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 1.8 by slacapra, Tue Aug 23 10:38:09 2005 UTC vs.
Revision 1.19 by slacapra, Tue Oct 18 14:11:12 2005 UTC

#	Line 4 | Line 4 | from crab_exceptions import *
4		from crab_util import *
5		import common
6
7	<	import os, sys, tempfile
7	>	import os, sys, time
8
9		class SchedulerEdg(Scheduler):
10		def __init__(self):
#	Line 21 | Line 21 | class SchedulerEdg(Scheduler):
21
22		def configure(self, cfg_params):
23
24	<	try: self.edg_ui_cfg = cfg_params["EDG.rb_config"]
25	<	except KeyError: self.edg_ui_cfg = ''
26	<
27	<	try:
28	<	self.edg_config = cfg_params["EDG.config"]
29	<	if not os.path.isfile(self.edg_config):
30	<	raise CrabException("File EDG.config "+self.edg_config+" does not exist")
24	>	try: self.edg_config = cfg_params["EDG.config"]
25		except KeyError: self.edg_config = ''
26
27	<	try:
34	<	self.edg_config_vo = cfg_params["EDG.config_vo"]
35	<	if not os.path.isfile(self.edg_config_vo):
36	<	raise CrabException("File EDG.config_vo "+self.edg_config_vo+" does not exist")
27	>	try: self.edg_config_vo = cfg_params["EDG.config_vo"]
28		except KeyError: self.edg_config_vo = ''
29
30		try: self.LCG_version = cfg_params["EDG.lcg_version"]
#	Line 45 | Line 36 | class SchedulerEdg(Scheduler):
36		try: self.EDG_retry_count = cfg_params['EDG.retry_count']
37		except KeyError: self.EDG_retry_count = ''
38
39	<	try:
40	<	self.VO = cfg_params['EDG.virtual_organization']
41	<	except KeyError:
42	<	self.VO = 'cms'
39	>	try: self.VO = cfg_params['EDG.virtual_organization']
40	>	except KeyError: self.VO = 'cms'
41	>
42	>	try: self.return_data = cfg_params['USER.return_data']
43	>	except KeyError: self.return_data = ''
44	>
45	>	try:
46	>	self.copy_data = cfg_params["USER.copy_data"]
47	>	try:
48	>	self.SE = cfg_params['USER.storage_element']
49	>	self.SE_PATH = cfg_params['USER.storage_path']
50	>	except KeyError:
51	>	msg = "Error. The [USER] section does not have 'storage_element'"
52	>	msg = msg + " and/or 'storage_path' entries, necessary to copy the output"
53	>	common.logger.message(msg)
54	>	raise CrabException(msg)
55	>	except KeyError: self.copy_data = ''
56	>
57	>	try:
58	>	self.register_data = cfg_params["USER.register_data"]
59	>	try:
60	>	self.LFN = cfg_params['USER.lfn_dir']
61	>	except KeyError:
62	>	msg = "Error. The [USER] section does not have 'lfn_dir' value"
63	>	msg = msg + " it's necessary for RLS registration"
64	>	common.logger.message(msg)
65	>	raise CrabException(msg)
66	>	except KeyError: self.register_data= ''
67	>
68	>	try: self.EDG_requirements = cfg_params['EDG.requirements']
69	>	except KeyError: self.EDG_requirements = ''
70	>
71	>	try: self.EDG_retry_count = cfg_params['EDG.retry_count']
72	>	except KeyError: self.EDG_retry_count = ''
73	>
74	>	try: self.EDG_clock_time = cfg_params['EDG.max_wall_clock_time']
75	>	except KeyError: self.EDG_clock_time= ''
76	>
77	>	try: self.EDG_cpu_time = cfg_params['EDG.max_cpu_time']
78	>	except KeyError: self.EDG_cpu_time = ''
79
80		# Add EDG_WL_LOCATION to the python path
81
#	Line 63 | Line 90 | class SchedulerEdg(Scheduler):
90		libPath=os.path.join(path, "lib", "python")
91		sys.path.append(libPath)
92
93	<	self.checkProxy_()
93	>	self.proxyValid=0
94		return
95
96	+
97	+	def sched_parameter(self):
98	+	"""
99	+	Returns file with scheduler-specific parameters
100	+	"""
101	+
102	+	if (self.edg_config and self.edg_config_vo != ''):
103	+	self.param='sched_param.clad'
104	+	param_file = open(common.work_space.shareDir()+'/'+self.param, 'w')
105	+	param_file.write('RBconfig = "'+self.edg_config+'";\n')
106	+	param_file.write('RBconfigVO = "'+self.edg_config_vo+'";')
107	+	param_file.close()
108	+	return 1
109	+	else:
110	+	return 0
111	+
112		def wsSetupEnvironment(self):
113		"""
