	Line 21 | Line 22 | class SchedulerEdg(Scheduler):
22
23		def configure(self, cfg_params):
24
25	<	try: self.edg_config = cfg_params["EDG.config"]
26	<	except KeyError: self.edg_config = ''
25	>	try:
26	>	RB = cfg_params["EDG.rb"]
27	>	edgConfig = EdgConfig(RB)
28	>	self.edg_config = edgConfig.config()
29	>	self.edg_config_vo = edgConfig.configVO()
30	>	except KeyError:
31	>	self.edg_config = ''
32	>	self.edg_config_vo = ''
33
34	<	try: self.edg_config_vo = cfg_params["EDG.config_vo"]
35	<	except KeyError: self.edg_config_vo = ''
34	>	try:
35	>	self.proxyServer = cfg_params["EDG.proxy_server"]
36	>	except KeyError:
37	>	self.proxyServer = 'myproxy.cern.ch'
38	>	common.logger.debug(5,'Setting myproxy server to '+self.proxyServer)
39
40	+	try:
41	+	self.role = cfg_params["EDG.rb"]
42	+	except KeyError:
43	+	self.role = None
44	+
45	+	try:
46	+	self.role = cfg_params["EDG.role"]
47	+	except KeyError:
48	+	self.role = None
49	+
50		try: self.LCG_version = cfg_params["EDG.lcg_version"]
51		except KeyError: self.LCG_version = '2'
52
#	Line 36 | Line 56 | class SchedulerEdg(Scheduler):
56		try: self.EDG_retry_count = cfg_params['EDG.retry_count']
57		except KeyError: self.EDG_retry_count = ''
58
59	+	try:
60	+	self.EDG_ce_black_list = cfg_params['EDG.ce_black_list']
61	+	#print "self.EDG_ce_black_list = ", self.EDG_ce_black_list
62	+	except KeyError:
63	+	self.EDG_ce_black_list  = ''
64	+
65	+	try:
66	+	self.EDG_ce_white_list = cfg_params['EDG.ce_white_list']
67	+	#print "self.EDG_ce_white_list = ", self.EDG_ce_white_list
68	+	except KeyError: self.EDG_ce_white_list = ''
69	+
70		try: self.VO = cfg_params['EDG.virtual_organization']
71		except KeyError: self.VO = 'cms'
72
73		try: self.return_data = cfg_params['USER.return_data']
74		except KeyError: self.return_data = 1
75
76	+	try:
77	+	self.copy_input_data = common.analisys_common_info['copy_input_data']
78	+	#print "self.copy_input_data = ", self.copy_input_data
79	+	except KeyError: self.copy_input_data = 0
80	+
81		try:
82		self.copy_data = cfg_params["USER.copy_data"]
83		if int(self.copy_data) == 1:
#	Line 59 | Line 95 | class SchedulerEdg(Scheduler):
95		msg = 'Warning: return_data = 0 and copy_data = 0 ==> your exe output will be lost\n'
96		msg = msg + 'Please modify return_data and copy_data value in your crab.cfg file\n'
97		raise CrabException(msg)
98	+
99	+	try:
100	+	self.lfc_host = cfg_params['EDG.lfc_host']
101	+	except KeyError:
102	+	msg = "Error. The [EDG] section does not have 'lfc_host' value"
103	+	msg = msg + " it's necessary to know the LFC host name"
104	+	common.logger.message(msg)
105	+	raise CrabException(msg)
106	+	try:
107	+	self.lcg_catalog_type = cfg_params['EDG.lcg_catalog_type']
108	+	except KeyError:
109	+	msg = "Error. The [EDG] section does not have 'lcg_catalog_type' value"
110	+	msg = msg + " it's necessary to know the catalog type"
111	+	common.logger.message(msg)
112	+	raise CrabException(msg)
113	+	try:
114	+	self.lfc_home = cfg_params['EDG.lfc_home']
115	+	except KeyError:
116	+	msg = "Error. The [EDG] section does not have 'lfc_home' value"
117	+	msg = msg + " it's necessary to know the home catalog dir"
118	+	common.logger.message(msg)
119	+	raise CrabException(msg)
120
121		try:
122		self.register_data = cfg_params["USER.register_data"]
#	Line 67 | Line 125 | class SchedulerEdg(Scheduler):
125		self.LFN = cfg_params['USER.lfn_dir']
126		except KeyError:
127		msg = "Error. The [USER] section does not have 'lfn_dir' value"
128	<	msg = msg + " it's necessary for RLS registration"
128	>	msg = msg + " it's necessary for LCF registration"
129		common.logger.message(msg)
130		raise CrabException(msg)
131		except KeyError: self.register_data = 0
#	Line 104 | Line 162 | class SchedulerEdg(Scheduler):
162		sys.path.append(libPath)
163
164		self.proxyValid=0
165	+
166	+	try:
167	+	self._taskId = cfg_params['taskId']
168	+	except:
169	+	self._taskId = ''
170	+
171		return
172
173
#	Line 126 | Line 190 | class SchedulerEdg(Scheduler):
190		"""
