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 1.1.2.3.2.1 by spiga, Fri Sep 15 07:51:02 2006 UTC vs.
Revision 1.14 by spiga, Mon Jun 25 14:21:56 2007 UTC

#	Line 1 | Line 1
1	<	from Scheduler import Scheduler
1	>	from SchedulerEdg import SchedulerEdg
2		from crab_logger import Logger
3		from crab_exceptions import *
4		from crab_util import *
5	<	from EdgConfig import *
5	>	from GliteConfig import *
6		import common
7
8		import os, sys, time
9
10	<	class SchedulerGlite(Scheduler):
10	>	class SchedulerGlite(SchedulerEdg):
11		def __init__(self):
12	<	Scheduler.__init__(self,"GLITE")
13	<	self.states = [ "Acl", "cancelReason", "cancelling","ce_node","children", \
14	<	"children_hist","children_num","children_states","condorId","condor_jdl", \
15	<	"cpuTime","destination", "done_code","exit_code","expectFrom", \
16	<	"expectUpdate","globusId","jdl","jobId","jobtype", \
17	<	"lastUpdateTime","localId","location", "matched_jdl","network_server", \
18	<	"owner","parent_job", "reason","resubmitted","rsl","seed",\
19	<	"stateEnterTime","stateEnterTimes","subjob_failed", \
20	<	"user tags" , "status" , "status_code","hierarchy"]
21	<	return
22	<
23	<	def configure(self, cfg_params):
24	<
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:
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'
42	<
43	<	try: self.EDG_requirements = cfg_params['EDG.requirements']
44	<	except KeyError: self.EDG_requirements = ''
45	<
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	<	except KeyError:
52	<	self.EDG_ce_black_list  = ''
53	<
54	<	try:
55	<	self.EDG_ce_white_list = cfg_params['EDG.ce_white_list']
56	<	except KeyError: self.EDG_ce_white_list = ''
57	<
58	<	try: self.VO = cfg_params['EDG.virtual_organization']
59	<	except KeyError: self.VO = 'cms'
60	<
61	<	try: self.return_data = cfg_params['USER.return_data']
62	<	except KeyError: self.return_data = 1
63	<
64	<	try:
65	<	self.copy_input_data = common.analisys_common_info['copy_input_data']
66	<	except KeyError: self.copy_input_data = 0
67	<
68	<	try:
69	<	self.copy_data = cfg_params["USER.copy_data"]
70	<	if int(self.copy_data) == 1:
71	<	try:
72	<	self.SE = cfg_params['USER.storage_element']
73	<	self.SE_PATH = cfg_params['USER.storage_path']
74	<	except KeyError:
75	<	msg = "Error. The [USER] section does not have 'storage_element'"
76	<	msg = msg + " and/or 'storage_path' entries, necessary to copy the output"
77	<	common.logger.message(msg)
78	<	raise CrabException(msg)
79	<	except KeyError: self.copy_data = 0
80	<
81	<	if ( int(self.return_data) == 0 and int(self.copy_data) == 0 ):
82	<	msg = 'Warning: return_data = 0 and copy_data = 0 ==> your exe output will be lost\n'
83	<	msg = msg + 'Please modify return_data and copy_data value in your crab.cfg file\n'
84	<	raise CrabException(msg)
85	<
86	<	try:
87	<	self.lfc_host = cfg_params['EDG.lfc_host']
88	<	except KeyError:
89	<	msg = "Error. The [EDG] section does not have 'lfc_host' value"
90	<	msg = msg + " it's necessary to know the LFC host name"
91	<	common.logger.message(msg)
92	<	raise CrabException(msg)
93	<	try:
94	<	self.lcg_catalog_type = cfg_params['EDG.lcg_catalog_type']
95	<	except KeyError:
96	<	msg = "Error. The [EDG] section does not have 'lcg_catalog_type' value"
97	<	msg = msg + " it's necessary to know the catalog type"
98	<	common.logger.message(msg)
99	<	raise CrabException(msg)
100	<	try:
101	<	self.lfc_home = cfg_params['EDG.lfc_home']
102	<	except KeyError:
103	<	msg = "Error. The [EDG] section does not have 'lfc_home' value"
104	<	msg = msg + " it's necessary to know the home catalog dir"
105	<	common.logger.message(msg)
106	<	raise CrabException(msg)
107	<
108	<	try:
109	<	self.register_data = cfg_params["USER.register_data"]
110	<	if int(self.register_data) == 1:
111	<	try:
112	<	self.LFN = cfg_params['USER.lfn_dir']
113	<	except KeyError:
114	<	msg = "Error. The [USER] section does not have 'lfn_dir' value"
115	<	msg = msg + " it's necessary for LCF registration"
116	<	common.logger.message(msg)
117	<	raise CrabException(msg)
118	<	except KeyError: self.register_data = 0
119	<
120	<	if ( int(self.copy_data) == 0 and int(self.register_data) == 1 ):
121	<	msg = 'Warning: register_data = 1 must be used with copy_data = 1\n'
122	<	msg = msg + 'Please modify copy_data value in your crab.cfg file\n'
123	<	common.logger.message(msg)
124	<	raise CrabException(msg)
125	<
126	<	try: self.EDG_requirements = cfg_params['EDG.requirements']
127	<	except KeyError: self.EDG_requirements = ''
128	<
129	<	try: self.EDG_retry_count = cfg_params['EDG.retry_count']
130	<	except KeyError: self.EDG_retry_count = ''
131	<
132	<	try: self.EDG_clock_time = cfg_params['EDG.max_wall_clock_time']
133	<	except KeyError: self.EDG_clock_time= ''
134	<
135	<	try: self.EDG_cpu_time = cfg_params['EDG.max_cpu_time']
136	<	except KeyError: self.EDG_cpu_time = ''
137	<
138	<
139	<	try: self.schedulerName = cfg_params['CRAB.scheduler']
140	<	except KeyError: self.schedulerName = ''
141	<
142	<	try: self.jobtypeName = cfg_params['CRAB.jobtype']
143	<	except KeyError: self.jobtypeName = ''
144	<
145	<	# Add EDG_WL_LOCATION to the python path
146	<	try:
147	<	path = os.environ['EDG_WL_LOCATION']
148	<	except:
149	<	msg = "Error: the EDG_WL_LOCATION variable is not set."