114		Returns part of a job script which does scheduler-specific work.
115		"""
116	<	txt = '\n'
116	>
117	>	txt = ''
118	>	if self.copy_data:
119	>	if self.SE:
120	>	txt += 'export SE='+self.SE+'\n'
121	>	txt += 'echo "SE = $SE"\n'
122	>	if self.SE_PATH:
123	>	if ( self.SE_PATH[-1] != '/' ) : self.SE_PATH = self.SE_PATH + '/'
124	>	txt += 'export SE_PATH='+self.SE_PATH+'\n'
125	>	txt += 'echo "SE_PATH = $SE_PATH"\n'
126	>
127	>	if self.register_data:
128	>	if self.VO:
129	>	txt += 'export VO='+self.VO+'\n'
130	>	if self.LFN:
131	>	txt += 'export LFN='+self.LFN+'\n'
132	>	txt += '\n'
133		txt += 'CloseCEs=`edg-brokerinfo getCE`\n'
134		txt += 'echo "CloseCEs = $CloseCEs"\n'
135		txt += 'CE=`echo $CloseCEs | sed -e "s/:.*//"`\n'
136		txt += 'echo "CE = $CE"\n'
137		return txt
138
139	+	def wsCopyOutput(self):
140	+	"""
141	+	Write a CopyResults part of a job script, e.g.
142	+	to copy produced output into a storage element.
143	+	"""
144	+	txt = ''
145	+	if self.copy_data:
146	+	copy = 'globus-url-copy file://`pwd`/$out_file gsiftp://${SE}${SE_PATH}$out_file'
147	+	txt += '#\n'
148	+	txt += '#   Copy output to SE = $SE\n'
149	+	txt += '#\n'
150	+	#### per orca l'exit_status non e' affidabile.....
151	+	#txt += 'if [ $executable_exit_status -eq 0 ]; then\n'
152	+	txt += 'if [ $exe_result -eq 0 ]; then\n'
153	+	txt += '  for out_file in $file_list ; do\n'
154	+	txt += '    echo "Trying to copy output file to $SE "\n'
155	+	txt += '    echo "'+copy+'"\n'
156	+	txt += '    '+copy+' 2>&1\n'
157	+	txt += '    copy_exit_status=$?\n'
158	+	txt += '    echo "COPY_EXIT_STATUS = $copy_exit_status"\n'
159	+	txt += '    echo "STAGE_OUT = $copy_exit_status"\n'
160	+	txt += '    if [ $copy_exit_status -ne 0 ]; then \n'
161	+	txt += '       echo "Problems with SE= $SE" \n'
162	+	txt += '    else \n'
163	+	txt += '       echo "output copied into $SE/$SE_PATH directory"\n'
164	+	txt += '    fi \n'
165	+	txt += '  done\n'
166	+	txt += 'fi \n'
167	+	return txt
168	+
169	+	def wsRegisterOutput(self):
170	+	"""
171	+	Returns part of a job script which does scheduler-specific work.
