Context Navigation

source: trunk/daemon/togad.py @ 179

Last change on this file since 179 was 179, checked in by bastiaans, 19 years ago

daemon/togad.py:

Rearranged thread locking, I suspect a conflict
Disabled error checking on rrd update:
- more important to have correct last update time!
- with correct update time there will be no errors anyway

File size: 33.6 KB

Rev	Line
[3]	1	#!/usr/bin/env python
	2
[70]	3	import xml.sax
[71]	4	import xml.sax.handler
[5]	5	import socket
	6	import sys
[169]	7	import syslog
[9]	8	import string
[12]	9	import os
	10	import os.path
[17]	11	import time
[176]	12	import thread
[36]	13	import threading
[38]	14	import random
[40]	15	from types import *
[60]	16	import DBClass
[3]	17
[169]	18	# Specify debugging level here (only when _not_ DAEMONIZE)
[8]	19	#
[38]	20	# 11 = XML: metrics
	21	# 10 = XML: host, cluster, grid, ganglia
	22	# 9 = RRD activity, gmetad config parsing
[40]	23	# 8 = RRD file activity
[84]	24	# 6 = SQL
[87]	25	# 1 = daemon threading
[169]	26	# 0 = errors
[8]	27	#
[169]	28	# default: 0
[102]	29	DEBUG_LEVEL = 1
[6]	30
[169]	31	# Enable logging to syslog?
	32	#
	33	USE_SYSLOG = 1
	34
	35	# What level msg'es should be logged to syslog?
	36	#
	37	# default: lvl 0 (errors)
	38	#
	39	SYSLOG_LEVEL = 0
	40
	41	# Which facility to use in syslog
	42	#
	43	# Syntax I.e.:
	44	# LOG_KERN, LOG_USER, LOG_MAIL, LOG_DAEMON, LOG_AUTH, LOG_LPR,
	45	# LOG_NEWS, LOG_UUCP, LOG_CRON and LOG_LOCAL0 to LOG_LOCAL7
	46	#
	47	SYSLOG_FACILITY = syslog.LOG_DAEMON
	48
[9]	49	# Where is the gmetad.conf located
	50	#
	51	GMETAD_CONF = '/etc/gmetad.conf'
	52
[60]	53	# Where to grab XML data from
	54	# Normally: local gmetad (port 8651)
	55	#
[63]	56	# Syntax: <hostname>:<port>
	57	#
	58	ARCHIVE_XMLSOURCE = "localhost:8651"
[60]	59
[9]	60	# List of data_source names to archive for
	61	#
[63]	62	# Syntax: [ "<clustername>", "<clustername>" ]
	63	#
[170]	64	ARCHIVE_DATASOURCES = [ "LISA Cluster", "LISA Test Cluster" ]
[9]	65
[60]	66	# Where to store the archived rrd's
	67	#
	68	ARCHIVE_PATH = '/data/toga/rrds'
	69
[13]	70	# Amount of hours to store in one single archived .rrd
[9]	71	#
[43]	72	ARCHIVE_HOURS_PER_RRD = 12
[13]	73
[63]	74	# Toga's SQL dbase to use
[60]	75	#
[63]	76	# Syntax: <hostname>/<database>
[60]	77	#
[63]	78	TOGA_SQL_DBASE = "localhost/toga"
[60]	79
[22]	80	# Wether or not to run as a daemon in background
	81	#
[169]	82	DAEMONIZE = 1
[22]	83
[17]	84	######################
	85	# #
	86	# Configuration ends #
	87	# #
	88	######################
[13]	89
[60]	90	###
	91	# You'll only want to change anything below here unless you
[63]	92	# know what you are doing (i.e. your name is Ramon Bastiaans :D )
[60]	93	###
	94
[17]	95	# What XML data types not to store
[13]	96	#
[17]	97	UNSUPPORTED_ARCHIVE_TYPES = [ 'string' ]
[9]	98
[47]	99	# Maximum time (in seconds) a parsethread may run
	100	#
	101	PARSE_TIMEOUT = 60
	102
	103	# Maximum time (in seconds) a storethread may run
	104	#
	105	STORE_TIMEOUT = 360
	106
[8]	107	"""
	108	This is TOrque-GAnglia's data Daemon
	109	"""
	110
[84]	111	class DataSQLStore:
	112
	113	db_vars = None
	114	dbc = None
	115
	116	def __init__( self, hostname, database ):
	117
	118	self.db_vars = DBClass.InitVars(DataBaseName=database,
	119	User='root',
	120	Host=hostname,
	121	Password='',
	122	Dictionary='true')
	123
	124	try:
	125	self.dbc = DBClass.DB(self.db_vars)
	126	except DBClass.DBError, details:
[169]	127	debug_msg( 0, 'FATAL ERROR: Unable to connect to database!: ' +str(details) )
[84]	128	sys.exit(1)
	129
	130	def setDatabase(self, statement):
	131	ret = self.doDatabase('set', statement)
	132	return ret
	133
	134	def getDatabase(self, statement):
	135	ret = self.doDatabase('get', statement)
	136	return ret
	137
	138	def doDatabase(self, type, statement):
	139
	140	debug_msg( 6, 'doDatabase(): %s: %s' %(type, statement) )
	141	try:
	142	if type == 'set':
	143	result = self.dbc.Set( statement )
	144	self.dbc.Commit()
	145	elif type == 'get':
	146	result = self.dbc.Get( statement )
	147
	148	except DBClass.DBError, detail:
	149	operation = statement.split(' ')[0]
[169]	150	debug_msg( 0, 'FATAL ERROR: ' +operation+ ' on database failed while doing ['+statement+'] full msg: '+str(detail) )
[84]	151	sys.exit(1)
	152
	153	debug_msg( 6, 'doDatabase(): result: %s' %(result) )
	154	return result
	155
	156	def getNodeId( self, hostname ):
	157
[89]	158	id = self.getDatabase( "SELECT node_id FROM nodes WHERE node_hostname = '%s'" %hostname )
[84]	159
[89]	160	if len( id ) > 0:
[84]	161
[89]	162	id = id[0][0]
	163
[84]	164	return id
	165	else:
	166	return None
	167
	168	def getNodeIds( self, hostnames ):
	169
	170	ids = [ ]
	171
	172	for node in hostnames:
	173
	174	id = self.getNodeId( node )
	175
	176	if id:
	177	ids.append( id )
	178
	179	return ids
	180
	181	def getJobId( self, jobid ):
	182
	183	id = self.getDatabase( "SELECT job_id FROM jobs WHERE job_id = '%s'" %jobid )
	184
	185	if id:
[89]	186	id = id[0][0]
[84]	187
	188	return id
	189	else:
	190	return None
	191
	192	def addJob( self, job_id, jobattrs ):
	193
	194	if not self.getJobId( job_id ):
	195
	196	self.mutateJob( 'insert', job_id, jobattrs )
	197	else:
	198	self.mutateJob( 'update', job_id, jobattrs )
	199
	200	def mutateJob( self, action, job_id, jobattrs ):
	201