150	<	raise CrabException(msg)
151	<
152	<	libPath=os.path.join(path, "lib")
153	<	sys.path.append(libPath)
154	<	libPath=os.path.join(path, "lib", "python")
155	<	sys.path.append(libPath)
156	<
157	<	self.proxyValid=0
158	<
159	<	try:
160	<	self._taskId = cfg_params['taskId']
161	<	except:
162	<	self._taskId = ''
163	<
164	<	try: self.jobtypeName = cfg_params['CRAB.jobtype']
165	<	except KeyError: self.jobtypeName = ''
166	<
167	<	try: self.schedulerName = cfg_params['CRAB.scheduler']
168	<	except KeyError: self.scheduler = ''
12	>	SchedulerEdg.__init__(self)
13
14	<	return
15	<
14	>	def rb_configure(self, RB):
15	>	self.glite_config = ''
16	>	self.rb_param_file = ''
17	>
18	>	gliteConfig = GliteConfig(RB)
19	>	self.glite_config = gliteConfig.config()
20	>
21	>	if (self.glite_config != ''):
22	>	self.rb_param_file = 'WMSconfig = '+self.glite_config+';\n'
23	>	#print "rb_param_file = ", self.rb_param_file
24	>	return self.rb_param_file
25
26		def sched_parameter(self):
27		"""
28	<	Returns file with scheduler-specific parameters
28	>	Returns file with requirements and scheduler-specific parameters
29		"""
30	<	if (self.edg_config and self.edg_config_vo != ''):
31	<	self.param='sched_param.clad'
30	>	index = int(common.jobDB.nJobs()) - 1
31	>	job = common.job_list[index]
32	>	jbt = job.type()
33	>
34	>	lastDest=''
35	>	first = []
36	>	last  = []
37	>	for n in range(common.jobDB.nJobs()):
38	>	currDest=common.jobDB.destination(n)
39	>	if (currDest!=lastDest):
40	>	lastDest = currDest
41	>	first.append(n)
42	>	if n != 0:last.append(n-1)
43	>	if len(first)>len(last) :last.append(common.jobDB.nJobs())
44	>
45	>	req = ''
46	>	req = req + jbt.getRequirements()
47	>
48	>
49	>	if self.EDG_requirements:
50	>	if (not req == ' '): req = req +  ' && '
51	>	req = req + self.EDG_requirements
52	>
53	>	if self.EDG_ce_white_list:
54	>	ce_white_list = string.split(self.EDG_ce_white_list,',')
55	>	tmpCe=[]
56	>	concString = '&&'
57	>	for ce in ce_white_list:
58	>	tmpCe.append('RegExp("' + string.strip(ce) + '", other.GlueCEUniqueId)')
59	>	if len(tmpCe): req = req + " && (" + concString.join(tmpCe) + ") "
60	>
61	>	if self.EDG_ce_black_list:
62	>	ce_black_list = string.split(self.EDG_ce_black_list,',')
63	>	tmpCe=[]
64	>	concString = '&&'
65	>	for ce in ce_black_list:
66	>	tmpCe.append('(!RegExp("' + string.strip(ce) + '", other.GlueCEUniqueId))')
67	>	if len(tmpCe): req = req + " && (" + concString.join(tmpCe) + ") "
68	>
69	>	if self.EDG_clock_time:
70	>	if (not req == ' '): req = req + ' && '
71	>	req = req + 'other.GlueCEPolicyMaxWallClockTime>='+self.EDG_clock_time
72	>
73	>	if self.EDG_cpu_time:
74	>	if (not req == ' '): req = req + ' && '
75	>	req = req + ' other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time
76	>
77	>	for i in range(len(first)): # Add loop DS
78	>	self.param='sched_param_'+str(i)+'.clad'
79		param_file = open(common.work_space.shareDir()+'/'+self.param, 'w')
80	<	param_file.write('RBconfig = "'+self.edg_config+'";\n')
81	<	param_file.write('RBconfigVO = "'+self.edg_config_vo+'";')
80	>
81	>	itr4=self.findSites_(first[i])
82	>	reqSites=''
83	>	reqtmp=[]
84	>	concString = '||'
85	>
86	>	#############
87	>	# MC Changed matching syntax to avoid gang matching
88	>	#############
89	>	for arg in itr4:
90	>	reqtmp.append(' Member("'+arg+'" , other.GlueCESEBindGroupSEUniqueID) ')
91	>
92	>	if len(reqtmp): reqSites = reqSites + " && (" + concString.join(reqtmp) + ") "
93	>
94	>	# requirement added to skip gliteCE
95	>	reqSites = reqSites + '&& (!RegExp("blah", other.GlueCEUniqueId));\n'
96	>
97	>	param_file.write('Requirements = ' + req + reqSites )
98	>
99	>	if (self.rb_param_file != ''):
100	>	param_file.write(self.rb_param_file)
101	>
102	>	if len(self.EDG_addJdlParam):
103	>	for p in self.EDG_addJdlParam:
104	>	param_file.write(p)
105	>
106		param_file.close()
183	–	return 1
184	–	else:
185	–	return 0
107
108		def wsSetupEnvironment(self):
109		"""
#	Line 195 | Line 116 | class SchedulerGlite(Scheduler):
116		txt += 'nargs=$#\n'
117		txt += 'shift $nargs\n'
118		txt += "# job number (first parameter for job wrapper)\n"
