[3] | 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 | # |
---|
[230] | 21 | # SVN $Id: jobarchived.py 324 2007-04-20 13:08:23Z bastiaans $ |
---|
| 22 | # |
---|
[3] | 23 | |
---|
[284] | 24 | DEFAULT_SEARCH_PATH = '/usr/share/jobarchived' |
---|
[3] | 25 | |
---|
[284] | 26 | import sys |
---|
| 27 | |
---|
| 28 | if DEFAULT_SEARCH_PATH not in sys.path: |
---|
| 29 | |
---|
| 30 | sys.path.append( DEFAULT_SEARCH_PATH ) |
---|
| 31 | |
---|
| 32 | import getopt, syslog, ConfigParser |
---|
| 33 | |
---|
[214] | 34 | def processArgs( args ): |
---|
[6] | 35 | |
---|
[292] | 36 | SHORT_L = 'c:' |
---|
| 37 | LONG_L = 'config=' |
---|
[169] | 38 | |
---|
[214] | 39 | config_filename = None |
---|
[169] | 40 | |
---|
[214] | 41 | try: |
---|
[169] | 42 | |
---|
[214] | 43 | opts, args = getopt.getopt( args, SHORT_L, LONG_L ) |
---|
[9] | 44 | |
---|
[214] | 45 | except getopt.error, detail: |
---|
[60] | 46 | |
---|
[214] | 47 | print detail |
---|
| 48 | sys.exit(1) |
---|
[9] | 49 | |
---|
[214] | 50 | for opt, value in opts: |
---|
[60] | 51 | |
---|
[214] | 52 | if opt in [ '--config', '-c' ]: |
---|
[13] | 53 | |
---|
[214] | 54 | config_filename = value |
---|
[198] | 55 | |
---|
[214] | 56 | if not config_filename: |
---|
[60] | 57 | |
---|
[214] | 58 | config_filename = '/etc/jobarchived.conf' |
---|
[22] | 59 | |
---|
[214] | 60 | try: |
---|
| 61 | return loadConfig( config_filename ) |
---|
[13] | 62 | |
---|
[214] | 63 | except ConfigParser.NoOptionError, detail: |
---|
| 64 | |
---|
| 65 | print detail |
---|
| 66 | sys.exit( 1 ) |
---|
| 67 | |
---|
| 68 | def loadConfig( filename ): |
---|
| 69 | |
---|
| 70 | def getlist( cfg_string ): |
---|
| 71 | |
---|
| 72 | my_list = [ ] |
---|
| 73 | |
---|
| 74 | for item_txt in cfg_string.split( ',' ): |
---|
| 75 | |
---|
| 76 | sep_char = None |
---|
| 77 | |
---|
| 78 | item_txt = item_txt.strip() |
---|
| 79 | |
---|
| 80 | for s_char in [ "'", '"' ]: |
---|
| 81 | |
---|
| 82 | if item_txt.find( s_char ) != -1: |
---|
| 83 | |
---|
| 84 | if item_txt.count( s_char ) != 2: |
---|
| 85 | |
---|
| 86 | print 'Missing quote: %s' %item_txt |
---|
| 87 | sys.exit( 1 ) |
---|
| 88 | |
---|
| 89 | else: |
---|
| 90 | |
---|
| 91 | sep_char = s_char |
---|
| 92 | break |
---|
| 93 | |
---|
| 94 | if sep_char: |
---|
| 95 | |
---|
| 96 | item_txt = item_txt.split( sep_char )[1] |
---|
| 97 | |
---|
| 98 | my_list.append( item_txt ) |
---|
| 99 | |
---|
| 100 | return my_list |
---|
| 101 | |
---|
| 102 | cfg = ConfigParser.ConfigParser() |
---|
| 103 | |
---|
| 104 | cfg.read( filename ) |
---|
| 105 | |
---|
[295] | 106 | global DEBUG_LEVEL, USE_SYSLOG, SYSLOG_LEVEL, SYSLOG_FACILITY, GMETAD_CONF, ARCHIVE_XMLSOURCE, ARCHIVE_DATASOURCES, ARCHIVE_PATH, ARCHIVE_HOURS_PER_RRD, ARCHIVE_EXCLUDE_METRICS, JOB_SQL_DBASE, DAEMONIZE, RRDTOOL, JOB_TIMEOUT |
---|
[214] | 107 | |
---|
[292] | 108 | ARCHIVE_PATH = cfg.get( 'DEFAULT', 'ARCHIVE_PATH' ) |
---|
[214] | 109 | |
---|
[292] | 110 | ARCHIVE_HOURS_PER_RRD = cfg.getint( 'DEFAULT', 'ARCHIVE_HOURS_PER_RRD' ) |
---|
[214] | 111 | |
---|
[292] | 112 | DEBUG_LEVEL = cfg.getint( 'DEFAULT', 'DEBUG_LEVEL' ) |
---|
[214] | 113 | |
---|
[292] | 114 | USE_SYSLOG = cfg.getboolean( 'DEFAULT', 'USE_SYSLOG' ) |
---|
[214] | 115 | |
---|
[292] | 116 | SYSLOG_LEVEL = cfg.getint( 'DEFAULT', 'SYSLOG_LEVEL' ) |
---|
[214] | 117 | |
---|
| 118 | try: |
---|
| 119 | |
---|
[292] | 120 | SYSLOG_FACILITY = eval( 'syslog.LOG_' + cfg.get( 'DEFAULT', 'SYSLOG_FACILITY' ) ) |
---|
[214] | 121 | |
---|
| 122 | except AttributeError, detail: |
---|
| 123 | |
---|
| 124 | print 'Unknown syslog facility' |
---|
| 125 | sys.exit( 1 ) |
---|
| 126 | |
---|
[292] | 127 | GMETAD_CONF = cfg.get( 'DEFAULT', 'GMETAD_CONF' ) |
---|
[214] | 128 | |
---|
[292] | 129 | ARCHIVE_XMLSOURCE = cfg.get( 'DEFAULT', 'ARCHIVE_XMLSOURCE' ) |
---|
[214] | 130 | |
---|
[292] | 131 | ARCHIVE_DATASOURCES = getlist( cfg.get( 'DEFAULT', 'ARCHIVE_DATASOURCES' ) ) |
---|
[214] | 132 | |
---|
[292] | 133 | ARCHIVE_EXCLUDE_METRICS = getlist( cfg.get( 'DEFAULT', 'ARCHIVE_EXCLUDE_METRICS' ) ) |
---|
[214] | 134 | |
---|
[292] | 135 | JOB_SQL_DBASE = cfg.get( 'DEFAULT', 'JOB_SQL_DBASE' ) |
---|
[214] | 136 | |
---|
[295] | 137 | JOB_TIMEOUT = cfg.getint( 'DEFAULT', 'JOB_TIMEOUT' ) |
---|
| 138 | |
---|
[292] | 139 | DAEMONIZE = cfg.getboolean( 'DEFAULT', 'DAEMONIZE' ) |
---|
[214] | 140 | |
---|
[292] | 141 | RRDTOOL = cfg.get( 'DEFAULT', 'RRDTOOL' ) |
---|
[224] | 142 | |
---|
[214] | 143 | return True |
---|
| 144 | |
---|
[17] | 145 | # What XML data types not to store |
---|
[13] | 146 | # |
---|
[17] | 147 | UNSUPPORTED_ARCHIVE_TYPES = [ 'string' ] |
---|
[9] | 148 | |
---|
[47] | 149 | # Maximum time (in seconds) a parsethread may run |
---|
| 150 | # |
---|
| 151 | PARSE_TIMEOUT = 60 |
---|
| 152 | |
---|
| 153 | # Maximum time (in seconds) a storethread may run |
---|
| 154 | # |
---|
| 155 | STORE_TIMEOUT = 360 |
---|
| 156 | |
---|
[8] | 157 | """ |
---|
[224] | 158 | The Job Archiving Daemon |
---|
[8] | 159 | """ |
---|
| 160 | |
---|
[214] | 161 | from types import * |
---|
| 162 | |
---|
| 163 | import DBClass |
---|
| 164 | import xml.sax, xml.sax.handler, socket, string, os, os.path, time, thread, threading, random, re |
---|
| 165 | |
---|
[84] | 166 | class DataSQLStore: |
---|
| 167 | |
---|
| 168 | db_vars = None |
---|
| 169 | dbc = None |
---|
| 170 | |
---|
| 171 | def __init__( self, hostname, database ): |
---|
| 172 | |
---|
| 173 | self.db_vars = DBClass.InitVars(DataBaseName=database, |
---|
| 174 | User='root', |
---|
| 175 | Host=hostname, |
---|
| 176 | Password='', |
---|
| 177 | Dictionary='true') |
---|
| 178 | |
---|
| 179 | try: |
---|
| 180 | self.dbc = DBClass.DB(self.db_vars) |
---|
| 181 | except DBClass.DBError, details: |
---|
[169] | 182 | debug_msg( 0, 'FATAL ERROR: Unable to connect to database!: ' +str(details) ) |
---|
[84] | 183 | sys.exit(1) |
---|
| 184 | |
---|
| 185 | def setDatabase(self, statement): |
---|
| 186 | ret = self.doDatabase('set', statement) |
---|
| 187 | return ret |
---|
| 188 | |
---|
| 189 | def getDatabase(self, statement): |
---|
| 190 | ret = self.doDatabase('get', statement) |
---|
| 191 | return ret |
---|
| 192 | |
---|
| 193 | def doDatabase(self, type, statement): |
---|
| 194 | |
---|
| 195 | debug_msg( 6, 'doDatabase(): %s: %s' %(type, statement) ) |
---|
| 196 | try: |
---|
| 197 | if type == 'set': |
---|
| 198 | result = self.dbc.Set( statement ) |
---|
| 199 | self.dbc.Commit() |
---|
| 200 | elif type == 'get': |
---|
| 201 | result = self.dbc.Get( statement ) |
---|
| 202 | |
---|
| 203 | except DBClass.DBError, detail: |
---|
| 204 | operation = statement.split(' ')[0] |
---|
[169] | 205 | debug_msg( 0, 'FATAL ERROR: ' +operation+ ' on database failed while doing ['+statement+'] full msg: '+str(detail) ) |
---|
[84] | 206 | sys.exit(1) |
---|
| 207 | |
---|
| 208 | debug_msg( 6, 'doDatabase(): result: %s' %(result) ) |
---|
| 209 | return result |
---|
| 210 | |
---|
[191] | 211 | def getJobNodeId( self, job_id, node_id ): |
---|
| 212 | |
---|
| 213 | id = self.getDatabase( "SELECT job_id,node_id FROM job_nodes WHERE job_id = '%s' AND node_id = '%s'" %(job_id, node_id) ) |
---|
| 214 | if len( id ) > 0: |
---|
| 215 | |
---|
| 216 | if len( id[0] ) > 0 and id[0] != '': |
---|
| 217 | |
---|
| 218 | return 1 |
---|
| 219 | |
---|
| 220 | return 0 |
---|
| 221 | |
---|
[84] | 222 | def getNodeId( self, hostname ): |
---|
| 223 | |
---|
[89] | 224 | id = self.getDatabase( "SELECT node_id FROM nodes WHERE node_hostname = '%s'" %hostname ) |
---|
[84] | 225 | |
---|
[89] | 226 | if len( id ) > 0: |
---|
[84] | 227 | |
---|
[89] | 228 | id = id[0][0] |
---|
| 229 | |
---|
[84] | 230 | return id |
---|
| 231 | else: |
---|
| 232 | return None |
---|
| 233 | |
---|
| 234 | def getNodeIds( self, hostnames ): |
---|
| 235 | |
---|
| 236 | ids = [ ] |
---|
| 237 | |
---|
| 238 | for node in hostnames: |
---|
| 239 | |
---|
| 240 | id = self.getNodeId( node ) |
---|
| 241 | |
---|
| 242 | if id: |
---|
| 243 | ids.append( id ) |
---|
| 244 | |
---|
| 245 | return ids |
---|
| 246 | |
---|
| 247 | def getJobId( self, jobid ): |
---|
| 248 | |
---|
| 249 | id = self.getDatabase( "SELECT job_id FROM jobs WHERE job_id = '%s'" %jobid ) |
---|
| 250 | |
---|
| 251 | if id: |