	202	job_values = [ 'name', 'queue', 'owner', 'requested_time', 'requested_memory', 'ppn', 'status', 'start_timestamp', 'stop_timestamp' ]
	203
	204	insert_col_str = 'job_id'
	205	insert_val_str = "'%s'" %job_id
	206	update_str = None
	207
	208	debug_msg( 6, 'mutateJob(): %s %s' %(action,job_id))
	209
[99]	210	ids = [ ]
	211
[84]	212	for valname, value in jobattrs.items():
	213
[96]	214	if valname in job_values and value != '':
[84]	215
	216	column_name = 'job_' + valname
	217
	218	if action == 'insert':
	219
	220	if not insert_col_str:
	221	insert_col_str = column_name
	222	else:
	223	insert_col_str = insert_col_str + ',' + column_name
	224
	225	if not insert_val_str:
	226	insert_val_str = value
	227	else:
	228	insert_val_str = insert_val_str + ",'%s'" %value
	229
	230	elif action == 'update':
	231
	232	if not update_str:
	233	update_str = "%s='%s'" %(column_name, value)
	234	else:
	235	update_str = update_str + ",%s='%s'" %(column_name, value)
	236
[90]	237	elif valname == 'nodes' and value:
[84]	238
[154]	239	ids = self.addNodes( value, jobattrs['domain'] )
[86]	240	node_list = value
[84]	241
	242	if action == 'insert':
	243
	244	self.setDatabase( "INSERT INTO jobs ( %s ) VALUES ( %s )" %( insert_col_str, insert_val_str ) )
[86]	245
[99]	246	if len( ids ) > 0:
	247	self.addJobNodes( job_id, ids )
	248
[84]	249	elif action == 'update':
	250
[89]	251	self.setDatabase( "UPDATE jobs SET %s WHERE job_id=%s" %(update_str, job_id) )
[84]	252
[154]	253	def addNodes( self, hostnames, domain ):
[84]	254
[98]	255	ids = [ ]
	256
[84]	257	for node in hostnames:
	258
[154]	259	node = '%s.%s' %( node, domain )
[84]	260	id = self.getNodeId( node )
	261
	262	if not id:
	263	self.setDatabase( "INSERT INTO nodes ( node_hostname ) VALUES ( '%s' )" %node )
[98]	264	id = self.getNodeId( node )
[84]	265
[98]	266	ids.append( id )
	267
	268	return ids
	269
[86]	270	def addJobNodes( self, jobid, nodes ):
	271
	272	for node in nodes:
	273	self.addJobNode( jobid, node )
	274
[84]	275	def addJobNode( self, jobid, nodeid ):
	276
	277	self.setDatabase( "INSERT INTO job_nodes (job_id,node_id) VALUES ( %s,%s )" %(jobid, nodeid) )
	278
	279	def storeJobInfo( self, jobid, jobattrs ):
	280
	281	self.addJob( jobid, jobattrs )
	282
[37]	283	class RRDMutator:
[63]	284	"""A class for performing RRD mutations"""
[37]	285
	286	binary = '/usr/bin/rrdtool'
	287
	288	def __init__( self, binary=None ):
[63]	289	"""Set alternate binary if supplied"""
[37]	290
	291	if binary:
	292	self.binary = binary
	293
	294	def create( self, filename, args ):
[63]	295	"""Create a new rrd with args"""
	296
[40]	297	return self.perform( 'create', '"' + filename + '"', args )
[37]	298
	299	def update( self, filename, args ):
[63]	300	"""Update a rrd with args"""
	301
[40]	302	return self.perform( 'update', '"' + filename + '"', args )
[37]	303
[42]	304	def grabLastUpdate( self, filename ):
[63]	305	"""Determine the last update time of filename rrd"""
[42]	306
	307	last_update = 0
	308
[53]	309	debug_msg( 8, self.binary + ' info "' + filename + '"' )
	310
[42]	311	for line in os.popen( self.binary + ' info "' + filename + '"' ).readlines():
	312
	313	if line.find( 'last_update') != -1:
	314
	315	last_update = line.split( ' = ' )[1]
	316
	317	if last_update:
	318	return last_update
	319	else:
	320	return 0
	321
[40]	322	def perform( self, action, filename, args ):
[63]	323	"""Perform action on rrd filename with args"""
[37]	324
	325	arg_string = None
	326
[40]	327	if type( args ) is not ListType:
	328	debug_msg( 8, 'Arguments needs to be of type List' )
	329	return 1
	330
[37]	331	for arg in args:
	332
	333	if not arg_string:
	334
	335	arg_string = arg
	336	else:
	337	arg_string = arg_string + ' ' + arg
	338
[54]	339	debug_msg( 8, self.binary + ' ' + action + ' ' + filename + ' ' + arg_string )
[37]	340
[146]	341	cmd = os.popen( self.binary + ' ' + action + ' ' + filename + ' ' + arg_string )
	342	lines = cmd.readlines()
	343	cmd.close()
[37]	344
[146]	345	for line in lines:
	346
[37]	347	if line.find( 'ERROR' ) != -1:
	348
	349	error_msg = string.join( line.split( ' ' )[1:] )
[40]	350	debug_msg( 8, error_msg )
[37]	351	return 1
	352
	353	return 0
	354
[78]	355	class XMLProcessor:
	356	"""Skeleton class for XML processor's"""
	357
	358	def run( self ):
	359	"""Do main processing of XML here"""
	360
	361	pass
	362
	363	class TorqueXMLProcessor( XMLProcessor ):
	364	"""Main class for processing XML and acting with it"""
	365
	366	def __init__( self ):
	367	"""Setup initial XML connection and handlers"""
	368
	369	self.myXMLGatherer = XMLGatherer( ARCHIVE_XMLSOURCE.split( ':' )[0], ARCHIVE_XMLSOURCE.split( ':' )[1] )
	370	self.myXMLSource = self.myXMLGatherer.getFileObject()
	371	self.myXMLHandler = TorqueXMLHandler()
	372	self.myXMLError = XMLErrorHandler()
[87]	373	self.config = GangliaConfigParser( GMETAD_CONF )
[78]	374
	375	def run( self ):
	376	"""Main XML processing"""
	377
[169]	378	debug_msg( 1, 'torque_xml_thread(): started.' )
[87]	379
[78]	380	while( 1 ):
	381
	382	self.myXMLSource = self.myXMLGatherer.getFileObject()
[169]	383	debug_msg( 1, 'torque_xml_thread(): Parsing..' )
[176]	384
	385	try:
	386	xml.sax.parse( self.myXMLSource, self.myXMLHandler, self.myXMLError )
	387	except socket.error, msg:
	388	debug_msg( 0, 'ERROR: Socket error in connect to datasource!: %s' %msg )
	389
[169]	390	debug_msg( 1, 'torque_xml_thread(): Done parsing.' )
	391	debug_msg( 1, 'torque_xml_thread(): Sleeping.. (%ss)' %(str( self.config.getLowestInterval() ) ) )
[87]	392	time.sleep( self.config.getLowestInterval() )
[78]	393
[71]	394	class TorqueXMLHandler( xml.sax.handler.ContentHandler ):
[63]	395	"""Parse Torque's jobinfo XML from our plugin"""
	396
[72]	397	jobAttrs = { }
	398
[84]	399	def __init__( self ):
	400
	401	self.ds = DataSQLStore( TOGA_SQL_DBASE.split( '/' )[0], TOGA_SQL_DBASE.split( '/' )[1] )
[98]	402	self.jobs_processed = [ ]
	403	self.jobs_to_store = [ ]
[84]	404
[63]	405	def startElement( self, name, attrs ):
	406	"""
	407	This XML will be all gmetric XML
	408	so there will be no specific start/end element
	409	just one XML statement with all info
	410	"""
	411
	412	heartbeat = 0
[73]	413
	414	jobinfo = { }
[63]	415
	416	if name == 'METRIC':
	417
[70]	418	metricname = attrs.get( 'NAME', "" )
[63]	419
	420	if metricname == 'TOGA-HEARTBEAT':
[70]	421	self.heartbeat = attrs.get( 'VAL', "" )
[63]	422
[70]	423	elif metricname.find( 'TOGA-JOB' ) != -1:
[63]	424
[72]	425	job_id = metricname.split( 'TOGA-JOB-' )[1]
[63]	426	val = attrs.get( 'VAL', "" )
	427
[96]	428	if not job_id in self.jobs_processed:
	429	self.jobs_processed.append( job_id )
	430
[73]	431	check_change = 0
	432
	433	if self.jobAttrs.has_key( job_id ):
	434	check_change = 1
	435
[63]	436	valinfo = val.split( ' ' )
	437
	438	for myval in valinfo:
	439
[84]	440	if len( myval.split( '=' ) ) > 1:
[63]	441
[84]	442	valname = myval.split( '=' )[0]
	443	value = myval.split( '=' )[1]
[70]	444
[84]	445	if valname == 'nodes':
	446	value = value.split( ';' )
[72]	447
[84]	448	jobinfo[ valname ] = value
	449
[73]	450	if check_change:
[100]	451	if self.jobinfoChanged( self.jobAttrs, job_id, jobinfo ) and self.jobAttrs[ job_id ]['status'] != 'E':
	452	self.jobAttrs[ job_id ]['stop_timestamp'] = ''
[82]	453	self.jobAttrs[ job_id ] = self.setJobAttrs( self.jobAttrs[ job_id ], jobinfo )
[84]	454	if not job_id in self.jobs_to_store:
	455	self.jobs_to_store.append( job_id )
	456
[102]	457	debug_msg( 6, 'jobinfo for job %s has changed' %job_id )
[73]	458	else:
	459	self.jobAttrs[ job_id ] = jobinfo
[84]	460
	461	if not job_id in self.jobs_to_store:
	462	self.jobs_to_store.append( job_id )
	463
[102]	464	debug_msg( 6, 'jobinfo for job %s has changed' %job_id )
[73]	465
[77]	466	def endDocument( self ):
[74]	467	"""When all metrics have gone, check if any jobs have finished"""
[72]	468
[74]	469	for jobid, jobinfo in self.jobAttrs.items():
	470
	471	# This is an old job, not in current jobinfo list anymore
	472	# it must have finished, since we _did_ get a new heartbeat
	473	#
[96]	474	mytime = int( jobinfo['reported'] ) + int( jobinfo['poll_interval'] )
[102]	475
[96]	476	if mytime < self.heartbeat and jobid not in self.jobs_processed and jobinfo['status'] == 'R':
[74]	477
[97]	478	if not jobid in self.jobs_processed:
[96]	479	self.jobs_processed.append( jobid )
	480
[74]	481	self.jobAttrs[ jobid ]['status'] = 'F'
[151]	482	self.jobAttrs[ jobid ]['stop_timestamp'] = str( mytime )
[96]	483
[84]	484	if not jobid in self.jobs_to_store:
	485	self.jobs_to_store.append( jobid )
[74]	486
[169]	487	debug_msg( 1, 'torque_xml_thread(): Storing..' )
[87]	488
[84]	489	for jobid in self.jobs_to_store:
[155]	490	if self.jobAttrs[ jobid ]['status'] in [ 'R', 'Q', 'F' ]:
[151]	491	self.ds.storeJobInfo( jobid, self.jobAttrs[ jobid ] )
[84]	492
[169]	493	debug_msg( 1, 'torque_xml_thread(): Done storing.' )
[87]	494
[96]	495	self.jobs_processed = [ ]
[84]	496	self.jobs_to_store = [ ]
	497
[82]	498	def setJobAttrs( self, old, new ):
	499	"""
	500	Set new job attributes in old, but not lose existing fields
	501	if old attributes doesn't have those
	502	"""
	503
	504	for valname, value in new.items():
	505	old[ valname ] = value
	506
	507	return old
	508
	509
[73]	510	def jobinfoChanged( self, jobattrs, jobid, jobinfo ):
[74]	511	"""
	512	Check if jobinfo has changed from jobattrs[jobid]
	513	if it's report time is bigger than previous one
	514	and it is report time is recent (equal to heartbeat)
	515	"""
[72]	516
[87]	517	ignore_changes = [ 'reported' ]
	518
[73]	519	if jobattrs.has_key( jobid ):
	520
	521	for valname, value in jobinfo.items():
	522
[87]	523	if valname not in ignore_changes:
[73]	524
[87]	525	if jobattrs[ jobid ].has_key( valname ):
[73]	526
[87]	527	if value != jobattrs[ jobid ][ valname ]:
[73]	528
[87]	529	if jobinfo['reported'] > jobattrs[ jobid ][ 'reported' ] and jobinfo['reported'] == self.heartbeat:
	530	return 1
[73]	531
[87]	532	else:
	533	return 1
	534
[73]	535	return 0
	536
[71]	537	class GangliaXMLHandler( xml.sax.handler.ContentHandler ):
[63]	538	"""Parse Ganglia's XML"""
[3]	539
[33]	540	def __init__( self, config ):
[63]	541	"""Setup initial variables and gather info on existing rrd archive"""
	542
[33]	543	self.config = config
[35]	544	self.clusters = { }
[169]	545	debug_msg( 1, 'Checking existing toga rrd archive..' )
[44]	546	self.gatherClusters()
[169]	547	debug_msg( 1, 'Check done.' )
[33]	548
[44]	549	def gatherClusters( self ):
[63]	550	"""Find all existing clusters in archive dir"""
[44]	551
[45]	552	archive_dir = check_dir(ARCHIVE_PATH)
[44]	553