191		Returns part of a job script which does scheduler-specific work.
192		"""
129	–
193		txt = ''
194	+	txt += '# strip arguments\n'
195	+	txt += 'echo "strip arguments"\n'
196	+	txt += 'args=("$@")\n'
197	+	txt += 'nargs=$#\n'
198	+	txt += 'shift $nargs\n'
199	+	txt += "# job number (first parameter for job wrapper)\n"
200	+	#txt += "NJob=$1\n"
201	+	txt += "NJob=${args[0]}\n"
202	+
203	+	txt += '# job identification to DashBoard \n'
204	+	txt += 'MonitorJobID=`echo ${NJob}_$EDG_WL_JOBID`\n'
205	+	txt += 'SyncGridJobId=`echo $EDG_WL_JOBID`\n'
206	+	txt += 'MonitorID=`echo ' + self._taskId + '`\n'
207	+	txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
208	+	txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
209	+	txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
210	+
211	+	txt += 'echo "middleware discovery " \n'
212	+	txt += 'if [ $VO_CMS_SW_DIR ]; then \n'
213	+	txt += '    middleware=LCG \n'
214	+	txt += '    echo "SyncCE=`edg-brokerinfo getCE`" | tee -a $RUNTIME_AREA/$repo \n'
215	+	txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
216	+	txt += '    echo "middleware =$middleware" \n'
217	+	txt += 'elif [ $GRID3_APP_DIR ]; then\n'
218	+	txt += '    middleware=OSG \n'
219	+	txt += '    echo "SyncCE=`echo $EDG_WL_LOG_DESTINATION`" | tee -a $RUNTIME_AREA/$repo \n'
220	+	txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
221	+	txt += '    echo "middleware =$middleware" \n'
222	+	txt += 'elif [ $OSG_APP ]; then \n'
223	+	txt += '    middleware=OSG \n'
224	+	txt += '    echo "SyncCE=`echo $EDG_WL_LOG_DESTINATION`" | tee -a $RUNTIME_AREA/$repo \n'
225	+	txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
226	+	txt += '    echo "middleware =$middleware" \n'
227	+	txt += 'else \n'
228	+	txt += '    echo "SET_CMS_ENV 10030 ==> middleware not identified" \n'
229	+	txt += '    echo "JOB_EXIT_STATUS = 10030" \n'
230	+	txt += '    echo "JobExitCode=10030" | tee -a $RUNTIME_AREA/$repo \n'
231	+	txt += '    dumpStatus $RUNTIME_AREA/$repo \n'
232	+	txt += '    rm -f $RUNTIME_AREA/$repo \n'
233	+	txt += '    echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
234	+	txt += '    echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
235	+	txt += '    exit 1 \n'
236	+	txt += 'fi \n'
237	+
238	+	txt += '# report first time to DashBoard \n'
239	+	txt += 'dumpStatus $RUNTIME_AREA/$repo \n'
240	+	txt += 'rm -f $RUNTIME_AREA/$repo \n'
241	+	txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
242	+	txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
243	+
244	+	txt += '\n\n'
245	+
246		if int(self.copy_data) == 1:
247		if self.SE:
248		txt += 'export SE='+self.SE+'\n'
#	Line 136 | Line 251 | class SchedulerEdg(Scheduler):
251		if ( self.SE_PATH[-1] != '/' ) : self.SE_PATH = self.SE_PATH + '/'
252		txt += 'export SE_PATH='+self.SE_PATH+'\n'
253		txt += 'echo "SE_PATH = $SE_PATH"\n'
254	<
254	>
255	>	txt += 'export VO='+self.VO+'\n'
256	>	### FEDE: add some line for LFC catalog setting
257	>	txt += 'if [ $middleware == LCG ]; then \n'
258	>	txt += '    if [[ $LCG_CATALOG_TYPE != \''+self.lcg_catalog_type+'\' ]]; then\n'
259	>	txt += '        export LCG_CATALOG_TYPE='+self.lcg_catalog_type+'\n'
260	>	txt += '    fi\n'
261	>	txt += '    if [[ $LFC_HOST != \''+self.lfc_host+'\' ]]; then\n'
262	>	txt += '        export LFC_HOST='+self.lfc_host+'\n'
263	>	txt += '    fi\n'
264	>	txt += '    if [[ $LFC_HOME != \''+self.lfc_home+'\' ]]; then\n'
265	>	txt += '        export LFC_HOME='+self.lfc_home+'\n'
266	>	txt += '    fi\n'
267	>	txt += 'elif [ $middleware == OSG ]; then\n'
268	>	txt += '    echo "LFC catalog setting to be implemented for OSG"\n'
269	>	txt += 'fi\n'
270	>	#####
271		if int(self.register_data) == 1:
272	+	txt += 'if [ $middleware == LCG ]; then \n'
273	+	txt += '    export LFN='+self.LFN+'\n'
274	+	txt += '    lfc-ls $LFN\n'
275	+	txt += '    result=$?\n'
276	+	txt += '    echo $result\n'
277	+	### creation of LFN dir in LFC catalog, under /grid/cms dir
278	+	txt += '    if [ $result != 0 ]; then\n'
279	+	txt += '       lfc-mkdir $LFN\n'
280	+	txt += '       result=$?\n'
281	+	txt += '       echo $result\n'
282	+	txt += '    fi\n'
283	+	txt += 'elif [ $middleware == OSG ]; then\n'
284	+	txt += '    echo " Files registration to be implemented for OSG"\n'
285	+	txt += 'fi\n'
286	+	txt += '\n'
287	+
288		if self.VO:
289		txt += 'export VO='+self.VO+'\n'
290		if self.LFN:
291	<	txt += 'export LFN='+self.LFN+'\n'
291	>	txt += 'if [ $middleware == LCG ]; then \n'
292	>	txt += '    export LFN='+self.LFN+'\n'
293	>	txt += 'fi\n'
294		txt += '\n'
295	<	txt += 'CloseCEs=`edg-brokerinfo getCE`\n'