198	–	#txt += "NJob=$1\n"
119		txt += "NJob=${args[0]}\n"
120
121		txt += '# job identification to DashBoard \n'
122	<	txt += 'MonitorJobID=`echo ${NJob}_$EDG_WL_JOBID`\n'
123	<	txt += 'SyncGridJobId=`echo $EDG_WL_JOBID`\n'
122	>	txt += 'MonitorJobID=`echo ${NJob}_$GLITE_WMS_JOBID`\n'
123	>	txt += 'SyncGridJobId=`echo $GLITE_WMS_JOBID`\n'
124		txt += 'MonitorID=`echo ' + self._taskId + '`\n'
125		txt += 'echo "MonitorJobID=`echo $MonitorJobID`" | tee -a $RUNTIME_AREA/$repo \n'
126		txt += 'echo "SyncGridJobId=`echo $SyncGridJobId`" | tee -a $RUNTIME_AREA/$repo \n'
127		txt += 'echo "MonitorID=`echo $MonitorID`" | tee -a $RUNTIME_AREA/$repo\n'
128
129		txt += 'echo "middleware discovery " \n'
130	<	txt += 'if [ $VO_CMS_SW_DIR ]; then\n'
130	>	txt += 'if [ $VO_CMS_SW_DIR ]; then \n'
131		txt += '    middleware=LCG \n'
132	<	txt += '    echo "SyncCE=`edg-brokerinfo getCE`" | tee -a $RUNTIME_AREA/$repo \n'
132	>	txt += '    echo "SyncCE=`glite-brokerinfo getCE`" | tee -a $RUNTIME_AREA/$repo \n'
133		txt += '    echo "GridFlavour=`echo $middleware`" | tee -a $RUNTIME_AREA/$repo \n'
134		txt += '    echo "middleware =$middleware" \n'
135		txt += 'elif [ $GRID3_APP_DIR ]; then\n'
#	Line 251 | Line 171 | class SchedulerGlite(Scheduler):
171		txt += 'echo "SE_PATH = $SE_PATH"\n'
172
173		txt += 'export VO='+self.VO+'\n'
174	<	### FEDE: add some line for LFC catalog setting
174	>	### some line for LFC catalog setting
175		txt += 'if [ $middleware == LCG ]; then \n'
176		txt += '    if [[ $LCG_CATALOG_TYPE != \''+self.lcg_catalog_type+'\' ]]; then\n'
177		txt += '        export LCG_CATALOG_TYPE='+self.lcg_catalog_type+'\n'
#	Line 312 | Line 232 | class SchedulerGlite(Scheduler):
232		txt += 'fi \n'
233
234		return txt
235	<
316	<	def wsCopyInput(self):
317	<	"""
318	<	Copy input data from SE to WN
319	<	"""
320	<	txt = ''
321	<	try:
322	<	self.copy_input_data = common.analisys_common_info['copy_input_data']
323	<	except KeyError: self.copy_input_data = 0
324	<	if int(self.copy_input_data) == 1:
325	<	## OLI_Daniele deactivate for OSG (wait for LCG UI installed on OSG)
326	<	txt += 'if [ $middleware == OSG ]; then\n'
327	<	txt += '   #\n'
328	<	txt += '   #   Copy Input Data from SE to this WN deactivated in OSG mode\n'
329	<	txt += '   #\n'
330	<	txt += '   echo "Copy Input Data from SE to this WN deactivated in OSG mode"\n'
331	<	txt += 'elif [ $middleware == LCG ]; then \n'
332	<	txt += '   #\n'
333	<	txt += '   #   Copy Input Data from SE to this WN\n'
334	<	txt += '   #\n'
335	<	### changed by georgia (put a loop copying more than one input files per jobs)
336	<	txt += '   for input_file in $cur_file_list \n'
337	<	txt += '   do \n'
338	<	txt += '      lcg-cp --vo $VO --verbose -t 1200 lfn:$input_lfn/$input_file file:`pwd`/$input_file 2>&1\n'
339	<	txt += '      copy_input_exit_status=$?\n'
340	<	txt += '      echo "COPY_INPUT_EXIT_STATUS = $copy_input_exit_status"\n'
341	<	txt += '      if [ $copy_input_exit_status -ne 0 ]; then \n'
342	<	txt += '         echo "Problems with copying to WN" \n'
343	<	txt += '      else \n'
344	<	txt += '         echo "input copied into WN" \n'
345	<	txt += '      fi \n'
346	<	txt += '   done \n'
347	<	### copy a set of PU ntuples (same for each jobs -- but accessed randomly)
348	<	txt += '   for file in $cur_pu_list \n'
349	<	txt += '   do \n'
350	<	txt += '      lcg-cp --vo $VO --verbose -t 1200 lfn:$pu_lfn/$file file:`pwd`/$file 2>&1\n'
351	<	txt += '      copy_input_pu_exit_status=$?\n'
352	<	txt += '      echo "COPY_INPUT_PU_EXIT_STATUS = $copy_input_pu_exit_status"\n'
353	<	txt += '      if [ $copy_input_pu_exit_status -ne 0 ]; then \n'
354	<	txt += '         echo "Problems with copying pu to WN" \n'
355	<	txt += '      else \n'
356	<	txt += '         echo "input pu files copied into WN" \n'
357	<	txt += '      fi \n'
358	<	txt += '   done \n'
359	<	txt += '   \n'
360	<	txt += '   ### Check SCRATCH space available on WN : \n'
361	<	txt += '   df -h \n'
362	<	txt += 'fi \n'
363	<
364	<	return txt
365	<
366	<	def wsCopyOutput(self):
367	<	"""
368	<	Write a CopyResults part of a job script, e.g.
