Context Navigation

source: trunk/jobmond/jobmond.py @ 354

Last change on this file since 354 was 354, checked in by bastiaans, 17 years ago

jobmond/jobmond.py:

code cleanup and rearrangement

Property svn:keywords set to Id

File size: 19.7 KB

Rev	Line
[23]	1	#!/usr/bin/env python
[225]	2	#
	3	# This file is part of Jobmonarch
	4	#
	5	# Copyright (C) 2006 Ramon Bastiaans
	6	#
	7	# Jobmonarch is free software; you can redistribute it and/or modify
	8	# it under the terms of the GNU General Public License as published by
	9	# the Free Software Foundation; either version 2 of the License, or
	10	# (at your option) any later version.
	11	#
	12	# Jobmonarch is distributed in the hope that it will be useful,
	13	# but WITHOUT ANY WARRANTY; without even the implied warranty of
	14	# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	15	# GNU General Public License for more details.
	16	#
	17	# You should have received a copy of the GNU General Public License
	18	# along with this program; if not, write to the Free Software
	19	# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
	20	#
[228]	21	# SVN $Id: jobmond.py 354 2007-05-03 12:47:18Z bastiaans $
[227]	22	#
[23]	23
[212]	24	import sys, getopt, ConfigParser
[354]	25	import time, os, socket, string, re
[318]	26	import xml, xml.sax
	27	from xml.sax import saxutils, make_parser
	28	from xml.sax import make_parser
	29	from xml.sax.handler import feature_namespaces
	30
[307]	31	def usage():
	32
	33	print
	34	print 'usage: jobmond [options]'
	35	print 'options:'
	36	print ' --config, -c configuration file'
	37	print ' --pidfile, -p pid file'
	38	print ' --help, -h help'
	39	print
	40
[212]	41	def processArgs( args ):
[26]	42
[354]	43	SHORT_L = 'hc:'
	44	LONG_L = [ 'help', 'config=' ]
[165]	45
[307]	46	global PIDFILE
[354]	47	PIDFILE = None
[61]	48
[354]	49	config_filename = '/etc/jobmond.conf'
	50
[212]	51	try:
[68]	52
[354]	53	opts, args = getopt.getopt( args, SHORT_L, LONG_L )
[185]	54
[307]	55	except getopt.GetoptError, detail:
[212]	56
	57	print detail
[307]	58	usage()
[354]	59	sys.exit( 1 )
[212]	60
	61	for opt, value in opts:
	62
	63	if opt in [ '--config', '-c' ]:
	64
[354]	65	config_filename = value
[212]	66
[307]	67	if opt in [ '--pidfile', '-p' ]:
[212]	68
[354]	69	PIDFILE = value
[307]	70
	71	if opt in [ '--help', '-h' ]:
	72
	73	usage()
[354]	74	sys.exit( 0 )
[212]	75
	76	return loadConfig( config_filename )
	77
	78	def loadConfig( filename ):
	79
[215]	80	def getlist( cfg_string ):
	81
	82	my_list = [ ]
	83
	84	for item_txt in cfg_string.split( ',' ):
	85
	86	sep_char = None
	87
	88	item_txt = item_txt.strip()
	89
	90	for s_char in [ "'", '"' ]:
	91
	92	if item_txt.find( s_char ) != -1:
	93
	94	if item_txt.count( s_char ) != 2:
	95
	96	print 'Missing quote: %s' %item_txt
	97	sys.exit( 1 )
	98
	99	else:
	100
	101	sep_char = s_char
	102	break
	103
	104	if sep_char:
	105
	106	item_txt = item_txt.split( sep_char )[1]
	107
	108	my_list.append( item_txt )
	109