296	<	txt += 'echo "CloseCEs = $CloseCEs"\n'
297	<	txt += 'CE=`echo $CloseCEs | sed -e "s/:.*//"`\n'
298	<	txt += 'echo "CE = $CE"\n'
295	>
296	>	txt += 'if [ $middleware == LCG ]; then\n'
297	>	txt += '    CloseCEs=`edg-brokerinfo getCE`\n'
298	>	txt += '    echo "CloseCEs = $CloseCEs"\n'
299	>	txt += '    CE=`echo $CloseCEs | sed -e "s/:.*//"`\n'
300	>	txt += '    echo "CE = $CE"\n'
301	>	txt += 'elif [ $middleware == OSG ]; then \n'
302	>	txt += '    if [ $OSG_JOB_CONTACT ]; then \n'
303	>	txt += '        CE=`echo $OSG_JOB_CONTACT | /usr/bin/awk -F\/ \'{print ${args[0]}}\'` \n'
304	>	txt += '    else \n'
305	>	txt += '        echo "SET_CMS_ENV 10099 ==> OSG mode: ERROR in setting CE name from OSG_JOB_CONTACT" \n'
306	>	txt += '        echo "JOB_EXIT_STATUS = 10099" \n'
307	>	txt += '        echo "JobExitCode=10099" | tee -a $RUNTIME_AREA/$repo \n'
308	>	txt += '        dumpStatus $RUNTIME_AREA/$repo \n'
309	>	txt += '        rm -f $RUNTIME_AREA/$repo \n'
310	>	txt += '        echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
311	>	txt += '        echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
312	>	txt += '        exit 1 \n'
313	>	txt += '    fi \n'
314	>	txt += 'fi \n'
315	>
316	>	return txt
317	>
318	>	def wsCopyInput(self):
319	>	"""
320	>	Copy input data from SE to WN
321	>	"""
322	>	txt = ''
323	>	try:
324	>	self.copy_input_data = common.analisys_common_info['copy_input_data']
325	>	#print "self.copy_input_data = ", self.copy_input_data
326	>	except KeyError: self.copy_input_data = 0
327	>	if int(self.copy_input_data) == 1:
328	>	## OLI_Daniele deactivate for OSG (wait for LCG UI installed on OSG)
329	>	txt += 'if [ $middleware == OSG ]; then\n'
330	>	txt += '   #\n'
331	>	txt += '   #   Copy Input Data from SE to this WN deactivated in OSG mode\n'
332	>	txt += '   #\n'
333	>	txt += '   echo "Copy Input Data from SE to this WN deactivated in OSG mode"\n'
334	>	txt += 'elif [ $middleware == LCG ]; then \n'
335	>	txt += '   #\n'
336	>	txt += '   #   Copy Input Data from SE to this WN\n'
337	>	txt += '   #\n'
338	>	### changed by georgia (put a loop copying more than one input files per jobs)
339	>	txt += '   for input_file in $cur_file_list \n'
340	>	txt += '   do \n'
341	>	#### FEDE
342	>	#txt += '      echo "which lcg-cp" \n'
343	>	#txt += '      which lcg-cp \n'
344	>	#########
345	>	txt += '      lcg-cp --vo $VO --verbose -t 1200 lfn:$input_lfn/$input_file file:`pwd`/$input_file 2>&1\n'
346	>	txt += '      copy_input_exit_status=$?\n'
347	>	txt += '      echo "COPY_INPUT_EXIT_STATUS = $copy_input_exit_status"\n'
348	>	txt += '      if [ $copy_input_exit_status -ne 0 ]; then \n'
349	>	txt += '         echo "Problems with copying to WN" \n'
350	>	txt += '      else \n'
351	>	txt += '         echo "input copied into WN" \n'
352	>	txt += '      fi \n'
353	>	txt += '   done \n'
354	>	### copy a set of PU ntuples (same for each jobs -- but accessed randomly)
355	>	txt += '   for file in $cur_pu_list \n'
356	>	txt += '   do \n'
357	>	#### FEDE
358	>	#txt += '      echo "which lcg-cp" \n'
359	>	#txt += '      which lcg-cp \n'
360	>	#########
361	>	txt += '      lcg-cp --vo $VO --verbose -t 1200 lfn:$pu_lfn/$file file:`pwd`/$file 2>&1\n'
362	>	txt += '      copy_input_pu_exit_status=$?\n'
363	>	txt += '      echo "COPY_INPUT_PU_EXIT_STATUS = $copy_input_pu_exit_status"\n'
364	>	txt += '      if [ $copy_input_pu_exit_status -ne 0 ]; then \n'
365	>	txt += '         echo "Problems with copying pu to WN" \n'
366	>	txt += '      else \n'
367	>	txt += '         echo "input pu files copied into WN" \n'
368	>	txt += '      fi \n'
369	>	txt += '   done \n'
370	>	txt += '   \n'
371	>	txt += '   ### Check SCRATCH space available on WN : \n'
372	>	txt += '   df -h \n'
373	>	txt += 'fi \n'
374	>
375		return txt
376
377		def wsCopyOutput(self):
#	Line 156 | Line 381 | class SchedulerEdg(Scheduler):
381		"""
382		txt = ''
383		if int(self.copy_data) == 1:
159	–	copy = 'globus-url-copy file://`pwd`/$out_file gsiftp://${SE}${SE_PATH}$out_file'
384		txt += '#\n'
385		txt += '#   Copy output to SE = $SE\n'
386		txt += '#\n'
387	<	txt += 'if [ $exe_result -eq 0 ]; then\n'
388	<	txt += '  for out_file in $file_list ; do\n'
389	<	txt += '    echo "Trying to copy output file to $SE "\n'
390	<	txt += '    echo "'+copy+'"\n'
391	<	txt += '    '+copy+' 2>&1\n'
392	<	txt += '    copy_exit_status=$?\n'
393	<	txt += '    echo "COPY_EXIT_STATUS = $copy_exit_status"\n'
170	<	txt += '    echo "STAGE_OUT = $copy_exit_status"\n'
171	<	txt += '    if [ $copy_exit_status -ne 0 ]; then \n'
172	<	txt += '       echo "Problems with SE= $SE" \n'