---|
[89] | 252 | id = id[0][0] |
---|
[84] | 253 | |
---|
| 254 | return id |
---|
| 255 | else: |
---|
| 256 | return None |
---|
| 257 | |
---|
| 258 | def addJob( self, job_id, jobattrs ): |
---|
| 259 | |
---|
| 260 | if not self.getJobId( job_id ): |
---|
| 261 | |
---|
| 262 | self.mutateJob( 'insert', job_id, jobattrs ) |
---|
| 263 | else: |
---|
| 264 | self.mutateJob( 'update', job_id, jobattrs ) |
---|
| 265 | |
---|
| 266 | def mutateJob( self, action, job_id, jobattrs ): |
---|
| 267 | |
---|
[292] | 268 | job_values = [ 'name', 'queue', 'owner', 'requested_time', 'requested_memory', 'ppn', 'status', 'start_timestamp', 'stop_timestamp' ] |
---|
[84] | 269 | |
---|
[292] | 270 | insert_col_str = 'job_id' |
---|
| 271 | insert_val_str = "'%s'" %job_id |
---|
| 272 | update_str = None |
---|
[84] | 273 | |
---|
| 274 | debug_msg( 6, 'mutateJob(): %s %s' %(action,job_id)) |
---|
| 275 | |
---|
[99] | 276 | ids = [ ] |
---|
| 277 | |
---|
[84] | 278 | for valname, value in jobattrs.items(): |
---|
| 279 | |
---|
[96] | 280 | if valname in job_values and value != '': |
---|
[84] | 281 | |
---|
| 282 | column_name = 'job_' + valname |
---|
| 283 | |
---|
| 284 | if action == 'insert': |
---|
| 285 | |
---|
| 286 | if not insert_col_str: |
---|
| 287 | insert_col_str = column_name |
---|
| 288 | else: |
---|
| 289 | insert_col_str = insert_col_str + ',' + column_name |
---|
| 290 | |
---|
| 291 | if not insert_val_str: |
---|
| 292 | insert_val_str = value |
---|
| 293 | else: |
---|
| 294 | insert_val_str = insert_val_str + ",'%s'" %value |
---|
| 295 | |
---|
| 296 | elif action == 'update': |
---|
| 297 | |
---|
| 298 | if not update_str: |
---|
| 299 | update_str = "%s='%s'" %(column_name, value) |
---|
| 300 | else: |
---|
| 301 | update_str = update_str + ",%s='%s'" %(column_name, value) |
---|
| 302 | |
---|
[90] | 303 | elif valname == 'nodes' and value: |
---|
[84] | 304 | |
---|
[191] | 305 | node_valid = 1 |
---|
[190] | 306 | |
---|
| 307 | if len(value) == 1: |
---|
| 308 | |
---|
[191] | 309 | if jobattrs['status'] == 'Q': |
---|
[190] | 310 | |
---|
[191] | 311 | node_valid = 0 |
---|
[190] | 312 | |
---|
[191] | 313 | else: |
---|
[190] | 314 | |
---|
[191] | 315 | node_valid = 0 |
---|
[190] | 316 | |
---|
[191] | 317 | for node_char in str(value[0]): |
---|
[190] | 318 | |
---|
[191] | 319 | if string.find( string.digits, node_char ) != -1 and not node_valid: |
---|
[190] | 320 | |
---|
[191] | 321 | node_valid = 1 |
---|
[84] | 322 | |
---|
[191] | 323 | if node_valid: |
---|
| 324 | |
---|
| 325 | ids = self.addNodes( value, jobattrs['domain'] ) |
---|
| 326 | |
---|
[84] | 327 | if action == 'insert': |
---|
| 328 | |
---|
| 329 | self.setDatabase( "INSERT INTO jobs ( %s ) VALUES ( %s )" %( insert_col_str, insert_val_str ) ) |
---|
[86] | 330 | |
---|
[84] | 331 | elif action == 'update': |
---|
| 332 | |
---|
[89] | 333 | self.setDatabase( "UPDATE jobs SET %s WHERE job_id=%s" %(update_str, job_id) ) |
---|
[84] | 334 | |
---|
[191] | 335 | if len( ids ) > 0: |
---|
| 336 | self.addJobNodes( job_id, ids ) |
---|
[190] | 337 | |
---|
[154] | 338 | def addNodes( self, hostnames, domain ): |
---|
[84] | 339 | |
---|
[98] | 340 | ids = [ ] |
---|
| 341 | |
---|
[84] | 342 | for node in hostnames: |
---|
| 343 | |
---|
[292] | 344 | node = '%s.%s' %( node, domain ) |
---|
| 345 | id = self.getNodeId( node ) |
---|
[84] | 346 | |
---|
| 347 | if not id: |
---|
| 348 | self.setDatabase( "INSERT INTO nodes ( node_hostname ) VALUES ( '%s' )" %node ) |
---|
[98] | 349 | id = self.getNodeId( node ) |
---|
[84] | 350 | |
---|
[98] | 351 | ids.append( id ) |
---|
| 352 | |
---|
| 353 | return ids |
---|
| 354 | |
---|
[86] | 355 | def addJobNodes( self, jobid, nodes ): |
---|
| 356 | |
---|
| 357 | for node in nodes: |
---|
| 358 | |
---|
[191] | 359 | if not self.getJobNodeId( jobid, node ): |
---|
| 360 | |
---|
| 361 | self.addJobNode( jobid, node ) |
---|
| 362 | |
---|
[84] | 363 | def addJobNode( self, jobid, nodeid ): |
---|
| 364 | |
---|
| 365 | self.setDatabase( "INSERT INTO job_nodes (job_id,node_id) VALUES ( %s,%s )" %(jobid, nodeid) ) |
---|
| 366 | |
---|
| 367 | def storeJobInfo( self, jobid, jobattrs ): |
---|
| 368 | |
---|
| 369 | self.addJob( jobid, jobattrs ) |
---|
| 370 | |
---|
[295] | 371 | def checkStaleJobs( self ): |
---|
| 372 | |
---|
| 373 | q = "SELECT * from jobs WHERE job_status != 'F'" |
---|
| 374 | |
---|
| 375 | r = self.getDatabase( q ) |
---|
| 376 | |
---|
| 377 | if len( r ) == 0: |
---|
| 378 | |
---|
| 379 | return None |
---|
| 380 | |
---|
| 381 | cleanjobs = [ ] |
---|
| 382 | timeoutjobs = [ ] |
---|
| 383 | |
---|
| 384 | jobtimeout_sec = JOB_TIMEOUT * (60 * 60) |
---|
| 385 | cur_time = time.time() |
---|
| 386 | |
---|
| 387 | for row in r: |
---|
| 388 | |
---|
| 389 | job_id = row[0] |
---|
| 390 | job_requested_time = row[4] |
---|
| 391 | job_status = row[7] |
---|
| 392 | job_start_timestamp = row[8] |
---|
| 393 | |
---|
| 394 | if job_status == 'Q' or not job_start_timestamp: |
---|
| 395 | |
---|
| 396 | cleanjobs.append( job_id ) |
---|
| 397 | |
---|
| 398 | else: |
---|
| 399 | |
---|
| 400 | start_timestamp = int( job_start_timestamp ) |
---|
| 401 | |
---|
| 402 | if ( cur_time - start_timestamp ) > jobtimeout_sec: |
---|
| 403 | |
---|
| 404 | if job_requested_time: |
---|
| 405 | |
---|
| 406 | rtime_epoch = reqtime2epoch( job_requested_time ) |
---|
| 407 | else: |
---|
| 408 | rtime_epoch = None |
---|
| 409 | |
---|
| 410 | timeoutjobs.append( (job_id, job_start_timestamp, rtime_epoch) ) |
---|
| 411 | |
---|
| 412 | debug_msg( 1, 'Found ' + str( len( cleanjobs ) ) + ' stale jobs in database: deleting entries' ) |
---|
| 413 | |
---|
| 414 | for j in cleanjobs: |
---|
| 415 | |
---|
| 416 | q = "DELETE FROM jobs WHERE job_id = '" + str( j ) + "'" |
---|
| 417 | self.setDatabase( q ) |
---|
| 418 | |
---|
| 419 | debug_msg( 1, 'Found ' + str( len( timeoutjobs ) ) + ' timed out jobs in database: closing entries' ) |
---|
| 420 | |
---|
| 421 | for j in timeoutjobs: |
---|
| 422 | |
---|
| 423 | ( i, s, r ) = j |
---|
| 424 | |
---|
| 425 | if r: |
---|
| 426 | new_end_timestamp = int( s ) + r |
---|
| 427 | |
---|
| 428 | q = "UPDATE jobs SET job_stop_timestamp = '" + str( new_end_timestamp ) + "' WHERE job_id = '" + str(i) + "'" |
---|
| 429 | self.setDatabase( q ) |
---|
| 430 | |
---|
[37] | 431 | class RRDMutator: |
---|
[63] | 432 | """A class for performing RRD mutations""" |
---|
[37] | 433 | |
---|
[277] | 434 | binary = None |
---|
[37] | 435 | |
---|
| 436 | def __init__( self, binary=None ): |
---|
[63] | 437 | """Set alternate binary if supplied""" |
---|
[37] | 438 | |
---|
| 439 | if binary: |
---|
| 440 | self.binary = binary |
---|
| 441 | |
---|
| 442 | def create( self, filename, args ): |
---|
[63] | 443 | """Create a new rrd with args""" |
---|
| 444 | |
---|
[40] | 445 | return self.perform( 'create', '"' + filename + '"', args ) |
---|
[37] | 446 | |
---|
| 447 | def update( self, filename, args ): |
---|
[63] | 448 | """Update a rrd with args""" |
---|
| 449 | |
---|
[40] | 450 | return self.perform( 'update', '"' + filename + '"', args ) |
---|
[37] | 451 | |
---|
[42] | 452 | def grabLastUpdate( self, filename ): |
---|
[63] | 453 | """Determine the last update time of filename rrd""" |
---|
[42] | 454 | |
---|
| 455 | last_update = 0 |
---|
| 456 | |
---|
[53] | 457 | debug_msg( 8, self.binary + ' info "' + filename + '"' ) |
---|
| 458 | |
---|
[292] | 459 | my_pipe = os.popen( self.binary + ' info "' + filename + '"' ) |
---|
[42] | 460 | |
---|
[292] | 461 | for line in my_pipe.readlines(): |
---|
| 462 | |
---|
[42] | 463 | if line.find( 'last_update') != -1: |
---|
| 464 | |
---|
| 465 | last_update = line.split( ' = ' )[1] |
---|
| 466 | |
---|
[292] | 467 | if my_pipe: |
---|
| 468 | |
---|
| 469 | my_pipe.close() |
---|
| 470 | |
---|
[42] | 471 | if last_update: |
---|
| 472 | return last_update |
---|
| 473 | else: |
---|
| 474 | return 0 |
---|
| 475 | |
---|
[40] | 476 | def perform( self, action, filename, args ): |
---|
[63] | 477 | """Perform action on rrd filename with args""" |
---|
[37] | 478 | |
---|
| 479 | arg_string = None |
---|
| 480 | |
---|
[40] | 481 | if type( args ) is not ListType: |
---|
| 482 | debug_msg( 8, 'Arguments needs to be of type List' ) |