	554	hosts = [ ]
	555
	556	if os.path.exists( archive_dir ):
	557
	558	dirlist = os.listdir( archive_dir )
	559
	560	for item in dirlist:
	561
	562	clustername = item
	563
[60]	564	if not self.clusters.has_key( clustername ) and clustername in ARCHIVE_DATASOURCES:
[44]	565
	566	self.clusters[ clustername ] = RRDHandler( self.config, clustername )
	567
[6]	568	def startElement( self, name, attrs ):
[63]	569	"""Memorize appropriate data from xml start tags"""
[3]	570
[7]	571	if name == 'GANGLIA_XML':
[32]	572
	573	self.XMLSource = attrs.get( 'SOURCE', "" )
	574	self.gangliaVersion = attrs.get( 'VERSION', "" )
	575
[12]	576	debug_msg( 10, 'Found XML data: source %s version %s' %( self.XMLSource, self.gangliaVersion ) )
[6]	577
[7]	578	elif name == 'GRID':
[32]	579
	580	self.gridName = attrs.get( 'NAME', "" )
	581	self.time = attrs.get( 'LOCALTIME', "" )
	582
[12]	583	debug_msg( 10, '`-Grid found: %s' %( self.gridName ) )
[6]	584
[7]	585	elif name == 'CLUSTER':
[32]	586
	587	self.clusterName = attrs.get( 'NAME', "" )
	588	self.time = attrs.get( 'LOCALTIME', "" )
	589
[60]	590	if not self.clusters.has_key( self.clusterName ) and self.clusterName in ARCHIVE_DATASOURCES:
[32]	591
[34]	592	self.clusters[ self.clusterName ] = RRDHandler( self.config, self.clusterName )
[33]	593
[35]	594	debug_msg( 10, ' \|-Cluster found: %s' %( self.clusterName ) )
[6]	595
[60]	596	elif name == 'HOST' and self.clusterName in ARCHIVE_DATASOURCES:
[32]	597
	598	self.hostName = attrs.get( 'NAME', "" )
	599	self.hostIp = attrs.get( 'IP', "" )
	600	self.hostReported = attrs.get( 'REPORTED', "" )
	601
[12]	602	debug_msg( 10, ' \| \|-Host found: %s - ip %s reported %s' %( self.hostName, self.hostIp, self.hostReported ) )
[6]	603
[60]	604	elif name == 'METRIC' and self.clusterName in ARCHIVE_DATASOURCES:
[6]	605
[32]	606	type = attrs.get( 'TYPE', "" )
[6]	607
[32]	608	if type not in UNSUPPORTED_ARCHIVE_TYPES:
[3]	609
[32]	610	myMetric = { }
	611	myMetric['name'] = attrs.get( 'NAME', "" )
	612	myMetric['val'] = attrs.get( 'VAL', "" )
	613	myMetric['time'] = self.hostReported
[3]	614
[34]	615	self.clusters[ self.clusterName ].memMetric( self.hostName, myMetric )
[3]	616
[34]	617	debug_msg( 11, ' \| \| \|-metric: %s:%s' %( myMetric['name'], myMetric['val'] ) )
[6]	618
[34]	619	def storeMetrics( self ):
[63]	620	"""Store metrics of each cluster rrd handler"""
[9]	621
[34]	622	for clustername, rrdh in self.clusters.items():
[16]	623
[38]	624	ret = rrdh.storeMetrics()
[9]	625
[38]	626	if ret:
	627	debug_msg( 9, 'An error occured while storing metrics for cluster %s' %clustername )
	628	return 1
	629
	630	return 0
	631
[71]	632	class XMLErrorHandler( xml.sax.handler.ErrorHandler ):
	633
	634	def error( self, exception ):
	635	"""Recoverable error"""
	636
[169]	637	debug_msg( 0, 'Recoverable XML error ' + str( exception ) + ' ignored.' )
[71]	638
	639	def fatalError( self, exception ):
	640	"""Non-recoverable error"""
	641
	642	exception_str = str( exception )
	643
	644	# Ignore 'no element found' errors
	645	if exception_str.find( 'no element found' ) != -1:
[169]	646	debug_msg( 0, 'No XML data found: Socket not (re)connected or datasource not available.' )
[71]	647	return 0
	648
[170]	649	debug_msg( 0, 'FATAL ERROR: Non-recoverable XML error ' + str( exception ) )
[71]	650	sys.exit( 1 )
	651
	652	def warning( self, exception ):
	653	"""Warning"""
	654
	655	debug_msg( 0, 'Warning ' + str( exception ) )
	656
[78]	657	class XMLGatherer:
[63]	658	"""Setup a connection and file object to Ganglia's XML"""
[3]	659
[8]	660	s = None
[70]	661	fd = None
[8]	662
	663	def __init__( self, host, port ):
[63]	664	"""Store host and port for connection"""
[8]	665
[5]	666	self.host = host
	667	self.port = port
[33]	668	self.connect()
[70]	669	self.makeFileDescriptor()
[3]	670
[33]	671	def connect( self ):
[63]	672	"""Setup connection to XML source"""
[8]	673
	674	for res in socket.getaddrinfo( self.host, self.port, socket.AF_UNSPEC, socket.SOCK_STREAM ):
[32]	675
[5]	676	af, socktype, proto, canonname, sa = res
[32]	677
[5]	678	try:
[32]	679
[8]	680	self.s = socket.socket( af, socktype, proto )
[32]	681
[5]	682	except socket.error, msg:
[32]	683
[8]	684	self.s = None
[5]	685	continue
[32]	686
[5]	687	try:
[32]	688
[8]	689	self.s.connect( sa )
[32]	690
[5]	691	except socket.error, msg:
[32]	692
[70]	693	self.disconnect()
[5]	694	continue
[32]	695
[5]	696	break
[3]	697
[8]	698	if self.s is None:
[32]	699
[170]	700	debug_msg( 0, 'FATAL ERROR: Could not open socket or unable to connect to datasource!' )
[33]	701	sys.exit( 1 )
[5]	702
[33]	703	def disconnect( self ):
[63]	704	"""Close socket"""
[33]	705
	706	if self.s:
[71]	707	self.s.shutdown( 2 )
[33]	708	self.s.close()
	709	self.s = None
	710
	711	def __del__( self ):
[63]	712	"""Kill the socket before we leave"""
[33]	713
	714	self.disconnect()
	715
[70]	716	def reconnect( self ):
	717	"""Reconnect"""
[33]	718
[38]	719	if self.s:
	720	self.disconnect()
[33]	721
[70]	722	self.connect()
[5]	723
[70]	724	def makeFileDescriptor( self ):
	725	"""Make file descriptor that points to our socket connection"""
	726
	727	self.reconnect()
	728
	729	if self.s:
	730	self.fd = self.s.makefile( 'r' )
	731
	732	def getFileObject( self ):
	733	"""Connect, and return a file object"""
	734
[78]	735	self.makeFileDescriptor()