173	<	txt += '    else \n'
174	<	txt += '       echo "output copied into $SE/$SE_PATH directory"\n'
387	>	#txt += 'if [ $exe_result -eq 0 ]; then\n'
388	>	txt += '    if [ $middleware == OSG ]; then\n'
389	>	txt += '        echo "X509_USER_PROXY = $X509_USER_PROXY"\n'
390	>	txt += '        echo "source $OSG_APP/glite/setup_glite_ui.sh"\n'
391	>	txt += '        source $OSG_APP/glite/setup_glite_ui.sh\n'
392	>	txt += '        export X509_CERT_DIR=$OSG_APP/glite/etc/grid-security/certificates\n'
393	>	txt += '        echo "export X509_CERT_DIR=$X509_CERT_DIR"\n'
394		txt += '    fi \n'
395	<	txt += '  done\n'
396	<	txt += 'fi \n'
395	>	txt += '    for out_file in $file_list ; do\n'
396	>	txt += '        echo "Trying to copy output file to $SE using lcg-cp"\n'
397	>	if common.logger.debugLevel() >= 5:
398	>	txt += '        echo "lcg-cp --vo $VO -t 2400 --verbose file://`pwd`/$out_file gsiftp://${SE}${SE_PATH}$out_file"\n'
399	>	txt += '        exitstring=`lcg-cp --vo $VO -t 2400 --verbose file://\`pwd\`/$out_file gsiftp://${SE}${SE_PATH}$out_file 2>&1`\n'
400	>	else:
401	>	txt += '        echo "lcg-cp --vo $VO -t 2400 file://`pwd`/$out_file gsiftp://${SE}${SE_PATH}$out_file"\n'
402	>	txt += '        exitstring=`lcg-cp --vo $VO -t 2400 file://\`pwd\`/$out_file gsiftp://${SE}${SE_PATH}$out_file 2>&1`\n'
403	>	txt += '        copy_exit_status=$?\n'
404	>	txt += '        echo "COPY_EXIT_STATUS for lcg-cp = $copy_exit_status"\n'
405	>	txt += '        echo "STAGE_OUT = $copy_exit_status"\n'
406	>	txt += '        if [ $copy_exit_status -ne 0 ]; then\n'
407	>	txt += '            echo "Possible problem with SE = $SE"\n'
408	>	txt += '            echo "StageOutExitStatus = 198" | tee -a $RUNTIME_AREA/$repo\n'
409	>	txt += '            echo "StageOutExitStatusReason = $exitstring" | tee -a $RUNTIME_AREA/$repo\n'
410	>	txt += '            echo "lcg-cp failed.  For verbose lcg-cp output, use command line option -debug 5."\n'
411	>	txt += '            echo "lcg-cp failed, attempting srmcp"\n'
412	>	txt += '            echo "mkdir -p $HOME/.srmconfig"\n'
413	>	txt += '            mkdir -p $HOME/.srmconfig\n'
414	>	txt += '            if [ $middleware == LCG ]; then\n'
415	>	txt += '               echo "srmcp -retry_num 5 -retry_timeout 480000 file:////`pwd`/$out_file srm://${SE}:8443${SE_PATH}$out_file"\n'
416	>	txt += '               exitstring=`srmcp -retry_num 5 -retry_timeout 480000 file:////\`pwd\`/$out_file srm://${SE}:8443${SE_PATH}$out_file 2>&1`\n'
417	>	txt += '            elif [ $middleware == OSG ]; then\n'
418	>	txt += '               echo "srmcp -retry_num 5 -retry_timeout 240000 -x509_user_trusted_certificates $OSG_APP/glite/etc/grid-security/certificates file:////`pwd`/$out_file srm://${SE}:8443${SE_PATH}$out_file"\n'
419	>	txt += '               exitstring=`srmcp -retry_num 5 -retry_timeout 240000 -x509_user_trusted_certificates $OSG_APP/glite/etc/grid-security/certificates file:////\`pwd\`/$out_file srm://${SE}:8443${SE_PATH}$out_file 2>&1`\n'
420	>	txt += '            fi \n'
421	>	txt += '            copy_exit_status=$?\n'
422	>	txt += '            echo "COPY_EXIT_STATUS for srm = $copy_exit_status"\n'
423	>	txt += '            echo "STAGE_OUT = $copy_exit_status"\n'
424	>	txt += '            if [ $copy_exit_status -ne 0 ]; then\n'
425	>	txt += '               echo "Problems with SE = $SE"\n'
426	>	txt += '               echo "StageOutExitStatus = 198" | tee -a $RUNTIME_AREA/$repo\n'
427	>	txt += '               echo "StageOutExitStatusReason = $exitstring" | tee -a $RUNTIME_AREA/$repo\n'
428	>	txt += '               echo "lcg-cp and srm failed"\n'
429	>	txt += '               echo "If storage_path in your config file contains a ? you may need a \? instead."\n'
430	>	txt += '            else\n'
431	>	txt += '               echo "StageOutSE = $SE" | tee -a $RUNTIME_AREA/$repo\n'
432	>	txt += '               echo "StageOutCatalog = " | tee -a $RUNTIME_AREA/$repo\n'
433	>	txt += '               echo "output copied into $SE/$SE_PATH directory"\n'
434	>	txt += '               echo "StageOutExitStatus = 0" | tee -a $RUNTIME_AREA/$repo\n'
435	>	txt += '               echo "srmcp succeeded"\n'
436	>	txt += '            fi\n'
437	>	txt += '        else\n'
438	>	txt += '            echo "StageOutSE = $SE" | tee -a $RUNTIME_AREA/$repo\n'
439	>	txt += '            echo "StageOutCatalog = " | tee -a $RUNTIME_AREA/$repo\n'
440	>	txt += '            echo "output copied into $SE/$SE_PATH directory"\n'
441	>	txt += '            echo "StageOutExitStatus = 0" | tee -a $RUNTIME_AREA/$repo\n'
442	>	txt += '            echo "lcg-cp succeeded"\n'
443	>	txt += '         fi\n'