	110	return my_list
	111
[354]	112	cfg = ConfigParser.ConfigParser()
[212]	113
	114	cfg.read( filename )
	115
[354]	116	global DEBUG_LEVEL, DAEMONIZE, BATCH_SERVER, BATCH_POLL_INTERVAL
	117	global GMOND_CONF, DETECT_TIME_DIFFS, BATCH_HOST_TRANSLATE
	118	global BATCH_API, QUEUE, GMETRIC_TARGET
[212]	119
[354]	120	DEBUG_LEVEL = cfg.getint( 'DEFAULT', 'DEBUG_LEVEL' )
[212]	121
[354]	122	DAEMONIZE = cfg.getboolean( 'DEFAULT', 'DAEMONIZE' )
[212]	123
[265]	124	try:
[212]	125
[354]	126	BATCH_SERVER = cfg.get( 'DEFAULT', 'BATCH_SERVER' )
[212]	127
[265]	128	except ConfigParser.NoOptionError:
	129
	130	# Backwards compatibility for old configs
	131	#
	132
[354]	133	BATCH_SERVER = cfg.get( 'DEFAULT', 'TORQUE_SERVER' )
	134	api_guess = 'pbs'
[265]	135
	136	try:
	137
[354]	138	BATCH_POLL_INTERVAL = cfg.getint( 'DEFAULT', 'BATCH_POLL_INTERVAL' )
[265]	139
	140	except ConfigParser.NoOptionError:
	141
	142	# Backwards compatibility for old configs
	143	#
	144
[354]	145	BATCH_POLL_INTERVAL = cfg.getint( 'DEFAULT', 'TORQUE_POLL_INTERVAL' )
	146	api_guess = 'pbs'
[353]	147
	148	try:
[212]	149
[354]	150	GMOND_CONF = cfg.get( 'DEFAULT', 'GMOND_CONF' )
[353]	151
	152	except ConfigParser.NoOptionError:
	153
[354]	154	GMOND_CONF = None
[353]	155
[354]	156	DETECT_TIME_DIFFS = cfg.getboolean( 'DEFAULT', 'DETECT_TIME_DIFFS' )
[212]	157
[354]	158	BATCH_HOST_TRANSLATE = getlist( cfg.get( 'DEFAULT', 'BATCH_HOST_TRANSLATE' ) )
[215]	159
[266]	160	try:
[256]	161
[354]	162	BATCH_API = cfg.get( 'DEFAULT', 'BATCH_API' )
[266]	163
	164	except ConfigParser.NoOptionError, detail:
	165
	166	if BATCH_SERVER and api_guess:
[354]	167
	168	BATCH_API = api_guess
[266]	169	else:
	170	debug_msg( 0, "fatal error: BATCH_API not set and can't make guess" )
	171	sys.exit( 1 )
[317]	172
	173	try:
	174
[354]	175	QUEUE = getlist( cfg.get( 'DEFAULT', 'QUEUE' ) )
[317]	176
	177	except ConfigParser.NoOptionError, detail:
	178
[354]	179	QUEUE = None
[353]	180
	181	try:
	182
[354]	183	GMETRIC_TARGET = cfg.get( 'DEFAULT', 'GMETRIC_TARGET' )
[353]	184
	185	except ConfigParser.NoOptionError:
	186
[354]	187	GMETRIC_TARGET = None
[353]	188
	189	if not GMOND_CONF:
	190
	191	debug_msg( 0, "fatal error: GMETRIC_TARGET or GMOND_CONF both not set!" )
	192	sys.exit( 1 )
	193	else:
	194
	195	debug_msg( 0, "error: GMETRIC_TARGET not set: internel Gmetric handling aborted. Failing back to DEPRECATED use of gmond.conf/gmetric binary. This will slow down jobmond significantly!" )
	196
[212]	197	return True
	198
[253]	199	METRIC_MAX_VAL_LEN = 900
	200
[61]	201	class DataProcessor:
[68]	202	"""Class for processing of data"""
[61]	203
	204	binary = '/usr/bin/gmetric'
	205
	206	def __init__( self, binary=None ):
[68]	207	"""Remember alternate binary location if supplied"""
[61]	208
	209	if binary:
	210	self.binary = binary
	211
[80]	212	# Timeout for XML
	213	#
	214	# From ganglia's documentation:
	215	#
	216	# 'A metric will be deleted DMAX seconds after it is received, and
	217	# DMAX=0 means eternal life.'