369	<	to copy produced output into a storage element.
370	<	"""
371	<	txt = ''
372	<	if int(self.copy_data) == 1:
373	<	txt += '#\n'
374	<	txt += '#   Copy output to SE = $SE\n'
375	<	txt += '#\n'
376	<	txt += '    if [ $middleware == OSG ]; then\n'
377	<	txt += '        echo "X509_USER_PROXY = $X509_USER_PROXY"\n'
378	<	txt += '        echo "source $OSG_APP/glite/setup_glite_ui.sh"\n'
379	<	txt += '        source $OSG_APP/glite/setup_glite_ui.sh\n'
380	<	txt += '        export X509_CERT_DIR=$OSG_APP/glite/etc/grid-security/certificates\n'
381	<	txt += '        echo "export X509_CERT_DIR=$X509_CERT_DIR"\n'
382	<	txt += '    fi \n'
383	<	txt += '    for out_file in $file_list ; do\n'
384	<	txt += '        echo "Trying to copy output file to $SE using lcg-cp"\n'
385	<	txt += '        echo "lcg-cp --vo $VO -t 1200 --verbose file://`pwd`/$out_file gsiftp://${SE}${SE_PATH}$out_file"\n'
386	<	txt += '        exitstring=`lcg-cp --vo $VO -t 1200 --verbose file://\`pwd\`/$out_file gsiftp://${SE}${SE_PATH}$out_file 2>&1`\n'
387	<	txt += '        copy_exit_status=$?\n'
388	<	txt += '        echo "COPY_EXIT_STATUS for lcg-cp = $copy_exit_status"\n'
389	<	txt += '        echo "STAGE_OUT = $copy_exit_status"\n'
390	<	txt += '        if [ $copy_exit_status -ne 0 ]; then\n'
391	<	txt += '            echo "Problems with SE = $SE"\n'
392	<	txt += '            echo "StageOutExitStatus = 198" | tee -a $RUNTIME_AREA/$repo\n'
393	<	txt += '            echo "StageOutExitStatusReason = $exitstring" | tee -a $RUNTIME_AREA/$repo\n'
394	<	txt += '            echo "lcg-cp failed, attempting srmcp"\n'
395	<	txt += '            echo "mkdir -p $HOME/.srmconfig"\n'
396	<	txt += '            mkdir -p $HOME/.srmconfig\n'
397	<	txt += '            if [ $middleware == LCG ]; then\n'
398	<	txt += '               echo "srmcp -retry_num 5 -retry_timeout 240000 file:////`pwd`/$out_file srm://${SE}:8443${SE_PATH}$out_file"\n'
399	<	txt += '               exitstring=`srmcp -retry_num 5 -retry_timeout 240000 file:////\`pwd\`/$out_file srm://${SE}:8443${SE_PATH}$out_file 2>&1`\n'
400	<	txt += '            elif [ $middleware == OSG ]; then\n'
401	<	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'
402	<	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}$out_file 2>&1`\n'
403	<	txt += '            fi \n'
404	<	txt += '            copy_exit_status=$?\n'
405	<	txt += '            echo "COPY_EXIT_STATUS for srm = $copy_exit_status"\n'
406	<	txt += '            echo "STAGE_OUT = $copy_exit_status"\n'
407	<	txt += '            if [ $copy_exit_status -ne 0 ]; then\n'
408	<	txt += '               echo "Problems with SE = $SE"\n'
409	<	txt += '               echo "StageOutExitStatus = 198" | tee -a $RUNTIME_AREA/$repo\n'
410	<	txt += '               echo "StageOutExitStatusReason = $exitstring" | tee -a $RUNTIME_AREA/$repo\n'
411	<	txt += '               echo "lcg-cp and srm failed"\n'
412	<	txt += '               echo "If storage_path in your config file contains a ? you may need a \? instead."\n'
413	<	txt += '            else\n'
414	<	txt += '               echo "StageOutSE = $SE" | tee -a $RUNTIME_AREA/$repo\n'
415	<	txt += '               echo "StageOutCatalog = " | tee -a $RUNTIME_AREA/$repo\n'
416	<	txt += '               echo "output copied into $SE/$SE_PATH directory"\n'
417	<	txt += '               echo "StageOutExitStatus = 0" | tee -a $RUNTIME_AREA/$repo\n'
418	<	txt += '               echo "srmcp succeeded"\n'
419	<	txt += '            fi\n'
420	<	txt += '        else\n'
421	<	txt += '            echo "StageOutSE = $SE" | tee -a $RUNTIME_AREA/$repo\n'
422	<	txt += '            echo "StageOutCatalog = " | tee -a $RUNTIME_AREA/$repo\n'
423	<	txt += '            echo "output copied into $SE/$SE_PATH directory"\n'
424	<	txt += '            echo "StageOutExitStatus = 0" | tee -a $RUNTIME_AREA/$repo\n'
425	<	txt += '            echo "lcg-cp succeeded"\n'
426	<	txt += '         fi\n'
427	<	txt += '     done\n'
428	<	return txt
429	<
430	<	def wsRegisterOutput(self):
431	<	"""
432	<	Returns part of a job script which does scheduler-specific work.