172	+	"""
173	+
174	+	txt = ''
175	+	if self.register_data:
176	+	txt += '#\n'
177	+	txt += '#  Register output to RLS\n'
178	+	txt += '#\n'
179	+	### analogo
180	+	#txt += 'if [[ $executable_exit_status -eq 0 && $copy_exit_status -eq 0 ]]; then\n'
181	+	txt += 'if [[ $exe_result -eq 0 && $copy_exit_status -eq 0 ]]; then\n'
182	+	txt += '   for out_file in $file_list ; do\n'
183	+	txt += '      echo "Trying to register the output file into RLS"\n'
184	+	txt += '      echo "lcg-rf -l $LFN/$out_file --vo $VO sfn://$SE$SE_PATH/$out_file"\n'
185	+	txt += '      lcg-rf -l $LFN/$out_file --vo $VO sfn://$SE$SE_PATH/$out_file 2>&1 \n'
186	+	txt += '      register_exit_status=$?\n'
187	+	txt += '      echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
188	+	txt += '      echo "STAGE_OUT = $register_exit_status"\n'
189	+	txt += '      if [ $register_exit_status -ne 0 ]; then \n'
190	+	txt += '         echo "Problems with the registration to RLS" \n'
191	+	txt += '         echo "Try with srm protocol" \n'
192	+	txt += '         echo "lcg-rf -l $LFN/$out_file --vo $VO srm://$SE$SE_PATH/$out_file"\n'
193	+	txt += '         lcg-rf -l $LFN/$out_file --vo $VO srm://$SE$SE_PATH/$out_file 2>&1 \n'
194	+	txt += '         register_exit_status=$?\n'
195	+	txt += '         echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
196	+	txt += '         echo "STAGE_OUT = $register_exit_status"\n'
197	+	txt += '         if [ $register_exit_status -ne 0 ]; then \n'
198	+	txt += '            echo "Problems with the registration into RLS" \n'
199	+	txt += '         fi \n'
200	+	txt += '      else \n'
201	+	txt += '         echo "output registered to RLS"\n'
202	+	txt += '      fi \n'
203	+	txt += '   done\n'
204	+	txt += 'elif [[ $exe_result -eq 0 && $copy_exit_status -ne 0 ]]; then \n'
205	+	txt += '   echo "Trying to copy output file to CloseSE"\n'
206	+	txt += '   CLOSE_SE=`edg-brokerinfo getCloseSEs | head -1`\n'
207	+	txt += '   for out_file in $file_list ; do\n'
208	+	txt += '      echo "lcg-cr -v -l lfn:${LFN}/$out_file -d $SE -P $LFN/$out_file --vo $VO file://`pwd`/$out_file" \n'
209	+	txt += '      lcg-cr -v -l lfn:${LFN}/$out_file -d $SE -P $LFN/$out_file --vo $VO file://`pwd`/$out_file 2>&1 \n'
210	+	txt += '      register_exit_status=$?\n'
211	+	txt += '      echo "REGISTER_EXIT_STATUS = $register_exit_status"\n'
212	+	txt += '      echo "STAGE_OUT = $register_exit_status"\n'
213	+	txt += '      if [ $register_exit_status -ne 0 ]; then \n'
214	+	txt += '         echo "Problems with CloseSE" \n'
215	+	txt += '      else \n'
216	+	txt += '         echo "The program was successfully executed"\n'
217	+	txt += '         echo "SE = $CLOSE_SE"\n'
218	+	txt += '         echo "LFN for the file is LFN=${LFN}/$out_file"\n'
219	+	txt += '      fi \n'
220	+	txt += '   done\n'
221	+	txt += 'else\n'
222	+	txt += '   echo "Problem with the executable"\n'
223	+	txt += 'fi \n'
224	+	return txt
225	+	#####################
226	+
227		def loggingInfo(self, nj):
228		"""
229		retrieve the logging info from logging and bookkeeping and return it
230		"""
231	+	self.checkProxy()
232		id = common.jobDB.jobId(nj)
233	<	edg_ui_cfg_opt = ''
86	<	if self.edg_config:
87	<	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
88	<	cmd = 'edg-job-get-logging-info -v 2 ' + edg_ui_cfg_opt + id
89	<	print cmd
233	>	cmd = 'edg-job-get-logging-info -v 2 ' + self.configOpt_() + id
234		myCmd = os.popen(cmd)
235		cmd_out = myCmd.readlines()
236		myCmd.close()
#	Line 96 | Line 240 | class SchedulerEdg(Scheduler):
240		"""
241		Check the compatibility of available resources
242		"""
243	+	self.checkProxy()
244		jdl = common.job_list[nj].jdlFilename()
245	<	#print ' jdl ',nj, jdl
101	<	edg_ui_cfg_opt = ''
102	<	if self.edg_config:
103	<	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
104	<	if self.edg_config_vo:
105	<	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
106	<	cmd = 'edg-job-list-match ' + edg_ui_cfg_opt + jdl
245	>	cmd = 'edg-job-list-match ' + self.configOpt_() + jdl
246		myCmd = os.popen(cmd)
247		cmd_out = myCmd.readlines()
248		myCmd.close()
#	Line 121 | Line 260 | class SchedulerEdg(Scheduler):
260		Match=0
261		for line in out:
262		line = line.strip()
124	–	#print line
263		if reComment.match( line ):