	736
[70]	737	if self.fd:
	738	return self.fd
	739
[78]	740	class GangliaXMLProcessor( XMLProcessor ):
[63]	741	"""Main class for processing XML and acting with it"""
[5]	742
[33]	743	def __init__( self ):
[63]	744	"""Setup initial XML connection and handlers"""
[33]	745
	746	self.config = GangliaConfigParser( GMETAD_CONF )
	747
[78]	748	self.myXMLGatherer = XMLGatherer( ARCHIVE_XMLSOURCE.split( ':' )[0], ARCHIVE_XMLSOURCE.split( ':' )[1] )
[70]	749	self.myXMLSource = self.myXMLGatherer.getFileObject()
[78]	750	self.myXMLHandler = GangliaXMLHandler( self.config )
	751	self.myXMLError = XMLErrorHandler()
[73]	752
[9]	753	def run( self ):
[63]	754	"""Main XML processing; start a xml and storethread"""
[8]	755
[102]	756	xml_thread = threading.Thread( None, self.processXML, 'xmlthread' )
	757	store_thread = threading.Thread( None, self.storeMetrics, 'storethread' )
[22]	758
[36]	759	while( 1 ):
	760
[102]	761	if not xml_thread.isAlive():
[36]	762	# Gather XML at the same interval as gmetad
	763
	764	# threaded call to: self.processXML()
	765	#
[169]	766	try:
	767	xml_thread = threading.Thread( None, self.processXML, 'xml_thread' )
	768	xml_thread.start()
[176]	769	except thread.error, msg:
[169]	770	debug_msg( 0, 'ERROR: Unable to start xml_thread!: '+str(msg))
	771	#return 1
[36]	772
[102]	773	if not store_thread.isAlive():
[55]	774	# Store metrics every .. sec
[36]	775
[55]	776	# threaded call to: self.storeMetrics()
	777	#
[169]	778	try:
	779	store_thread = threading.Thread( None, self.storeMetrics, 'store_thread' )
	780	store_thread.start()
[176]	781	except thread.error, msg:
[169]	782	debug_msg( 0, 'ERROR: Unable to start store_thread!: '+str(msg))
	783	#return 1
[36]	784
	785	# Just sleep a sec here, to prevent daemon from going mad. We're all threads here anyway
	786	time.sleep( 1 )
	787
[33]	788	def storeMetrics( self ):
[63]	789	"""Store metrics retained in memory to disk"""
[22]	790
[63]	791	# Store metrics somewhere between every 360 and 640 seconds
[38]	792	#
[55]	793	STORE_INTERVAL = random.randint( 360, 640 )
[22]	794
[169]	795	try:
	796	store_metric_thread = threading.Thread( None, self.storeThread, 'store_metric_thread' )
	797	store_metric_thread.start()
[176]	798	except thread.error, msg:
[169]	799	debug_msg( 0, 'ERROR: Unable to start ganglia_store_thread()!: '+str(msg) )
	800	return 1
[36]	801
[169]	802	debug_msg( 1, 'ganglia_store_thread(): started.' )
	803
	804	debug_msg( 1, 'ganglia_store_thread(): Sleeping.. (%ss)' %STORE_INTERVAL )
[36]	805	time.sleep( STORE_INTERVAL )
[169]	806	debug_msg( 1, 'ganglia_store_thread(): Done sleeping.' )
[36]	807
[102]	808	if store_metric_thread.isAlive():
[36]	809
[169]	810	debug_msg( 1, 'ganglia_store_thread(): storemetricthread() still running, waiting to finish..' )
[136]	811	store_metric_thread.join( STORE_TIMEOUT ) # Maximum time is for storing thread to finish
[169]	812	debug_msg( 1, 'ganglia_store_thread(): Done waiting.' )
[36]	813
[169]	814	debug_msg( 1, 'ganglia_store_thread(): finished.' )
[36]	815
	816	return 0
	817
[39]	818	def storeThread( self ):
[63]	819	"""Actual metric storing thread"""
[39]	820
[169]	821	debug_msg( 1, 'ganglia_store_metric_thread(): started.' )
	822	debug_msg( 1, 'ganglia_store_metric_thread(): Storing data..' )
[78]	823	ret = self.myXMLHandler.storeMetrics()
[176]	824	if ret > 0:
	825	debug_msg( 0, 'ganglia_store_metric_thread(): UNKNOWN ERROR %s while storing Metrics!' %str(ret) )
[169]	826	debug_msg( 1, 'ganglia_store_metric_thread(): Done storing.' )
	827	debug_msg( 1, 'ganglia_store_metric_thread(): finished.' )
[39]	828
[176]	829	return 0
[39]	830
[8]	831	def processXML( self ):
[63]	832	"""Process XML"""
[8]	833
[169]	834	try:
	835	parsethread = threading.Thread( None, self.parseThread, 'parsethread' )
	836	parsethread.start()
[176]	837	except thread.error, msg:
[169]	838	debug_msg( 0, 'ERROR: Unable to start ganglia_xml_thread()!: ' + str(msg) )
	839	return 1
[8]	840
[169]	841	debug_msg( 1, 'ganglia_xml_thread(): started.' )
[36]	842
[169]	843	debug_msg( 1, 'ganglia_xml_thread(): Sleeping.. (%ss)' %self.config.getLowestInterval() )
[36]	844	time.sleep( float( self.config.getLowestInterval() ) )
[169]	845	debug_msg( 1, 'ganglia_xml_thread(): Done sleeping.' )
[36]	846
	847	if parsethread.isAlive():
	848
[169]	849	debug_msg( 1, 'ganglia_xml_thread(): parsethread() still running, waiting (%ss) to finish..' %PARSE_TIMEOUT )
[47]	850	parsethread.join( PARSE_TIMEOUT ) # Maximum time for XML thread to finish
[169]	851	debug_msg( 1, 'ganglia_xml_thread(): Done waiting.' )
[36]	852
[169]	853	debug_msg( 1, 'ganglia_xml_thread(): finished.' )
[36]	854
	855	return 0
	856
[39]	857	def parseThread( self ):
[63]	858	"""Actual parsing thread"""
[39]	859
[169]	860	debug_msg( 1, 'ganglia_parse_thread(): started.' )
	861	debug_msg( 1, 'ganglia_parse_thread(): Parsing XML..' )
[78]	862	self.myXMLSource = self.myXMLGatherer.getFileObject()
[176]	863
	864	try:
	865	xml.sax.parse( self.myXMLSource, self.myXMLHandler, self.myXMLError )
	866	except socket.error, msg:
	867	debug_msg( 0, 'ERROR: Socket error in connect to datasource!: %s' %msg )
	868
[169]	869	debug_msg( 1, 'ganglia_parse_thread(): Done parsing.' )
	870	debug_msg( 1, 'ganglia_parse_thread(): finished.' )
[39]	871
[176]	872	return 0
[39]	873