444	>	txt += '     done\n'
445	>	#txt += 'fi\n'
446		return txt
447
448		def wsRegisterOutput(self):
#	Line 184 | Line 452 | class SchedulerEdg(Scheduler):
452
453		txt = ''
454		if int(self.register_data) == 1:
455	+	## OLI_Daniele deactivate for OSG (wait for LCG UI installed on OSG)
456	+	txt += 'if [ $middleware == OSG ]; then\n'
457	+	txt += '   #\n'
458	+	txt += '   #   Register output to LFC deactivated in OSG mode\n'
459	+	txt += '   #\n'
460	+	txt += '   echo "Register output to LFC deactivated in OSG mode"\n'
461	+	txt += 'elif [ $middleware == LCG ]; then \n'
462		txt += '#\n'
463	<	txt += '#  Register output to RLS\n'
463	>	txt += '#  Register output to LFC\n'
464		txt += '#\n'
465	<	txt += 'if [[ $exe_result -eq 0 && $copy_exit_status -eq 0 ]]; then\n'
466	<	txt += '   for out_file in $file_list ; do\n'
467	<	txt += '      echo "Trying to register the output file into RLS"\n'
468	<	txt += '      echo "lcg-rf -l $LFN/$out_file --vo $VO sfn://$SE$SE_PATH/$out_file"\n'
469	<	txt += '      lcg-rf -l $LFN/$out_file --vo $VO sfn://$SE$SE_PATH/$out_file 2>&1 \n'
470	<	txt += '      register_exit_status=$?\n'
471	<	txt += '      echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
472	<	txt += '      echo "STAGE_OUT = $register_exit_status"\n'
473	<	txt += '      if [ $register_exit_status -ne 0 ]; then \n'
474	<	txt += '         echo "Problems with the registration to RLS" \n'
200	<	txt += '         echo "Try with srm protocol" \n'
201	<	txt += '         echo "lcg-rf -l $LFN/$out_file --vo $VO srm://$SE$SE_PATH/$out_file"\n'
202	<	txt += '         lcg-rf -l $LFN/$out_file --vo $VO srm://$SE$SE_PATH/$out_file 2>&1 \n'
465	>	#txt += '   if [[ $exe_result -eq 0 && $copy_exit_status -eq 0 ]]; then\n'
466	>	txt += '   if [ $copy_exit_status -eq 0 ]; then\n'
467	>	txt += '      for out_file in $file_list ; do\n'
468	>	txt += '         echo "Trying to register the output file into LFC"\n'
469	>	#### FEDE
470	>	#txt += '         echo "which lcg-rf" \n'
471	>	#txt += '         which lcg-rf \n'
472	>	#########
473	>	txt += '         echo "lcg-rf -l $LFN/$out_file --vo $VO -t 1200 sfn://$SE$SE_PATH/$out_file 2>&1"\n'
474	>	txt += '         lcg-rf -l $LFN/$out_file --vo $VO -t 1200 sfn://$SE$SE_PATH/$out_file 2>&1 \n'
475		txt += '         register_exit_status=$?\n'
476		txt += '         echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
477		txt += '         echo "STAGE_OUT = $register_exit_status"\n'
478		txt += '         if [ $register_exit_status -ne 0 ]; then \n'
479	<	txt += '            echo "Problems with the registration into RLS" \n'
479	>	txt += '            echo "Problems with the registration to LFC" \n'
480	>	txt += '            echo "Try with srm protocol" \n'
481	>	#### FEDE
482	>	#txt += '            echo "which lcg-rf" \n'
483	>	#txt += '            which lcg-rf \n'
484	>	#########
485	>	txt += '            echo "lcg-rf -l $LFN/$out_file --vo $VO -t 1200 srm://$SE$SE_PATH/$out_file 2>&1"\n'
486	>	txt += '            lcg-rf -l $LFN/$out_file --vo $VO -t 1200 srm://$SE$SE_PATH/$out_file 2>&1 \n'
487	>	txt += '            register_exit_status=$?\n'
488	>	txt += '            echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
489	>	txt += '            echo "STAGE_OUT = $register_exit_status"\n'
490	>	txt += '            if [ $register_exit_status -ne 0 ]; then \n'
491	>	txt += '               echo "Problems with the registration into LFC" \n'
492	>	txt += '            fi \n'
493	>	txt += '         else \n'
494	>	txt += '            echo "output registered to LFC"\n'
495		txt += '         fi \n'
496	<	txt += '      else \n'
497	<	txt += '         echo "output registered to RLS"\n'
498	<	txt += '      fi \n'
499	<	txt += '   done\n'
500	<	txt += 'elif [[ $exe_result -eq 0 && $copy_exit_status -ne 0 ]]; then \n'
501	<	txt += '   echo "Trying to copy output file to CloseSE"\n'
502	<	txt += '   CLOSE_SE=`edg-brokerinfo getCloseSEs | head -1`\n'
503	<	txt += '   for out_file in $file_list ; do\n'
504	<	txt += '      echo "lcg-cr -v -l lfn:${LFN}/$out_file -d $CLOSE_SE -P $LFN/$out_file --vo $VO file://`pwd`/$out_file" \n'
505	<	txt += '      lcg-cr -v -l lfn:${LFN}/$out_file -d $CLOSE_SE -P $LFN/$out_file --vo $VO file://`pwd`/$out_file 2>&1 \n'
506	<	txt += '      register_exit_status=$?\n'
507	<	txt += '      echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
508	<	txt += '      echo "STAGE_OUT = $register_exit_status"\n'
509	<	txt += '      if [ $register_exit_status -ne 0 ]; then \n'
510	<	txt += '         echo "Problems with CloseSE" \n'
511	<	txt += '      else \n'