[61]	218
[256]	219	self.dmax = str( int( int( BATCH_POLL_INTERVAL ) * 2 ) )
[80]	220
[353]	221	if GMOND_CONF:
[354]	222
[353]	223	try:
	224	gmond_file = GMOND_CONF
[68]	225
[353]	226	except NameError:
	227	gmond_file = '/etc/gmond.conf'
[68]	228
[353]	229	if not os.path.exists( gmond_file ):
	230	debug_msg( 0, 'fatal error: ' + gmond_file + ' does not exist' )
	231	sys.exit( 1 )
[68]	232
[353]	233	incompatible = self.checkGmetricVersion()
[61]	234
[353]	235	if incompatible:
	236	debug_msg( 0, 'Gmetric version not compatible, pls upgrade to at least 3.0.1' )
	237	sys.exit( 1 )
[65]	238
	239	def checkGmetricVersion( self ):
[68]	240	"""
	241	Check version of gmetric is at least 3.0.1
	242	for the syntax we use
	243	"""
[65]	244
[255]	245	global METRIC_MAX_VAL_LEN
	246
[341]	247	incompatible = 0
	248
[65]	249	for line in os.popen( self.binary + ' --version' ).readlines():
	250
	251	line = line.split( ' ' )
	252
[69]	253	if len( line ) == 2 and str(line).find( 'gmetric' ) != -1:
[65]	254
[69]	255	gmetric_version = line[1].split( '\n' )[0]
[65]	256
[69]	257	version_major = int( gmetric_version.split( '.' )[0] )
	258	version_minor = int( gmetric_version.split( '.' )[1] )
	259	version_patch = int( gmetric_version.split( '.' )[2] )
[65]	260
	261	incompatible = 0
	262
	263	if version_major < 3:
	264
	265	incompatible = 1
	266
	267	elif version_major == 3:
	268
	269	if version_minor == 0:
	270
	271	if version_patch < 1:
	272
[91]	273	incompatible = 1
[65]	274
[255]	275	if version_patch < 3:
	276
	277	METRIC_MAX_VAL_LEN = 900
	278
	279	elif version_patch >= 3:
	280
	281	METRIC_MAX_VAL_LEN = 1400
	282
[65]	283	return incompatible
	284
[75]	285	def multicastGmetric( self, metricname, metricval, valtype='string' ):
[68]	286	"""Call gmetric binary and multicast"""
[65]	287
	288	cmd = self.binary
	289
[353]	290	if GMETRIC_TARGET:
[61]	291
[353]	292	from gmetric import Gmetric
[61]	293
[353]	294	if GMETRIC_TARGET:
[61]	295
[353]	296	GMETRIC_TARGET_HOST = GMETRIC_TARGET.split( ':' )[0]
	297	GMETRIC_TARGET_PORT = GMETRIC_TARGET.split( ':' )[1]
	298
	299	metric_debug = "[gmetric] name: %s - val: %s - dmax: %s" %( str( metricname ), str( metricval ), str( self.dmax ) )
	300
	301	debug_msg( 10, printTime() + ' ' + metric_debug)
	302
	303	gm = Gmetric( GMETRIC_TARGET_HOST, GMETRIC_TARGET_PORT )
	304
	305	gm.send( str( metricname ), str( metricval ), str( self.dmax ) )
	306
	307	else:
	308	try:
	309	cmd = cmd + ' -c' + GMOND_CONF
	310
	311	except NameError:
	312
	313	debug_msg( 10, 'Assuming /etc/gmond.conf for gmetric cmd (ommitting)' )
	314
	315	cmd = cmd + ' -n' + str( metricname )+ ' -v"' + str( metricval )+ '" -t' + str( valtype ) + ' -d' + str( self.dmax )
	316
	317	debug_msg( 10, printTime() + ' ' + cmd )
	318
	319	os.system( cmd )
	320
[318]	321	class DataGatherer:
[23]	322
[318]	323	"""Skeleton class for batch system DataGatherer"""