---|
| 483 | return 1 |
---|
| 484 | |
---|
[37] | 485 | for arg in args: |
---|
| 486 | |
---|
| 487 | if not arg_string: |
---|
| 488 | |
---|
| 489 | arg_string = arg |
---|
| 490 | else: |
---|
| 491 | arg_string = arg_string + ' ' + arg |
---|
| 492 | |
---|
[54] | 493 | debug_msg( 8, self.binary + ' ' + action + ' ' + filename + ' ' + arg_string ) |
---|
[37] | 494 | |
---|
[292] | 495 | cmd = os.popen( self.binary + ' ' + action + ' ' + filename + ' ' + arg_string ) |
---|
| 496 | lines = cmd.readlines() |
---|
| 497 | |
---|
[146] | 498 | cmd.close() |
---|
[37] | 499 | |
---|
[146] | 500 | for line in lines: |
---|
| 501 | |
---|
[37] | 502 | if line.find( 'ERROR' ) != -1: |
---|
| 503 | |
---|
| 504 | error_msg = string.join( line.split( ' ' )[1:] ) |
---|
[40] | 505 | debug_msg( 8, error_msg ) |
---|
[37] | 506 | return 1 |
---|
| 507 | |
---|
| 508 | return 0 |
---|
| 509 | |
---|
[78] | 510 | class XMLProcessor: |
---|
| 511 | """Skeleton class for XML processor's""" |
---|
| 512 | |
---|
| 513 | def run( self ): |
---|
| 514 | """Do main processing of XML here""" |
---|
| 515 | |
---|
| 516 | pass |
---|
| 517 | |
---|
| 518 | class TorqueXMLProcessor( XMLProcessor ): |
---|
| 519 | """Main class for processing XML and acting with it""" |
---|
| 520 | |
---|
[295] | 521 | def __init__( self, XMLSource, DataStore ): |
---|
[78] | 522 | """Setup initial XML connection and handlers""" |
---|
| 523 | |
---|
[293] | 524 | #self.myXMLGatherer = XMLGatherer( ARCHIVE_XMLSOURCE.split( ':' )[0], ARCHIVE_XMLSOURCE.split( ':' )[1] ) |
---|
[287] | 525 | #self.myXMLSource = self.myXMLGatherer.getFileObject() |
---|
| 526 | self.myXMLSource = XMLSource |
---|
[295] | 527 | self.myXMLHandler = TorqueXMLHandler( DataStore ) |
---|
[287] | 528 | self.myXMLError = XMLErrorHandler() |
---|
[78] | 529 | |
---|
[287] | 530 | self.config = GangliaConfigParser( GMETAD_CONF ) |
---|
| 531 | |
---|
[78] | 532 | def run( self ): |
---|
| 533 | """Main XML processing""" |
---|
| 534 | |
---|
[169] | 535 | debug_msg( 1, 'torque_xml_thread(): started.' ) |
---|
[87] | 536 | |
---|
[78] | 537 | while( 1 ): |
---|
| 538 | |
---|
[287] | 539 | #self.myXMLSource = self.mXMLGatherer.getFileObject() |
---|
[169] | 540 | debug_msg( 1, 'torque_xml_thread(): Parsing..' ) |
---|
[176] | 541 | |
---|
[287] | 542 | my_data = self.myXMLSource.getData() |
---|
| 543 | |
---|
[176] | 544 | try: |
---|
[287] | 545 | xml.sax.parseString( my_data, self.myXMLHandler, self.myXMLError ) |
---|
[176] | 546 | except socket.error, msg: |
---|
| 547 | debug_msg( 0, 'ERROR: Socket error in connect to datasource!: %s' %msg ) |
---|
| 548 | |
---|
[169] | 549 | debug_msg( 1, 'torque_xml_thread(): Done parsing.' ) |
---|
| 550 | debug_msg( 1, 'torque_xml_thread(): Sleeping.. (%ss)' %(str( self.config.getLowestInterval() ) ) ) |
---|
[87] | 551 | time.sleep( self.config.getLowestInterval() ) |
---|
[78] | 552 | |
---|
[71] | 553 | class TorqueXMLHandler( xml.sax.handler.ContentHandler ): |
---|
[63] | 554 | """Parse Torque's jobinfo XML from our plugin""" |
---|
| 555 | |
---|
[72] | 556 | jobAttrs = { } |
---|
| 557 | |
---|
[295] | 558 | def __init__( self, datastore ): |
---|
[84] | 559 | |
---|
[295] | 560 | #self.ds = DataSQLStore( JOB_SQL_DBASE.split( '/' )[0], JOB_SQL_DBASE.split( '/' )[1] ) |
---|
| 561 | self.ds = datastore |
---|
[292] | 562 | self.jobs_processed = [ ] |
---|
| 563 | self.jobs_to_store = [ ] |
---|
[84] | 564 | |
---|
[183] | 565 | def startDocument( self ): |
---|
| 566 | |
---|
[292] | 567 | self.heartbeat = 0 |
---|
[183] | 568 | |
---|
[63] | 569 | def startElement( self, name, attrs ): |
---|
| 570 | """ |
---|
| 571 | This XML will be all gmetric XML |
---|
| 572 | so there will be no specific start/end element |
---|
| 573 | just one XML statement with all info |
---|
| 574 | """ |
---|
| 575 | |
---|
[73] | 576 | jobinfo = { } |
---|
[63] | 577 | |
---|
[199] | 578 | if name == 'CLUSTER': |
---|
[63] | 579 | |
---|
[199] | 580 | self.clustername = attrs.get( 'NAME', "" ) |
---|
| 581 | |
---|
| 582 | elif name == 'METRIC' and self.clustername in ARCHIVE_DATASOURCES: |
---|
| 583 | |
---|
[70] | 584 | metricname = attrs.get( 'NAME', "" ) |
---|
[63] | 585 | |
---|
[285] | 586 | if metricname == 'MONARCH-HEARTBEAT': |
---|
[70] | 587 | self.heartbeat = attrs.get( 'VAL', "" ) |
---|
[63] | 588 | |
---|
[285] | 589 | elif metricname.find( 'MONARCH-JOB' ) != -1: |
---|
[63] | 590 | |
---|
[292] | 591 | job_id = metricname.split( 'MONARCH-JOB-' )[1].split( '-' )[0] |
---|
| 592 | val = attrs.get( 'VAL', "" ) |
---|
[63] | 593 | |
---|
[96] | 594 | if not job_id in self.jobs_processed: |
---|
[292] | 595 | |
---|
[96] | 596 | self.jobs_processed.append( job_id ) |
---|
| 597 | |
---|
[73] | 598 | check_change = 0 |
---|
| 599 | |
---|
| 600 | if self.jobAttrs.has_key( job_id ): |
---|
[292] | 601 | |
---|
[73] | 602 | check_change = 1 |
---|
| 603 | |
---|
[63] | 604 | valinfo = val.split( ' ' ) |
---|
| 605 | |
---|
| 606 | for myval in valinfo: |
---|
| 607 | |
---|
[84] | 608 | if len( myval.split( '=' ) ) > 1: |
---|
[63] | 609 | |
---|
[292] | 610 | valname = myval.split( '=' )[0] |
---|
| 611 | value = myval.split( '=' )[1] |
---|
[70] | 612 | |
---|
[84] | 613 | if valname == 'nodes': |
---|
| 614 | value = value.split( ';' ) |
---|
[72] | 615 | |
---|
[84] | 616 | jobinfo[ valname ] = value |
---|
| 617 | |
---|
[73] | 618 | if check_change: |
---|
[182] | 619 | if self.jobinfoChanged( self.jobAttrs, job_id, jobinfo ) and self.jobAttrs[ job_id ]['status'] in [ 'R', 'Q' ]: |
---|
[292] | 620 | self.jobAttrs[ job_id ]['stop_timestamp'] = '' |
---|
| 621 | self.jobAttrs[ job_id ] = self.setJobAttrs( self.jobAttrs[ job_id ], jobinfo ) |
---|
[84] | 622 | if not job_id in self.jobs_to_store: |
---|
| 623 | self.jobs_to_store.append( job_id ) |
---|
| 624 | |
---|
[102] | 625 | debug_msg( 6, 'jobinfo for job %s has changed' %job_id ) |
---|
[73] | 626 | else: |
---|
| 627 | self.jobAttrs[ job_id ] = jobinfo |
---|
[84] | 628 | |
---|
| 629 | if not job_id in self.jobs_to_store: |
---|
| 630 | self.jobs_to_store.append( job_id ) |
---|
| 631 | |
---|
[102] | 632 | debug_msg( 6, 'jobinfo for job %s has changed' %job_id ) |
---|
[73] | 633 | |
---|
[77] | 634 | def endDocument( self ): |
---|
[74] | 635 | """When all metrics have gone, check if any jobs have finished""" |
---|
[72] | 636 | |
---|
[182] | 637 | if self.heartbeat: |
---|
| 638 | for jobid, jobinfo in self.jobAttrs.items(): |
---|
[74] | 639 | |
---|
[182] | 640 | # This is an old job, not in current jobinfo list anymore |
---|
| 641 | # it must have finished, since we _did_ get a new heartbeat |
---|
| 642 | # |
---|
| 643 | mytime = int( jobinfo['reported'] ) + int( jobinfo['poll_interval'] ) |
---|
[102] | 644 | |
---|
[184] | 645 | if (mytime < self.heartbeat) and (jobid not in self.jobs_processed) and (jobinfo['status'] == 'R'): |
---|
[74] | 646 | |
---|
[182] | 647 | if not jobid in self.jobs_processed: |
---|
| 648 | self.jobs_processed.append( jobid ) |
---|
[96] | 649 | |
---|
[182] | 650 | self.jobAttrs[ jobid ]['status'] = 'F' |
---|
| 651 | self.jobAttrs[ jobid ]['stop_timestamp'] = str( mytime ) |
---|
[96] | 652 | |
---|
[182] | 653 | if not jobid in self.jobs_to_store: |
---|
| 654 | self.jobs_to_store.append( jobid ) |
---|
[74] | 655 | |
---|
[182] | 656 | debug_msg( 1, 'torque_xml_thread(): Storing..' ) |
---|
[87] | 657 | |
---|
[182] | 658 | for jobid in self.jobs_to_store: |
---|
[295] | 659 | if self.jobAttrs[ jobid ]['status'] in [ 'R', 'F' ]: |
---|
[84] | 660 | |
---|
[184] | 661 | self.ds.storeJobInfo( jobid, self.jobAttrs[ jobid ] ) |
---|
| 662 | |
---|
| 663 | if self.jobAttrs[ jobid ]['status'] == 'F': |
---|
| 664 | del self.jobAttrs[ jobid ] |
---|
| 665 | |
---|
[182] | 666 | debug_msg( 1, 'torque_xml_thread(): Done storing.' ) |
---|
[87] | 667 | |
---|
[292] | 668 | self.jobs_processed = [ ] |
---|
| 669 | self.jobs_to_store = [ ] |
---|
[84] | 670 | |
---|
[82] | 671 | def setJobAttrs( self, old, new ): |
---|
| 672 | """ |
---|
| 673 | Set new job attributes in old, but not lose existing fields |
---|
| 674 | if old attributes doesn't have those |
---|
| 675 | """ |
---|
| 676 | |
---|
| 677 | for valname, value in new.items(): |
---|
| 678 | old[ valname ] = value |
---|
| 679 | |
---|