512	<	txt += '         echo "The program was successfully executed"\n'
513	<	txt += '         echo "SE = $CLOSE_SE"\n'
514	<	txt += '         echo "LFN for the file is LFN=${LFN}/$out_file"\n'
515	<	txt += '      fi \n'
516	<	txt += '   done\n'
517	<	txt += 'else\n'
518	<	txt += '   echo "Problem with the executable"\n'
496	>	txt += '         echo "StageOutExitStatus = $register_exit_status" | tee -a $RUNTIME_AREA/$repo\n'
497	>	txt += '      done\n'
498	>	#txt += '   elif [[ $exe_result -eq 0 && $copy_exit_status -ne 0 ]]; then \n'
499	>	txt += '   else \n'
500	>	txt += '      echo "Trying to copy output file to CloseSE"\n'
501	>	txt += '      CLOSE_SE=`edg-brokerinfo getCloseSEs | head -1`\n'
502	>	txt += '      for out_file in $file_list ; do\n'
503	>	#### FEDE
504	>	#txt += '         echo "which lcg-cr" \n'
505	>	#txt += '         which lcg-cr \n'
506	>	#########
507	>	txt += '         echo "lcg-cr -v -l lfn:${LFN}/$out_file -d $CLOSE_SE -P $LFN/$out_file --vo $VO file://$RUNTIME_AREA/$out_file 2>&1" \n'
508	>	txt += '         lcg-cr -v -l lfn:${LFN}/$out_file -d $CLOSE_SE -P $LFN/$out_file --vo $VO file://$RUNTIME_AREA/$out_file 2>&1 \n'
509	>	txt += '         register_exit_status=$?\n'
510	>	txt += '         echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
511	>	txt += '         echo "STAGE_OUT = $register_exit_status"\n'
512	>	txt += '         if [ $register_exit_status -ne 0 ]; then \n'
513	>	txt += '            echo "Problems with CloseSE or Catalog" \n'
514	>	txt += '         else \n'
515	>	txt += '            echo "The program was successfully executed"\n'
516	>	txt += '            echo "SE = $CLOSE_SE"\n'
517	>	txt += '            echo "LFN for the file is LFN=${LFN}/$out_file"\n'
518	>	txt += '         fi \n'
519	>	txt += '         echo "StageOutExitStatus = $register_exit_status" | tee -a $RUNTIME_AREA/$repo\n'
520	>	txt += '      done\n'
521	>	#txt += '   else\n'
522	>	#txt += '      echo "Problem with the executable"\n'
523	>	txt += '   fi \n'
524	>	txt += '   exit_status=$register_exit_status\n'
525		txt += 'fi \n'
526		return txt
527
#	Line 237 | Line 530 | class SchedulerEdg(Scheduler):
530		retrieve the logging info from logging and bookkeeping and return it
531		"""
532		self.checkProxy()
240	–	#  id = common.jobDB.jobId(nj)
533		cmd = 'edg-job-get-logging-info -v 2 ' + id
534	<	cmd_out = os.popen(cmd)
535	<	#  cmd_out = runCommand(cmd)
534	>	#cmd_out = os.popen(cmd)
535	>	cmd_out = runCommand(cmd)
536		return cmd_out
537
538		def listMatch(self, nj):
#	Line 250 | Line 542 | class SchedulerEdg(Scheduler):
542		self.checkProxy()
543		jdl = common.job_list[nj].jdlFilename()
544		cmd = 'edg-job-list-match ' + self.configOpt_() + jdl
545	<	# myCmd = os.popen(cmd)
546	<	# cmd_out = myCmd.readlines()
547	<	# myCmd.close()
548	<	cmd_out = runCommand(cmd,0,240)
545	>	cmd_out = runCommand(cmd,0,45)
546	>	if not cmd_out:
547	>	raise CrabException("ERROR: "+cmd+" failed!")
548	>
549		return self.parseListMatch_(cmd_out, jdl)
550
551		def parseListMatch_(self, out, jdl):
#	Line 311 | Line 603 | class SchedulerEdg(Scheduler):
603		[sites.append(it) for it in CEs if not sites.count(it)]
604
605		common.logger.debug(5,"All Sites :"+str(sites))
606	+	common.logger.message("Matched Sites :"+str(sites))
607		return len(sites)
608
609		def noMatchFound_(self, jdl):
#	Line 349 | Line 642 | class SchedulerEdg(Scheduler):
642		pass
643		return jid
644
645	+	def resubmit(self, nj_list):
646	+	"""
647	+	Prepare jobs to be submit
648	+	"""
649	+	return
650	+
651		def getExitStatus(self, id):
652		return self.getStatusAttribute_(id, 'exit_code')
653
#	Line 389 | Line 688 | class SchedulerEdg(Scheduler):
688		cmd_out = runCommand(cmd)
689		return cmd_out
690
691	<	def getOutput(self, id):
691	>	##### FEDE ######
692	>	def findSites_(self, n_tot_job):
693	>	itr4 = []
694	>	# print "n_tot_job = ", n_tot_job
695	>	for n in range(n_tot_job):
696	>	sites = common.jobDB.destination(n)
697	>	if len(sites)>0 and sites[0]=="Any": continue
698	>
699	>	#job = common.job_list[n]
700	>	#jbt = job.type()
701	>	# print "common.jobDB.destination(n) = ", common.jobDB.destination(n)
702	>	# print "sites = ", sites
703	>	itr = ''
704	>	for site in sites:
705	>	#itr = itr + 'target.GlueSEUniqueID==&quot;'+site+'&quot; || '
706	>	itr = itr + 'target.GlueSEUniqueID=="'+site+'" || '
707	>	pass
708	>	# remove last ||
709	>	itr = itr[0:-4]
710	>	itr4.append( itr )
711	>	# remove last ,
712	>	# print "itr4 = ", itr4
713	>	return itr4
714	>
715	>	def createXMLSchScript(self, nj, argsList):
716	>	# def createXMLSchScript(self, nj):
717	>	"""
718	>	Create a XML-file for BOSS4.