[256]	324
[318]	325	def printJobs( self, jobs ):
	326	"""Print a jobinfo overview"""
	327
	328	for name, attrs in self.jobs.items():
	329
	330	print 'job %s' %(name)
	331
	332	for name, val in attrs.items():
	333
	334	print '\t%s = %s' %( name, val )
	335
	336	def printJob( self, jobs, job_id ):
	337	"""Print job with job_id from jobs"""
	338
	339	print 'job %s' %(job_id)
	340
	341	for name, val in jobs[ job_id ].items():
	342
	343	print '\t%s = %s' %( name, val )
	344
[256]	345	def daemon( self ):
[318]	346	"""Run as daemon forever"""
[256]	347
[318]	348	# Fork the first child
	349	#
	350	pid = os.fork()
	351	if pid > 0:
	352	sys.exit(0) # end parent
[256]	353
[318]	354	# creates a session and sets the process group ID
	355	#
	356	os.setsid()
	357
	358	# Fork the second child
	359	#
	360	pid = os.fork()
	361	if pid > 0:
	362	sys.exit(0) # end parent
	363
	364	write_pidfile()
	365
	366	# Go to the root directory and set the umask
	367	#
	368	os.chdir('/')
	369	os.umask(0)
	370
	371	sys.stdin.close()
	372	sys.stdout.close()
	373	sys.stderr.close()
	374
	375	os.open('/dev/null', os.O_RDWR)
	376	os.dup2(0, 1)
	377	os.dup2(0, 2)
	378
	379	self.run()
	380
[256]	381	def run( self ):
[318]	382	"""Main thread"""
[256]	383
[318]	384	while ( 1 ):
	385
[348]	386	self.getJobData()
	387	self.submitJobData()
[318]	388	time.sleep( BATCH_POLL_INTERVAL )
[256]	389
[318]	390	class SgeQstatXMLParser(xml.sax.handler.ContentHandler):
[256]	391
[318]	392	"""Babu Sundaram's experimental SGE qstat XML parser"""
[256]	393
[318]	394	def __init__(self, qstatinxml):
	395
	396	self.qstatfile = qstatinxml
	397	self.attribs = {}
	398	self.value = ''
	399	self.jobID = ''
	400	self.currentJobInfo = ''
	401	self.job_list = []
	402	self.EOFFlag = 0
	403	self.jobinfoCount = 0
	404
	405
	406	def startElement(self, name, attrs):
	407
	408	if name == 'job_list':
	409	self.currentJobInfo = 'Status=' + attrs.get('state', None) + ' '
	410	elif name == 'job_info':
	411	self.job_list = []
	412	self.jobinfoCount += 1
	413
	414	def characters(self, ch):
	415
	416	self.value = self.value + ch
	417
	418	def endElement(self, name):
	419
	420	if len(self.value.strip()) > 0 :
	421
	422	self.currentJobInfo += name + '=' + self.value.strip() + ' '
	423	elif name != 'job_list':
	424
	425	self.currentJobInfo += name + '=Unknown '
	426
	427	if name == 'JB_job_number':
	428
	429	self.jobID = self.value.strip()
	430	self.job_list.append(self.jobID)
	431
	432	if name == 'job_list':
	433
	434	if self.attribs.has_key(self.jobID) == False:
	435	self.attribs[self.jobID] = self.currentJobInfo
	436	elif self.attribs.has_key(self.jobID) and self.attribs[self.jobID] != self.currentJobInfo:
	437	self.attribs[self.jobID] = self.currentJobInfo
	438	self.currentJobInfo = ''
	439	self.jobID = ''
	440
	441	elif name == 'job_info' and self.jobinfoCount == 2:
	442
	443	deljobs = []
	444	for id in self.attribs:
	445	try:
	446	self.job_list.index(str(id))
	447	except ValueError:
	448	deljobs.append(id)
	449	for i in deljobs:
	450	del self.attribs[i]
	451	deljobs = []
	452	self.jobinfoCount = 0
	453
	454	self.value = ''
	455
	456	class SgeDataGatherer(DataGatherer):
	457
[61]	458	jobs = { }
[347]	459	SGE_QSTAT_XML_FILE = '/tmp/.jobmonarch.sge.qstat'
[61]	460
[318]	461	def __init__( self ):
	462	"""Setup appropriate variables"""
	463
	464	self.jobs = { }
	465	self.timeoffset = 0
	466	self.dp = DataProcessor()
	467	self.initSgeJobInfo()
	468
	469	def initSgeJobInfo( self ):
	470	"""This is outside the scope of DRMAA; Get the current jobs in SGE"""
	471	"""This is a hack because we cant get info about jobs beyond"""
	472	"""those in the current DRMAA session"""
	473
[347]	474	self.qstatparser = SgeQstatXMLParser( self.SGE_QSTAT_XML_FILE )
[318]	475
	476	# Obtain the qstat information from SGE in XML format
	477	# This would change to DRMAA-specific calls from 6.0u9
	478
	479	def getJobData(self):
	480	"""Gather all data on current jobs in SGE"""
	481
	482	# Get the information about the current jobs in the SGE queue
	483	info = os.popen("qstat -ext -xml").readlines()
[347]	484	f = open(self.SGE_QSTAT_XML_FILE,'w')
[318]	485	for lines in info:
	486	f.write(lines)
	487	f.close()
	488
	489	# Parse the input
	490	f = open(self.qstatparser.qstatfile, 'r')
	491	xml.sax.parse(f, self.qstatparser)
	492	f.close()
	493
	494	self.cur_time = time.time()
	495
	496	return self.qstatparser.attribs
	497
	498	def submitJobData(self):
	499	"""Submit job info list"""
	500
	501	self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) )
	502	# Now let's spread the knowledge
	503	#
	504	metric_increment = 0
	505	for jobid, jobattrs in self.qstatparser.attribs.items():
	506
	507	self.dp.multicastGmetric( 'MONARCH-JOB-' + jobid + '-' + str(metric_increment), jobattrs)
	508
	509	class PbsDataGatherer(DataGatherer):
	510
	511	"""This is the DataGatherer for PBS and Torque"""
	512
[256]	513	global PBSQuery
	514
[23]	515	def __init__( self ):
[354]	516
[68]	517	"""Setup appropriate variables"""
[23]	518
[354]	519	self.jobs = { }
	520	self.timeoffset = 0
	521	self.dp = DataProcessor()
	522
[91]	523	self.initPbsQuery()
[23]	524
[91]	525	def initPbsQuery( self ):
	526
[354]	527	self.pq = None
	528
[256]	529	if( BATCH_SERVER ):
[354]	530
	531	self.pq = PBSQuery( BATCH_SERVER )
[174]	532	else:
[354]	533	self.pq = PBSQuery()
[91]	534
[26]	535	def getAttr( self, attrs, name ):
[354]	536
[68]	537	"""Return certain attribute from dictionary, if exists"""
[26]	538
	539	if attrs.has_key( name ):
[354]	540
	541	return attrs[ name ]
[26]	542	else:
	543	return ''
	544
	545	def jobDataChanged( self, jobs, job_id, attrs ):
[354]	546
[68]	547	"""Check if job with attrs and job_id in jobs has changed"""