264		next = 0
265		continue
#	Line 175 | Line 313 | class SchedulerEdg(Scheduler):
313		Submit one EDG job.
314		"""
315
316	+	self.checkProxy()
317		jid = None
318		jdl = common.job_list[nj].jdlFilename()
319	<	id_tmp = tempfile.mktemp()
320	<	edg_ui_cfg_opt = ' '
182	<	if self.edg_config:
183	<	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
184	<	if self.edg_config_vo:
185	<	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
186	<	cmd = 'edg-job-submit -o ' + id_tmp + edg_ui_cfg_opt + jdl
319	>
320	>	cmd = 'edg-job-submit ' + self.configOpt_() + jdl
321		cmd_out = runCommand(cmd)
322		if cmd_out != None:
323	<	idfile = open(id_tmp)
324	<	jid_line = idfile.readline()
191	<	while jid_line[0] == '#':
192	<	jid_line = idfile.readline()
193	<	pass
194	<	jid = string.strip(jid_line)
195	<	os.unlink(id_tmp)
323	>	reSid = re.compile( r'https.+' )
324	>	jid = reSid.search(cmd).group()
325		pass
326		return jid
327
328		def getExitStatus(self, id):
329		return self.getStatusAttribute_(id, 'exit_code')
330	+
331		def queryStatus(self, id):
332		return self.getStatusAttribute_(id, 'status')
333
334	+	def queryDest(self, id):
335	+	return self.getStatusAttribute_(id, 'destination')
336	+
337	+
338		def getStatusAttribute_(self, id, attr):
339		""" Query a status of the job with id """
340
341	+	self.checkProxy()
342		hstates = {}
343		Status = importName('edg_wl_userinterface_common_LbWrapper', 'Status')
344		# Bypass edg-job-status interfacing directly to C++ API
#	Line 220 | Line 355 | class SchedulerEdg(Scheduler):
355		return None
356		else:
357		for i in range(len(self.states)):
223	–	#print "states = ", states
358		# Fill an hash table with all information retrieved from LB API
359		hstates[ self.states[i] ] = jobStat.loadStatus(st)[i]
360		result = jobStat.loadStatus(st)[ self.states.index(attr) ]
#	Line 238 | Line 372 | class SchedulerEdg(Scheduler):
372		Returns the name of directory with results.
373		"""
374
375	+	self.checkProxy()
376		cmd = 'edg-job-get-output --dir ' + common.work_space.resDir() + ' ' + id
377		cmd_out = runCommand(cmd)
378
#	Line 249 | Line 384 | class SchedulerEdg(Scheduler):
384
385		def cancel(self, id):
386		""" Cancel the EDG job with id """
387	+	self.checkProxy()
388		cmd = 'edg-job-cancel --noint ' + id
389		cmd_out = runCommand(cmd)
390		return cmd_out
391
392	<	def checkProxy_(self):
257	<	"""
258	<	Function to check the Globus proxy.
