[3] | 1 | #!/usr/bin/env python |
---|
| 2 | |
---|
[5] | 3 | from xml.sax import make_parser |
---|
| 4 | from xml.sax.handler import ContentHandler |
---|
| 5 | import socket |
---|
| 6 | import sys |
---|
[9] | 7 | import rrdtool |
---|
| 8 | import string |
---|
[12] | 9 | import os |
---|
| 10 | import os.path |
---|
[17] | 11 | import time |
---|
| 12 | import re |
---|
[3] | 13 | |
---|
[8] | 14 | # Specify debugging level here; |
---|
| 15 | # |
---|
[22] | 16 | # >=11 = metric XML |
---|
| 17 | # >=10 = host,cluster,grid,ganglia XML |
---|
| 18 | # >=9 = RRD activity,gmetad config parsing |
---|
| 19 | # >=7 = daemon threading |
---|
[8] | 20 | # |
---|
[19] | 21 | DEBUG_LEVEL = 7 |
---|
[6] | 22 | |
---|
[9] | 23 | # Where is the gmetad.conf located |
---|
| 24 | # |
---|
| 25 | GMETAD_CONF = '/etc/gmetad.conf' |
---|
| 26 | |
---|
| 27 | # Where to store the archived rrd's |
---|
| 28 | # |
---|
| 29 | ARCHIVE_PATH = '/data/toga/rrds' |
---|
| 30 | |
---|
| 31 | # List of data_source names to archive for |
---|
| 32 | # |
---|
| 33 | ARCHIVE_SOURCES = [ "LISA Cluster" ] |
---|
| 34 | |
---|
[13] | 35 | # Amount of hours to store in one single archived .rrd |
---|
[9] | 36 | # |
---|
[22] | 37 | ARCHIVE_HOURS_PER_RRD = 12 |
---|
[13] | 38 | |
---|
[21] | 39 | # Interval at which to grab&store XML |
---|
| 40 | # |
---|
| 41 | GRAB_INTERVAL = 15 |
---|
| 42 | |
---|
[22] | 43 | # Wether or not to run as a daemon in background |
---|
| 44 | # |
---|
| 45 | DAEMONIZE = 0 |
---|
| 46 | |
---|
[17] | 47 | ###################### |
---|
| 48 | # # |
---|
| 49 | # Configuration ends # |
---|
| 50 | # # |
---|
| 51 | ###################### |
---|
[13] | 52 | |
---|
[17] | 53 | # What XML data types not to store |
---|
[13] | 54 | # |
---|
[17] | 55 | UNSUPPORTED_ARCHIVE_TYPES = [ 'string' ] |
---|
[9] | 56 | |
---|
[8] | 57 | """ |
---|
| 58 | This is TOrque-GAnglia's data Daemon |
---|
| 59 | """ |
---|
| 60 | |
---|
[6] | 61 | class GangliaXMLHandler( ContentHandler ): |
---|
[8] | 62 | "Parse Ganglia's XML" |
---|
[3] | 63 | |
---|
[6] | 64 | metrics = [ ] |
---|
| 65 | |
---|
| 66 | def startElement( self, name, attrs ): |
---|
[8] | 67 | "Store appropriate data from xml start tags" |
---|
[3] | 68 | |
---|
[7] | 69 | if name == 'GANGLIA_XML': |
---|
| 70 | self.XMLSource = attrs.get('SOURCE',"") |
---|
| 71 | self.gangliaVersion = attrs.get('VERSION',"") |
---|
[12] | 72 | debug_msg( 10, 'Found XML data: source %s version %s' %( self.XMLSource, self.gangliaVersion ) ) |
---|
[6] | 73 | |
---|
[7] | 74 | elif name == 'GRID': |
---|
| 75 | self.gridName = attrs.get('NAME',"") |
---|
[12] | 76 | self.time = attrs.get('LOCALTIME',"") |
---|
| 77 | debug_msg( 10, '`-Grid found: %s' %( self.gridName ) ) |
---|
[6] | 78 | |
---|
[7] | 79 | elif name == 'CLUSTER': |
---|
| 80 | self.clusterName = attrs.get('NAME',"") |
---|
[12] | 81 | self.time = attrs.get('LOCALTIME',"") |
---|
[9] | 82 | self.rrd = RRDHandler( self.clusterName ) |
---|
[12] | 83 | debug_msg( 10, ' |-Cluster found: %s' %( self.clusterName ) ) |
---|
[6] | 84 | |
---|
[14] | 85 | elif name == 'HOST' and self.clusterName in ARCHIVE_SOURCES: |
---|
[7] | 86 | self.hostName = attrs.get('NAME',"") |
---|
| 87 | self.hostIp = attrs.get('IP',"") |
---|
| 88 | self.hostReported = attrs.get('REPORTED',"") |
---|
[9] | 89 | # Reset the metrics list for each host |
---|
| 90 | self.metrics = [ ] |
---|
[12] | 91 | debug_msg( 10, ' | |-Host found: %s - ip %s reported %s' %( self.hostName, self.hostIp, self.hostReported ) ) |
---|
[6] | 92 | |
---|
[14] | 93 | elif name == 'METRIC' and self.clusterName in ARCHIVE_SOURCES: |
---|
[6] | 94 | myMetric = { } |
---|
[7] | 95 | myMetric['name'] = attrs.get('NAME',"") |
---|
| 96 | myMetric['val'] = attrs.get('VAL',"") |
---|
[12] | 97 | myMetric['time'] = self.time |
---|
[16] | 98 | myMetric['type'] = attrs.get('TYPE',"") |
---|
[6] | 99 | |
---|
| 100 | self.metrics.append( myMetric ) |
---|
[12] | 101 | debug_msg( 11, ' | | |-metric: %s:%s' %( myMetric['name'], myMetric['val'] ) ) |
---|
[6] | 102 | |
---|
[5] | 103 | return |
---|
[3] | 104 | |
---|
[9] | 105 | def endElement( self, name ): |
---|
[12] | 106 | #if name == 'GANGLIA_XML': |
---|
[3] | 107 | |
---|
[12] | 108 | #if name == 'GRID': |
---|
[3] | 109 | |
---|
[12] | 110 | #if name == 'CLUSTER': |
---|
[6] | 111 | |
---|
[14] | 112 | if name == 'HOST' and self.clusterName in ARCHIVE_SOURCES: |
---|
[6] | 113 | |
---|
[18] | 114 | # Determine time here, so all use same time in this run |
---|
| 115 | mytime = self.rrd.makeTimeSerial() |
---|
[20] | 116 | correct_serial = self.rrd.checkNewRrdPeriod( self.hostName, mytime ) |
---|
[18] | 117 | |
---|
[20] | 118 | self.storeMetrics( self.hostName, correct_serial ) |
---|
[18] | 119 | |
---|
[12] | 120 | #if name == 'METRIC': |
---|
[6] | 121 | |
---|
[18] | 122 | def storeMetrics( self, hostname, timeserial ): |
---|
[9] | 123 | |
---|
| 124 | for metric in self.metrics: |
---|
[17] | 125 | if metric['type'] not in UNSUPPORTED_ARCHIVE_TYPES: |
---|
[16] | 126 | |
---|
[18] | 127 | self.rrd.createCheck( hostname, metric, timeserial ) |
---|
| 128 | self.rrd.update( hostname, metric, timeserial ) |
---|
[16] | 129 | debug_msg( 9, 'stored metric %s for %s: %s' %( hostname, metric['name'], metric['val'] ) ) |
---|
[19] | 130 | #sys.exit(1) |
---|
[9] | 131 | |
---|
| 132 | |
---|
[5] | 133 | class GangliaXMLGatherer: |
---|
[8] | 134 | "Setup a connection and file object to Ganglia's XML" |
---|
[3] | 135 | |
---|
[8] | 136 | s = None |
---|
| 137 | |
---|
| 138 | def __init__( self, host, port ): |
---|
| 139 | "Store host and port for connection" |
---|
| 140 | |
---|
[5] | 141 | self.host = host |
---|
| 142 | self.port = port |
---|
[3] | 143 | |
---|
[8] | 144 | def __del__( self ): |
---|
| 145 | "Kill the socket before we leave" |
---|
| 146 | |
---|
| 147 | self.s.close() |
---|
| 148 | |
---|
| 149 | def getFileObject( self ): |
---|
| 150 | "Connect, and return a file object" |
---|
| 151 | |
---|
| 152 | for res in socket.getaddrinfo( self.host, self.port, socket.AF_UNSPEC, socket.SOCK_STREAM ): |
---|
[5] | 153 | af, socktype, proto, canonname, sa = res |
---|
| 154 | try: |
---|
[8] | 155 | self.s = socket.socket( af, socktype, proto ) |
---|
[5] | 156 | except socket.error, msg: |
---|
[8] | 157 | self.s = None |
---|
[5] | 158 | continue |
---|
| 159 | try: |
---|
[8] | 160 | self.s.connect( sa ) |
---|
[5] | 161 | except socket.error, msg: |
---|
[8] | 162 | self.s.close() |
---|
| 163 | self.s = None |
---|
[5] | 164 | continue |
---|
| 165 | break |
---|
[3] | 166 | |
---|
[8] | 167 | if self.s is None: |
---|
| 168 | print 'Could not open socket' |
---|
[5] | 169 | sys.exit(1) |
---|
| 170 | |
---|
[8] | 171 | return self.s.makefile( 'r' ) |
---|
[5] | 172 | |
---|
[8] | 173 | class GangliaXMLProcessor: |
---|
[5] | 174 | |
---|
[9] | 175 | def daemon( self ): |
---|
[8] | 176 | "Run as daemon forever" |
---|
[5] | 177 | |
---|
[8] | 178 | self.DAEMON = 1 |
---|
[5] | 179 | |
---|
[8] | 180 | # Fork the first child |
---|
| 181 | # |
---|
| 182 | pid = os.fork() |
---|
| 183 | if pid > 0: |
---|
| 184 | sys.exit(0) # end parrent |
---|
[7] | 185 | |
---|
[8] | 186 | # creates a session and sets the process group ID |
---|
| 187 | # |
---|
| 188 | os.setsid() |
---|
[7] | 189 | |
---|
[8] | 190 | # Fork the second child |
---|
| 191 | # |
---|
| 192 | pid = os.fork() |
---|
| 193 | if pid > 0: |
---|
| 194 | sys.exit(0) # end parrent |
---|
[5] | 195 | |
---|
[8] | 196 | # Go to the root directory and set the umask |
---|
| 197 | # |
---|
| 198 | os.chdir('/') |
---|
| 199 | os.umask(0) |
---|
| 200 | |
---|
| 201 | sys.stdin.close() |
---|
| 202 | sys.stdout.close() |
---|
[21] | 203 | sys.stderr.close() |
---|
[8] | 204 | |
---|
| 205 | os.open('/dev/null', 0) |
---|
| 206 | os.dup(0) |
---|
| 207 | os.dup(0) |
---|
| 208 | |
---|
| 209 | self.run() |
---|
| 210 | |
---|
[22] | 211 | def printTime( self ): |
---|
| 212 | |
---|
| 213 | return time.strftime("%a, %d %b %Y %H:%M:%S") |
---|
| 214 | |
---|
[9] | 215 | def run( self ): |
---|
[8] | 216 | "Main thread" |
---|
| 217 | |
---|
| 218 | while ( 1 ): |
---|
| 219 | |
---|
[22] | 220 | debug_msg( 7, self.printTime() + ' - mainthread() - xmlthread() started' ) |
---|
| 221 | pid = os.fork() |
---|
| 222 | |
---|
| 223 | if pid == 0: |
---|
| 224 | # Child - XML Thread |
---|
| 225 | # |
---|
| 226 | # Process XML and exit |
---|
| 227 | |
---|
| 228 | debug_msg( 7, self.printTime() + ' - xmlthread() - Start XML processing..' ) |
---|
| 229 | self.processXML() |
---|
| 230 | debug_msg( 7, self.printTime() + ' - xmlthread() - Done processing; exiting.' ) |
---|
| 231 | sys.exit(0) |
---|
| 232 | |
---|
| 233 | elif pid > 0: |
---|
| 234 | # Parent - Daemon Thread |
---|
| 235 | |
---|
| 236 | debug_msg( 7, self.printTime() + ' - mainthread() - Sleep '+ str(GRAB_INTERVAL) +'s: zzzzz..' ) |
---|
| 237 | time.sleep( GRAB_INTERVAL ) |
---|
| 238 | debug_msg( 7, self.printTime() + ' - mainthread() - Awoken: waiting for XML thread..' ) |
---|
| 239 | |
---|
| 240 | r = os.wait() |
---|
| 241 | |
---|
| 242 | debug_msg( 7, self.printTime() + ' - mainthread() - Done waiting.' ) |
---|
| 243 | |
---|
[8] | 244 | def processXML( self ): |
---|
| 245 | "Process XML" |
---|
| 246 | |
---|
| 247 | myXMLGatherer = GangliaXMLGatherer( 'localhost', 8651 ) |
---|
| 248 | |
---|
| 249 | myParser = make_parser() |
---|
| 250 | myHandler = GangliaXMLHandler() |
---|
| 251 | myParser.setContentHandler( myHandler ) |
---|
| 252 | |
---|
| 253 | myParser.parse( myXMLGatherer.getFileObject() ) |
---|
| 254 | |
---|
[9] | 255 | class GangliaConfigParser: |
---|
| 256 | |
---|
| 257 | sources = [ ] |
---|
| 258 | |
---|
| 259 | def __init__( self, config ): |
---|
| 260 | self.config = config |
---|
| 261 | self.parseValues() |
---|
| 262 | |
---|
| 263 | def parseValues(self): |
---|
| 264 | "Parse certain values from gmetad.conf" |
---|
| 265 | |
---|
| 266 | readcfg = open( self.config, 'r' ) |
---|
| 267 | |
---|
| 268 | for line in readcfg.readlines(): |
---|
| 269 | |
---|
| 270 | if line.count( '"' ) > 1: |
---|
| 271 | |
---|
[10] | 272 | if line.find( 'data_source' ) != -1 and line[0] != '#': |
---|
[9] | 273 | |
---|
[11] | 274 | source = { } |
---|
| 275 | source['name'] = line.split( '"' )[1] |
---|
[9] | 276 | source_words = line.split( '"' )[2].split( ' ' ) |
---|
| 277 | |
---|
| 278 | for word in source_words: |
---|
| 279 | |
---|
| 280 | valid_interval = 1 |
---|
| 281 | |
---|
| 282 | for letter in word: |
---|
| 283 | if letter not in string.digits: |
---|
| 284 | valid_interval = 0 |
---|
| 285 | |
---|
[10] | 286 | if valid_interval and len(word) > 0: |
---|
[9] | 287 | source['interval'] = word |
---|
[12] | 288 | debug_msg( 9, 'polling interval for %s = %s' %(source['name'], source['interval'] ) ) |
---|
[9] | 289 | |
---|
| 290 | # No interval found, use Ganglia's default |
---|
| 291 | if not source.has_key( 'interval' ): |
---|
| 292 | source['interval'] = 15 |
---|
[12] | 293 | debug_msg( 9, 'polling interval for %s defaulted to 15' %(source['name']) ) |
---|
[9] | 294 | |
---|
| 295 | self.sources.append( source ) |
---|
| 296 | |
---|
| 297 | def getInterval( self, source_name ): |
---|
| 298 | for source in self.sources: |
---|
[12] | 299 | if source['name'] == source_name: |
---|
[9] | 300 | return source['interval'] |
---|
| 301 | return None |
---|
| 302 | |
---|
| 303 | class RRDHandler: |
---|
| 304 | |
---|
| 305 | def __init__( self, cluster ): |
---|
| 306 | self.cluster = cluster |
---|
| 307 | self.gmetad_conf = GangliaConfigParser( GMETAD_CONF ) |
---|
| 308 | |
---|
[17] | 309 | def makeTimeSerial( self ): |
---|
| 310 | |
---|
| 311 | # YYYYMMDDhhmmss: 20050321143411 |
---|
| 312 | #mytime = time.strftime( "%Y%m%d%H%M%S" ) |
---|
| 313 | |
---|
| 314 | # Seconds since epoch |
---|
| 315 | mytime = int( time.time() ) |
---|
| 316 | |
---|
| 317 | return mytime |
---|
| 318 | |
---|
[20] | 319 | def makeRrdPath( self, host, metric=None, timeserial=None ): |
---|
[17] | 320 | |
---|
[20] | 321 | if not timeserial: |
---|
| 322 | rrd_dir = '%s/%s/%s' %( check_dir(ARCHIVE_PATH), self.cluster, host ) |
---|
| 323 | else: |
---|
| 324 | rrd_dir = '%s/%s/%s/%s' %( check_dir(ARCHIVE_PATH), self.cluster, host, timeserial ) |
---|
| 325 | if metric: |
---|
| 326 | rrd_file = '%s/%s.rrd' %( rrd_dir, metric['name'] ) |
---|
| 327 | else: |
---|
| 328 | rrd_file = None |
---|
[17] | 329 | |
---|
| 330 | return rrd_dir, rrd_file |
---|
| 331 | |
---|
[20] | 332 | def getLastRrdTimeSerial( self, host ): |
---|
[17] | 333 | |
---|
[20] | 334 | rrd_dir, rrd_file = self.makeRrdPath( host ) |
---|
[17] | 335 | |
---|
[19] | 336 | newest_timeserial = 0 |
---|
| 337 | |
---|
[17] | 338 | if os.path.exists( rrd_dir ): |
---|
| 339 | for root, dirs, files in os.walk( rrd_dir ): |
---|
| 340 | |
---|
[20] | 341 | for dir in dirs: |
---|
[17] | 342 | |
---|
[20] | 343 | valid_dir = 1 |
---|
[17] | 344 | |
---|
[20] | 345 | for letter in dir: |
---|
| 346 | if letter not in string.digits: |
---|
| 347 | valid_dir = 0 |
---|
[17] | 348 | |
---|
[20] | 349 | if valid_dir: |
---|
| 350 | timeserial = dir |
---|
[17] | 351 | if timeserial > newest_timeserial: |
---|
| 352 | newest_timeserial = timeserial |
---|
| 353 | |
---|
| 354 | if newest_timeserial: |
---|
[18] | 355 | return newest_timeserial |
---|
[17] | 356 | else: |
---|
| 357 | return 0 |
---|
| 358 | |
---|
[20] | 359 | def checkNewRrdPeriod( self, host, current_timeserial ): |
---|
[17] | 360 | |
---|
[20] | 361 | last_timeserial = int( self.getLastRrdTimeSerial( host ) ) |
---|
| 362 | debug_msg( 8, 'last timeserial of %s is %s' %( host, last_timeserial ) ) |
---|
[17] | 363 | |
---|
| 364 | if not last_timeserial: |
---|
[18] | 365 | serial = current_timeserial |
---|
| 366 | else: |
---|
[17] | 367 | |
---|
[18] | 368 | archive_secs = ARCHIVE_HOURS_PER_RRD * (60 * 60) |
---|
[17] | 369 | |
---|
[18] | 370 | if (current_timeserial - last_timeserial) >= archive_secs: |
---|
| 371 | serial = current_timeserial |
---|
| 372 | else: |
---|
| 373 | serial = last_timeserial |
---|
[17] | 374 | |
---|
[18] | 375 | return serial |
---|
| 376 | |
---|
[20] | 377 | def createCheck( self, host, metric, timeserial ): |
---|
[9] | 378 | "Check if an .rrd allready exists for this metric, create if not" |
---|
| 379 | |
---|
[18] | 380 | debug_msg( 8, 'rrdcreate: using timeserial %s for %s/%s' %( timeserial, host, metric['name'] ) ) |
---|
[9] | 381 | |
---|
[17] | 382 | rrd_dir, rrd_file = self.makeRrdPath( host, metric, timeserial ) |
---|
| 383 | |
---|
[9] | 384 | if not os.path.exists( rrd_dir ): |
---|
| 385 | os.makedirs( rrd_dir ) |
---|
[14] | 386 | debug_msg( 9, 'created dir %s' %( str(rrd_dir) ) ) |
---|
[9] | 387 | |
---|
[14] | 388 | if not os.path.exists( rrd_file ): |
---|
[9] | 389 | |
---|
[14] | 390 | interval = self.gmetad_conf.getInterval( self.cluster ) |
---|
| 391 | heartbeat = 8 * int(interval) |
---|
[9] | 392 | |
---|
[14] | 393 | param_step1 = '--step' |
---|
| 394 | param_step2 = str( interval ) |
---|
[12] | 395 | |
---|
[14] | 396 | param_start1 = '--start' |
---|
| 397 | param_start2 = str( int( metric['time'] ) - 1 ) |
---|
[12] | 398 | |
---|
[14] | 399 | param_ds = 'DS:sum:GAUGE:%d:U:U' %heartbeat |
---|
| 400 | param_rra = 'RRA:AVERAGE:0.5:1:%s' %(ARCHIVE_HOURS_PER_RRD * 240) |
---|
[12] | 401 | |
---|
[14] | 402 | rrdtool.create( str(rrd_file), param_step1, param_step2, param_start1, param_start2, param_ds, param_rra ) |
---|
[13] | 403 | |
---|
[14] | 404 | debug_msg( 9, 'created rrd %s' %( str(rrd_file) ) ) |
---|
| 405 | |
---|
[20] | 406 | def update( self, host, metric, timeserial ): |
---|
[9] | 407 | |
---|
[18] | 408 | debug_msg( 8, 'rrdupdate: using timeserial %s for %s/%s' %( timeserial, host, metric['name'] ) ) |
---|
[9] | 409 | |
---|
[18] | 410 | rrd_dir, rrd_file = self.makeRrdPath( host, metric, timeserial ) |
---|
| 411 | |
---|
[15] | 412 | timestamp = metric['time'] |
---|
| 413 | val = metric['val'] |
---|
[9] | 414 | |
---|
[15] | 415 | update_string = '%s:%s' %(timestamp, val) |
---|
| 416 | |
---|
| 417 | rrdtool.update( str(rrd_file), str(update_string) ) |
---|
| 418 | debug_msg( 9, 'updated rrd %s with %s' %( str(rrd_file), update_string ) ) |
---|
| 419 | |
---|
[8] | 420 | def main(): |
---|
| 421 | "Program startup" |
---|
| 422 | |
---|
| 423 | myProcessor = GangliaXMLProcessor() |
---|
| 424 | |
---|
[22] | 425 | if DAEMONIZE: |
---|
| 426 | myProcessor.daemon() |
---|
| 427 | else: |
---|
| 428 | myProcessor.run() |
---|
| 429 | |
---|
[9] | 430 | def check_dir( directory ): |
---|
| 431 | "Check if directory is a proper directory. I.e.: Does _not_ end with a '/'" |
---|
| 432 | |
---|
| 433 | if directory[-1] == '/': |
---|
| 434 | directory = directory[:-1] |
---|
| 435 | |
---|
| 436 | return directory |
---|
| 437 | |
---|
[12] | 438 | def debug_msg( level, msg ): |
---|
| 439 | |
---|
| 440 | if (DEBUG_LEVEL >= level): |
---|
| 441 | sys.stderr.write( msg + '\n' ) |
---|
| 442 | |
---|
[5] | 443 | # Let's go |
---|
[9] | 444 | if __name__ == '__main__': |
---|
| 445 | main() |
---|