[26]	548
	549	if jobs.has_key( job_id ):
[354]	550
[26]	551	oldData = jobs[ job_id ]
	552	else:
	553	return 1
	554
	555	for name, val in attrs.items():
	556
	557	if oldData.has_key( name ):
	558
	559	if oldData[ name ] != attrs[ name ]:
	560
	561	return 1
	562
	563	else:
	564	return 1
	565
	566	return 0
	567
[348]	568	def getJobData( self ):
[354]	569
[68]	570	"""Gather all data on current jobs in Torque"""
[26]	571
[354]	572	joblist = {}
[349]	573
[354]	574	while len( joblist ) == 0:
	575
[282]	576	try:
	577	joblist = self.pq.getjobs()
[354]	578
[349]	579	except PBSError, detail:
[354]	580
	581	debug_msg( 10, "Caught PBS unavailable, skipping until next polling interval: " + str( detail ) )
[349]	582	return None
[26]	583
[354]	584	self.cur_time = time.time()
[68]	585
[354]	586	jobs_processed = [ ]
[26]	587
	588	for name, attrs in joblist.items():
	589
[354]	590	job_id = name.split( '.' )[0]
[26]	591
	592	jobs_processed.append( job_id )
[61]	593
[354]	594	name = self.getAttr( attrs, 'Job_Name' )
	595	queue = self.getAttr( attrs, 'queue' )
[317]	596
	597	if QUEUE:
	598
	599	if QUEUE != queue:
	600
	601	continue
	602
[354]	603	owner = self.getAttr( attrs, 'Job_Owner' ).split( '@' )[0]
	604	requested_time = self.getAttr( attrs, 'Resource_List.walltime' )
	605	requested_memory = self.getAttr( attrs, 'Resource_List.mem' )
[95]	606
[354]	607	mynoderequest = self.getAttr( attrs, 'Resource_List.nodes' )
[95]	608
[354]	609	ppn = ''
[281]	610
[26]	611	if mynoderequest.find( ':' ) != -1 and mynoderequest.find( 'ppn' ) != -1:
[95]	612
[354]	613	mynoderequest_fields = mynoderequest.split( ':' )
[281]	614
	615	for mynoderequest_field in mynoderequest_fields:
	616
	617	if mynoderequest_field.find( 'ppn' ) != -1:
	618
[354]	619	ppn = mynoderequest_field.split( 'ppn=' )[1]
[281]	620
[354]	621	status = self.getAttr( attrs, 'job_state' )
[25]	622
[354]	623	queued_timestamp = self.getAttr( attrs, 'ctime' )
[243]	624
[95]	625	if status == 'R':
[133]	626
[354]	627	start_timestamp = self.getAttr( attrs, 'mtime' )
	628	nodes = self.getAttr( attrs, 'exec_host' ).split( '+' )
[133]	629
[354]	630	nodeslist = [ ]
	631
[133]	632	for node in nodes:
	633
[354]	634	host = node.split( '/' )[0]
	635
[133]	636	if nodeslist.count( host ) == 0:
[215]	637
	638	for translate_pattern in BATCH_HOST_TRANSLATE:
	639
[220]	640	if translate_pattern.find( '/' ) != -1:
[215]	641
[354]	642	translate_orig = translate_pattern.split( '/' )[1]
	643	translate_new = translate_pattern.split( '/' )[2]
[220]	644
[354]	645	host = re.sub( translate_orig, translate_new, host )
[216]	646
[217]	647	if not host in nodeslist:
[216]	648
	649	nodeslist.append( host )
[133]	650
[185]	651	if DETECT_TIME_DIFFS:
	652
	653	# If a job start if later than our current date,
	654	# that must mean the Torque server's time is later
	655	# than our local time.
	656
[354]	657	if int( start_timestamp ) > int( int( self.cur_time ) + int( self.timeoffset ) ):
[185]	658
[354]	659	self.timeoffset = int( int(start_timestamp) - int(self.cur_time) )
[185]	660
[133]	661	elif status == 'Q':
[95]	662
[354]	663	start_timestamp = ''
	664	count_mynodes = 0
	665	numeric_node = 1
	666
[133]	667	for node in mynoderequest.split( '+' ):
[67]	668
[354]	669	nodepart = node.split( ':' )[0]
[67]	670
[133]	671	for letter in nodepart:
[67]	672
[133]	673	if letter not in string.digits:
	674
[354]	675	numeric_node = 0
[133]	676
	677	if not numeric_node:
[354]	678
	679	count_mynodes = count_mynodes + 1
[133]	680	else:
[327]	681	try:
[354]	682	count_mynodes = count_mynodes + int( nodepart )
	683
[327]	684	except ValueError, detail:
[354]	685
[327]	686	debug_msg( 10, str( detail ) )
	687	debug_msg( 10, "Encountered weird node in Resources_List?!" )
	688	debug_msg( 10, 'nodepart = ' + str( nodepart ) )
	689	debug_msg( 10, 'job = ' + str( name ) )
	690	debug_msg( 10, 'attrs = ' + str( attrs ) )
[133]	691
[354]	692	nodeslist = str( count_mynodes )
[172]	693	else:
[354]	694	start_timestamp = ''
	695	nodeslist = ''
[133]	696
[354]	697	myAttrs = { }
[26]	698
[354]	699	myAttrs[ 'name' ] = str( name )
	700	myAttrs[ 'queue' ] = str( queue )
	701	myAttrs[ 'owner' ] = str( owner )
	702	myAttrs[ 'requested_time' ] = str( requested_time )
	703	myAttrs[ 'requested_memory' ] = str( requested_memory )
	704	myAttrs[ 'ppn' ] = str( ppn )
	705	myAttrs[ 'status' ] = str( status )
	706	myAttrs[ 'start_timestamp' ] = str( start_timestamp )
	707	myAttrs[ 'queued_timestamp' ] = str( queued_timestamp )
	708	myAttrs[ 'reported' ] = str( int( int( self.cur_time ) + int( self.timeoffset ) ) )
	709	myAttrs[ 'nodes' ] = nodeslist
	710	myAttrs[ 'domain' ] = string.join( socket.getfqdn().split( '.' )[1:], '.' )
	711	myAttrs[ 'poll_interval' ] = str( BATCH_POLL_INTERVAL )
	712
[348]	713	if self.jobDataChanged( self.jobs, job_id, myAttrs ) and myAttrs['status'] in [ 'R', 'Q' ]:
[61]	714
[354]	715	self.jobs[ job_id ] = myAttrs
[26]	716
[348]	717	for id, attrs in self.jobs.items():
[76]	718
	719	if id not in jobs_processed:
	720
	721	# This one isn't there anymore; toedeledoki!
	722	#
[348]	723	del self.jobs[ id ]
[76]	724
[348]	725	def submitJobData( self ):
[354]	726
[65]	727	"""Submit job info list"""
	728
[219]	729	self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) )
[69]	730
[61]	731	# Now let's spread the knowledge
	732	#
[348]	733	for jobid, jobattrs in self.jobs.items():
[61]	734
[354]	735	gmetric_val = self.compileGmetricVal( jobid, jobattrs )
	736	metric_increment = 0
[61]	737
[354]	738	for val in gmetric_val:
[253]	739
	740	self.dp.multicastGmetric( 'MONARCH-JOB-' + jobid + '-' + str(metric_increment), val )
[61]	741
[354]	742	metric_increment = metric_increment + 1
	743
[253]	744	def compileGmetricVal( self, jobid, jobattrs ):
[354]	745
[253]	746	"""Create a val string for gmetric of jobinfo"""
[67]	747
[354]	748	gval_lists = [ ]
	749	mystr = None
	750	val_list = { }
[67]	751
[253]	752	for val_name, val_value in jobattrs.items():
[61]	753
[253]	754	val_list_names_len = len( string.join( val_list.keys() ) ) + len(val_list.keys())
