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 1.12 by fanzago, Mon Sep 26 08:59:06 2005 UTC vs.
Revision 1.64 by gutsche, Tue Jun 27 02:31:31 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, tempfile
8	>	import os, sys, time
9
10		class SchedulerEdg(Scheduler):
11		def __init__(self):
#	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: self.LCG_version = cfg_params["EDG.lcg_version"]
41		except KeyError: self.LCG_version = '2'
#	Line 36 | Line 46 | class SchedulerEdg(Scheduler):
46		try: self.EDG_retry_count = cfg_params['EDG.retry_count']
47		except KeyError: self.EDG_retry_count = ''
48
49	+	try:
50	+	self.EDG_ce_black_list = cfg_params['EDG.ce_black_list']
51	+	#print "self.EDG_ce_black_list = ", self.EDG_ce_black_list
52	+	except KeyError:
53	+	self.EDG_ce_black_list  = ''
54	+
55	+	try:
56	+	self.EDG_ce_white_list = cfg_params['EDG.ce_white_list']
57	+	#print "self.EDG_ce_white_list = ", self.EDG_ce_white_list
58	+	except KeyError: self.EDG_ce_white_list = ''
59	+
60	+	try: self.VO = cfg_params['EDG.virtual_organization']
61	+	except KeyError: self.VO = 'cms'
62	+
63	+	try: self.return_data = cfg_params['USER.return_data']
64	+	except KeyError: self.return_data = 1
65	+
66	+	try:
67	+	self.copy_input_data = common.analisys_common_info['copy_input_data']
68	+	#print "self.copy_input_data = ", self.copy_input_data
69	+	except KeyError: self.copy_input_data = 0
70	+
71	+	try:
72	+	self.copy_data = cfg_params["USER.copy_data"]
73	+	if int(self.copy_data) == 1:
74	+	try:
75	+	self.SE = cfg_params['USER.storage_element']
76	+	self.SE_PATH = cfg_params['USER.storage_path']
77	+	except KeyError:
78	+	msg = "Error. The [USER] section does not have 'storage_element'"
79	+	msg = msg + " and/or 'storage_path' entries, necessary to copy the output"
80	+	common.logger.message(msg)
81	+	raise CrabException(msg)
82	+	except KeyError: self.copy_data = 0
83	+
84	+	if ( int(self.return_data) == 0 and int(self.copy_data) == 0 ):
85	+	msg = 'Warning: return_data = 0 and copy_data = 0 ==> your exe output will be lost\n'
86	+	msg = msg + 'Please modify return_data and copy_data value in your crab.cfg file\n'
87	+	raise CrabException(msg)
88	+
89	+	try:
90	+	self.lfc_host = cfg_params['EDG.lfc_host']
91	+	except KeyError:
92	+	msg = "Error. The [EDG] section does not have 'lfc_host' value"
93	+	msg = msg + " it's necessary to know the LFC host name"
94	+	common.logger.message(msg)
95	+	raise CrabException(msg)
96	+	try:
97	+	self.lcg_catalog_type = cfg_params['EDG.lcg_catalog_type']
98	+	except KeyError:
99	+	msg = "Error. The [EDG] section does not have 'lcg_catalog_type' value"
100	+	msg = msg + " it's necessary to know the catalog type"
101	+	common.logger.message(msg)
102	+	raise CrabException(msg)
103		try:
104	<	self.VO = cfg_params['EDG.virtual_organization']
104	>	self.lfc_home = cfg_params['EDG.lfc_home']
105		except KeyError:
106	<	self.VO = 'cms'
106	>	msg = "Error. The [EDG] section does not have 'lfc_home' value"
107	>	msg = msg + " it's necessary to know the home catalog dir"
108	>	common.logger.message(msg)
109	>	raise CrabException(msg)
110	>
111	>	try:
112	>	self.register_data = cfg_params["USER.register_data"]
113	>	if int(self.register_data) == 1:
114	>	try:
115	>	self.LFN = cfg_params['USER.lfn_dir']
116	>	except KeyError:
117	>	msg = "Error. The [USER] section does not have 'lfn_dir' value"
118	>	msg = msg + " it's necessary for LCF registration"
119	>	common.logger.message(msg)
120	>	raise CrabException(msg)
121	>	except KeyError: self.register_data = 0
122	>
123	>	if ( int(self.copy_data) == 0 and int(self.register_data) == 1 ):
124	>	msg = 'Warning: register_data = 1 must be used with copy_data = 1\n'
125	>	msg = msg + 'Please modify copy_data value in your crab.cfg file\n'
126	>	common.logger.message(msg)
127	>	raise CrabException(msg)
128	>
129	>	try: self.EDG_requirements = cfg_params['EDG.requirements']
130	>	except KeyError: self.EDG_requirements = ''
131	>
132	>	try: self.EDG_retry_count = cfg_params['EDG.retry_count']
133	>	except KeyError: self.EDG_retry_count = ''
134	>
135	>	try: self.EDG_clock_time = cfg_params['EDG.max_wall_clock_time']
136	>	except KeyError: self.EDG_clock_time= ''
137	>
138	>	try: self.EDG_cpu_time = cfg_params['EDG.max_cpu_time']
139	>	except KeyError: self.EDG_cpu_time = ''
140
141		# Add EDG_WL_LOCATION to the python path
142
143		try:
47	–	#            path = os.environ['GLITE_WMS_LOCATION']
144		path = os.environ['EDG_WL_LOCATION']
145		except:
50	–	#            msg = "Error: the GLITE_WMS_LOCATION variable is not set."