| 680 | return old |
---|
| 681 | |
---|
| 682 | |
---|
[73] | 683 | def jobinfoChanged( self, jobattrs, jobid, jobinfo ): |
---|
[74] | 684 | """ |
---|
| 685 | Check if jobinfo has changed from jobattrs[jobid] |
---|
| 686 | if it's report time is bigger than previous one |
---|
| 687 | and it is report time is recent (equal to heartbeat) |
---|
| 688 | """ |
---|
[72] | 689 | |
---|
[87] | 690 | ignore_changes = [ 'reported' ] |
---|
| 691 | |
---|
[73] | 692 | if jobattrs.has_key( jobid ): |
---|
| 693 | |
---|
| 694 | for valname, value in jobinfo.items(): |
---|
| 695 | |
---|
[87] | 696 | if valname not in ignore_changes: |
---|
[73] | 697 | |
---|
[87] | 698 | if jobattrs[ jobid ].has_key( valname ): |
---|
[73] | 699 | |
---|
[87] | 700 | if value != jobattrs[ jobid ][ valname ]: |
---|
[73] | 701 | |
---|
[87] | 702 | if jobinfo['reported'] > jobattrs[ jobid ][ 'reported' ] and jobinfo['reported'] == self.heartbeat: |
---|
| 703 | return 1 |
---|
[73] | 704 | |
---|
[87] | 705 | else: |
---|
| 706 | return 1 |
---|
| 707 | |
---|
[73] | 708 | return 0 |
---|
| 709 | |
---|
[71] | 710 | class GangliaXMLHandler( xml.sax.handler.ContentHandler ): |
---|
[63] | 711 | """Parse Ganglia's XML""" |
---|
[3] | 712 | |
---|
[295] | 713 | def __init__( self, config, datastore ): |
---|
[63] | 714 | """Setup initial variables and gather info on existing rrd archive""" |
---|
| 715 | |
---|
[292] | 716 | self.config = config |
---|
| 717 | self.clusters = { } |
---|
[295] | 718 | self.ds = datastore |
---|
[324] | 719 | |
---|
[295] | 720 | debug_msg( 1, 'Checking database..' ) |
---|
| 721 | self.ds.checkStaleJobs() |
---|
| 722 | debug_msg( 1, 'Check done.' ) |
---|
[296] | 723 | debug_msg( 1, 'Checking rrd archive..' ) |
---|
[293] | 724 | self.gatherClusters() |
---|
[169] | 725 | debug_msg( 1, 'Check done.' ) |
---|
[33] | 726 | |
---|
[44] | 727 | def gatherClusters( self ): |
---|
[63] | 728 | """Find all existing clusters in archive dir""" |
---|
[44] | 729 | |
---|
[292] | 730 | archive_dir = check_dir(ARCHIVE_PATH) |
---|
[44] | 731 | |
---|
[292] | 732 | hosts = [ ] |
---|
[44] | 733 | |
---|
| 734 | if os.path.exists( archive_dir ): |
---|
| 735 | |
---|
[292] | 736 | dirlist = os.listdir( archive_dir ) |
---|
[44] | 737 | |
---|
| 738 | for item in dirlist: |
---|
| 739 | |
---|
| 740 | clustername = item |
---|
| 741 | |
---|
[60] | 742 | if not self.clusters.has_key( clustername ) and clustername in ARCHIVE_DATASOURCES: |
---|
[44] | 743 | |
---|
| 744 | self.clusters[ clustername ] = RRDHandler( self.config, clustername ) |
---|
| 745 | |
---|
[6] | 746 | def startElement( self, name, attrs ): |
---|
[63] | 747 | """Memorize appropriate data from xml start tags""" |
---|
[3] | 748 | |
---|
[7] | 749 | if name == 'GANGLIA_XML': |
---|
[32] | 750 | |
---|
[292] | 751 | self.XMLSource = attrs.get( 'SOURCE', "" ) |
---|
| 752 | self.gangliaVersion = attrs.get( 'VERSION', "" ) |
---|
[32] | 753 | |
---|
[12] | 754 | debug_msg( 10, 'Found XML data: source %s version %s' %( self.XMLSource, self.gangliaVersion ) ) |
---|
[6] | 755 | |
---|
[7] | 756 | elif name == 'GRID': |
---|
[32] | 757 | |
---|
[292] | 758 | self.gridName = attrs.get( 'NAME', "" ) |
---|
| 759 | self.time = attrs.get( 'LOCALTIME', "" ) |
---|
[32] | 760 | |
---|
[12] | 761 | debug_msg( 10, '`-Grid found: %s' %( self.gridName ) ) |
---|
[6] | 762 | |
---|
[7] | 763 | elif name == 'CLUSTER': |
---|
[32] | 764 | |
---|
[292] | 765 | self.clusterName = attrs.get( 'NAME', "" ) |
---|
| 766 | self.time = attrs.get( 'LOCALTIME', "" ) |
---|
[32] | 767 | |
---|
[60] | 768 | if not self.clusters.has_key( self.clusterName ) and self.clusterName in ARCHIVE_DATASOURCES: |
---|
[32] | 769 | |
---|
[34] | 770 | self.clusters[ self.clusterName ] = RRDHandler( self.config, self.clusterName ) |
---|
[33] | 771 | |
---|
[35] | 772 | debug_msg( 10, ' |-Cluster found: %s' %( self.clusterName ) ) |
---|
[6] | 773 | |
---|
[60] | 774 | elif name == 'HOST' and self.clusterName in ARCHIVE_DATASOURCES: |
---|
[32] | 775 | |
---|
[292] | 776 | self.hostName = attrs.get( 'NAME', "" ) |
---|
| 777 | self.hostIp = attrs.get( 'IP', "" ) |
---|
| 778 | self.hostReported = attrs.get( 'REPORTED', "" ) |
---|
[32] | 779 | |
---|
[12] | 780 | debug_msg( 10, ' | |-Host found: %s - ip %s reported %s' %( self.hostName, self.hostIp, self.hostReported ) ) |
---|
[6] | 781 | |
---|
[60] | 782 | elif name == 'METRIC' and self.clusterName in ARCHIVE_DATASOURCES: |
---|
[6] | 783 | |
---|
[32] | 784 | type = attrs.get( 'TYPE', "" ) |
---|
[198] | 785 | |
---|
| 786 | exclude_metric = False |
---|
| 787 | |
---|
| 788 | for ex_metricstr in ARCHIVE_EXCLUDE_METRICS: |
---|
[6] | 789 | |
---|
[198] | 790 | orig_name = attrs.get( 'NAME', "" ) |
---|
[3] | 791 | |
---|
[198] | 792 | if string.lower( orig_name ) == string.lower( ex_metricstr ): |
---|
| 793 | |
---|
| 794 | exclude_metric = True |
---|
| 795 | |
---|
| 796 | elif re.match( ex_metricstr, orig_name ): |
---|
| 797 | |
---|
| 798 | exclude_metric = True |
---|
| 799 | |
---|
| 800 | if type not in UNSUPPORTED_ARCHIVE_TYPES and not exclude_metric: |
---|
| 801 | |
---|
[292] | 802 | myMetric = { } |
---|
| 803 | myMetric['name'] = attrs.get( 'NAME', "" ) |
---|
| 804 | myMetric['val'] = attrs.get( 'VAL', "" ) |
---|
| 805 | myMetric['time'] = self.hostReported |
---|
[3] | 806 | |
---|
[34] | 807 | self.clusters[ self.clusterName ].memMetric( self.hostName, myMetric ) |
---|
[3] | 808 | |
---|
[34] | 809 | debug_msg( 11, ' | | |-metric: %s:%s' %( myMetric['name'], myMetric['val'] ) ) |
---|
[6] | 810 | |
---|
[34] | 811 | def storeMetrics( self ): |
---|
[63] | 812 | """Store metrics of each cluster rrd handler""" |
---|
[9] | 813 | |
---|
[34] | 814 | for clustername, rrdh in self.clusters.items(): |
---|
[16] | 815 | |
---|
[38] | 816 | ret = rrdh.storeMetrics() |
---|
[9] | 817 | |
---|
[38] | 818 | if ret: |
---|
| 819 | debug_msg( 9, 'An error occured while storing metrics for cluster %s' %clustername ) |
---|
| 820 | return 1 |
---|
| 821 | |
---|
| 822 | return 0 |
---|
| 823 | |
---|
[71] | 824 | class XMLErrorHandler( xml.sax.handler.ErrorHandler ): |
---|
| 825 | |
---|
| 826 | def error( self, exception ): |
---|
| 827 | """Recoverable error""" |
---|
| 828 | |
---|
[169] | 829 | debug_msg( 0, 'Recoverable XML error ' + str( exception ) + ' ignored.' ) |
---|
[71] | 830 | |
---|
| 831 | def fatalError( self, exception ): |
---|
| 832 | """Non-recoverable error""" |
---|
| 833 | |
---|
| 834 | exception_str = str( exception ) |
---|
| 835 | |
---|
| 836 | # Ignore 'no element found' errors |
---|
| 837 | if exception_str.find( 'no element found' ) != -1: |
---|
[169] | 838 | debug_msg( 0, 'No XML data found: Socket not (re)connected or datasource not available.' ) |
---|
[71] | 839 | return 0 |
---|
| 840 | |
---|
[170] | 841 | debug_msg( 0, 'FATAL ERROR: Non-recoverable XML error ' + str( exception ) ) |
---|
[71] | 842 | sys.exit( 1 ) |
---|
| 843 | |
---|
| 844 | def warning( self, exception ): |
---|
| 845 | """Warning""" |
---|
| 846 | |
---|
| 847 | debug_msg( 0, 'Warning ' + str( exception ) ) |
---|
| 848 | |
---|
[78] | 849 | class XMLGatherer: |
---|
[63] | 850 | """Setup a connection and file object to Ganglia's XML""" |
---|
[3] | 851 | |
---|
[287] | 852 | s = None |
---|
| 853 | fd = None |
---|
| 854 | data = None |
---|
[293] | 855 | slot = None |
---|
[8] | 856 | |
---|
[287] | 857 | # Time since the last update |
---|
| 858 | # |
---|
| 859 | LAST_UPDATE = 0 |
---|
| 860 | |
---|
| 861 | # Minimum interval between updates |
---|
| 862 | # |
---|
| 863 | MIN_UPDATE_INT = 10 |
---|
| 864 | |
---|
| 865 | # Is a update occuring now |
---|
| 866 | # |
---|
| 867 | update_now = False |
---|
| 868 | |
---|
[8] | 869 | def __init__( self, host, port ): |
---|
[63] | 870 | """Store host and port for connection""" |
---|
[8] | 871 | |
---|
[293] | 872 | self.host = host |
---|
| 873 | self.port = port |
---|
| 874 | self.slot = threading.Lock() |
---|
[3] | 875 | |
---|
[287] | 876 | self.retrieveData() |
---|
| 877 | |
---|
| 878 | def retrieveData( self ): |
---|
[63] | 879 | """Setup connection to XML source""" |
---|
[8] | 880 | |
---|
[287] | 881 | self.update_now = True |
---|
| 882 | |
---|
[293] | 883 | self.slot.acquire() |
---|
| 884 | |
---|
[8] | 885 | for res in socket.getaddrinfo( self.host, self.port, socket.AF_UNSPEC, socket.SOCK_STREAM ): |
---|
[32] | 886 | |
---|
[5] | 887 | af, socktype, proto, canonname, sa = res |
---|
[32] | 888 | |
---|
[5] | 889 | try: |
---|
[32] | 890 | |
---|
[8] | 891 | self.s = socket.socket( af, socktype, proto ) |
---|
[32] | 892 | |
---|
[5] | 893 | except socket.error, msg: |
---|
[32] | 894 | |
---|
[8] | 895 | self.s = None |
---|
[5] | 896 | continue |
---|
[32] | 897 | |
---|
[5] | 898 | try: |
---|
[32] | 899 | |
---|
[8] | 900 | self.s.connect( sa ) |
---|
[32] | 901 | |
---|
[5] | 902 | except socket.error, msg: |
---|
[32] | 903 | |
---|
[70] | 904 | self.disconnect() |
---|
[5] | 905 | continue |
---|
[32] | 906 | |
---|
[5] | 907 | break |
---|
[3] | 908 | |
---|
[8] | 909 | if self.s is None: |
---|
[32] | 910 | |
---|
[170] | 911 | debug_msg( 0, 'FATAL ERROR: Could not open socket or unable to connect to datasource!' ) |
---|
[287] | 912 | self.update_now = False |
---|
[33] | 913 | sys.exit( 1 ) |
---|
[5] | 914 | |
---|
[287] | 915 | else: |
---|
[324] | 916 | #self.s.send( '\n' ) |
---|
[287] | 917 | |
---|
| 918 | my_fp = self.s.makefile( 'r' ) |
---|
| 919 | my_data = my_fp.readlines() |
---|
| 920 | my_data = string.join( my_data, '' ) |
---|
| 921 | |
---|
| 922 | self.data = my_data |
---|
| 923 | |
---|
| 924 | self.LAST_UPDATE = time.time() |
---|
| 925 | |
---|
[293] | 926 | self.slot.release() |
---|
| 927 | |
---|
[287] | 928 | self.update_now = False |
---|
| 929 | |
---|
[33] | 930 | def disconnect( self ): |
---|
[63] | 931 | """Close socket""" |
---|
[33] | 932 | |
---|
| 933 | if self.s: |
---|
[287] | 934 | #self.s.shutdown( 2 ) |
---|
[33] | 935 | self.s.close() |
---|
| 936 | self.s = None |
---|
| 937 | |
---|
| 938 | def __del__( self ): |
---|
[63] | 939 | """Kill the socket before we leave""" |
---|
[33] | 940 | |
---|
| 941 | self.disconnect() |
---|
| 942 | |
---|
[287] | 943 | def reGetData( self ): |
---|
[70] | 944 | """Reconnect""" |
---|
[33] | 945 | |
---|
[287] | 946 | while self.update_now: |
---|
| 947 | |
---|
| 948 | # Must be another update in progress: |
---|
| 949 | # Wait until the update is complete |
---|
| 950 | # |
---|
| 951 | time.sleep( 1 ) |
---|
| 952 | |
---|
[38] | 953 | if self.s: |
---|
| 954 | self.disconnect() |
---|
[33] | 955 | |
---|
[287] | 956 | self.retrieveData() |
---|
[5] | 957 | |
---|
[287] | 958 | def getData( self ): |
---|
| 959 | |
---|
| 960 | """Return the XML data""" |
---|
| 961 | |
---|
| 962 | # If more than MIN_UPDATE_INT seconds passed since last data update |
---|
| 963 | # update the XML first before returning it |
---|
| 964 | # |
---|
| 965 | |
---|
| 966 | cur_time = time.time() |
---|
| 967 | |
---|
| 968 | if ( cur_time - self.LAST_UPDATE ) > self.MIN_UPDATE_INT: |
---|
| 969 | |
---|
| 970 | self.reGetData() |
---|
| 971 | |
---|
| 972 | while self.update_now: |
---|
| 973 | |
---|
| 974 | # Must be another update in progress: |
---|
| 975 | # Wait until the update is complete |
---|
| 976 | # |
---|
| 977 | time.sleep( 1 ) |
---|
| 978 | |
---|
| 979 | return self.data |
---|
| 980 | |
---|
[70] | 981 | def makeFileDescriptor( self ): |
---|
| 982 | """Make file descriptor that points to our socket connection""" |
---|
| 983 | |
---|
| 984 | self.reconnect() |
---|
| 985 | |
---|
| 986 | if self.s: |
---|
| 987 | self.fd = self.s.makefile( 'r' ) |
---|
| 988 | |
---|
| 989 | def getFileObject( self ): |
---|
| 990 | """Connect, and return a file object""" |
---|
| 991 | |
---|
[78] | 992 | self.makeFileDescriptor() |
---|
| 993 | |
---|
[70] | 994 | if self.fd: |
---|
| 995 | return self.fd |
---|
| 996 | |
---|
[78] | 997 | class GangliaXMLProcessor( XMLProcessor ): |
---|
[63] | 998 | """Main class for processing XML and acting with it""" |
---|
[5] | 999 | |
---|
[295] | 1000 | def __init__( self, XMLSource, DataStore ): |
---|
[63] | 1001 | """Setup initial XML connection and handlers""" |
---|
[33] | 1002 | |
---|
[287] | 1003 | self.config = GangliaConfigParser( GMETAD_CONF ) |
---|
[33] | 1004 | |
---|
[293] | 1005 | #self.myXMLGatherer = XMLGatherer( ARCHIVE_XMLSOURCE.split( ':' )[0], ARCHIVE_XMLSOURCE.split( ':' )[1] ) |
---|
[287] | 1006 | #self.myXMLSource = self.myXMLGatherer.getFileObject() |
---|
| 1007 | self.myXMLSource = XMLSource |
---|
[295] | 1008 | self.ds = DataStore |
---|
| 1009 | self.myXMLHandler = GangliaXMLHandler( self.config, self.ds ) |
---|
[287] | 1010 | self.myXMLError = XMLErrorHandler() |
---|
[73] | 1011 | |
---|
[9] | 1012 | def run( self ): |
---|
[63] | 1013 | """Main XML processing; start a xml and storethread""" |
---|
[8] | 1014 | |
---|
[102] | 1015 | xml_thread = threading.Thread( None, self.processXML, 'xmlthread' ) |
---|
| 1016 | store_thread = threading.Thread( None, self.storeMetrics, 'storethread' ) |
---|
[22] | 1017 | |
---|
[36] | 1018 | while( 1 ): |
---|
| 1019 | |
---|
[102] | 1020 | if not xml_thread.isAlive(): |
---|
[36] | 1021 | # Gather XML at the same interval as gmetad |
---|
| 1022 | |
---|
| 1023 | # threaded call to: self.processXML() |
---|
| 1024 | # |
---|
[169] | 1025 | try: |
---|
| 1026 | xml_thread = threading.Thread( None, self.processXML, 'xml_thread' ) |
---|
| 1027 | xml_thread.start() |
---|
[176] | 1028 | except thread.error, msg: |
---|
[169] | 1029 | debug_msg( 0, 'ERROR: Unable to start xml_thread!: '+str(msg)) |
---|
| 1030 | #return 1 |
---|
[36] | 1031 | |
---|
[102] | 1032 | if not store_thread.isAlive(): |
---|
[55] | 1033 | # Store metrics every .. sec |
---|
[36] | 1034 | |
---|
[55] | 1035 | # threaded call to: self.storeMetrics() |
---|
| 1036 | # |
---|
[169] | 1037 | try: |
---|
| 1038 | store_thread = threading.Thread( None, self.storeMetrics, 'store_thread' ) |
---|
| 1039 | store_thread.start() |
---|
[176] | 1040 | except thread.error, msg: |
---|
[169] | 1041 | debug_msg( 0, 'ERROR: Unable to start store_thread!: '+str(msg)) |
---|
| 1042 | #return 1 |
---|
[36] | 1043 | |
---|
| 1044 | # Just sleep a sec here, to prevent daemon from going mad. We're all threads here anyway |
---|
| 1045 | time.sleep( 1 ) |
---|
| 1046 | |
---|
[33] | 1047 | def storeMetrics( self ): |
---|
[63] | 1048 | """Store metrics retained in memory to disk""" |
---|
[22] | 1049 | |
---|
[63] | 1050 | # Store metrics somewhere between every 360 and 640 seconds |
---|
[38] | 1051 | # |
---|
[55] | 1052 | STORE_INTERVAL = random.randint( 360, 640 ) |
---|
[22] | 1053 | |
---|
[169] | 1054 | try: |
---|
| 1055 | store_metric_thread = threading.Thread( None, self.storeThread, 'store_metric_thread' ) |
---|
| 1056 | store_metric_thread.start() |
---|
[176] | 1057 | except thread.error, msg: |
---|
[169] | 1058 | debug_msg( 0, 'ERROR: Unable to start ganglia_store_thread()!: '+str(msg) ) |
---|
| 1059 | return 1 |
---|
[36] | 1060 | |
---|
[169] | 1061 | debug_msg( 1, 'ganglia_store_thread(): started.' ) |
---|
| 1062 | |
---|
| 1063 | debug_msg( 1, 'ganglia_store_thread(): Sleeping.. (%ss)' %STORE_INTERVAL ) |
---|
[36] | 1064 | time.sleep( STORE_INTERVAL ) |
---|
[169] | 1065 | debug_msg( 1, 'ganglia_store_thread(): Done sleeping.' ) |
---|
[36] | 1066 | |
---|
[102] | 1067 | if store_metric_thread.isAlive(): |
---|
[36] | 1068 | |
---|
[169] | 1069 | debug_msg( 1, 'ganglia_store_thread(): storemetricthread() still running, waiting to finish..' ) |
---|
[136] | 1070 | store_metric_thread.join( STORE_TIMEOUT ) # Maximum time is for storing thread to finish |
---|
[169] | 1071 | debug_msg( 1, 'ganglia_store_thread(): Done waiting.' ) |
---|
[36] | 1072 | |
---|
[169] | 1073 | debug_msg( 1, 'ganglia_store_thread(): finished.' ) |
---|
[36] | 1074 | |
---|
| 1075 | return 0 |
---|
| 1076 | |
---|
[39] | 1077 | def storeThread( self ): |
---|
[63] | 1078 | """Actual metric storing thread""" |
---|
[39] | 1079 | |
---|
[169] | 1080 | debug_msg( 1, 'ganglia_store_metric_thread(): started.' ) |
---|
| 1081 | debug_msg( 1, 'ganglia_store_metric_thread(): Storing data..' ) |
---|
[78] | 1082 | ret = self.myXMLHandler.storeMetrics() |
---|
[176] | 1083 | if ret > 0: |
---|
| 1084 | debug_msg( 0, 'ganglia_store_metric_thread(): UNKNOWN ERROR %s while storing Metrics!' %str(ret) ) |
---|
[169] | 1085 | debug_msg( 1, 'ganglia_store_metric_thread(): Done storing.' ) |
---|
| 1086 | debug_msg( 1, 'ganglia_store_metric_thread(): finished.' ) |
---|
[39] | 1087 | |
---|
[176] | 1088 | return 0 |
---|
[39] | 1089 | |
---|
[8] | 1090 | def processXML( self ): |
---|
[63] | 1091 | """Process XML""" |
---|
[8] | 1092 | |
---|
[169] | 1093 | try: |
---|
| 1094 | parsethread = threading.Thread( None, self.parseThread, 'parsethread' ) |
---|
| 1095 | parsethread.start() |
---|
[176] | 1096 | except thread.error, msg: |
---|
[169] | 1097 | debug_msg( 0, 'ERROR: Unable to start ganglia_xml_thread()!: ' + str(msg) ) |
---|
| 1098 | return 1 |
---|
[8] | 1099 | |
---|
[169] | 1100 | debug_msg( 1, 'ganglia_xml_thread(): started.' ) |
---|
[36] | 1101 | |
---|
[169] | 1102 | debug_msg( 1, 'ganglia_xml_thread(): Sleeping.. (%ss)' %self.config.getLowestInterval() ) |
---|
[36] | 1103 | time.sleep( float( self.config.getLowestInterval() ) ) |
---|
[169] | 1104 | debug_msg( 1, 'ganglia_xml_thread(): Done sleeping.' ) |
---|
[36] | 1105 | |
---|
| 1106 | if parsethread.isAlive(): |
---|
| 1107 | |
---|
[169] | 1108 | debug_msg( 1, 'ganglia_xml_thread(): parsethread() still running, waiting (%ss) to finish..' %PARSE_TIMEOUT ) |
---|
[47] | 1109 | parsethread.join( PARSE_TIMEOUT ) # Maximum time for XML thread to finish |
---|
[169] | 1110 | debug_msg( 1, 'ganglia_xml_thread(): Done waiting.' ) |
---|
[36] | 1111 | |
---|
[169] | 1112 | debug_msg( 1, 'ganglia_xml_thread(): finished.' ) |
---|
[36] | 1113 | |
---|
| 1114 | return 0 |
---|
| 1115 | |
---|
[39] | 1116 | def parseThread( self ): |
---|
[63] | 1117 | """Actual parsing thread""" |
---|
[39] | 1118 | |
---|
[169] | 1119 | debug_msg( 1, 'ganglia_parse_thread(): started.' ) |
---|
| 1120 | debug_msg( 1, 'ganglia_parse_thread(): Parsing XML..' ) |
---|
[287] | 1121 | #self.myXMLSource = self.myXMLGatherer.getFileObject() |
---|
| 1122 | |
---|
| 1123 | my_data = self.myXMLSource.getData() |
---|
[176] | 1124 | |
---|
[293] | 1125 | #print my_data |
---|
| 1126 | |
---|
[176] | 1127 | try: |
---|
[287] | 1128 | xml.sax.parseString( my_data, self.myXMLHandler, self.myXMLError ) |
---|
[176] | 1129 | except socket.error, msg: |
---|
| 1130 | debug_msg( 0, 'ERROR: Socket error in connect to datasource!: %s' %msg ) |
---|
| 1131 | |
---|
[169] | 1132 | debug_msg( 1, 'ganglia_parse_thread(): Done parsing.' ) |
---|
| 1133 | debug_msg( 1, 'ganglia_parse_thread(): finished.' ) |
---|
[39] | 1134 | |
---|
[176] | 1135 | return 0 |
---|
[39] | 1136 | |
---|
[9] | 1137 | class GangliaConfigParser: |
---|
| 1138 | |
---|
[34] | 1139 | sources = [ ] |
---|
[9] | 1140 | |
---|
| 1141 | def __init__( self, config ): |
---|
[63] | 1142 | """Parse some stuff from our gmetad's config, such as polling interval""" |
---|
[32] | 1143 | |
---|
[9] | 1144 | self.config = config |
---|
| 1145 | self.parseValues() |
---|
| 1146 | |
---|
[32] | 1147 | def parseValues( self ): |
---|
[63] | 1148 | """Parse certain values from gmetad.conf""" |
---|
[9] | 1149 | |
---|
| 1150 | readcfg = open( self.config, 'r' ) |
---|
| 1151 | |
---|
| 1152 | for line in readcfg.readlines(): |
---|
| 1153 | |
---|
| 1154 | if line.count( '"' ) > 1: |
---|
| 1155 | |
---|
[10] | 1156 | if line.find( 'data_source' ) != -1 and line[0] != '#': |
---|
[9] | 1157 | |
---|
[292] | 1158 | source = { } |
---|
| 1159 | source['name'] = line.split( '"' )[1] |
---|
| 1160 | source_words = line.split( '"' )[2].split( ' ' ) |
---|
[9] | 1161 | |
---|
| 1162 | for word in source_words: |
---|
| 1163 | |
---|
| 1164 | valid_interval = 1 |
---|
| 1165 | |
---|
| 1166 | for letter in word: |
---|
[32] | 1167 | |
---|
[9] | 1168 | if letter not in string.digits: |
---|
[32] | 1169 | |
---|
[9] | 1170 | valid_interval = 0 |
---|
| 1171 | |
---|
[10] | 1172 | if valid_interval and len(word) > 0: |
---|
[32] | 1173 | |
---|
[9] | 1174 | source['interval'] = word |
---|
[12] | 1175 | debug_msg( 9, 'polling interval for %s = %s' %(source['name'], source['interval'] ) ) |
---|
[33] | 1176 | |
---|
| 1177 | # No interval found, use Ganglia's default |
---|
| 1178 | if not source.has_key( 'interval' ): |
---|
| 1179 | source['interval'] = 15 |
---|
| 1180 | debug_msg( 9, 'polling interval for %s defaulted to 15' %(source['name']) ) |
---|
[32] | 1181 | |
---|
[33] | 1182 | self.sources.append( source ) |
---|
[9] | 1183 | |
---|
| 1184 | def getInterval( self, source_name ): |
---|
[63] | 1185 | """Return interval for source_name""" |
---|
[32] | 1186 | |
---|
[9] | 1187 | for source in self.sources: |
---|
[32] | 1188 | |
---|
[12] | 1189 | if source['name'] == source_name: |
---|
[32] | 1190 | |
---|
[9] | 1191 | return source['interval'] |
---|
[32] | 1192 | |
---|
[9] | 1193 | return None |
---|
| 1194 | |
---|
[34] | 1195 | def getLowestInterval( self ): |
---|
[63] | 1196 | """Return the lowest interval of all clusters""" |
---|
[34] | 1197 | |
---|
| 1198 | lowest_interval = 0 |
---|
| 1199 | |
---|
| 1200 | for source in self.sources: |
---|
| 1201 | |
---|
| 1202 | if not lowest_interval or source['interval'] <= lowest_interval: |
---|
| 1203 | |
---|
| 1204 | lowest_interval = source['interval'] |
---|
| 1205 | |
---|
| 1206 | # Return 15 when nothing is found, so that the daemon won't go insane with 0 sec delays |
---|
| 1207 | if lowest_interval: |
---|
| 1208 | return lowest_interval |
---|
| 1209 | else: |
---|
| 1210 | return 15 |
---|
| 1211 | |
---|
[9] | 1212 | class RRDHandler: |
---|
[63] | 1213 | """Class for handling RRD activity""" |
---|
[9] | 1214 | |
---|
[32] | 1215 | myMetrics = { } |
---|
[40] | 1216 | lastStored = { } |
---|
[47] | 1217 | timeserials = { } |
---|
[36] | 1218 | slot = None |
---|
[32] | 1219 | |
---|
[33] | 1220 | def __init__( self, config, cluster ): |
---|
[63] | 1221 | """Setup initial variables""" |
---|
[78] | 1222 | |
---|
[292] | 1223 | self.block = 0 |
---|
| 1224 | self.cluster = cluster |
---|
| 1225 | self.config = config |
---|
| 1226 | self.slot = threading.Lock() |
---|
| 1227 | self.rrdm = RRDMutator( RRDTOOL ) |
---|
| 1228 | |
---|
[42] | 1229 | self.gatherLastUpdates() |
---|
[9] | 1230 | |
---|
[42] | 1231 | def gatherLastUpdates( self ): |
---|
[63] | 1232 | """Populate the lastStored list, containing timestamps of all last updates""" |
---|
[42] | 1233 | |
---|
| 1234 | cluster_dir = '%s/%s' %( check_dir(ARCHIVE_PATH), self.cluster ) |
---|
| 1235 | |
---|
| 1236 | hosts = [ ] |
---|
| 1237 | |
---|
| 1238 | if os.path.exists( cluster_dir ): |
---|
| 1239 | |
---|
[44] | 1240 | dirlist = os.listdir( cluster_dir ) |
---|
[42] | 1241 | |
---|
[44] | 1242 | for dir in dirlist: |
---|
[42] | 1243 | |
---|
[44] | 1244 | hosts.append( dir ) |
---|
[42] | 1245 | |
---|
| 1246 | for host in hosts: |
---|
| 1247 | |
---|
[292] | 1248 | host_dir = cluster_dir + '/' + host |
---|
| 1249 | dirlist = os.listdir( host_dir ) |
---|
[47] | 1250 | |
---|
| 1251 | for dir in dirlist: |
---|
| 1252 | |
---|
| 1253 | if not self.timeserials.has_key( host ): |
---|
| 1254 | |
---|
| 1255 | self.timeserials[ host ] = [ ] |
---|
| 1256 | |
---|
| 1257 | self.timeserials[ host ].append( dir ) |
---|
| 1258 | |
---|
[42] | 1259 | last_serial = self.getLastRrdTimeSerial( host ) |
---|
[292] | 1260 | |
---|
[42] | 1261 | if last_serial: |
---|
| 1262 | |
---|
| 1263 | metric_dir = cluster_dir + '/' + host + '/' + last_serial |
---|
[292] | 1264 | |
---|
[42] | 1265 | if os.path.exists( metric_dir ): |
---|
| 1266 | |
---|
[44] | 1267 | dirlist = os.listdir( metric_dir ) |
---|
[42] | 1268 | |
---|
[44] | 1269 | for file in dirlist: |
---|
[42] | 1270 | |
---|
[44] | 1271 | metricname = file.split( '.rrd' )[0] |
---|
[42] | 1272 | |
---|
[44] | 1273 | if not self.lastStored.has_key( host ): |
---|
[42] | 1274 | |
---|
[44] | 1275 | self.lastStored[ host ] = { } |
---|
[42] | 1276 | |
---|
[44] | 1277 | self.lastStored[ host ][ metricname ] = self.rrdm.grabLastUpdate( metric_dir + '/' + file ) |
---|
[42] | 1278 | |
---|
[32] | 1279 | def getClusterName( self ): |
---|
[63] | 1280 | """Return clustername""" |
---|
| 1281 | |
---|
[32] | 1282 | return self.cluster |
---|
| 1283 | |
---|
| 1284 | def memMetric( self, host, metric ): |
---|
[63] | 1285 | """Store metric from host in memory""" |
---|
[32] | 1286 | |
---|
[179] | 1287 | # <ATOMIC> |
---|
| 1288 | # |
---|
| 1289 | self.slot.acquire() |
---|
| 1290 | |
---|
[34] | 1291 | if self.myMetrics.has_key( host ): |
---|
[32] | 1292 | |
---|
[34] | 1293 | if self.myMetrics[ host ].has_key( metric['name'] ): |
---|
[32] | 1294 | |
---|
[34] | 1295 | for mymetric in self.myMetrics[ host ][ metric['name'] ]: |
---|
[32] | 1296 | |
---|
[34] | 1297 | if mymetric['time'] == metric['time']: |
---|
[32] | 1298 | |
---|
[34] | 1299 | # Allready have this metric, abort |
---|
[179] | 1300 | self.slot.release() |
---|
[34] | 1301 | return 1 |
---|
| 1302 | else: |
---|
| 1303 | self.myMetrics[ host ][ metric['name'] ] = [ ] |
---|
| 1304 | else: |
---|
[292] | 1305 | self.myMetrics[ host ] = { } |
---|
| 1306 | self.myMetrics[ host ][ metric['name'] ] = [ ] |
---|
[32] | 1307 | |
---|
[63] | 1308 | # Push new metric onto stack |
---|
| 1309 | # atomic code; only 1 thread at a time may access the stack |
---|
| 1310 | |
---|
[32] | 1311 | self.myMetrics[ host ][ metric['name'] ].append( metric ) |
---|
| 1312 | |
---|
[40] | 1313 | self.slot.release() |
---|
[53] | 1314 | # |
---|
| 1315 | # </ATOMIC> |
---|
[40] | 1316 | |
---|
[47] | 1317 | def makeUpdateList( self, host, metriclist ): |
---|
[63] | 1318 | """ |
---|
| 1319 | Make a list of update values for rrdupdate |
---|
| 1320 | but only those that we didn't store before |
---|
| 1321 | """ |
---|
[37] | 1322 | |
---|
[292] | 1323 | update_list = [ ] |
---|
| 1324 | metric = None |
---|
[37] | 1325 | |
---|
[47] | 1326 | while len( metriclist ) > 0: |
---|
[37] | 1327 | |
---|
[53] | 1328 | metric = metriclist.pop( 0 ) |
---|
[37] | 1329 | |
---|
[53] | 1330 | if self.checkStoreMetric( host, metric ): |
---|
[292] | 1331 | |
---|
[53] | 1332 | update_list.append( '%s:%s' %( metric['time'], metric['val'] ) ) |
---|
[40] | 1333 | |
---|
[37] | 1334 | return update_list |
---|
| 1335 | |
---|
[49] | 1336 | def checkStoreMetric( self, host, metric ): |
---|
[63] | 1337 | """Check if supplied metric if newer than last one stored""" |
---|
[40] | 1338 | |
---|
| 1339 | if self.lastStored.has_key( host ): |
---|
| 1340 | |
---|
[47] | 1341 | if self.lastStored[ host ].has_key( metric['name'] ): |
---|
[40] | 1342 | |
---|
[47] | 1343 | if metric['time'] <= self.lastStored[ host ][ metric['name'] ]: |
---|
[40] | 1344 | |
---|
[50] | 1345 | # This is old |
---|
[40] | 1346 | return 0 |
---|
| 1347 | |
---|
[50] | 1348 | return 1 |
---|
| 1349 | |
---|
[54] | 1350 | def memLastUpdate( self, host, metricname, metriclist ): |
---|
[63] | 1351 | """ |
---|
| 1352 | Memorize the time of the latest metric from metriclist |
---|
| 1353 | but only if it wasn't allready memorized |
---|
| 1354 | """ |
---|
[50] | 1355 | |
---|
[54] | 1356 | if not self.lastStored.has_key( host ): |
---|
| 1357 | self.lastStored[ host ] = { } |
---|
| 1358 | |
---|
[50] | 1359 | last_update_time = 0 |
---|
| 1360 | |
---|
| 1361 | for metric in metriclist: |
---|
| 1362 | |
---|
[54] | 1363 | if metric['name'] == metricname: |
---|
[50] | 1364 | |
---|
[54] | 1365 | if metric['time'] > last_update_time: |
---|
[50] | 1366 | |
---|
[54] | 1367 | last_update_time = metric['time'] |
---|
[40] | 1368 | |
---|
[54] | 1369 | if self.lastStored[ host ].has_key( metricname ): |
---|
[52] | 1370 | |
---|
[54] | 1371 | if last_update_time <= self.lastStored[ host ][ metricname ]: |
---|
[52] | 1372 | return 1 |
---|
[40] | 1373 | |
---|
[54] | 1374 | self.lastStored[ host ][ metricname ] = last_update_time |
---|
[52] | 1375 | |
---|
[33] | 1376 | def storeMetrics( self ): |
---|
[63] | 1377 | """ |
---|
| 1378 | Store all metrics from memory to disk |
---|
| 1379 | and do it to the RRD's in appropriate timeperiod directory |
---|
| 1380 | """ |
---|
[33] | 1381 | |
---|
| 1382 | for hostname, mymetrics in self.myMetrics.items(): |
---|
| 1383 | |
---|
| 1384 | for metricname, mymetric in mymetrics.items(): |
---|
| 1385 | |
---|
[53] | 1386 | metrics_to_store = [ ] |
---|
| 1387 | |
---|
[63] | 1388 | # Pop metrics from stack for storing until none is left |
---|
| 1389 | # atomic code: only 1 thread at a time may access myMetrics |
---|
| 1390 | |
---|
[53] | 1391 | # <ATOMIC> |
---|
[50] | 1392 | # |
---|
[47] | 1393 | self.slot.acquire() |
---|
[33] | 1394 | |
---|
[54] | 1395 | while len( self.myMetrics[ hostname ][ metricname ] ) > 0: |
---|
[53] | 1396 | |
---|
[54] | 1397 | if len( self.myMetrics[ hostname ][ metricname ] ) > 0: |
---|
[53] | 1398 | |
---|
[176] | 1399 | try: |
---|
| 1400 | metrics_to_store.append( self.myMetrics[ hostname ][ metricname ].pop( 0 ) ) |
---|
| 1401 | except IndexError, msg: |
---|
| 1402 | |
---|
[179] | 1403 | # Somehow sometimes myMetrics[ hostname ][ metricname ] |
---|
| 1404 | # is still len 0 when the statement is executed. |
---|
| 1405 | # Just ignore indexerror's.. |
---|
[176] | 1406 | pass |
---|
| 1407 | |
---|
[53] | 1408 | self.slot.release() |
---|
| 1409 | # |
---|
| 1410 | # </ATOMIC> |
---|
| 1411 | |
---|
[47] | 1412 | # Create a mapping table, each metric to the period where it should be stored |
---|
| 1413 | # |
---|
[53] | 1414 | metric_serial_table = self.determineSerials( hostname, metricname, metrics_to_store ) |
---|
[33] | 1415 | |
---|
[50] | 1416 | update_rets = [ ] |
---|
| 1417 | |
---|
[47] | 1418 | for period, pmetric in metric_serial_table.items(): |
---|
| 1419 | |
---|
[146] | 1420 | create_ret = self.createCheck( hostname, metricname, period ) |
---|
[47] | 1421 | |
---|
| 1422 | update_ret = self.update( hostname, metricname, period, pmetric ) |
---|
| 1423 | |
---|
| 1424 | if update_ret == 0: |
---|
| 1425 | |
---|
| 1426 | debug_msg( 9, 'stored metric %s for %s' %( hostname, metricname ) ) |
---|
| 1427 | else: |
---|
| 1428 | debug_msg( 9, 'metric update failed' ) |
---|
| 1429 | |
---|
[146] | 1430 | update_rets.append( create_ret ) |
---|
[50] | 1431 | update_rets.append( update_ret ) |
---|
[47] | 1432 | |
---|
[179] | 1433 | # Lets ignore errors here for now, we need to make sure last update time |
---|
| 1434 | # is correct! |
---|
| 1435 | # |
---|
| 1436 | #if not (1) in update_rets: |
---|
[50] | 1437 | |
---|
[179] | 1438 | self.memLastUpdate( hostname, metricname, metrics_to_store ) |
---|
[50] | 1439 | |
---|
[17] | 1440 | def makeTimeSerial( self ): |
---|
[63] | 1441 | """Generate a time serial. Seconds since epoch""" |
---|
[17] | 1442 | |
---|
| 1443 | # Seconds since epoch |
---|
| 1444 | mytime = int( time.time() ) |
---|
| 1445 | |
---|
| 1446 | return mytime |
---|
| 1447 | |
---|
[50] | 1448 | def makeRrdPath( self, host, metricname, timeserial ): |
---|
[63] | 1449 | """Make a RRD location/path and filename""" |
---|
[17] | 1450 | |
---|
[292] | 1451 | rrd_dir = '%s/%s/%s/%s' %( check_dir(ARCHIVE_PATH), self.cluster, host, timeserial ) |
---|
| 1452 | rrd_file = '%s/%s.rrd' %( rrd_dir, metricname ) |
---|
[17] | 1453 | |
---|
| 1454 | return rrd_dir, rrd_file |
---|
| 1455 | |
---|
[20] | 1456 | def getLastRrdTimeSerial( self, host ): |
---|
[63] | 1457 | """Find the last timeserial (directory) for this host""" |
---|
[17] | 1458 | |
---|
[19] | 1459 | newest_timeserial = 0 |
---|
| 1460 | |
---|
[47] | 1461 | for dir in self.timeserials[ host ]: |
---|
[32] | 1462 | |
---|
[47] | 1463 | valid_dir = 1 |
---|
[17] | 1464 | |
---|
[47] | 1465 | for letter in dir: |
---|
| 1466 | if letter not in string.digits: |
---|
| 1467 | valid_dir = 0 |
---|
[17] | 1468 | |
---|
[47] | 1469 | if valid_dir: |
---|
| 1470 | timeserial = dir |
---|
| 1471 | if timeserial > newest_timeserial: |
---|
| 1472 | newest_timeserial = timeserial |
---|
[17] | 1473 | |
---|
| 1474 | if newest_timeserial: |
---|
[18] | 1475 | return newest_timeserial |
---|
[17] | 1476 | else: |
---|
| 1477 | return 0 |
---|
| 1478 | |
---|
[47] | 1479 | def determinePeriod( self, host, check_serial ): |
---|
[63] | 1480 | """Determine to which period (directory) this time(serial) belongs""" |
---|
[47] | 1481 | |
---|
| 1482 | period_serial = 0 |
---|
| 1483 | |
---|
[56] | 1484 | if self.timeserials.has_key( host ): |
---|
[47] | 1485 | |
---|
[56] | 1486 | for serial in self.timeserials[ host ]: |
---|
[47] | 1487 | |
---|
[56] | 1488 | if check_serial >= serial and period_serial < serial: |
---|
[47] | 1489 | |
---|
[56] | 1490 | period_serial = serial |
---|
| 1491 | |
---|
[47] | 1492 | return period_serial |
---|
| 1493 | |
---|
| 1494 | def determineSerials( self, host, metricname, metriclist ): |
---|
| 1495 | """ |
---|
| 1496 | Determine the correct serial and corresponding rrd to store |
---|
| 1497 | for a list of metrics |
---|
| 1498 | """ |
---|
| 1499 | |
---|
| 1500 | metric_serial_table = { } |
---|
| 1501 | |
---|
| 1502 | for metric in metriclist: |
---|
| 1503 | |
---|
| 1504 | if metric['name'] == metricname: |
---|
| 1505 | |
---|
[292] | 1506 | period = self.determinePeriod( host, metric['time'] ) |
---|
[47] | 1507 | |
---|
[292] | 1508 | archive_secs = ARCHIVE_HOURS_PER_RRD * (60 * 60) |
---|
[47] | 1509 | |
---|
[49] | 1510 | if (int( metric['time'] ) - int( period ) ) > archive_secs: |
---|
[47] | 1511 | |
---|
| 1512 | # This one should get it's own new period |
---|
| 1513 | period = metric['time'] |
---|
[57] | 1514 | |
---|
| 1515 | if not self.timeserials.has_key( host ): |
---|
| 1516 | self.timeserials[ host ] = [ ] |
---|
| 1517 | |
---|
[50] | 1518 | self.timeserials[ host ].append( period ) |
---|
[47] | 1519 | |
---|
| 1520 | if not metric_serial_table.has_key( period ): |
---|
| 1521 | |
---|
[49] | 1522 | metric_serial_table[ period ] = [ ] |
---|
[47] | 1523 | |
---|
| 1524 | metric_serial_table[ period ].append( metric ) |
---|
| 1525 | |
---|
| 1526 | return metric_serial_table |
---|
| 1527 | |
---|
[33] | 1528 | def createCheck( self, host, metricname, timeserial ): |
---|
[63] | 1529 | """Check if an rrd allready exists for this metric, create if not""" |
---|
[9] | 1530 | |
---|
[35] | 1531 | debug_msg( 9, 'rrdcreate: using timeserial %s for %s/%s' %( timeserial, host, metricname ) ) |
---|
[47] | 1532 | |
---|
[33] | 1533 | rrd_dir, rrd_file = self.makeRrdPath( host, metricname, timeserial ) |
---|
[17] | 1534 | |
---|
[9] | 1535 | if not os.path.exists( rrd_dir ): |
---|
[58] | 1536 | |
---|
| 1537 | try: |
---|
| 1538 | os.makedirs( rrd_dir ) |
---|
| 1539 | |
---|
[169] | 1540 | except os.OSError, msg: |
---|
[58] | 1541 | |
---|
| 1542 | if msg.find( 'File exists' ) != -1: |
---|
| 1543 | |
---|
| 1544 | # Ignore exists errors |
---|
| 1545 | pass |
---|
| 1546 | |
---|
| 1547 | else: |
---|
| 1548 | |
---|
| 1549 | print msg |
---|
| 1550 | return |
---|
| 1551 | |
---|
[14] | 1552 | debug_msg( 9, 'created dir %s' %( str(rrd_dir) ) ) |
---|
[9] | 1553 | |
---|
[14] | 1554 | if not os.path.exists( rrd_file ): |
---|
[9] | 1555 | |
---|
[292] | 1556 | interval = self.config.getInterval( self.cluster ) |
---|
| 1557 | heartbeat = 8 * int( interval ) |
---|
[9] | 1558 | |
---|
[292] | 1559 | params = [ ] |
---|
[12] | 1560 | |
---|
[37] | 1561 | params.append( '--step' ) |
---|
| 1562 | params.append( str( interval ) ) |
---|
[12] | 1563 | |
---|
[37] | 1564 | params.append( '--start' ) |
---|
[47] | 1565 | params.append( str( int( timeserial ) - 1 ) ) |
---|
[12] | 1566 | |
---|
[37] | 1567 | params.append( 'DS:sum:GAUGE:%d:U:U' %heartbeat ) |
---|
| 1568 | params.append( 'RRA:AVERAGE:0.5:1:%s' %(ARCHIVE_HOURS_PER_RRD * 240) ) |
---|
[13] | 1569 | |
---|
[37] | 1570 | self.rrdm.create( str(rrd_file), params ) |
---|
| 1571 | |
---|
[14] | 1572 | debug_msg( 9, 'created rrd %s' %( str(rrd_file) ) ) |
---|
| 1573 | |
---|
[47] | 1574 | def update( self, host, metricname, timeserial, metriclist ): |
---|
[63] | 1575 | """ |
---|
| 1576 | Update rrd file for host with metricname |
---|
| 1577 | in directory timeserial with metriclist |
---|
| 1578 | """ |
---|
[9] | 1579 | |
---|
[35] | 1580 | debug_msg( 9, 'rrdupdate: using timeserial %s for %s/%s' %( timeserial, host, metricname ) ) |
---|
[9] | 1581 | |
---|
[292] | 1582 | rrd_dir, rrd_file = self.makeRrdPath( host, metricname, timeserial ) |
---|
[18] | 1583 | |
---|
[292] | 1584 | update_list = self.makeUpdateList( host, metriclist ) |
---|
[15] | 1585 | |
---|
[41] | 1586 | if len( update_list ) > 0: |
---|
| 1587 | ret = self.rrdm.update( str(rrd_file), update_list ) |
---|
[32] | 1588 | |
---|
[41] | 1589 | if ret: |
---|
| 1590 | return 1 |
---|
[27] | 1591 | |
---|
[41] | 1592 | debug_msg( 9, 'updated rrd %s with %s' %( str(rrd_file), string.join( update_list ) ) ) |
---|
[15] | 1593 | |
---|
[36] | 1594 | return 0 |
---|
| 1595 | |
---|
[169] | 1596 | def daemon(): |
---|
| 1597 | """daemonized threading""" |
---|
[8] | 1598 | |
---|
[169] | 1599 | # Fork the first child |
---|
| 1600 | # |
---|
| 1601 | pid = os.fork() |
---|
| 1602 | |
---|
| 1603 | if pid > 0: |
---|
| 1604 | |
---|
| 1605 | sys.exit(0) # end parent |
---|
| 1606 | |
---|
| 1607 | # creates a session and sets the process group ID |
---|
| 1608 | # |
---|
| 1609 | os.setsid() |
---|
| 1610 | |
---|
| 1611 | # Fork the second child |
---|
| 1612 | # |
---|
| 1613 | pid = os.fork() |
---|
| 1614 | |
---|
| 1615 | if pid > 0: |
---|
| 1616 | |
---|
| 1617 | sys.exit(0) # end parent |
---|
| 1618 | |
---|
| 1619 | # Go to the root directory and set the umask |
---|
| 1620 | # |
---|
| 1621 | os.chdir('/') |
---|
| 1622 | os.umask(0) |
---|
| 1623 | |
---|
| 1624 | sys.stdin.close() |
---|
| 1625 | sys.stdout.close() |
---|
| 1626 | sys.stderr.close() |
---|
| 1627 | |
---|
[273] | 1628 | os.open('/dev/null', os.O_RDWR) |
---|
| 1629 | os.dup2(0, 1) |
---|
| 1630 | os.dup2(0, 2) |
---|
[169] | 1631 | |
---|
| 1632 | run() |
---|
| 1633 | |
---|
| 1634 | def run(): |
---|
| 1635 | """Threading start""" |
---|
| 1636 | |
---|
[287] | 1637 | myXMLSource = XMLGatherer( ARCHIVE_XMLSOURCE.split( ':' )[0], ARCHIVE_XMLSOURCE.split( ':' )[1] ) |
---|
[295] | 1638 | myDataStore = DataSQLStore( JOB_SQL_DBASE.split( '/' )[0], JOB_SQL_DBASE.split( '/' )[1] ) |
---|
[8] | 1639 | |
---|
[295] | 1640 | myTorqueProcessor = TorqueXMLProcessor( myXMLSource, myDataStore ) |
---|
| 1641 | myGangliaProcessor = GangliaXMLProcessor( myXMLSource, myDataStore ) |
---|
[287] | 1642 | |
---|
[169] | 1643 | try: |
---|
[292] | 1644 | torque_xml_thread = threading.Thread( None, myTorqueProcessor.run, 'torque_proc_thread' ) |
---|
| 1645 | ganglia_xml_thread = threading.Thread( None, myGangliaProcessor.run, 'ganglia_proc_thread' ) |
---|
[22] | 1646 | |
---|
[169] | 1647 | torque_xml_thread.start() |
---|
| 1648 | ganglia_xml_thread.start() |
---|
| 1649 | |
---|
[176] | 1650 | except thread.error, msg: |
---|
[169] | 1651 | debug_msg( 0, 'FATAL ERROR: Unable to start main threads!: '+ str(msg) ) |
---|
| 1652 | syslog.closelog() |
---|
| 1653 | sys.exit(1) |
---|
| 1654 | |
---|
| 1655 | debug_msg( 0, 'main threading started.' ) |
---|
[78] | 1656 | |
---|
[169] | 1657 | def main(): |
---|
| 1658 | """Program startup""" |
---|
| 1659 | |
---|
[214] | 1660 | if not processArgs( sys.argv[1:] ): |
---|
| 1661 | sys.exit( 1 ) |
---|
| 1662 | |
---|
[169] | 1663 | if( DAEMONIZE and USE_SYSLOG ): |
---|
| 1664 | syslog.openlog( 'jobarchived', syslog.LOG_NOWAIT, SYSLOG_FACILITY ) |
---|
| 1665 | |
---|
| 1666 | if DAEMONIZE: |
---|
| 1667 | daemon() |
---|
| 1668 | else: |
---|
| 1669 | run() |
---|
| 1670 | |
---|
| 1671 | # |
---|
[81] | 1672 | # Global functions |
---|
[169] | 1673 | # |
---|
[81] | 1674 | |
---|
[9] | 1675 | def check_dir( directory ): |
---|
[63] | 1676 | """Check if directory is a proper directory. I.e.: Does _not_ end with a '/'""" |
---|
[9] | 1677 | |
---|
| 1678 | if directory[-1] == '/': |
---|
| 1679 | directory = directory[:-1] |
---|
| 1680 | |
---|
| 1681 | return directory |
---|
| 1682 | |
---|
[295] | 1683 | def reqtime2epoch( rtime ): |
---|
| 1684 | |
---|
| 1685 | (hours, minutes, seconds ) = rtime.split( ':' ) |
---|
| 1686 | |
---|
| 1687 | etime = int(seconds) |
---|
| 1688 | etime = etime + ( int(minutes) * 60 ) |
---|
| 1689 | etime = etime + ( int(hours) * 60 * 60 ) |
---|
| 1690 | |
---|
| 1691 | return etime |
---|
| 1692 | |
---|
[12] | 1693 | def debug_msg( level, msg ): |
---|
[169] | 1694 | """Only print msg if correct levels""" |
---|
[12] | 1695 | |
---|
[169] | 1696 | if (not DAEMONIZE and DEBUG_LEVEL >= level): |
---|
| 1697 | sys.stderr.write( printTime() + ' - ' + msg + '\n' ) |
---|
| 1698 | |
---|
| 1699 | if (DAEMONIZE and USE_SYSLOG and SYSLOG_LEVEL >= level): |
---|
| 1700 | syslog.syslog( msg ) |
---|
[12] | 1701 | |
---|
[46] | 1702 | def printTime( ): |
---|
[63] | 1703 | """Print current time in human readable format""" |
---|
[46] | 1704 | |
---|
| 1705 | return time.strftime("%a %d %b %Y %H:%M:%S") |
---|
| 1706 | |
---|
[63] | 1707 | # Ooohh, someone started me! Let's go.. |
---|
[9] | 1708 | if __name__ == '__main__': |
---|
| 1709 | main() |
---|