719	>	"""
720	>	#      job = common.job_list[nj]
721	>	"""
722	>	INDY
723	>	[begin] da rivedere:
724	>	in particolare passerei il jobType ed eliminerei le dipendenze da job
725	>	"""
726	>	index = nj - 1
727	>	job = common.job_list[index]
728	>	jbt = job.type()
729	>
730	>	inp_sandbox = jbt.inputSandbox(index)
731	>	out_sandbox = jbt.outputSandbox(index)
732		"""
733		Get output for a finished job with id.
734		Returns the name of directory with results.
#	Line 401 | Line 740 | class SchedulerEdg(Scheduler):
740
741		# Determine the output directory name
742		dir = common.work_space.resDir()
743	<	dir += os.getlogin()
743	>	dir += os.environ['USER']
744		dir += '_' + os.path.basename(id)
745		return dir
746
408	–	def cancel(self, id):
409	–	""" Cancel the EDG job with id """
410	–	self.checkProxy()
411	–	cmd = 'edg-job-cancel --noint ' + id
412	–	cmd_out = runCommand(cmd)
413	–	return cmd_out
414	–
747		def createSchScript(self, nj):
748		"""
749		Create a JDL-file for EDG.
#	Line 441 | Line 773 | class SchedulerEdg(Scheduler):
773		jdl.write(jt_string)
774
775		### only one .sh  JDL has arguments:
444	–	firstEvent = common.jobDB.firstEvent(nj)
445	–	maxEvents = common.jobDB.maxEvents(nj)
446	–	jdl.write('Arguments = "' + str(nj+1)+' '+str(firstEvent)+' '+str(maxEvents)+'";\n')
776
777	+	### Fabio
778	+	jdl.write('Arguments = "' + str(nj+1)+' '+ jbt.getJobTypeArguments(nj, "EDG") +'";\n')
779		inp_box = 'InputSandbox = { '
780		inp_box = inp_box + '"' + script + '",'
781
#	Line 457 | Line 788 | class SchedulerEdg(Scheduler):
788		#if common.use_jam:
789		#   inp_box = inp_box+' "'+common.bin_dir+'/'+common.run_jam+'",'
790
791	<	for addFile in jbt.additional_inbox_files:
792	<	addFile = os.path.abspath(addFile)
793	<	inp_box = inp_box+' "'+addFile+'",'
794	<	pass
791	>	# Marco (VERY TEMPORARY ML STUFF)
792	>	inp_box = inp_box+' "' + os.path.abspath(os.environ['CRABDIR']+'/python/'+'report.py') + '", "' +\
793	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'DashboardAPI.py') + '", "'+\
794	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'Logger.py') + '", "'+\
795	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'ProcInfo.py') + '", "'+\
796	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'apmon.py') + '"'
797	>	# End Marco
798	>
799	>	if (not jbt.additional_inbox_files == []):
800	>	inp_box = inp_box + ', '
801	>	for addFile in jbt.additional_inbox_files:
802	>	addFile = os.path.abspath(addFile)
803	>	inp_box = inp_box+' "'+addFile+'",'
804	>	pass
805
806		if inp_box[-1] == ',' : inp_box = inp_box[:-1]
807		inp_box = inp_box + ' };\n'
#	Line 490 | Line 831 | class SchedulerEdg(Scheduler):
831		out_box = out_box + ' };'
832		jdl.write(out_box+'\n')
833
834	<	### if at least a CE exists ...
835	<	if common.analisys_common_info['sites']:
836	<	if common.analisys_common_info['sw_version']:
837	<	req='Requirements = '
838	<	req=req + 'Member("VO-cms-' + \
839	<	common.analisys_common_info['sw_version'] + \
840	<	'", other.GlueHostApplicationSoftwareRunTimeEnvironment)'
841	<	if len(common.analisys_common_info['sites'])>0:
842	<	req = req + ' && ('
843	<	for i in range(len(common.analisys_common_info['sites'])):
844	<	req = req + 'other.GlueCEInfoHostName == "' \
845	<	+ common.analisys_common_info['sites'][i] + '"'
846	<	if ( i < (int(len(common.analisys_common_info['sites']) - 1)) ):
847	<	req = req + ' || '
834	>
835	>	req='Requirements = '
836	>	noreq=req
837	>	req = req + jbt.getRequirements(nj)
838	>	#### and USER REQUIREMENT
839	>	if self.EDG_requirements:
840	>	if (req != noreq):
841	>	req = req +  ' && '
842	>	req = req + self.EDG_requirements
843	>	#### FEDE #####
844	>	if self.EDG_ce_white_list:
845	>	ce_white_list = string.split(self.EDG_ce_white_list,',')
846	>	#print "req = ", req
847	>	for i in range(len(ce_white_list)):
848	>	if i == 0:
849	>	if (req != noreq):
850	>	req = req +  ' && '
851	>	req = req + '((RegExp("' + ce_white_list[i] + '", other.GlueCEUniqueId))'
852	>	pass
853	>	else:
854	>	req = req +  ' || (RegExp("' + ce_white_list[i] + '", other.GlueCEUniqueId))'
855		req = req + ')'
856	+
857	+	if self.EDG_ce_black_list:
858	+	ce_black_list = string.split(self.EDG_ce_black_list,',')
859	+	for ce in ce_black_list:
860	+	if (req != noreq):
861	+	req = req +  ' && '
862	+	req = req + '(!RegExp("' + ce + '", other.GlueCEUniqueId))'
863	+	pass
864	+
865	+	###############
866	+	clockTime=480
867	+	if self.EDG_clock_time:
868	+	clockTime= self.EDG_clock_time
869	+	if (req != noreq):