259	<	"""
260	<	cmd = 'grid-proxy-info -timeleft'
261	<	cmd_out = runCommand(cmd)
262	<	ok = 1
263	<	timeleft = -999
264	<	try: timeleft = int(cmd_out)
265	<	except ValueError: ok=0
266	<	except TypeError: ok=0
267	<	if timeleft < 1:  ok=0
268	<
269	<	if ok==0:
270	<	msg = 'No valid proxy found !\n'
271	<	msg += "Please do 'grid-proxy-init'."
272	<	raise CrabException(msg)
273	<	return
274	<
275	<	def createJDL(self, nj):
392	>	def createSchScript(self, nj):
393		"""
394		Create a JDL-file for EDG.
395		"""
396
397		job = common.job_list[nj]
398		jbt = job.type()
282	–	#        jbt.loadJobInfo()
399		inp_sandbox = jbt.inputSandbox(nj)
400		out_sandbox = jbt.outputSandbox(nj)
401	<	inp_storage_subdir = ''#jbt.inputStorageSubdir()
401	>	inp_storage_subdir = ''
402
403		title = '# This JDL was generated by '+\
404		common.prog_name+' (version '+common.prog_version_str+')\n'
405		jt_string = ''
406	+
407	+
408
409		SPL = inp_storage_subdir
410		if ( SPL and SPL[-1] != '/' ) : SPL = SPL + '/'
#	Line 299 | Line 417 | class SchedulerEdg(Scheduler):
417		jdl.write('Executable = "' + os.path.basename(script) +'";\n')
418		jdl.write(jt_string)
419
420	+	### only one .sh  JDL has arguments:
421		firstEvent = common.jobDB.firstEvent(nj)
422		maxEvents = common.jobDB.maxEvents(nj)
423		jdl.write('Arguments = "' + str(nj+1)+' '+str(firstEvent)+' '+str(maxEvents)+'";\n')
#	Line 326 | Line 445 | class SchedulerEdg(Scheduler):
445
446		jdl.write('StdOutput     = "' + job.stdout() + '";\n')
447		jdl.write('StdError      = "' + job.stderr() + '";\n')
448	<
449	<	#if common.flag_return_data :
450	<	#    for fl in job.outputDataFiles():
451	<	#        out_box = out_box + ' "' + fl + '",'
452	<	#        pass
453	<	#    pass
454	<
455	<	out_box = 'OutputSandbox = { '
456	<	if out_sandbox != None:
457	<	for fl in out_sandbox:
458	<	out_box = out_box + ' "' + fl + '",'
448	>
449	>
450	>	if job.stdout() == job.stderr():
451	>	out_box = 'OutputSandbox = { "' + \
452	>	job.stdout() + '", ".BrokerInfo",'
453	>	else:
454	>	out_box = 'OutputSandbox = { "' + \
455	>	job.stdout() + '", "' + \
456	>	job.stderr() + '", ".BrokerInfo",'
457	>
458	>	if self.return_data :
459	>	if out_sandbox != None:
460	>	for fl in out_sandbox:
461	>	out_box = out_box + ' "' + fl + '",'
462	>	pass
463		pass
464		pass
465	<
465	>
466		if out_box[-1] == ',' : out_box = out_box[:-1]
467		out_box = out_box + ' };'
468		jdl.write(out_box+'\n')
469
470	<	# If CloseCE is used ...
348	<	#if common.flag_usecloseCE and job.inputDataFiles():
349	<	#    indata = 'InputData = { '
350	<	#    for fl in job.inputDataFiles():
351	<	#       indata = indata + ' "lfn:' + SPL + fl + '",'
352	<	#    if indata[-1] == ',' : indata = indata[:-1]
353	<	#    indata = indata + ' };'
354	<	#    jdl.write(indata+'\n')
355	<	#    jdl.write('DataAccessProtocol = { "gsiftp" };\n')
356	<
470	>	### if at least a CE exists ...