	755	val_list_vals_len = len( string.join( val_list.values() ) ) + len(val_list.values())
[95]	756
[254]	757	if val_name == 'nodes' and jobattrs['status'] == 'R':
[95]	758
[253]	759	node_str = None
[95]	760
[253]	761	for node in val_value:
[65]	762
[253]	763	if node_str:
[354]	764
[253]	765	node_str = node_str + ';' + node
	766	else:
	767	node_str = node
[65]	768
[253]	769	if (val_list_names_len + len(val_name) ) + (val_list_vals_len + len(node_str) ) > METRIC_MAX_VAL_LEN:
[65]	770
[354]	771	val_list[ val_name ] = node_str
	772
[253]	773	gval_lists.append( val_list )
[65]	774
[354]	775	val_list = { }
	776	node_str = None
	777
	778	val_list[ val_name ] = node_str
	779
[253]	780	gval_lists.append( val_list )
[65]	781
[354]	782	val_list = { }
	783
[254]	784	elif val_value != '':
	785
	786	if (val_list_names_len + len(val_name) ) + (val_list_vals_len + len(str(val_value)) ) > METRIC_MAX_VAL_LEN:
	787
	788	gval_lists.append( val_list )
	789
[354]	790	val_list = { }
[254]	791
[354]	792	val_list[ val_name ] = val_value
	793
	794	if len( val_list ) > 0:
	795
[254]	796	gval_lists.append( val_list )
	797
[354]	798	str_list = [ ]
[65]	799
[253]	800	for val_list in gval_lists:
[65]	801
[354]	802	my_val_str = None
[65]	803
[253]	804	for val_name, val_value in val_list.items():
[65]	805
[253]	806	if my_val_str:
[65]	807
[253]	808	my_val_str = my_val_str + ' ' + val_name + '=' + val_value
	809	else:
	810	my_val_str = val_name + '=' + val_value
	811
	812	str_list.append( my_val_str )
	813
	814	return str_list
	815
[26]	816	def printTime( ):
[354]	817
[65]	818	"""Print current time/date in human readable format for log/debug"""
[26]	819
	820	return time.strftime("%a, %d %b %Y %H:%M:%S")
	821
	822	def debug_msg( level, msg ):
[354]	823
[65]	824	"""Print msg if at or above current debug level"""
[26]	825
	826	if (DEBUG_LEVEL >= level):
	827	sys.stderr.write( msg + '\n' )
	828
[307]	829	def write_pidfile():
	830
	831	# Write pidfile if PIDFILE exists
	832	if PIDFILE:
	833
[354]	834	pid = os.getpid()
	835
	836	pidfile = open(PIDFILE, 'w')
	837
	838	pidfile.write( str( pid ) )
[307]	839	pidfile.close()
	840
[23]	841	def main():
[354]	842
[65]	843	"""Application start"""
[23]	844
[352]	845	global PBSQuery, PBSError
[256]	846
[212]	847	if not processArgs( sys.argv[1:] ):
[354]	848
[212]	849	sys.exit( 1 )
	850
[256]	851	if BATCH_API == 'pbs':
	852
	853	try:
[282]	854	from PBSQuery import PBSQuery, PBSError
[256]	855
	856	except ImportError:
	857
	858	debug_msg( 0, "fatal error: BATCH_API set to 'pbs' but python module 'pbs_python' is not installed" )
	859	sys.exit( 1 )
	860
	861	gather = PbsDataGatherer()
	862
	863	elif BATCH_API == 'sge':
	864
[347]	865	gather = SgeDataGatherer()
[256]	866
	867	else:
	868	debug_msg( 0, "fatal error: unknown BATCH_API '" + BATCH_API + "' is not supported" )
[354]	869
[256]	870	sys.exit( 1 )
	871
[26]	872	if DAEMONIZE:
[354]	873
[26]	874	gather.daemon()
	875	else:
	876	gather.run()
[23]	877
[256]	878	# wh00t? someone started me! :)
[65]	879	#
[23]	880	if __name__ == '__main__':
	881	main()

Note: See TracBrowser for help on using the repository browser.

Download in other formats: