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