433	<	"""
434	<
435	<	txt = ''
436	<	if int(self.register_data) == 1:
437	<	## OLI_Daniele deactivate for OSG (wait for LCG UI installed on OSG)
438	<	txt += 'if [ $middleware == OSG ]; then\n'
439	<	txt += '   #\n'
440	<	txt += '   #   Register output to LFC deactivated in OSG mode\n'
441	<	txt += '   #\n'
442	<	txt += '   echo "Register output to LFC deactivated in OSG mode"\n'
443	<	txt += 'elif [ $middleware == LCG ]; then \n'
444	<	txt += '#\n'
445	<	txt += '#  Register output to LFC\n'
446	<	txt += '#\n'
447	<	txt += '   if [ $copy_exit_status -eq 0 ]; then\n'
448	<	txt += '      for out_file in $file_list ; do\n'
449	<	txt += '         echo "Trying to register the output file into LFC"\n'
450	<	txt += '         echo "lcg-rf -l $LFN/$out_file --vo $VO -t 1200 sfn://$SE$SE_PATH/$out_file 2>&1"\n'
451	<	txt += '         lcg-rf -l $LFN/$out_file --vo $VO -t 1200 sfn://$SE$SE_PATH/$out_file 2>&1 \n'
452	<	txt += '         register_exit_status=$?\n'
453	<	txt += '         echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
454	<	txt += '         echo "STAGE_OUT = $register_exit_status"\n'
455	<	txt += '         if [ $register_exit_status -ne 0 ]; then \n'
456	<	txt += '            echo "Problems with the registration to LFC" \n'
457	<	txt += '            echo "Try with srm protocol" \n'
458	<	txt += '            echo "lcg-rf -l $LFN/$out_file --vo $VO -t 1200 srm://$SE$SE_PATH/$out_file 2>&1"\n'
459	<	txt += '            lcg-rf -l $LFN/$out_file --vo $VO -t 1200 srm://$SE$SE_PATH/$out_file 2>&1 \n'
460	<	txt += '            register_exit_status=$?\n'
461	<	txt += '            echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
462	<	txt += '            echo "STAGE_OUT = $register_exit_status"\n'
463	<	txt += '            if [ $register_exit_status -ne 0 ]; then \n'
464	<	txt += '               echo "Problems with the registration into LFC" \n'
465	<	txt += '            fi \n'
466	<	txt += '         else \n'
467	<	txt += '            echo "output registered to LFC"\n'
468	<	txt += '         fi \n'
469	<	txt += '         echo "StageOutExitStatus = $register_exit_status" | tee -a $RUNTIME_AREA/$repo\n'
470	<	txt += '      done\n'
471	<	txt += '   else \n'
472	<	txt += '      echo "Trying to copy output file to CloseSE"\n'
473	<	txt += '      CLOSE_SE=`glite-brokerinfo getCloseSEs | head -1`\n'
474	<	txt += '      for out_file in $file_list ; do\n'
475	<	txt += '         echo "lcg-cr -v -l lfn:${LFN}/$out_file -d $CLOSE_SE -P $LFN/$out_file --vo $VO file://$RUNTIME_AREA/$out_file 2>&1" \n'
476	<	txt += '         lcg-cr -v -l lfn:${LFN}/$out_file -d $CLOSE_SE -P $LFN/$out_file --vo $VO file://$RUNTIME_AREA/$out_file 2>&1 \n'
477	<	txt += '         register_exit_status=$?\n'
478	<	txt += '         echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
479	<	txt += '         echo "STAGE_OUT = $register_exit_status"\n'
480	<	txt += '         if [ $register_exit_status -ne 0 ]; then \n'
481	<	txt += '            echo "Problems with CloseSE or Catalog" \n'
482	<	txt += '         else \n'
483	<	txt += '            echo "The program was successfully executed"\n'
484	<	txt += '            echo "SE = $CLOSE_SE"\n'
485	<	txt += '            echo "LFN for the file is LFN=${LFN}/$out_file"\n'
486	<	txt += '         fi \n'
487	<	txt += '         echo "StageOutExitStatus = $register_exit_status" | tee -a $RUNTIME_AREA/$repo\n'
488	<	txt += '      done\n'
489	<	txt += '   fi \n'
490	<	txt += '   exit_status=$register_exit_status\n'
491	<	txt += 'fi \n'
492	<	return txt
493	<
235	>
236		def loggingInfo(self, id):
237		"""
238		retrieve the logging info from logging and bookkeeping and return it
239		"""
240		self.checkProxy()
241	<	cmd = 'glite-job-logging-info -v 2 ' + id
500	<	#cmd_out = os.popen(cmd)
241	>	cmd = 'glite-job-logging-info -v 3 ' + id
242		cmd_out = runCommand(cmd)
243		return cmd_out
244
504	–
505	–	def getExitStatus(self, id):
506	–	return self.getStatusAttribute_(id, 'exit_code')
507	–
508	–	def queryStatus(self, id):
509	–	return self.getStatusAttribute_(id, 'status')
510	–
511	–	def queryDest(self, id):
512	–	return self.getStatusAttribute_(id, 'destination')
513	–