146		msg = "Error: the EDG_WL_LOCATION variable is not set."
147		raise CrabException(msg)
148
#	Line 56 | Line 151 | class SchedulerEdg(Scheduler):
151		libPath=os.path.join(path, "lib", "python")
152		sys.path.append(libPath)
153
154	<	self.checkProxy_()
154	>	self.proxyValid=0
155	>
156	>	try:
157	>	self._taskId = cfg_params['taskId']
158	>	except:
159	>	self._taskId = ''
160	>
161		return
162
163
#	Line 74 | Line 175 | class SchedulerEdg(Scheduler):
175		return 1
176		else:
177		return 0
178	+
179		def wsSetupEnvironment(self):
180		"""
181		Returns part of a job script which does scheduler-specific work.
182		"""
183	<	txt = '\n'
184	<	txt += 'CloseCEs=`edg-brokerinfo getCE`\n'
185	<	# MARCO
186	<	#txt += 'CloseCEs=`glite-brokerinfo getCE`\n'
187	<	# MARCO
188	<	txt += 'echo "CloseCEs = $CloseCEs"\n'
189	<	txt += 'CE=`echo $CloseCEs | sed -e "s/:.*//"`\n'
190	<	txt += 'echo "CE = $CE"\n'
183	>	txt = ''
184	>	txt += "# job number (first parameter for job wrapper)\n"
185	>	txt += "NJob=$1\n"
186	>
187	>	txt += '# job identification to DashBoard \n'
188	>	txt += 'MonitorJobID=`echo ${NJob}_$EDG_WL_JOBID`\n'
189	>	txt += 'SyncGridJobId=`echo $EDG_WL_JOBID`\n'
190	>	txt += 'MonitorID=`echo ' + self._taskId + '`\n'
191	>	txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
192	>	txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
193	>	txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
194	>
195	>	txt += 'echo "middleware discovery " \n'
196	>	txt += 'if [ $VO_CMS_SW_DIR ]; then \n'
197	>	txt += '    middleware=LCG \n'
198	>	txt += '    echo "SyncCE=`edg-brokerinfo getCE`" | tee -a $RUNTIME_AREA/$repo \n'
199	>	txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
200	>	txt += '    echo "middleware =$middleware" \n'
201	>	txt += 'elif [ $GRID3_APP_DIR ]; then\n'
202	>	txt += '    middleware=OSG \n'
203	>	txt += '    echo "SyncCE=`echo $EDG_WL_LOG_DESTINATION`" | tee -a $RUNTIME_AREA/$repo \n'
204	>	txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
205	>	txt += '    echo "middleware =$middleware" \n'
206	>	txt += 'elif [ $OSG_APP ]; then \n'
207	>	txt += '    middleware=OSG \n'
208	>	txt += '    echo "SyncCE=`echo $EDG_WL_LOG_DESTINATION`" | tee -a $RUNTIME_AREA/$repo \n'
209	>	txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
210	>	txt += '    echo "middleware =$middleware" \n'
211	>	txt += 'else \n'
212	>	txt += '    echo "SET_CMS_ENV 10030 ==> middleware not identified" \n'
213	>	txt += '    echo "JOB_EXIT_STATUS = 10030" \n'
214	>	txt += '    echo "JobExitCode=10030" | tee -a $RUNTIME_AREA/$repo \n'
215	>	txt += '    dumpStatus $RUNTIME_AREA/$repo \n'
216	>	txt += '    rm -f $RUNTIME_AREA/$repo \n'
217	>	txt += '    echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
218	>	txt += '    echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
219	>	txt += '    echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
220	>	txt += '    exit 1 \n'
221	>	txt += 'fi \n'
222	>
223	>	txt += '# report first time to DashBoard \n'
224	>	txt += 'dumpStatus $RUNTIME_AREA/$repo \n'
225	>	txt += 'rm -f $RUNTIME_AREA/$repo \n'
226	>	txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
227	>	txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
228	>	txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
229	>
230	>	txt += '\n\n'
231	>
232	>	if int(self.copy_data) == 1:
233	>	if self.SE:
234	>	txt += 'export SE='+self.SE+'\n'
235	>	txt += 'echo "SE = $SE"\n'
236	>	if self.SE_PATH:
237	>	if ( self.SE_PATH[-1] != '/' ) : self.SE_PATH = self.SE_PATH + '/'
238	>	txt += 'export SE_PATH='+self.SE_PATH+'\n'
239	>	txt += 'echo "SE_PATH = $SE_PATH"\n'
240	>
241	>	txt += 'export VO='+self.VO+'\n'
242	>	### FEDE: add some line for LFC catalog setting
243	>	txt += 'if [ $middleware == LCG ]; then \n'
244	>	txt += '    if [[ $LCG_CATALOG_TYPE != \''+self.lcg_catalog_type+'\' ]]; then\n'
245	>	txt += '        export LCG_CATALOG_TYPE='+self.lcg_catalog_type+'\n'
246	>	txt += '    fi\n'
247	>	txt += '    if [[ $LFC_HOST != \''+self.lfc_host+'\' ]]; then\n'
248	>	txt += '        export LFC_HOST='+self.lfc_host+'\n'
249	>	txt += '    fi\n'
250	>	txt += '    if [[ $LFC_HOME != \''+self.lfc_home+'\' ]]; then\n'
251	>	txt += '        export LFC_HOME='+self.lfc_home+'\n'
252	>	txt += '    fi\n'
253	>	txt += 'elif [ $middleware == OSG ]; then\n'
254	>	txt += '    echo "LFC catalog setting to be implemented for OSG"\n'
255	>	txt += 'fi\n'
256	>	#####
257	>	if int(self.register_data) == 1:
258	>	txt += 'if [ $middleware == LCG ]; then \n'
259	>	txt += '    export LFN='+self.LFN+'\n'
260	>	txt += '    lfc-ls $LFN\n'
261	>	txt += '    result=$?\n'
262	>	txt += '    echo $result\n'
263	>	### creation of LFN dir in LFC catalog, under /grid/cms dir
264	>	txt += '    if [ $result != 0 ]; then\n'
265	>	txt += '       lfc-mkdir $LFN\n'
266	>	txt += '       result=$?\n'
267	>	txt += '       echo $result\n'
268	>	txt += '    fi\n'
269	>	txt += 'elif [ $middleware == OSG ]; then\n'
270	>	txt += '    echo " Files registration to be implemented for OSG"\n'
271	>	txt += 'fi\n'
272	>	txt += '\n'
273	>
274	>	if self.VO:
275	>	txt += 'export VO='+self.VO+'\n'
276	>	if self.LFN:
277	>	txt += 'if [ $middleware == LCG ]; then \n'
278	>	txt += '    export LFN='+self.LFN+'\n'
279	>	txt += 'fi\n'
280	>	txt += '\n'
281	>
282	>	txt += 'if [ $middleware == LCG ]; then\n'
283	>	txt += '    CloseCEs=`edg-brokerinfo getCE`\n'
284	>	txt += '    echo "CloseCEs = $CloseCEs"\n'
285	>	txt += '    CE=`echo $CloseCEs | sed -e "s/:.*//"`\n'
286	>	txt += '    echo "CE = $CE"\n'
287	>	txt += 'elif [ $middleware == OSG ]; then \n'
288	>	txt += '    if [ $OSG_JOB_CONTACT ]; then \n'
289	>	txt += '        CE=`echo $OSG_JOB_CONTACT | /usr/bin/awk -F\/ \'{print $1}\'` \n'
290	>	txt += '    else \n'
291	>	txt += '        echo "SET_CMS_ENV 10099 ==> OSG mode: ERROR in setting CE name from OSG_JOB_CONTACT" \n'
292	>	txt += '        echo "JOB_EXIT_STATUS = 10099" \n'
293	>	txt += '        echo "JobExitCode=10099" | tee -a $RUNTIME_AREA/$repo \n'
294	>	txt += '        dumpStatus $RUNTIME_AREA/$repo \n'
295	>	txt += '        rm -f $RUNTIME_AREA/$repo \n'
296	>	txt += '        echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
297	>	txt += '        echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
298	>	txt += '        echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
299	>	txt += '        exit 1 \n'
300	>	txt += '    fi \n'
301	>	txt += 'fi \n'
302	>
303	>	return txt
304	>
305	>	def wsCopyInput(self):
306	>	"""
307	>	Copy input data from SE to WN
308	>	"""
309	>	txt = ''
310	>	try:
311	>	self.copy_input_data = common.analisys_common_info['copy_input_data']
312	>	#print "self.copy_input_data = ", self.copy_input_data
313	>	except KeyError: self.copy_input_data = 0
314	>	if int(self.copy_input_data) == 1:
315	>	## OLI_Daniele deactivate for OSG (wait for LCG UI installed on OSG)
316	>	txt += 'if [ $middleware == OSG ]; then\n'
317	>	txt += '   #\n'
318	>	txt += '   #   Copy Input Data from SE to this WN deactivated in OSG mode\n'
319	>	txt += '   #\n'
320	>	txt += '   echo "Copy Input Data from SE to this WN deactivated in OSG mode"\n'
321	>	txt += 'elif [ $middleware == LCG ]; then \n'
322	>	txt += '   #\n'
323	>	txt += '   #   Copy Input Data from SE to this WN\n'
324	>	txt += '   #\n'
325	>	### changed by georgia (put a loop copying more than one input files per jobs)
326	>	txt += '   for input_file in $cur_file_list \n'
327	>	txt += '   do \n'
328	>	txt += '    lcg-cp --vo $VO lfn:$input_lfn/$input_file file:`pwd`/$input_file 2>&1\n'
329	>	txt += '    copy_input_exit_status=$?\n'
330	>	txt += '    echo "COPY_INPUT_EXIT_STATUS = $copy_input_exit_status"\n'
331	>	txt += '    if [ $copy_input_exit_status -ne 0 ]; then \n'
332	>	txt += '       echo "Problems with copying to WN" \n'
333	>	txt += '    else \n'
334	>	txt += '       echo "input copied into WN" \n'
335	>	txt += '    fi \n'
336	>	txt += '   done \n'
337	>	### copy a set of PU ntuples (same for each jobs -- but accessed randomly)
338	>	txt += '   for file in $cur_pu_list \n'
339	>	txt += '   do \n'
340	>	txt += '    lcg-cp --vo $VO lfn:$pu_lfn/$file file:`pwd`/$file 2>&1\n'
341	>	txt += '    copy_input_exit_status=$?\n'
342	>	txt += '    echo "COPY_INPUT_PU_EXIT_STATUS = $copy_input_pu_exit_status"\n'
343	>	txt += '    if [ $copy_input_pu_exit_status -ne 0 ]; then \n'
344	>	txt += '       echo "Problems with copying pu to WN" \n'
345	>	txt += '    else \n'
346	>	txt += '       echo "input pu files copied into WN" \n'
347	>	txt += '    fi \n'
348	>	txt += '   done \n'
349	>	txt += '   \n'
350	>	txt += '   ### Check SCRATCH space available on WN : \n'
351	>	txt += '   df -h \n'
352	>	txt += 'fi \n'
353	>
354	>	return txt
355	>
356	>	def wsCopyOutput(self):
357	>	"""
358	>	Write a CopyResults part of a job script, e.g.
359	>	to copy produced output into a storage element.
360	>	"""
361	>	txt = ''
362	>	if int(self.copy_data) == 1:
363	>	txt += '#\n'
364	>	txt += '#   Copy output to SE = $SE\n'
365	>	txt += '#\n'
366	>	txt += 'if [ $exe_result -eq 0 ]; then\n'
367	>	txt += '    for out_file in $file_list ; do\n'
368	>	txt += '        echo "Trying to copy output file to $SE "\n'
369	>	## OLI_Daniele globus-* for OSG, lcg-* for LCG
370	>	txt += '        if [ $middleware == OSG ]; then\n'
371	>	txt += '           echo "globus-url-copy file://`pwd`/$out_file gsiftp://${SE}${SE_PATH}$out_file"\n'
372	>	txt += '           globus-url-copy file://\`pwd\`/$out_file gsiftp://${SE}${SE_PATH}$out_file 2>&1 \n'
373	>	txt += '           copy_exit_status=$? \n'
374	>	txt += '        elif [ $middleware == LCG ]; then \n'
375	>	txt += '           echo "lcg-cp --vo cms -t 1200 file://`pwd`/$out_file gsiftp://${SE}${SE_PATH}$out_file"\n'
376	>	txt += '           lcg-cp --vo cms -t 1200 file://\`pwd\`/$out_file gsiftp://${SE}${SE_PATH}$out_file 2>&1\n'
377	>	txt += '           copy_exit_status=$? \n'
378	>	txt += '        fi \n'
379	>	txt += '        echo "COPY_EXIT_STATUS = $copy_exit_status"\n'
380	>	txt += '        echo "STAGE_OUT = $copy_exit_status"\n'
381	>	txt += '        if [ $copy_exit_status -ne 0 ]; then\n'
382	>	txt += '            echo "Problems with SE = $SE"\n'
383	>	txt += '            echo "StageOutExitStatus = 198" | tee -a $RUNTIME_AREA/$repo\n'
384	>	txt += '            echo "StageOutExitStatusReason = $exitstring" | tee -a $RUNTIME_AREA/$repo\n'
385	>	txt += '        else\n'
386	>	txt += '            echo "StageOutSE = $SE" | tee -a $RUNTIME_AREA/$repo\n'
387	>	txt += '            echo "StageOutCatalog = " | tee -a $RUNTIME_AREA/$repo\n'
388	>	txt += '            echo "output copied into $SE/$SE_PATH directory"\n'
389	>	txt += '            echo "StageOutExitStatus = 0" | tee -a $RUNTIME_AREA/$repo\n'
390	>	txt += '         fi\n'
391	>	txt += '     done\n'
392	>	txt += 'fi\n'
393	>	return txt
394	>
395	>	def wsRegisterOutput(self):
396	>	"""
397	>	Returns part of a job script which does scheduler-specific work.
398	>	"""
399	>
400	>	txt = ''
401	>	if int(self.register_data) == 1:
402	>	## OLI_Daniele deactivate for OSG (wait for LCG UI installed on OSG)
403	>	txt += 'if [ $middleware == OSG ]; then\n'
404	>	txt += '   #\n'
405	>	txt += '   #   Register output to LFC deactivated in OSG mode\n'
406	>	txt += '   #\n'
407	>	txt += '   echo "Register output to LFC deactivated in OSG mode"\n'
408	>	txt += 'elif [ $middleware == LCG ]; then \n'
409	>	txt += '#\n'
410	>	txt += '#  Register output to LFC\n'
411	>	txt += '#\n'
412	>	txt += '   if [[ $exe_result -eq 0 && $copy_exit_status -eq 0 ]]; then\n'
413	>	txt += '      for out_file in $file_list ; do\n'
414	>	txt += '         echo "Trying to register the output file into LFC"\n'
415	>	txt += '         echo "lcg-rf -l $LFN/$out_file --vo $VO sfn://$SE$SE_PATH/$out_file"\n'
416	>	txt += '         lcg-rf -l $LFN/$out_file --vo $VO sfn://$SE$SE_PATH/$out_file 2>&1 \n'
417	>	txt += '         register_exit_status=$?\n'
418	>	txt += '         echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
419	>	txt += '         echo "STAGE_OUT = $register_exit_status"\n'
420	>	txt += '         if [ $register_exit_status -ne 0 ]; then \n'
421	>	txt += '            echo "Problems with the registration to LFC" \n'
422	>	txt += '            echo "Try with srm protocol" \n'
423	>	txt += '            echo "lcg-rf -l $LFN/$out_file --vo $VO srm://$SE$SE_PATH/$out_file"\n'
424	>	txt += '            lcg-rf -l $LFN/$out_file --vo $VO srm://$SE$SE_PATH/$out_file 2>&1 \n'
425	>	txt += '            register_exit_status=$?\n'
426	>	txt += '            echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
427	>	txt += '            echo "STAGE_OUT = $register_exit_status"\n'
428	>	txt += '            if [ $register_exit_status -ne 0 ]; then \n'
429	>	txt += '               echo "Problems with the registration into LFC" \n'
430	>	txt += '            fi \n'
431	>	txt += '         else \n'
432	>	txt += '            echo "output registered to LFC"\n'
433	>	txt += '         fi \n'
434	>	txt += '         echo "StageOutExitStatus = $register_exit_status" | tee -a $RUNTIME_AREA/$repo\n'
435	>	txt += '      done\n'
436	>	txt += '   elif [[ $exe_result -eq 0 && $copy_exit_status -ne 0 ]]; then \n'
437	>	txt += '      echo "Trying to copy output file to CloseSE"\n'
438	>	txt += '      CLOSE_SE=`edg-brokerinfo getCloseSEs | head -1`\n'
439	>	txt += '      for out_file in $file_list ; do\n'
440	>	txt += '         echo "lcg-cr -v -l lfn:${LFN}/$out_file -d $CLOSE_SE -P $LFN/$out_file --vo $VO file://`pwd`/$out_file" \n'
441	>	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'
442	>	txt += '         register_exit_status=$?\n'
443	>	txt += '         echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
444	>	txt += '         echo "STAGE_OUT = $register_exit_status"\n'
445	>	txt += '         if [ $register_exit_status -ne 0 ]; then \n'
446	>	txt += '            echo "Problems with CloseSE" \n'
447	>	txt += '         else \n'
448	>	txt += '            echo "The program was successfully executed"\n'
449	>	txt += '            echo "SE = $CLOSE_SE"\n'
450	>	txt += '            echo "LFN for the file is LFN=${LFN}/$out_file"\n'
451	>	txt += '         fi \n'
452	>	txt += '         echo "StageOutExitStatus = $register_exit_status" | tee -a $RUNTIME_AREA/$repo\n'
453	>	txt += '      done\n'
454	>	txt += '   else\n'
455	>	txt += '      echo "Problem with the executable"\n'
456	>	txt += '   fi \n'
457	>	txt += 'fi \n'
458		return txt
459
460	<	def loggingInfo(self, nj):
460	>	def loggingInfo(self, id):
461		"""
462		retrieve the logging info from logging and bookkeeping and return it
463		"""
464	<	id = common.jobDB.jobId(nj)
465	<	edg_ui_cfg_opt = ''
466	<	if self.edg_config:
467	<	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
99	<	cmd = 'edg-job-get-logging-info -v 2 ' + edg_ui_cfg_opt + id
100	<	print cmd
101	<	myCmd = os.popen(cmd)
102	<	cmd_out = myCmd.readlines()
103	<	myCmd.close()
464	>	self.checkProxy()
465	>	cmd = 'edg-job-get-logging-info -v 2 ' + id
466	>	#cmd_out = os.popen(cmd)
467	>	cmd_out = runCommand(cmd)
468		return cmd_out
469
470		def listMatch(self, nj):
471		"""
472		Check the compatibility of available resources
473		"""
474	+	self.checkProxy()
475		jdl = common.job_list[nj].jdlFilename()
476	<	edg_ui_cfg_opt = ''
477	<	if self.edg_config:
478	<	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
479	<	if self.edg_config_vo:
480	<	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
116	<	cmd = 'edg-job-list-match ' + edg_ui_cfg_opt + jdl
117	<	myCmd = os.popen(cmd)
118	<	cmd_out = myCmd.readlines()
119	<	myCmd.close()
476	>	cmd = 'edg-job-list-match ' + self.configOpt_() + jdl
477	>	cmd_out = runCommand(cmd,0,10)
478	>	if not cmd_out:
479	>	raise CrabException("ERROR: "+cmd+" failed!")
480	>
481		return self.parseListMatch_(cmd_out, jdl)
482
483		def parseListMatch_(self, out, jdl):
484	+	"""
485	+	Parse the f* output of edg-list-match and produce something sensible
486	+	"""
487		reComment = re.compile( r'^\**$' )
488		reEmptyLine = re.compile( r'^$' )
489		reVO = re.compile( r'Selected Virtual Organisation name.*' )
490	<	reCE = re.compile( r'CEId' )
490	>	reLine = re.compile( r'.*')
491	>	reCE = re.compile( r'(.*:.*)')
492	>	reCEId = re.compile( r'CEId.*')
493		reNO = re.compile( r'No Computing Element matching' )
494		reRB = re.compile( r'Connecting to host' )
495		next = 0
496		CEs=[]
497		Match=0
498	<	for line in out:
499	<	line = line.strip()
500	<	if reComment.match( line ):
501	<	next = 0
502	<	continue
503	<	if reEmptyLine.match(line):
504	<	continue
498	>
499	>	#print out
500	>	lines = reLine.findall(out)
501	>
502	>	i=0
503	>	CEs=[]
504	>	for line in lines:
505	>	string.strip(line)
506	>	#print line
507	>	if reNO.match( line ):
508	>	common.logger.debug(5,line)
509	>	return 0
510	>	pass
511		if reVO.match( line ):
512	<	VO =line.split()[-1]
513	<	common.logger.debug(5, 'VO           :'+VO)
512	>	VO =reVO.match( line ).group()
513	>	common.logger.debug(5,"VO "+VO)
514		pass
515	+
516		if reRB.match( line ):
517	<	RB =line.split()[3]
518	<	common.logger.debug(5, 'Using RB     :'+RB)
517	>	RB = reRB.match(line).group()
518	>	common.logger.debug(5,"RB "+RB)
519		pass
520	<	if reCE.search( line ):
521	<	next = 1
522	<	continue
523	<	if next:
524	<	CE=line.split(':')[0]
525	<	CEs.append(CE)
526	<	common.logger.debug(5, 'Matched CE   :'+CE)
527	<	Match=Match+1
155	<	pass
156	<	if reNO.match( line ):
157	<	common.logger.debug(5,line)
158	<	self.noMatchFound_(jdl)
159	<	Match=0
520	>
521	>	if reCEId.search( line ):
522	>	for lineCE in lines[i:-1]:
523	>	if reCE.match( lineCE ):
524	>	CE = string.strip(reCE.search(lineCE).group(1))
525	>	CEs.append(CE.split(':')[0])
526	>	pass
527	>	pass
528		pass
529	<	return Match
529	>	i=i+1
530	>	pass
531	>
532	>	common.logger.debug(5,"All CE :"+str(CEs))
533	>
534	>	sites = []
535	>	[sites.append(it) for it in CEs if not sites.count(it)]
536	>
537	>	common.logger.debug(5,"All Sites :"+str(sites))
538	>	common.logger.message("Matched Sites :"+str(sites))
539	>	return len(sites)
540
541		def noMatchFound_(self, jdl):
542		reReq = re.compile( r'Requirements' )
#	Line 184 | Line 562 | class SchedulerEdg(Scheduler):
562		Submit one EDG job.
563		"""
564
565	+	self.checkProxy()
566		jid = None
567		jdl = common.job_list[nj].jdlFilename()
568	<	id_tmp = tempfile.mktemp()
569	<	edg_ui_cfg_opt = ' '
191	<	if self.edg_config:
192	<	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
193	<	if self.edg_config_vo:
194	<	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
195	<	cmd = 'edg-job-submit -o ' + id_tmp + edg_ui_cfg_opt + jdl
568	>
569	>	cmd = 'edg-job-submit ' + self.configOpt_() + jdl
570		cmd_out = runCommand(cmd)
571		if cmd_out != None:
572	<	idfile = open(id_tmp)
573	<	jid_line = idfile.readline()
200	<	while jid_line[0] == '#':
201	<	jid_line = idfile.readline()
202	<	pass
203	<	jid = string.strip(jid_line)
204	<	os.unlink(id_tmp)
572	>	reSid = re.compile( r'https.+' )
573	>	jid = reSid.search(cmd_out).group()
574		pass
575		return jid
576
577	+	def resubmit(self, nj_list):
578	+	"""
579	+	Prepare jobs to be submit
580	+	"""
581	+	return
582	+
583		def getExitStatus(self, id):
584		return self.getStatusAttribute_(id, 'exit_code')
585
#	Line 218 | Line 593 | class SchedulerEdg(Scheduler):
593		def getStatusAttribute_(self, id, attr):
594		""" Query a status of the job with id """
595
596	+	self.checkProxy()
597		hstates = {}
598		Status = importName('edg_wl_userinterface_common_LbWrapper', 'Status')
223	–	#        Status = importName('glite_wmsui_LbWrapper', 'Status')
599		# Bypass edg-job-status interfacing directly to C++ API
600		# Job attribute vector to retrieve status without edg-job-status
601		level = 0
#	Line 230 | Line 605 | class SchedulerEdg(Scheduler):
605		jobStat.getStatus(id, level)
606		err, apiMsg = jobStat.get_error()
607		if err:
608	<	print 'Error caught', apiMsg
234	<	common.log.message(apiMsg)
608	>	common.logger.debug(5,'Error caught' + apiMsg)
609		return None
610		else:
611		for i in range(len(self.states)):
238	–	#print "states = ", states
612		# Fill an hash table with all information retrieved from LB API
613		hstates[ self.states[i] ] = jobStat.loadStatus(st)[i]
614		result = jobStat.loadStatus(st)[ self.states.index(attr) ]
#	Line 253 | Line 626 | class SchedulerEdg(Scheduler):
626		Returns the name of directory with results.
627		"""
628
629	+	self.checkProxy()
630		cmd = 'edg-job-get-output --dir ' + common.work_space.resDir() + ' ' + id
631		cmd_out = runCommand(cmd)
632
633		# Determine the output directory name
634		dir = common.work_space.resDir()
635	<	dir += os.getlogin()
635	>	dir += os.environ['USER']
636		dir += '_' + os.path.basename(id)
637		return dir
638
639		def cancel(self, id):
640		""" Cancel the EDG job with id """
641	+	self.checkProxy()
642		cmd = 'edg-job-cancel --noint ' + id
643		cmd_out = runCommand(cmd)
644		return cmd_out
645
271	–	def checkProxy_(self):
272	–	"""
273	–	Function to check the Globus proxy.
274	–	"""
275	–	cmd = 'grid-proxy-info -timeleft'
276	–	cmd_out = runCommand(cmd)
277	–	ok = 1
278	–	timeleft = -999
279	–	try: timeleft = int(cmd_out)
280	–	except ValueError: ok=0
281	–	except TypeError: ok=0
282	–	if timeleft < 1:  ok=0
283	–
284	–	if ok==0:
285	–	print "No valid proxy found !\n"
286	–	print "Creating a user proxy with default length of 100h\n"
287	–	msg = "Unable to create a valid proxy!\n"
288	–	if os.system("grid-proxy-init -valid 100:00"):
289	–	raise CrabException(msg)
290	–	return
291	–
646		def createSchScript(self, nj):
647		"""
648		Create a JDL-file for EDG.
#	Line 303 | Line 657 | class SchedulerEdg(Scheduler):
657		title = '# This JDL was generated by '+\
658		common.prog_name+' (version '+common.prog_version_str+')\n'
659		jt_string = ''
660	+
661	+
662
663		SPL = inp_storage_subdir
664		if ( SPL and SPL[-1] != '/' ) : SPL = SPL + '/'
#	Line 315 | Line 671 | class SchedulerEdg(Scheduler):
671		jdl.write('Executable = "' + os.path.basename(script) +'";\n')
672		jdl.write(jt_string)
673
674	<	firstEvent = common.jobDB.firstEvent(nj)
319	<	maxEvents = common.jobDB.maxEvents(nj)
320	<	jdl.write('Arguments = "' + str(nj+1)+' '+str(firstEvent)+' '+str(maxEvents)+'";\n')
674	>	### only one .sh  JDL has arguments:
675
676	+	### Fabio
677	+	jdl.write('Arguments = "' + str(nj+1)+' '+ jbt.getJobTypeArguments(nj, "EDG") +'";\n')
678		inp_box = 'InputSandbox = { '
679		inp_box = inp_box + '"' + script + '",'
680
#	Line 331 | Line 687 | class SchedulerEdg(Scheduler):
687		#if common.use_jam:
688		#   inp_box = inp_box+' "'+common.bin_dir+'/'+common.run_jam+'",'
689
690	<	for addFile in jbt.additional_inbox_files:
691	<	addFile = os.path.abspath(addFile)
692	<	inp_box = inp_box+' "'+addFile+'",'
693	<	pass
690	>	# Marco (VERY TEMPORARY ML STUFF)
691	>	inp_box = inp_box+' "' + os.path.abspath(os.environ['CRABDIR']+'/python/'+'report.py') + '", "' +\
692	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'DashboardAPI.py') + '", "'+\
693	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'Logger.py') + '", "'+\
694	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'ProcInfo.py') + '", "'+\
695	>	os.path.abspath(os.environ['CRABDIR']+'/python/'+'apmon.py') + '"'
696	>	# End Marco
697	>
698	>	if (not jbt.additional_inbox_files == []):
699	>	inp_box = inp_box + ', '
700	>	for addFile in jbt.additional_inbox_files:
701	>	addFile = os.path.abspath(addFile)
702	>	inp_box = inp_box+' "'+addFile+'",'
703	>	pass
704
705		if inp_box[-1] == ',' : inp_box = inp_box[:-1]
706		inp_box = inp_box + ' };\n'
#	Line 342 | Line 708 | class SchedulerEdg(Scheduler):
708
709		jdl.write('StdOutput     = "' + job.stdout() + '";\n')
710		jdl.write('StdError      = "' + job.stderr() + '";\n')
711	<
712	<	#if common.flag_return_data :
713	<	#    for fl in job.outputDataFiles():
714	<	#        out_box = out_box + ' "' + fl + '",'
715	<	#        pass
716	<	#    pass
717	<
718	<	out_box = 'OutputSandbox = { '
719	<	if out_sandbox != None:
720	<	for fl in out_sandbox:
721	<	out_box = out_box + ' "' + fl + '",'
711	>
712	>
713	>	if job.stdout() == job.stderr():
714	>	out_box = 'OutputSandbox = { "' + \
715	>	job.stdout() + '", ".BrokerInfo",'
716	>	else:
717	>	out_box = 'OutputSandbox = { "' + \
718	>	job.stdout() + '", "' + \
719	>	job.stderr() + '", ".BrokerInfo",'
720	>
721	>	if int(self.return_data) == 1:
722	>	if out_sandbox != None:
723	>	for fl in out_sandbox:
724	>	out_box = out_box + ' "' + fl + '",'
725	>	pass
726		pass
727		pass
728	<
728	>
729		if out_box[-1] == ',' : out_box = out_box[:-1]
730		out_box = out_box + ' };'
731		jdl.write(out_box+'\n')
732
363	–	# If CloseCE is used ...
364	–	#if common.flag_usecloseCE and job.inputDataFiles():
365	–	#    indata = 'InputData = { '
366	–	#    for fl in job.inputDataFiles():
367	–	#       indata = indata + ' "lfn:' + SPL + fl + '",'
368	–	#    if indata[-1] == ',' : indata = indata[:-1]
369	–	#    indata = indata + ' };'
370	–	#    jdl.write(indata+'\n')
371	–	#    jdl.write('DataAccessProtocol = { "gsiftp" };\n')
372	–
373	–	if common.analisys_common_info['sites']:
374	–	if common.analisys_common_info['sw_version']:
375	–
376	–	req='Requirements = '
377	–	### First ORCA version
378	–	req=req + 'Member("VO-cms-' + \
379	–	common.analisys_common_info['sw_version'] + \
380	–	'", other.GlueHostApplicationSoftwareRunTimeEnvironment)'
381	–	## then sites
382	–	if len(common.analisys_common_info['sites'])>0:
383	–	req = req + ' && ('
384	–	for i in range(len(common.analisys_common_info['sites'])):
385	–	req = req + 'other.GlueCEInfoHostName == "' \
386	–	+ common.analisys_common_info['sites'][i] + '"'
387	–	if ( i < (int(len(common.analisys_common_info['sites']) - 1)) ):
388	–	req = req + ' || '
389	–	req = req + ')'
390	–	## then user requirement
391	–	if self.EDG_requirements:
392	–	req = req +  ' && ' + self.EDG_requirements
393	–	req = req + ';\n'
394	–	jdl.write(req)
733
734	+	req='Requirements = '
735	+	noreq=req
736	+	req = req + jbt.getRequirements()
737	+	#### and USER REQUIREMENT
738	+	if self.EDG_requirements:
739	+	if (req != noreq):
740	+	req = req +  ' && '
741	+	req = req + self.EDG_requirements
742	+	#### FEDE #####
743	+	if self.EDG_ce_white_list:
744	+	ce_white_list = string.split(self.EDG_ce_white_list,',')
745	+	#print "req = ", req
746	+	for i in range(len(ce_white_list)):
747	+	if i == 0:
748	+	if (req != noreq):
749	+	req = req +  ' && '
750	+	req = req + '((RegExp("' + ce_white_list[i] + '", other.GlueCEUniqueId))'
751	+	pass
752	+	else:
753	+	req = req +  ' || (RegExp("' + ce_white_list[i] + '", other.GlueCEUniqueId))'
754	+	req = req + ')'
755	+
756	+	if self.EDG_ce_black_list:
757	+	ce_black_list = string.split(self.EDG_ce_black_list,',')
758	+	for ce in ce_black_list:
759	+	if (req != noreq):
760	+	req = req +  ' && '
761	+	req = req + '(!RegExp("' + ce + '", other.GlueCEUniqueId))'
762	+	pass
763	+
764	+	###############
765	+	clockTime=480
766	+	if self.EDG_clock_time:
767	+	clockTime= self.EDG_clock_time
768	+	if (req != noreq):
769	+	req = req + ' && '
770	+	req = req + '((other.GlueCEPolicyMaxWallClockTime == 0) || (other.GlueCEPolicyMaxWallClockTime>='+str(clockTime)+'))'
771	+
772	+	cpuTime=1000
773	+	if self.EDG_cpu_time:
774	+	cpuTime=self.EDG_cpu_time
775	+	if (req != noreq):
776	+	req = req + ' && '
777	+	req = req + '((other.GlueCEPolicyMaxCPUTime == 0) || (other.GlueCEPolicyMaxCPUTime>='+str(cpuTime)+'))'
778	+
779	+	if (req != noreq):
780	+	req = req + ';\n'
781	+	jdl.write(req)
782	+
783		jdl.write('VirtualOrganisation = "' + self.VO + '";\n')
784
785		if ( self.EDG_retry_count ):
786		jdl.write('RetryCount = '+self.EDG_retry_count+';\n')
787		pass
788
789	+	jdl.write('MyProxyServer = "' + self.proxyServer + '";\n')
790	+
791		jdl.close()
792		return
793	+
794	+	def checkProxy(self):
795	+	"""
796	+	Function to check the Globus proxy.
797	+	"""
798	+	if (self.proxyValid): return
799	+	timeleft = -999
800	+	minTimeLeft=10*3600 # in seconds
801	+
802	+	minTimeLeftServer = 100 # in hours
803	+
804	+	#cmd = 'voms-proxy-info -exists -valid '+str(minTimeLeft)+':00'
805	+	#cmd = 'voms-proxy-info -timeleft'
806	+	mustRenew = 0
807	+	timeLeftLocal = runCommand('voms-proxy-info -timeleft')
808	+	timeLeftServer = -999
809	+	if not timeLeftLocal or int(timeLeftLocal) <= 0 or not isInt(timeLeftLocal):
810	+	mustRenew = 1
811	+	else:
812	+	timeLeftServer = runCommand('voms-proxy-info -actimeleft | head -1')
813	+	if not timeLeftServer or not isInt(timeLeftServer):
814	+	mustRenew = 1
815	+	elif timeLeftLocal<minTimeLeft or timeLeftServer<minTimeLeft:
816	+	mustRenew = 1
817	+	pass
818	+	pass
819	+
820	+	if mustRenew:
821	+	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")
822	+	cmd = 'voms-proxy-init -voms cms -valid 96:00'
823	+	try:
824	+	# SL as above: damn it!
825	+	out = os.system(cmd)
826	+	if (out>0): raise CrabException("Unable to create a valid proxy!\n")
827	+	except:
828	+	msg = "Unable to create a valid proxy!\n"
829	+	raise CrabException(msg)
830	+	# cmd = 'grid-proxy-info -timeleft'
831	+	# cmd_out = runCommand(cmd,0,20)
832	+	pass
833	+
834	+	## now I do have a voms proxy valid, and I check the myproxy server
835	+	renewProxy = 0
836	+	cmd = 'myproxy-info -d -s '+self.proxyServer
837	+	cmd_out = runCommand(cmd,0,20)
838	+	if not cmd_out:
839	+	common.logger.message('No credential delegated to myproxy server '+self.proxyServer+' will do now')
840	+	renewProxy = 1
841	+	else:
842	+	# if myproxy exist but not long enough, renew
843	+	reTime = re.compile( r'timeleft: (\d+)' )
844	+	#print "<"+str(reTime.search( cmd_out ).group(1))+">"
845	+	if reTime.match( cmd_out ):
846	+	time = reTime.search( line ).group(1)
847	+	if time < minTimeLeftServer:
848	+	renewProxy = 1
849	+	common.logger.message('No credential delegation will expire in '+time+' hours: renew it')
850	+	pass
851	+	pass
852	+
853	+	# if not, create one.
854	+	if renewProxy:
855	+	cmd = 'myproxy-init -d -n -s '+self.proxyServer
856	+	out = os.system(cmd)
857	+	if (out>0):
858	+	raise CrabException("Unable to delegate the proxy to myproxyserver "+self.proxyServer+" !\n")
859	+	pass
860	+
861	+	# cache proxy validity
862	+	self.proxyValid=1
863	+	return
864	+
865	+	def configOpt_(self):
866	+	edg_ui_cfg_opt = ' '
867	+	if self.edg_config:
868	+	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
869	+	if self.edg_config_vo:
870	+	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
871	+	return edg_ui_cfg_opt

Diff Legend

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