870	+	req = req + ' && '
871	+	req = req + '((other.GlueCEPolicyMaxWallClockTime == 0) || (other.GlueCEPolicyMaxWallClockTime>='+str(clockTime)+'))'
872	+
873	+	cpuTime=1000
874	+	if self.EDG_cpu_time:
875	+	cpuTime=self.EDG_cpu_time
876	+	if (req != noreq):
877	+	req = req + ' && '
878	+	req = req + '((other.GlueCEPolicyMaxCPUTime == 0) || (other.GlueCEPolicyMaxCPUTime>='+str(cpuTime)+'))'
879
880	<	#### and USER REQUIREMENT
510	<	if self.EDG_requirements:
511	<	req = req +  ' && ' + self.EDG_requirements
512	<	if self.EDG_clock_time:
513	<	req = req + ' && other.GlueCEPolicyMaxWallClockTime>='+self.EDG_clock_time
514	<	if self.EDG_cpu_time:
515	<	req = req + ' && other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time
880	>	if (req != noreq):
881		req = req + ';\n'
882		jdl.write(req)
883
#	Line 522 | Line 887 | class SchedulerEdg(Scheduler):
887		jdl.write('RetryCount = '+self.EDG_retry_count+';\n')
888		pass
889
890	+	jdl.write('MyProxyServer = "' + self.proxyServer + '";\n')
891	+
892		jdl.close()
893		return
894
#	Line 531 | Line 898 | class SchedulerEdg(Scheduler):
898		"""
899		if (self.proxyValid): return
900		timeleft = -999
901	<	minTimeLeft=10 # in hours
902	<	cmd = 'grid-proxy-info -e -v '+str(minTimeLeft)+':00'
903	<	try: cmd_out = runCommand(cmd,0)
904	<	except: print cmd_out
905	<	if (cmd_out == None or cmd_out=='1'):
906	<	common.logger.message( "No valid proxy found or timeleft too short!\n Creating a user proxy with default length of 100h\n")
907	<	cmd = 'grid-proxy-init -valid 100:00'
901	>	minTimeLeft=10*3600 # in seconds
902	>
903	>	minTimeLeftServer = 100 # in hours
904	>
905	>	#cmd = 'voms-proxy-info -exists -valid '+str(minTimeLeft)+':00'
906	>	#cmd = 'voms-proxy-info -timeleft'
907	>	mustRenew = 0
908	>	timeLeftLocal = runCommand('voms-proxy-info -timeleft 2>/dev/null')
909	>	timeLeftServer = -999
910	>	if not timeLeftLocal or int(timeLeftLocal) <= 0 or not isInt(timeLeftLocal):
911	>	mustRenew = 1
912	>	else:
913	>	timeLeftServer = runCommand('voms-proxy-info -actimeleft 2>/dev/null | head -1')
914	>	if not timeLeftServer or not isInt(timeLeftServer):
915	>	mustRenew = 1
916	>	elif timeLeftLocal<minTimeLeft or timeLeftServer<minTimeLeft:
917	>	mustRenew = 1
918	>	pass
919	>	pass
920	>
921	>	if mustRenew:
922	>	common.logger.message( "No valid proxy found or remaining time of validity of already existing proxy shorter than 10 hours!\n Creating a user proxy with default length of 96h\n")
923	>	cmd = 'voms-proxy-init -voms '+self.VO+' -valid 96:00'
924	>	if self.role:
925	>	cmd = 'voms-proxy-init -voms '+self.VO+':/'+self.VO+'/role='+self.role+' -valid 96:00'
926		try:
927	+	# SL as above: damn it!
928		out = os.system(cmd)
929		if (out>0): raise CrabException("Unable to create a valid proxy!\n")
930		except:
931		msg = "Unable to create a valid proxy!\n"
932		raise CrabException(msg)
547	–	cmd = 'grid-proxy-info -timeleft'
548	–	cmd_out = runCommand(cmd,0)
933		pass
934	+
935	+	## now I do have a voms proxy valid, and I check the myproxy server
936	+	renewProxy = 0
937	+	cmd = 'myproxy-info -d -s '+self.proxyServer
938	+	cmd_out = runCommand(cmd,0,20)
939	+	if not cmd_out:
940	+	common.logger.message('No credential delegated to myproxy server '+self.proxyServer+' will do now')
941	+	renewProxy = 1
942	+	else:
943	+	# if myproxy exist but not long enough, renew
944	+	reTime = re.compile( r'timeleft: (\d+)' )
945	+	#print "<"+str(reTime.search( cmd_out ).group(1))+">"
946	+	if reTime.match( cmd_out ):
947	+	time = reTime.search( line ).group(1)
948	+	if time < minTimeLeftServer:
949	+	renewProxy = 1
950	+	common.logger.message('No credential delegation will expire in '+time+' hours: renew it')
951	+	pass
952	+	pass
953	+
954	+	# if not, create one.
955	+	if renewProxy:
956	+	cmd = 'myproxy-init -d -n -s '+self.proxyServer
957	+	out = os.system(cmd)
958	+	if (out>0):
959	+	raise CrabException("Unable to delegate the proxy to myproxyserver "+self.proxyServer+" !\n")
960	+	pass
961	+
962	+	# cache proxy validity
963		self.proxyValid=1
964		return
965	<
965	>
966		def configOpt_(self):
967		edg_ui_cfg_opt = ' '
968		if self.edg_config:
969	<	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
969	>	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
970		if self.edg_config_vo:
971	<	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
971	>	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
972		return edg_ui_cfg_opt

Diff Legend

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