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