[9]	874	class GangliaConfigParser:
	875
[34]	876	sources = [ ]
[9]	877
	878	def __init__( self, config ):
[63]	879	"""Parse some stuff from our gmetad's config, such as polling interval"""
[32]	880
[9]	881	self.config = config
	882	self.parseValues()
	883
[32]	884	def parseValues( self ):
[63]	885	"""Parse certain values from gmetad.conf"""
[9]	886
	887	readcfg = open( self.config, 'r' )
	888
	889	for line in readcfg.readlines():
	890
	891	if line.count( '"' ) > 1:
	892
[10]	893	if line.find( 'data_source' ) != -1 and line[0] != '#':
[9]	894
[11]	895	source = { }
	896	source['name'] = line.split( '"' )[1]
[9]	897	source_words = line.split( '"' )[2].split( ' ' )
	898
	899	for word in source_words:
	900
	901	valid_interval = 1
	902
	903	for letter in word:
[32]	904
[9]	905	if letter not in string.digits:
[32]	906
[9]	907	valid_interval = 0
	908
[10]	909	if valid_interval and len(word) > 0:
[32]	910
[9]	911	source['interval'] = word
[12]	912	debug_msg( 9, 'polling interval for %s = %s' %(source['name'], source['interval'] ) )
[33]	913
	914	# No interval found, use Ganglia's default
	915	if not source.has_key( 'interval' ):
	916	source['interval'] = 15
	917	debug_msg( 9, 'polling interval for %s defaulted to 15' %(source['name']) )
[32]	918
[33]	919	self.sources.append( source )
[9]	920
	921	def getInterval( self, source_name ):
[63]	922	"""Return interval for source_name"""
[32]	923
[9]	924	for source in self.sources:
[32]	925
[12]	926	if source['name'] == source_name:
[32]	927
[9]	928	return source['interval']
[32]	929
[9]	930	return None
	931
[34]	932	def getLowestInterval( self ):
[63]	933	"""Return the lowest interval of all clusters"""
[34]	934
	935	lowest_interval = 0
	936
	937	for source in self.sources:
	938
	939	if not lowest_interval or source['interval'] <= lowest_interval:
	940
	941	lowest_interval = source['interval']
	942
	943	# Return 15 when nothing is found, so that the daemon won't go insane with 0 sec delays
	944	if lowest_interval:
	945	return lowest_interval
	946	else:
	947	return 15
	948
[9]	949	class RRDHandler:
[63]	950	"""Class for handling RRD activity"""
[9]	951
[32]	952	myMetrics = { }
[40]	953	lastStored = { }
[47]	954	timeserials = { }
[36]	955	slot = None
[32]	956
[33]	957	def __init__( self, config, cluster ):
[63]	958	"""Setup initial variables"""
[78]	959
[44]	960	self.block = 0
[9]	961	self.cluster = cluster
[33]	962	self.config = config
[40]	963	self.slot = threading.Lock()
[37]	964	self.rrdm = RRDMutator()
[42]	965	self.gatherLastUpdates()
[9]	966
[42]	967	def gatherLastUpdates( self ):
[63]	968	"""Populate the lastStored list, containing timestamps of all last updates"""
[42]	969
	970	cluster_dir = '%s/%s' %( check_dir(ARCHIVE_PATH), self.cluster )
	971
	972	hosts = [ ]
	973
	974	if os.path.exists( cluster_dir ):
	975
[44]	976	dirlist = os.listdir( cluster_dir )
[42]	977
[44]	978	for dir in dirlist:
[42]	979
[44]	980	hosts.append( dir )
[42]	981
	982	for host in hosts:
	983
[47]	984	host_dir = cluster_dir + '/' + host
	985	dirlist = os.listdir( host_dir )
	986
	987	for dir in dirlist:
	988
	989	if not self.timeserials.has_key( host ):
	990
	991	self.timeserials[ host ] = [ ]
	992
	993	self.timeserials[ host ].append( dir )
	994
[42]	995	last_serial = self.getLastRrdTimeSerial( host )
	996	if last_serial:
	997
	998	metric_dir = cluster_dir + '/' + host + '/' + last_serial
	999	if os.path.exists( metric_dir ):
	1000
[44]	1001	dirlist = os.listdir( metric_dir )
[42]	1002
[44]	1003	for file in dirlist:
[42]	1004
[44]	1005	metricname = file.split( '.rrd' )[0]
[42]	1006
[44]	1007	if not self.lastStored.has_key( host ):
[42]	1008
[44]	1009	self.lastStored[ host ] = { }
[42]	1010
[44]	1011	self.lastStored[ host ][ metricname ] = self.rrdm.grabLastUpdate( metric_dir + '/' + file )
[42]	1012
[32]	1013	def getClusterName( self ):
[63]	1014	"""Return clustername"""
	1015
[32]	1016	return self.cluster
	1017
	1018	def memMetric( self, host, metric ):
[63]	1019	"""Store metric from host in memory"""
[32]	1020
[179]	1021	# <ATOMIC>
	1022	#
	1023	self.slot.acquire()
	1024
[34]	1025	if self.myMetrics.has_key( host ):
[32]	1026
[34]	1027	if self.myMetrics[ host ].has_key( metric['name'] ):
[32]	1028
[34]	1029	for mymetric in self.myMetrics[ host ][ metric['name'] ]:
[32]	1030
[34]	1031	if mymetric['time'] == metric['time']:
[32]	1032
[34]	1033	# Allready have this metric, abort
[179]	1034	self.slot.release()
[34]	1035	return 1
	1036	else:
	1037	self.myMetrics[ host ][ metric['name'] ] = [ ]
	1038	else:
[32]	1039	self.myMetrics[ host ] = { }
[34]	1040	self.myMetrics[ host ][ metric['name'] ] = [ ]
[32]	1041
[63]	1042	# Push new metric onto stack
	1043	# atomic code; only 1 thread at a time may access the stack
	1044
[32]	1045	self.myMetrics[ host ][ metric['name'] ].append( metric )
	1046
[40]	1047	self.slot.release()
[53]	1048	#
	1049	# </ATOMIC>
[40]	1050
[47]	1051	def makeUpdateList( self, host, metriclist ):
[63]	1052	"""
	1053	Make a list of update values for rrdupdate
	1054	but only those that we didn't store before
	1055	"""
[37]	1056
	1057	update_list = [ ]
[41]	1058	metric = None
[37]	1059
[47]	1060	while len( metriclist ) > 0:
[37]	1061
[53]	1062	metric = metriclist.pop( 0 )
[37]	1063
[53]	1064	if self.checkStoreMetric( host, metric ):