514	–
515	–	def getStatusAttribute_(self, id, attr):
516	–	""" Query a status of the job with id """
517	–
518	–	self.checkProxy()
519	–	hstates = {}
520	–	Status = importName('edg_wl_userinterface_common_LbWrapper', 'Status')
521	–	# Bypass edg-job-status interfacing directly to C++ API
522	–	# Job attribute vector to retrieve status without edg-job-status
523	–	level = 0
524	–	# Instance of the Status class provided by LB API
525	–	jobStat = Status()
526	–	st = 0
527	–	jobStat.getStatus(id, level)
528	–	err, apiMsg = jobStat.get_error()
529	–	if err:
530	–	common.logger.debug(5,'Error caught' + apiMsg)
531	–	return None
532	–	else:
533	–	for i in range(len(self.states)):
534	–	# Fill an hash table with all information retrieved from LB API
535	–	hstates[ self.states[i] ] = jobStat.loadStatus(st)[i]
536	–	result = jobStat.loadStatus(st)[ self.states.index(attr) ]
537	–	return result
538	–
245		def queryDetailedStatus(self, id):
246		""" Query a detailed status of the job with id """
247		cmd = 'glite-job-status '+id
248		cmd_out = runCommand(cmd)
249		return cmd_out
250
251	<	##### FEDE ######
252	<	def findSites_(self, n_tot_job):
253	<	itr4 = []
254	<	# print "n_tot_job = ", n_tot_job
255	<	for n in range(n_tot_job):
550	<	sites = common.jobDB.destination(n)
551	<	#job = common.job_list[n]
552	<	#jbt = job.type()
553	<	# print "common.jobDB.destination(n) = ", common.jobDB.destination(n)
554	<	# print "sites = ", sites
555	<	itr = ''
556	<	for site in sites:
557	<	#itr = itr + 'target.GlueSEUniqueID==&quot;'+site+'&quot; || '
558	<	itr = itr + 'target.GlueSEUniqueID=="'+site+'" || '
559	<	pass
560	<	# remove last ||
561	<	itr = itr[0:-4]
562	<	itr4.append( itr )
563	<	# remove last ,
564	<	# print "itr4 = ", itr4
565	<	return itr4
566	<
567	<	def createXMLSchScript(self, nj, argsList):
568	<	# def createXMLSchScript(self, nj):
569	<	"""
570	<	Create a XML-file for BOSS4.
571	<	"""
572	<	#      job = common.job_list[nj]
573	<	"""
574	<	INDY
575	<	[begin] da rivedere:
576	<	in particolare passerei il jobType ed eliminerei le dipendenze da job
577	<	"""
578	<	index = nj - 1
579	<	job = common.job_list[index]
580	<	jbt = job.type()
581	<
582	<	inp_sandbox = jbt.inputSandbox(index)
583	<	out_sandbox = jbt.outputSandbox(index)
584	<	"""
585	<	[end] da rivedere
586	<	"""
587	<
588	<
589	<	title = '<?xml version="1.0" encoding="UTF-8" standalone="no"?>\n'
590	<	jt_string = ''
591	<
592	<	xml_fname = str(self.jobtypeName)+'.xml'
593	<	xml = open(common.work_space.shareDir()+'/'+xml_fname, 'a')
594	<
595	<	#TaskName
596	<	dir = string.split(common.work_space.topDir(), '/')
597	<	taskName = dir[len(dir)-2]
598	<
599	<	to_writeReq = ''
600	<	to_write = ''
601	<
602	<	req=' '
603	<	req = req + jbt.getRequirements(nj)
604	<
605	<
606	<	#sites = common.jobDB.destination(nj)
607	<	#if len(sites)>0 and sites[0]!="Any":
608	<	#    req = req + ' && anyMatch(other.storage.CloseSEs, (_ITR4_))'
609	<	#req = req
610	<
611	<	if self.EDG_requirements:
612	<	if (req == ' '):
613	<	req = req + self.EDG_requirements
614	<	else:
615	<	req = req +  ' && ' + self.EDG_requirements
616	<	if self.EDG_ce_white_list:
617	<	ce_white_list = string.split(self.EDG_ce_white_list,',')
618	<	for i in range(len(ce_white_list)):
619	<	if i == 0:
620	<	if (req == ' '):
621	<	req = req + '((RegExp("' + ce_white_list[i] + '", other.GlueCEUniqueId))'
622	<	else:
623	<	req = req +  ' && ((RegExp("' + ce_white_list[i] + '", other.GlueCEUniqueId))'
624	<	pass
625	<	else:
626	<	req = req +  ' || (RegExp("' + ce_white_list[i] + '", other.GlueCEUniqueId))'
627	<	req = req + ')'
628	<
629	<	if self.EDG_ce_black_list:
630	<	ce_black_list = string.split(self.EDG_ce_black_list,',')
631	<	for ce in ce_black_list:
632	<	if (req == ' '):
633	<	req = req + '(!RegExp("' + ce + '", other.GlueCEUniqueId))'
634	<	else:
635	<	req = req +  ' && (!RegExp("' + ce + '", other.GlueCEUniqueId))'
636	<	pass
637	<	if self.EDG_clock_time:
638	<	if (req == ' '):
639	<	req = req + 'other.GlueCEPolicyMaxWallClockTime>='+self.EDG_clock_time
640	<	else:
641	<	req = req + ' && other.GlueCEPolicyMaxWallClockTime>='+self.EDG_clock_time
642	<
643	<	if self.EDG_cpu_time:
644	<	if (req == ' '):
645	<	req = req + ' other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time
646	<	else:
647	<	req = req + ' && other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time
648	<
649	<	#if (req != ' '):
650	<	#    req = req + '\n'
651	<	#    to_writeReq = req
652	<
653	<	if ( self.EDG_retry_count ):
654	<	to_write = to_write + 'RetryCount = "'+self.EDG_retry_count+'"\n'
655	<	pass
656	<
657	<	to_write = to_write + 'MyProxyServer = "&quot;' + self.proxyServer + '&quot;"\n'
658	<	to_write = to_write + 'VirtualOrganisation = "&quot;' + self.VO + '&quot;"\n'
659	<
660	<
661	<	#TaskName
662	<	dir = string.split(common.work_space.topDir(), '/')
663	<	taskName = dir[len(dir)-2]
664	<
665	<	xml.write(str(title))
666	<	xml.write('<task name="' +str(taskName)+'">\n')
667	<	xml.write(jt_string)
668	<
669	<	xml.write('<iterator>\n')
670	<
671	<	#print str(nj)
672	<	#        xml.write('\t<iteratorRule name="ITR1" rule="1:'+ str(nj) + '" />\n')
673	<	#print argsList
674	<	#        xml.write('\t<iteratorRule name="ITR2" rule="'+ argsList + '" />\n')
675	<	#print jobList
676	<	#        xml.write('\t<iteratorRule name="ITR3" rule="1:'+ str(nj) + ':6" />\n')        #print str(nj)
677	<	xml.write('\t<iteratorRule name="ITR1">\n')
678	<	xml.write('\t\t<ruleElement> 1:'+ str(nj) + ' </ruleElement>\n')
679	<	xml.write('\t</iteratorRule>\n')
680	<	xml.write('\t<iteratorRule name="ITR2">\n')
681	<	#print argsList
682	<	for arg in argsList:
683	<	xml.write('\t\t<ruleElement> <![CDATA[\n'+ arg + '\n\t\t]]> </ruleElement>\n')
684	<	pass
685	<	xml.write('\t</iteratorRule>\n')
686	<	#print jobList
687	<	xml.write('\t<iteratorRule name="ITR3">\n')
688	<	xml.write('\t\t<ruleElement> 1:'+ str(nj) + ':1:6 </ruleElement>\n')
689	<	xml.write('\t</iteratorRule>\n')
690	<
691	<	#### FEDE #####
692	<	'''
693	<	indy: qui sotto ci sta itr4
694	<	'''
695	<
696	<	itr4=self.findSites_(nj)
697	<	#print "--->>> itr4 = ", itr4
698	<	if (itr4 != ''):
699	<	xml.write('\t<iteratorRule name="ITR4">\n')
700	<	#print argsList
701	<	for arg in itr4:
702	<	xml.write('\t\t<ruleElement> <![CDATA[\n'+ arg + '\n\t\t]]> </ruleElement>\n')
703	<	pass
704	<	xml.write('\t</iteratorRule>\n')
705	<	req = req + ' && anyMatch(other.storage.CloseSEs, (_ITR4_))'
706	<	pass
707	<	#    print "--->>> req= ", req
708	<
709	<	if (to_write != ''):
710	<	xml.write('<extraTags\n')
711	<	xml.write(to_write)
712	<	xml.write('/>\n')
713	<	pass
714	<
715	<	xml.write('<chain scheduler="'+str(self.schedulerName)+'">\n')
716	<	xml.write(jt_string)
717	<
718	<	if (req != ' '):
719	<	req = req + '\n'
720	<	xml.write('<extraTags>\n')
721	<	xml.write('<Requirements>\n')
722	<	xml.write('<![CDATA[\n')
723	<	xml.write(req)
724	<	xml.write(']]>\n')
725	<	xml.write('</Requirements>\n')
726	<	xml.write('</extraTags>\n')
727	<	pass
251	>	def findSites_(self, n):
252	>	sites = common.jobDB.destination(n)
253	>	if len(sites)>0 and sites[0]=="":
254	>	return []
255	>	return sites
256
257	<	#executable
257	>	def submitTout(self, list):
258	>	return 120
259
731	–	"""
732	–	INDY
733	–	script dipende dal jobType: dovrebbe essere semplice tirarlo fuori in altro modo
734	–	"""
735	–	script = job.scriptFilename()
736	–	xml.write('<program>\n')
737	–	xml.write('<exec> ' + os.path.basename(script) +' </exec>\n')
738	–	xml.write(jt_string)
739	–
740	–
741	–	### only one .sh  JDL has arguments:
742	–	### Fabio
743	–	#        xml.write('args = "' + str(nj+1)+' '+ jbt.getJobTypeArguments(nj, "EDG") +'"\n')
744	–	xml.write('<args> <![CDATA[\n _ITR2_ \n]]> </args>\n')
745	–	xml.write('<program_types> crabjob </program_types>\n')
746	–	inp_box = script + ','
747	–
748	–	if inp_sandbox != None:
749	–	for fl in inp_sandbox:
750	–	inp_box = inp_box + '' + fl + ','
751	–	pass
752	–	pass
753	–
754	–	# Marco (VERY TEMPORARY ML STUFF)
755	–	inp_box = inp_box + os.path.abspath(os.environ['CRABDIR']+'/python/'+'report.py') + ',' +\
756	–	os.path.abspath(os.environ['CRABDIR']+'/python/'+'DashboardAPI.py') + ','+\
757	–	os.path.abspath(os.environ['CRABDIR']+'/python/'+'Logger.py') + ','+\
758	–	os.path.abspath(os.environ['CRABDIR']+'/python/'+'ProcInfo.py') + ','+\
759	–	os.path.abspath(os.environ['CRABDIR']+'/python/'+'apmon.py')
760	–	# End Marco
761	–
762	–	if (not jbt.additional_inbox_files == []):
763	–	inp_box = inp_box + ', '
764	–	for addFile in jbt.additional_inbox_files:
765	–	addFile = os.path.abspath(addFile)
766	–	inp_box = inp_box+''+addFile+','
767	–	pass
768	–
769	–	if inp_box[-1] == ',' : inp_box = inp_box[:-1]
770	–	inp_box = '<infiles> <![CDATA[\n' + inp_box + '\n]]> </infiles>\n'
771	–	xml.write(inp_box)
772	–
773	–	base = jbt.name()
774	–	stdout = base + '__ITR3_.stdout'
775	–	stderr = base + '__ITR3_.stderr'
776	–
777	–	xml.write('<stderr> ' + stderr + '</stderr>\n')
778	–	xml.write('<stdout> ' + stdout + '</stdout>\n')
779	–
780	–
781	–	out_box = stdout + ',' + \
782	–	stderr + ',.BrokerInfo,'
783	–
784	–	"""
785	–	if int(self.return_data) == 1:
786	–	if out_sandbox != None:
787	–	for fl in out_sandbox:
788	–	out_box = out_box + '' + fl + ','
789	–	pass
790	–	pass
791	–	pass
792	–	"""
793	–
794	–	"""
795	–	INDY
796	–	qualcosa del genere andrebbe fatta per gli infiles
797	–	"""
798	–	if int(self.return_data) == 1:
799	–	for fl in jbt.output_file:
800	–	out_box = out_box + '' + jbt.numberFile_(fl, '_ITR1_') + ','
801	–	pass
802	–	pass
803	–
804	–	if out_box[-1] == ',' : out_box = out_box[:-1]
805	–	out_box = '<outfiles> <![CDATA[\n' + out_box + '\n]]></outfiles>\n'
806	–	xml.write(out_box)
807	–
808	–	xml.write('<BossAttr> crabjob.INTERNAL_ID=_ITR1_ </BossAttr>\n')
809	–
810	–	xml.write('</program>\n')
811	–	xml.write('</chain>\n')
812	–
813	–	xml.write('</iterator>\n')
814	–	xml.write('</task>\n')
815	–
816	–	xml.close()
817	–	return
818	–
819	–	def checkProxy(self):
820	–	"""
821	–	Function to check the Globus proxy.
822	–	"""
823	–	if (self.proxyValid): return
824	–	timeleft = -999
825	–	minTimeLeft=10*3600 # in seconds
826	–
827	–	minTimeLeftServer = 100 # in hours
828	–
829	–	mustRenew = 0
830	–	timeLeftLocal = runCommand('voms-proxy-info -timeleft 2>/dev/null')
831	–	timeLeftServer = -999
832	–	if not timeLeftLocal or int(timeLeftLocal) <= 0 or not isInt(timeLeftLocal):
833	–	mustRenew = 1
834	–	else:
835	–	timeLeftServer = runCommand('voms-proxy-info -actimeleft 2>/dev/null | head -1')
836	–	if not timeLeftServer or not isInt(timeLeftServer):
837	–	mustRenew = 1
838	–	elif timeLeftLocal<minTimeLeft or timeLeftServer<minTimeLeft:
839	–	mustRenew = 1
840	–	pass
841	–	pass
842	–
843	–	if mustRenew:
844	–	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")
845	–	cmd = 'voms-proxy-init -voms cms -valid 96:00'
846	–	try:
847	–	# SL as above: damn it!
848	–	out = os.system(cmd)
849	–	if (out>0): raise CrabException("Unable to create a valid proxy!\n")
850	–	except:
851	–	msg = "Unable to create a valid proxy!\n"
852	–	raise CrabException(msg)
853	–	pass
854	–
855	–	## now I do have a voms proxy valid, and I check the myproxy server
856	–	renewProxy = 0
857	–	cmd = 'myproxy-info -d -s '+self.proxyServer
858	–	cmd_out = runCommand(cmd,0,20)
859	–	if not cmd_out:
860	–	common.logger.message('No credential delegated to myproxy server '+self.proxyServer+' will do now')
861	–	renewProxy = 1
862	–	else:
863	–	# if myproxy exist but not long enough, renew
864	–	reTime = re.compile( r'timeleft: (\d+)' )
865	–	if reTime.match( cmd_out ):
866	–	time = reTime.search( line ).group(1)
867	–	if time < minTimeLeftServer:
868	–	renewProxy = 1
869	–	common.logger.message('No credential delegation will expire in '+time+' hours: renew it')
870	–	pass
871	–	pass
872	–
873	–	# if not, create one.
874	–	if renewProxy:
875	–	cmd = 'myproxy-init -d -n -s '+self.proxyServer
876	–	out = os.system(cmd)
877	–	if (out>0):
878	–	raise CrabException("Unable to delegate the proxy to myproxyserver "+self.proxyServer+" !\n")
879	–	pass
880	–
881	–	# cache proxy validity
882	–	self.proxyValid=1
883	–	return
884	–
885	–	def configOpt_(self):
886	–	edg_ui_cfg_opt = ' '
887	–	if self.edg_config:
888	–	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
889	–	if self.edg_config_vo:
890	–	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
891	–	return edg_ui_cfg_opt

Diff Legend

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