471		if common.analisys_common_info['sites']:
472	<	if common.analisys_common_info['sw_version']:
473	<
474	<	req='Requirements = '
475	<	### First ORCA version
476	<	req=req + 'Member("VO-cms-' + \
477	<	common.analisys_common_info['sw_version'] + \
478	<	'", other.GlueHostApplicationSoftwareRunTimeEnvironment)'
479	<	## then sites
480	<	if len(common.analisys_common_info['sites'])>0:
481	<	req = req + ' && ('
482	<	for i in range(len(common.analisys_common_info['sites'])):
483	<	req = req + 'other.GlueCEInfoHostName == "' \
484	<	+ common.analisys_common_info['sites'][i] + '"'
485	<	if ( i < (int(len(common.analisys_common_info['sites']) - 1)) ):
486	<	req = req + ' || '
487	<	req = req + ')'
488	<	## then user requirement
489	<	if self.EDG_requirements:
490	<	req = req +  ' && ' + self.EDG_requirements
491	<	req = req + ';\n'
492	<	jdl.write(req)
493	<
472	>	if common.analisys_common_info['sw_version']:
473	>	req='Requirements = '
474	>	req=req + 'Member("VO-cms-' + \
475	>	common.analisys_common_info['sw_version'] + \
476	>	'", other.GlueHostApplicationSoftwareRunTimeEnvironment)'
477	>	if len(common.analisys_common_info['sites'])>0:
478	>	req = req + ' && ('
479	>	for i in range(len(common.analisys_common_info['sites'])):
480	>	req = req + 'other.GlueCEInfoHostName == "' \
481	>	+ common.analisys_common_info['sites'][i] + '"'
482	>	if ( i < (int(len(common.analisys_common_info['sites']) - 1)) ):
483	>	req = req + ' || '
484	>	req = req + ')'
485	>
486	>	#### and USER REQUIREMENT
487	>	if self.EDG_requirements:
488	>	req = req +  ' && ' + self.EDG_requirements
489	>	if self.EDG_clock_time:
490	>	req = req + ' && other.GlueCEPolicyMaxWallClockTime>='+self.EDG_clock_time
491	>	if self.EDG_cpu_time:
492	>	req = req + ' && other.GlueCEPolicyMaxCPUTime>='+self.EDG_cpu_time
493	>	req = req + ';\n'
494	>	jdl.write(req)
495	>
496		jdl.write('VirtualOrganisation = "' + self.VO + '";\n')
497
498		if ( self.EDG_retry_count ):
#	Line 385 | Line 501 | class SchedulerEdg(Scheduler):
501
502		jdl.close()
503		return
504	+
505	+	def checkProxy(self):
506	+	"""
507	+	Function to check the Globus proxy.
508	+	"""
509	+	if (self.proxyValid): return
510	+	timeleft = -999
511	+	minTimeLeft=10 # in hours
512	+	cmd = 'grid-proxy-info -e -v '+str(minTimeLeft)+':00'
513	+	try: cmd_out = runCommand(cmd,0)
514	+	except: print cmd_out
515	+	if (cmd_out == None or cmd_out=='1'):
516	+	common.logger.message( "No valid proxy found or timeleft too short!\n Creating a user proxy with default length of 100h\n")
517	+	cmd = 'grid-proxy-init -valid 100:00'
518	+	try:
519	+	out = os.system(cmd)
520	+	if (out>0): raise CrabException("Unable to create a valid proxy!\n")
521	+	except:
522	+	msg = "Unable to create a valid proxy!\n"
523	+	raise CrabException(msg)
524	+	cmd = 'grid-proxy-info -timeleft'
525	+	cmd_out = runCommand(cmd,0)
526	+	print cmd_out, time.time()
527	+	#time.time(cms_out)
528	+	pass
529	+	self.proxyValid=1
530	+	return
531	+
532	+	def configOpt_(self):
533	+	edg_ui_cfg_opt = ' '
534	+	if self.edg_config:
535	+	edg_ui_cfg_opt = ' -c ' + self.edg_config + ' '
536	+	if self.edg_config_vo:
537	+	edg_ui_cfg_opt += ' --config-vo ' + self.edg_config_vo + ' '
538	+	return edg_ui_cfg_opt

Diff Legend

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