	1065	update_list.append( '%s:%s' %( metric['time'], metric['val'] ) )
[40]	1066
[37]	1067	return update_list
	1068
[49]	1069	def checkStoreMetric( self, host, metric ):
[63]	1070	"""Check if supplied metric if newer than last one stored"""
[40]	1071
	1072	if self.lastStored.has_key( host ):
	1073
[47]	1074	if self.lastStored[ host ].has_key( metric['name'] ):
[40]	1075
[47]	1076	if metric['time'] <= self.lastStored[ host ][ metric['name'] ]:
[40]	1077
[50]	1078	# This is old
[40]	1079	return 0
	1080
[50]	1081	return 1
	1082
[54]	1083	def memLastUpdate( self, host, metricname, metriclist ):
[63]	1084	"""
	1085	Memorize the time of the latest metric from metriclist
	1086	but only if it wasn't allready memorized
	1087	"""
[50]	1088
[54]	1089	if not self.lastStored.has_key( host ):
	1090	self.lastStored[ host ] = { }
	1091
[50]	1092	last_update_time = 0
	1093
	1094	for metric in metriclist:
	1095
[54]	1096	if metric['name'] == metricname:
[50]	1097
[54]	1098	if metric['time'] > last_update_time:
[50]	1099
[54]	1100	last_update_time = metric['time']
[40]	1101
[54]	1102	if self.lastStored[ host ].has_key( metricname ):
[52]	1103
[54]	1104	if last_update_time <= self.lastStored[ host ][ metricname ]:
[52]	1105	return 1
[40]	1106
[54]	1107	self.lastStored[ host ][ metricname ] = last_update_time
[52]	1108
[33]	1109	def storeMetrics( self ):
[63]	1110	"""
	1111	Store all metrics from memory to disk
	1112	and do it to the RRD's in appropriate timeperiod directory
	1113	"""
[33]	1114
	1115	for hostname, mymetrics in self.myMetrics.items():
	1116
	1117	for metricname, mymetric in mymetrics.items():
	1118
[53]	1119	metrics_to_store = [ ]
	1120
[63]	1121	# Pop metrics from stack for storing until none is left
	1122	# atomic code: only 1 thread at a time may access myMetrics
	1123
[53]	1124	# <ATOMIC>
[50]	1125	#
[47]	1126	self.slot.acquire()
[33]	1127
[54]	1128	while len( self.myMetrics[ hostname ][ metricname ] ) > 0:
[53]	1129
[54]	1130	if len( self.myMetrics[ hostname ][ metricname ] ) > 0:
[53]	1131
[176]	1132	try:
	1133	metrics_to_store.append( self.myMetrics[ hostname ][ metricname ].pop( 0 ) )
	1134	except IndexError, msg:
	1135
[179]	1136	# Somehow sometimes myMetrics[ hostname ][ metricname ]
	1137	# is still len 0 when the statement is executed.
	1138	# Just ignore indexerror's..
[176]	1139	pass
	1140
[53]	1141	self.slot.release()
	1142	#
	1143	# </ATOMIC>
	1144
[47]	1145	# Create a mapping table, each metric to the period where it should be stored
	1146	#
[53]	1147	metric_serial_table = self.determineSerials( hostname, metricname, metrics_to_store )
[33]	1148
[50]	1149	update_rets = [ ]
	1150
[47]	1151	for period, pmetric in metric_serial_table.items():
	1152
[146]	1153	create_ret = self.createCheck( hostname, metricname, period )
[47]	1154
	1155	update_ret = self.update( hostname, metricname, period, pmetric )
	1156
	1157	if update_ret == 0:
	1158
	1159	debug_msg( 9, 'stored metric %s for %s' %( hostname, metricname ) )
	1160	else:
	1161	debug_msg( 9, 'metric update failed' )
	1162
[146]	1163	update_rets.append( create_ret )
[50]	1164	update_rets.append( update_ret )
[47]	1165
[179]	1166	# Lets ignore errors here for now, we need to make sure last update time
	1167	# is correct!
	1168	#
	1169	#if not (1) in update_rets:
[50]	1170
[179]	1171	self.memLastUpdate( hostname, metricname, metrics_to_store )
[50]	1172
[17]	1173	def makeTimeSerial( self ):
[63]	1174	"""Generate a time serial. Seconds since epoch"""
[17]	1175
	1176	# Seconds since epoch
	1177	mytime = int( time.time() )
	1178
	1179	return mytime
	1180
[50]	1181	def makeRrdPath( self, host, metricname, timeserial ):
[63]	1182	"""Make a RRD location/path and filename"""
[17]	1183
[50]	1184	rrd_dir = '%s/%s/%s/%s' %( check_dir(ARCHIVE_PATH), self.cluster, host, timeserial )
	1185	rrd_file = '%s/%s.rrd' %( rrd_dir, metricname )
[17]	1186
	1187	return rrd_dir, rrd_file
	1188
[20]	1189	def getLastRrdTimeSerial( self, host ):
[63]	1190	"""Find the last timeserial (directory) for this host"""
[17]	1191
[19]	1192	newest_timeserial = 0
	1193
[47]	1194	for dir in self.timeserials[ host ]:
[32]	1195
[47]	1196	valid_dir = 1
[17]	1197
[47]	1198	for letter in dir:
	1199	if letter not in string.digits:
	1200	valid_dir = 0
[17]	1201
[47]	1202	if valid_dir:
	1203	timeserial = dir
	1204	if timeserial > newest_timeserial:
	1205	newest_timeserial = timeserial
[17]	1206
	1207	if newest_timeserial:
[18]	1208	return newest_timeserial
[17]	1209	else:
	1210	return 0
	1211
[47]	1212	def determinePeriod( self, host, check_serial ):
[63]	1213	"""Determine to which period (directory) this time(serial) belongs"""
[47]	1214
	1215	period_serial = 0
	1216
[56]	1217	if self.timeserials.has_key( host ):
[47]	1218
[56]	1219	for serial in self.timeserials[ host ]:
[47]	1220
[56]	1221	if check_serial >= serial and period_serial < serial:
[47]	1222
[56]	1223	period_serial = serial
	1224
[47]	1225	return period_serial
	1226
	1227	def determineSerials( self, host, metricname, metriclist ):
	1228	"""
	1229	Determine the correct serial and corresponding rrd to store
	1230	for a list of metrics
	1231	"""
	1232
	1233	metric_serial_table = { }
	1234
	1235	for metric in metriclist:
	1236
	1237	if metric['name'] == metricname:
	1238
	1239	period = self.determinePeriod( host, metric['time'] )
	1240
	1241	archive_secs = ARCHIVE_HOURS_PER_RRD * (60 * 60)
	1242
[49]	1243	if (int( metric['time'] ) - int( period ) ) > archive_secs:
[47]	1244
	1245	# This one should get it's own new period
	1246	period = metric['time']
[57]	1247
	1248	if not self.timeserials.has_key( host ):
	1249	self.timeserials[ host ] = [ ]
	1250
[50]	1251	self.timeserials[ host ].append( period )
[47]	1252
	1253	if not metric_serial_table.has_key( period ):
	1254
[49]	1255	metric_serial_table[ period ] = [ ]
[47]	1256
	1257	metric_serial_table[ period ].append( metric )
	1258
	1259	return metric_serial_table
	1260
[33]	1261	def createCheck( self, host, metricname, timeserial ):
[63]	1262	"""Check if an rrd allready exists for this metric, create if not"""
[9]	1263
[35]	1264	debug_msg( 9, 'rrdcreate: using timeserial %s for %s/%s' %( timeserial, host, metricname ) )
[47]	1265
[33]	1266	rrd_dir, rrd_file = self.makeRrdPath( host, metricname, timeserial )
[17]	1267
[9]	1268	if not os.path.exists( rrd_dir ):
[58]	1269
	1270	try:
	1271	os.makedirs( rrd_dir )
	1272
[169]	1273	except os.OSError, msg:
[58]	1274
	1275	if msg.find( 'File exists' ) != -1:
	1276
	1277	# Ignore exists errors
	1278	pass
	1279
	1280	else:
	1281
	1282	print msg
	1283	return
	1284
[14]	1285	debug_msg( 9, 'created dir %s' %( str(rrd_dir) ) )
[9]	1286
[14]	1287	if not os.path.exists( rrd_file ):
[9]	1288
[33]	1289	interval = self.config.getInterval( self.cluster )
[47]	1290	heartbeat = 8 * int( interval )
[9]	1291
[37]	1292	params = [ ]
[12]	1293
[37]	1294	params.append( '--step' )
	1295	params.append( str( interval ) )
[12]	1296
[37]	1297	params.append( '--start' )
[47]	1298	params.append( str( int( timeserial ) - 1 ) )
[12]	1299
[37]	1300	params.append( 'DS:sum:GAUGE:%d:U:U' %heartbeat )
	1301	params.append( 'RRA:AVERAGE:0.5:1:%s' %(ARCHIVE_HOURS_PER_RRD * 240) )
[13]	1302
[37]	1303	self.rrdm.create( str(rrd_file), params )
	1304
[14]	1305	debug_msg( 9, 'created rrd %s' %( str(rrd_file) ) )
	1306
[47]	1307	def update( self, host, metricname, timeserial, metriclist ):
[63]	1308	"""
	1309	Update rrd file for host with metricname
	1310	in directory timeserial with metriclist
	1311	"""
[9]	1312
[35]	1313	debug_msg( 9, 'rrdupdate: using timeserial %s for %s/%s' %( timeserial, host, metricname ) )
[9]	1314
[33]	1315	rrd_dir, rrd_file = self.makeRrdPath( host, metricname, timeserial )
[18]	1316
[47]	1317	update_list = self.makeUpdateList( host, metriclist )
[15]	1318
[41]	1319	if len( update_list ) > 0:
	1320	ret = self.rrdm.update( str(rrd_file), update_list )
[32]	1321
[41]	1322	if ret:
	1323	return 1
[27]	1324
[41]	1325	debug_msg( 9, 'updated rrd %s with %s' %( str(rrd_file), string.join( update_list ) ) )
[15]	1326
[36]	1327	return 0
	1328
[169]	1329	def daemon():
	1330	"""daemonized threading"""
[8]	1331
[169]	1332	# Fork the first child
	1333	#
	1334	pid = os.fork()
	1335
	1336	if pid > 0:
	1337
	1338	sys.exit(0) # end parent
	1339
	1340	# creates a session and sets the process group ID
	1341	#
	1342	os.setsid()
	1343
	1344	# Fork the second child
	1345	#
	1346	pid = os.fork()
	1347
	1348	if pid > 0:
	1349
	1350	sys.exit(0) # end parent
	1351
	1352	# Go to the root directory and set the umask
	1353	#
	1354	os.chdir('/')
	1355	os.umask(0)
	1356
	1357	sys.stdin.close()
	1358	sys.stdout.close()
	1359	sys.stderr.close()
	1360
	1361	os.open('/dev/null', 0)
	1362	os.dup(0)
	1363	os.dup(0)
	1364
	1365	run()
	1366
	1367	def run():
	1368	"""Threading start"""
	1369
[102]	1370	myTorqueProcessor = TorqueXMLProcessor()
	1371	myGangliaProcessor = GangliaXMLProcessor()
[8]	1372
[169]	1373	try:
[102]	1374	torque_xml_thread = threading.Thread( None, myTorqueProcessor.run, 'torque_proc_thread' )
	1375	ganglia_xml_thread = threading.Thread( None, myGangliaProcessor.run, 'ganglia_proc_thread' )
[22]	1376
[169]	1377	torque_xml_thread.start()
	1378	ganglia_xml_thread.start()
	1379
[176]	1380	except thread.error, msg:
[169]	1381	debug_msg( 0, 'FATAL ERROR: Unable to start main threads!: '+ str(msg) )
	1382	syslog.closelog()
	1383	sys.exit(1)
	1384
	1385	debug_msg( 0, 'main threading started.' )
[78]	1386
[169]	1387	def main():
	1388	"""Program startup"""
	1389
	1390	if( DAEMONIZE and USE_SYSLOG ):
	1391	syslog.openlog( 'jobarchived', syslog.LOG_NOWAIT, SYSLOG_FACILITY )
	1392
	1393	if DAEMONIZE:
	1394	daemon()
	1395	else:
	1396	run()
	1397
	1398	#
[81]	1399	# Global functions
[169]	1400	#
[81]	1401
[9]	1402	def check_dir( directory ):
[63]	1403	"""Check if directory is a proper directory. I.e.: Does _not_ end with a '/'"""
[9]	1404
	1405	if directory[-1] == '/':
	1406	directory = directory[:-1]
	1407
	1408	return directory
	1409
[12]	1410	def debug_msg( level, msg ):
[169]	1411	"""Only print msg if correct levels"""
[12]	1412
[169]	1413	if (not DAEMONIZE and DEBUG_LEVEL >= level):
	1414	sys.stderr.write( printTime() + ' - ' + msg + '\n' )
	1415
	1416	if (DAEMONIZE and USE_SYSLOG and SYSLOG_LEVEL >= level):
	1417	syslog.syslog( msg )
[12]	1418
[46]	1419	def printTime( ):
[63]	1420	"""Print current time in human readable format"""
[46]	1421
	1422	return time.strftime("%a %d %b %Y %H:%M:%S")
	1423
[63]	1424	# Ooohh, someone started me! Let's go..
[9]	1425	if __name__ == '__main__':
	1426	main()

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

Download in other formats: