[23] | 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 | # |
---|
[228] | 21 | # SVN $Id: jobmond.py 355 2007-05-03 13:09:25Z bastiaans $ |
---|
[227] | 22 | # |
---|
[23] | 23 | |
---|
[212] | 24 | import sys, getopt, ConfigParser |
---|
[354] | 25 | import time, os, socket, string, re |
---|
[318] | 26 | import xml, xml.sax |
---|
| 27 | from xml.sax import saxutils, make_parser |
---|
| 28 | from xml.sax import make_parser |
---|
| 29 | from xml.sax.handler import feature_namespaces |
---|
| 30 | |
---|
[307] | 31 | def usage(): |
---|
| 32 | |
---|
| 33 | print |
---|
| 34 | print 'usage: jobmond [options]' |
---|
| 35 | print 'options:' |
---|
| 36 | print ' --config, -c configuration file' |
---|
| 37 | print ' --pidfile, -p pid file' |
---|
| 38 | print ' --help, -h help' |
---|
| 39 | print |
---|
| 40 | |
---|
[212] | 41 | def processArgs( args ): |
---|
[26] | 42 | |
---|
[354] | 43 | SHORT_L = 'hc:' |
---|
| 44 | LONG_L = [ 'help', 'config=' ] |
---|
[165] | 45 | |
---|
[307] | 46 | global PIDFILE |
---|
[354] | 47 | PIDFILE = None |
---|
[61] | 48 | |
---|
[354] | 49 | config_filename = '/etc/jobmond.conf' |
---|
| 50 | |
---|
[212] | 51 | try: |
---|
[68] | 52 | |
---|
[354] | 53 | opts, args = getopt.getopt( args, SHORT_L, LONG_L ) |
---|
[185] | 54 | |
---|
[307] | 55 | except getopt.GetoptError, detail: |
---|
[212] | 56 | |
---|
| 57 | print detail |
---|
[307] | 58 | usage() |
---|
[354] | 59 | sys.exit( 1 ) |
---|
[212] | 60 | |
---|
| 61 | for opt, value in opts: |
---|
| 62 | |
---|
| 63 | if opt in [ '--config', '-c' ]: |
---|
| 64 | |
---|
[354] | 65 | config_filename = value |
---|
[212] | 66 | |
---|
[307] | 67 | if opt in [ '--pidfile', '-p' ]: |
---|
[212] | 68 | |
---|
[354] | 69 | PIDFILE = value |
---|
[307] | 70 | |
---|
| 71 | if opt in [ '--help', '-h' ]: |
---|
| 72 | |
---|
| 73 | usage() |
---|
[354] | 74 | sys.exit( 0 ) |
---|
[212] | 75 | |
---|
| 76 | return loadConfig( config_filename ) |
---|
| 77 | |
---|
| 78 | def loadConfig( filename ): |
---|
| 79 | |
---|
[215] | 80 | def getlist( cfg_string ): |
---|
| 81 | |
---|
| 82 | my_list = [ ] |
---|
| 83 | |
---|
| 84 | for item_txt in cfg_string.split( ',' ): |
---|
| 85 | |
---|
| 86 | sep_char = None |
---|
| 87 | |
---|
| 88 | item_txt = item_txt.strip() |
---|
| 89 | |
---|
| 90 | for s_char in [ "'", '"' ]: |
---|
| 91 | |
---|
| 92 | if item_txt.find( s_char ) != -1: |
---|
| 93 | |
---|
| 94 | if item_txt.count( s_char ) != 2: |
---|
| 95 | |
---|
| 96 | print 'Missing quote: %s' %item_txt |
---|
| 97 | sys.exit( 1 ) |
---|
| 98 | |
---|
| 99 | else: |
---|
| 100 | |
---|
| 101 | sep_char = s_char |
---|
| 102 | break |
---|
| 103 | |
---|
| 104 | if sep_char: |
---|
| 105 | |
---|
| 106 | item_txt = item_txt.split( sep_char )[1] |
---|
| 107 | |
---|
| 108 | my_list.append( item_txt ) |
---|
| 109 | |
---|
| 110 | return my_list |
---|
| 111 | |
---|
[354] | 112 | cfg = ConfigParser.ConfigParser() |
---|
[212] | 113 | |
---|
| 114 | cfg.read( filename ) |
---|
| 115 | |
---|
[354] | 116 | global DEBUG_LEVEL, DAEMONIZE, BATCH_SERVER, BATCH_POLL_INTERVAL |
---|
| 117 | global GMOND_CONF, DETECT_TIME_DIFFS, BATCH_HOST_TRANSLATE |
---|
| 118 | global BATCH_API, QUEUE, GMETRIC_TARGET |
---|
[212] | 119 | |
---|
[354] | 120 | DEBUG_LEVEL = cfg.getint( 'DEFAULT', 'DEBUG_LEVEL' ) |
---|
[212] | 121 | |
---|
[354] | 122 | DAEMONIZE = cfg.getboolean( 'DEFAULT', 'DAEMONIZE' ) |
---|
[212] | 123 | |
---|
[265] | 124 | try: |
---|
[212] | 125 | |
---|
[354] | 126 | BATCH_SERVER = cfg.get( 'DEFAULT', 'BATCH_SERVER' ) |
---|
[212] | 127 | |
---|
[265] | 128 | except ConfigParser.NoOptionError: |
---|
| 129 | |
---|
| 130 | # Backwards compatibility for old configs |
---|
| 131 | # |
---|
| 132 | |
---|
[354] | 133 | BATCH_SERVER = cfg.get( 'DEFAULT', 'TORQUE_SERVER' ) |
---|
| 134 | api_guess = 'pbs' |
---|
[265] | 135 | |
---|
| 136 | try: |
---|
| 137 | |
---|
[354] | 138 | BATCH_POLL_INTERVAL = cfg.getint( 'DEFAULT', 'BATCH_POLL_INTERVAL' ) |
---|
[265] | 139 | |
---|
| 140 | except ConfigParser.NoOptionError: |
---|
| 141 | |
---|
| 142 | # Backwards compatibility for old configs |
---|
| 143 | # |
---|
| 144 | |
---|
[354] | 145 | BATCH_POLL_INTERVAL = cfg.getint( 'DEFAULT', 'TORQUE_POLL_INTERVAL' ) |
---|
| 146 | api_guess = 'pbs' |
---|
[353] | 147 | |
---|
| 148 | try: |
---|
[212] | 149 | |
---|
[354] | 150 | GMOND_CONF = cfg.get( 'DEFAULT', 'GMOND_CONF' ) |
---|
[353] | 151 | |
---|
| 152 | except ConfigParser.NoOptionError: |
---|
| 153 | |
---|
[354] | 154 | GMOND_CONF = None |
---|
[353] | 155 | |
---|
[354] | 156 | DETECT_TIME_DIFFS = cfg.getboolean( 'DEFAULT', 'DETECT_TIME_DIFFS' ) |
---|
[212] | 157 | |
---|
[354] | 158 | BATCH_HOST_TRANSLATE = getlist( cfg.get( 'DEFAULT', 'BATCH_HOST_TRANSLATE' ) ) |
---|
[215] | 159 | |
---|
[266] | 160 | try: |
---|
[256] | 161 | |
---|
[354] | 162 | BATCH_API = cfg.get( 'DEFAULT', 'BATCH_API' ) |
---|
[266] | 163 | |
---|
| 164 | except ConfigParser.NoOptionError, detail: |
---|
| 165 | |
---|
| 166 | if BATCH_SERVER and api_guess: |
---|
[354] | 167 | |
---|
| 168 | BATCH_API = api_guess |
---|
[266] | 169 | else: |
---|
| 170 | debug_msg( 0, "fatal error: BATCH_API not set and can't make guess" ) |
---|
| 171 | sys.exit( 1 ) |
---|
[317] | 172 | |
---|
| 173 | try: |
---|
| 174 | |
---|
[354] | 175 | QUEUE = getlist( cfg.get( 'DEFAULT', 'QUEUE' ) ) |
---|
[317] | 176 | |
---|
| 177 | except ConfigParser.NoOptionError, detail: |
---|
| 178 | |
---|
[354] | 179 | QUEUE = None |
---|
[353] | 180 | |
---|
| 181 | try: |
---|
| 182 | |
---|
[354] | 183 | GMETRIC_TARGET = cfg.get( 'DEFAULT', 'GMETRIC_TARGET' ) |
---|
[353] | 184 | |
---|
| 185 | except ConfigParser.NoOptionError: |
---|
| 186 | |
---|
[354] | 187 | GMETRIC_TARGET = None |
---|
[353] | 188 | |
---|
| 189 | if not GMOND_CONF: |
---|
| 190 | |
---|
| 191 | debug_msg( 0, "fatal error: GMETRIC_TARGET or GMOND_CONF both not set!" ) |
---|
| 192 | sys.exit( 1 ) |
---|
| 193 | else: |
---|
| 194 | |
---|
| 195 | debug_msg( 0, "error: GMETRIC_TARGET not set: internel Gmetric handling aborted. Failing back to DEPRECATED use of gmond.conf/gmetric binary. This will slow down jobmond significantly!" ) |
---|
| 196 | |
---|
[212] | 197 | return True |
---|
| 198 | |
---|
[253] | 199 | METRIC_MAX_VAL_LEN = 900 |
---|
| 200 | |
---|
[61] | 201 | class DataProcessor: |
---|
[355] | 202 | |
---|
[68] | 203 | """Class for processing of data""" |
---|
[61] | 204 | |
---|
| 205 | binary = '/usr/bin/gmetric' |
---|
| 206 | |
---|
| 207 | def __init__( self, binary=None ): |
---|
[355] | 208 | |
---|
[68] | 209 | """Remember alternate binary location if supplied""" |
---|
[61] | 210 | |
---|
| 211 | if binary: |
---|
| 212 | self.binary = binary |
---|
| 213 | |
---|
[80] | 214 | # Timeout for XML |
---|
| 215 | # |
---|
| 216 | # From ganglia's documentation: |
---|
| 217 | # |
---|
| 218 | # 'A metric will be deleted DMAX seconds after it is received, and |
---|
| 219 | # DMAX=0 means eternal life.' |
---|
[61] | 220 | |
---|
[256] | 221 | self.dmax = str( int( int( BATCH_POLL_INTERVAL ) * 2 ) ) |
---|
[80] | 222 | |
---|
[353] | 223 | if GMOND_CONF: |
---|
[354] | 224 | |
---|
[353] | 225 | try: |
---|
| 226 | gmond_file = GMOND_CONF |
---|
[68] | 227 | |
---|
[353] | 228 | except NameError: |
---|
| 229 | gmond_file = '/etc/gmond.conf' |
---|
[68] | 230 | |
---|
[353] | 231 | if not os.path.exists( gmond_file ): |
---|
| 232 | debug_msg( 0, 'fatal error: ' + gmond_file + ' does not exist' ) |
---|
| 233 | sys.exit( 1 ) |
---|
[68] | 234 | |
---|
[353] | 235 | incompatible = self.checkGmetricVersion() |
---|
[61] | 236 | |
---|
[353] | 237 | if incompatible: |
---|
[355] | 238 | |
---|
| 239 | debug_msg( 0, 'Gmetric version not compatible, please upgrade to at least 3.0.1' ) |
---|
[353] | 240 | sys.exit( 1 ) |
---|
[65] | 241 | |
---|
| 242 | def checkGmetricVersion( self ): |
---|
[355] | 243 | |
---|
[68] | 244 | """ |
---|
| 245 | Check version of gmetric is at least 3.0.1 |
---|
| 246 | for the syntax we use |
---|
| 247 | """ |
---|
[65] | 248 | |
---|
[255] | 249 | global METRIC_MAX_VAL_LEN |
---|
| 250 | |
---|
[341] | 251 | incompatible = 0 |
---|
| 252 | |
---|
[355] | 253 | gfp = os.popen( self.binary + ' --version' ) |
---|
| 254 | lines = gfp.readlines() |
---|
[65] | 255 | |
---|
[355] | 256 | gfp.close() |
---|
| 257 | |
---|
| 258 | for line in lines: |
---|
| 259 | |
---|
[65] | 260 | line = line.split( ' ' ) |
---|
| 261 | |
---|
[355] | 262 | if len( line ) == 2 and str( line ).find( 'gmetric' ) != -1: |
---|
[65] | 263 | |
---|
[355] | 264 | gmetric_version = line[1].split( '\n' )[0] |
---|
[65] | 265 | |
---|
[355] | 266 | version_major = int( gmetric_version.split( '.' )[0] ) |
---|
| 267 | version_minor = int( gmetric_version.split( '.' )[1] ) |
---|
| 268 | version_patch = int( gmetric_version.split( '.' )[2] ) |
---|
[65] | 269 | |
---|
[355] | 270 | incompatible = 0 |
---|
[65] | 271 | |
---|
| 272 | if version_major < 3: |
---|
| 273 | |
---|
| 274 | incompatible = 1 |
---|
| 275 | |
---|
| 276 | elif version_major == 3: |
---|
| 277 | |
---|
| 278 | if version_minor == 0: |
---|
| 279 | |
---|
| 280 | if version_patch < 1: |
---|
| 281 | |
---|
[91] | 282 | incompatible = 1 |
---|
[65] | 283 | |
---|
[255] | 284 | if version_patch < 3: |
---|
| 285 | |
---|
| 286 | METRIC_MAX_VAL_LEN = 900 |
---|
| 287 | |
---|
| 288 | elif version_patch >= 3: |
---|
| 289 | |
---|
| 290 | METRIC_MAX_VAL_LEN = 1400 |
---|
| 291 | |
---|
[65] | 292 | return incompatible |
---|
| 293 | |
---|
[75] | 294 | def multicastGmetric( self, metricname, metricval, valtype='string' ): |
---|
[355] | 295 | |
---|
[68] | 296 | """Call gmetric binary and multicast""" |
---|
[65] | 297 | |
---|
| 298 | cmd = self.binary |
---|
| 299 | |
---|
[353] | 300 | if GMETRIC_TARGET: |
---|
[61] | 301 | |
---|
[353] | 302 | from gmetric import Gmetric |
---|
[61] | 303 | |
---|
[353] | 304 | if GMETRIC_TARGET: |
---|
[61] | 305 | |
---|
[353] | 306 | GMETRIC_TARGET_HOST = GMETRIC_TARGET.split( ':' )[0] |
---|
| 307 | GMETRIC_TARGET_PORT = GMETRIC_TARGET.split( ':' )[1] |
---|
| 308 | |
---|
| 309 | metric_debug = "[gmetric] name: %s - val: %s - dmax: %s" %( str( metricname ), str( metricval ), str( self.dmax ) ) |
---|
| 310 | |
---|
| 311 | debug_msg( 10, printTime() + ' ' + metric_debug) |
---|
| 312 | |
---|
| 313 | gm = Gmetric( GMETRIC_TARGET_HOST, GMETRIC_TARGET_PORT ) |
---|
| 314 | |
---|
| 315 | gm.send( str( metricname ), str( metricval ), str( self.dmax ) ) |
---|
| 316 | |
---|
| 317 | else: |
---|
| 318 | try: |
---|
| 319 | cmd = cmd + ' -c' + GMOND_CONF |
---|
| 320 | |
---|
| 321 | except NameError: |
---|
| 322 | |
---|
| 323 | debug_msg( 10, 'Assuming /etc/gmond.conf for gmetric cmd (ommitting)' ) |
---|
| 324 | |
---|
| 325 | cmd = cmd + ' -n' + str( metricname )+ ' -v"' + str( metricval )+ '" -t' + str( valtype ) + ' -d' + str( self.dmax ) |
---|
| 326 | |
---|
| 327 | debug_msg( 10, printTime() + ' ' + cmd ) |
---|
| 328 | |
---|
| 329 | os.system( cmd ) |
---|
| 330 | |
---|
[318] | 331 | class DataGatherer: |
---|
[23] | 332 | |
---|
[318] | 333 | """Skeleton class for batch system DataGatherer""" |
---|
[256] | 334 | |
---|
[318] | 335 | def printJobs( self, jobs ): |
---|
[355] | 336 | |
---|
[318] | 337 | """Print a jobinfo overview""" |
---|
| 338 | |
---|
| 339 | for name, attrs in self.jobs.items(): |
---|
| 340 | |
---|
| 341 | print 'job %s' %(name) |
---|
| 342 | |
---|
| 343 | for name, val in attrs.items(): |
---|
| 344 | |
---|
| 345 | print '\t%s = %s' %( name, val ) |
---|
| 346 | |
---|
| 347 | def printJob( self, jobs, job_id ): |
---|
[355] | 348 | |
---|
[318] | 349 | """Print job with job_id from jobs""" |
---|
| 350 | |
---|
| 351 | print 'job %s' %(job_id) |
---|
| 352 | |
---|
| 353 | for name, val in jobs[ job_id ].items(): |
---|
| 354 | |
---|
| 355 | print '\t%s = %s' %( name, val ) |
---|
| 356 | |
---|
[256] | 357 | def daemon( self ): |
---|
[355] | 358 | |
---|
[318] | 359 | """Run as daemon forever""" |
---|
[256] | 360 | |
---|
[318] | 361 | # Fork the first child |
---|
| 362 | # |
---|
| 363 | pid = os.fork() |
---|
| 364 | if pid > 0: |
---|
| 365 | sys.exit(0) # end parent |
---|
[256] | 366 | |
---|
[318] | 367 | # creates a session and sets the process group ID |
---|
| 368 | # |
---|
| 369 | os.setsid() |
---|
| 370 | |
---|
| 371 | # Fork the second child |
---|
| 372 | # |
---|
| 373 | pid = os.fork() |
---|
| 374 | if pid > 0: |
---|
| 375 | sys.exit(0) # end parent |
---|
| 376 | |
---|
| 377 | write_pidfile() |
---|
| 378 | |
---|
| 379 | # Go to the root directory and set the umask |
---|
| 380 | # |
---|
| 381 | os.chdir('/') |
---|
| 382 | os.umask(0) |
---|
| 383 | |
---|
| 384 | sys.stdin.close() |
---|
| 385 | sys.stdout.close() |
---|
| 386 | sys.stderr.close() |
---|
| 387 | |
---|
| 388 | os.open('/dev/null', os.O_RDWR) |
---|
| 389 | os.dup2(0, 1) |
---|
| 390 | os.dup2(0, 2) |
---|
| 391 | |
---|
| 392 | self.run() |
---|
| 393 | |
---|
[256] | 394 | def run( self ): |
---|
[355] | 395 | |
---|
[318] | 396 | """Main thread""" |
---|
[256] | 397 | |
---|
[318] | 398 | while ( 1 ): |
---|
| 399 | |
---|
[348] | 400 | self.getJobData() |
---|
| 401 | self.submitJobData() |
---|
[318] | 402 | time.sleep( BATCH_POLL_INTERVAL ) |
---|
[256] | 403 | |
---|
[318] | 404 | class SgeQstatXMLParser(xml.sax.handler.ContentHandler): |
---|
[256] | 405 | |
---|
[318] | 406 | """Babu Sundaram's experimental SGE qstat XML parser""" |
---|
[256] | 407 | |
---|
[318] | 408 | def __init__(self, qstatinxml): |
---|
| 409 | |
---|
| 410 | self.qstatfile = qstatinxml |
---|
| 411 | self.attribs = {} |
---|
| 412 | self.value = '' |
---|
| 413 | self.jobID = '' |
---|
| 414 | self.currentJobInfo = '' |
---|
| 415 | self.job_list = [] |
---|
| 416 | self.EOFFlag = 0 |
---|
| 417 | self.jobinfoCount = 0 |
---|
| 418 | |
---|
| 419 | |
---|
| 420 | def startElement(self, name, attrs): |
---|
| 421 | |
---|
| 422 | if name == 'job_list': |
---|
| 423 | self.currentJobInfo = 'Status=' + attrs.get('state', None) + ' ' |
---|
| 424 | elif name == 'job_info': |
---|
| 425 | self.job_list = [] |
---|
| 426 | self.jobinfoCount += 1 |
---|
| 427 | |
---|
| 428 | def characters(self, ch): |
---|
| 429 | |
---|
| 430 | self.value = self.value + ch |
---|
| 431 | |
---|
| 432 | def endElement(self, name): |
---|
| 433 | |
---|
| 434 | if len(self.value.strip()) > 0 : |
---|
| 435 | |
---|
| 436 | self.currentJobInfo += name + '=' + self.value.strip() + ' ' |
---|
| 437 | elif name != 'job_list': |
---|
| 438 | |
---|
| 439 | self.currentJobInfo += name + '=Unknown ' |
---|
| 440 | |
---|
| 441 | if name == 'JB_job_number': |
---|
| 442 | |
---|
| 443 | self.jobID = self.value.strip() |
---|
| 444 | self.job_list.append(self.jobID) |
---|
| 445 | |
---|
| 446 | if name == 'job_list': |
---|
| 447 | |
---|
| 448 | if self.attribs.has_key(self.jobID) == False: |
---|
| 449 | self.attribs[self.jobID] = self.currentJobInfo |
---|
| 450 | elif self.attribs.has_key(self.jobID) and self.attribs[self.jobID] != self.currentJobInfo: |
---|
| 451 | self.attribs[self.jobID] = self.currentJobInfo |
---|
| 452 | self.currentJobInfo = '' |
---|
| 453 | self.jobID = '' |
---|
| 454 | |
---|
| 455 | elif name == 'job_info' and self.jobinfoCount == 2: |
---|
| 456 | |
---|
| 457 | deljobs = [] |
---|
| 458 | for id in self.attribs: |
---|
| 459 | try: |
---|
| 460 | self.job_list.index(str(id)) |
---|
| 461 | except ValueError: |
---|
| 462 | deljobs.append(id) |
---|
| 463 | for i in deljobs: |
---|
| 464 | del self.attribs[i] |
---|
| 465 | deljobs = [] |
---|
| 466 | self.jobinfoCount = 0 |
---|
| 467 | |
---|
| 468 | self.value = '' |
---|
| 469 | |
---|
| 470 | class SgeDataGatherer(DataGatherer): |
---|
| 471 | |
---|
[61] | 472 | jobs = { } |
---|
[347] | 473 | SGE_QSTAT_XML_FILE = '/tmp/.jobmonarch.sge.qstat' |
---|
[61] | 474 | |
---|
[318] | 475 | def __init__( self ): |
---|
| 476 | """Setup appropriate variables""" |
---|
| 477 | |
---|
| 478 | self.jobs = { } |
---|
| 479 | self.timeoffset = 0 |
---|
| 480 | self.dp = DataProcessor() |
---|
| 481 | self.initSgeJobInfo() |
---|
| 482 | |
---|
| 483 | def initSgeJobInfo( self ): |
---|
| 484 | """This is outside the scope of DRMAA; Get the current jobs in SGE""" |
---|
| 485 | """This is a hack because we cant get info about jobs beyond""" |
---|
| 486 | """those in the current DRMAA session""" |
---|
| 487 | |
---|
[347] | 488 | self.qstatparser = SgeQstatXMLParser( self.SGE_QSTAT_XML_FILE ) |
---|
[318] | 489 | |
---|
| 490 | # Obtain the qstat information from SGE in XML format |
---|
| 491 | # This would change to DRMAA-specific calls from 6.0u9 |
---|
| 492 | |
---|
| 493 | def getJobData(self): |
---|
| 494 | """Gather all data on current jobs in SGE""" |
---|
| 495 | |
---|
| 496 | # Get the information about the current jobs in the SGE queue |
---|
| 497 | info = os.popen("qstat -ext -xml").readlines() |
---|
[347] | 498 | f = open(self.SGE_QSTAT_XML_FILE,'w') |
---|
[318] | 499 | for lines in info: |
---|
| 500 | f.write(lines) |
---|
| 501 | f.close() |
---|
| 502 | |
---|
| 503 | # Parse the input |
---|
| 504 | f = open(self.qstatparser.qstatfile, 'r') |
---|
| 505 | xml.sax.parse(f, self.qstatparser) |
---|
| 506 | f.close() |
---|
| 507 | |
---|
| 508 | self.cur_time = time.time() |
---|
| 509 | |
---|
| 510 | return self.qstatparser.attribs |
---|
| 511 | |
---|
| 512 | def submitJobData(self): |
---|
| 513 | """Submit job info list""" |
---|
| 514 | |
---|
| 515 | self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) ) |
---|
| 516 | # Now let's spread the knowledge |
---|
| 517 | # |
---|
| 518 | metric_increment = 0 |
---|
| 519 | for jobid, jobattrs in self.qstatparser.attribs.items(): |
---|
| 520 | |
---|
| 521 | self.dp.multicastGmetric( 'MONARCH-JOB-' + jobid + '-' + str(metric_increment), jobattrs) |
---|
| 522 | |
---|
[355] | 523 | class PbsDataGatherer( DataGatherer ): |
---|
[318] | 524 | |
---|
| 525 | """This is the DataGatherer for PBS and Torque""" |
---|
| 526 | |
---|
[256] | 527 | global PBSQuery |
---|
| 528 | |
---|
[23] | 529 | def __init__( self ): |
---|
[354] | 530 | |
---|
[68] | 531 | """Setup appropriate variables""" |
---|
[23] | 532 | |
---|
[354] | 533 | self.jobs = { } |
---|
| 534 | self.timeoffset = 0 |
---|
| 535 | self.dp = DataProcessor() |
---|
| 536 | |
---|
[91] | 537 | self.initPbsQuery() |
---|
[23] | 538 | |
---|
[91] | 539 | def initPbsQuery( self ): |
---|
| 540 | |
---|
[354] | 541 | self.pq = None |
---|
| 542 | |
---|
[256] | 543 | if( BATCH_SERVER ): |
---|
[354] | 544 | |
---|
| 545 | self.pq = PBSQuery( BATCH_SERVER ) |
---|
[174] | 546 | else: |
---|
[354] | 547 | self.pq = PBSQuery() |
---|
[91] | 548 | |
---|
[26] | 549 | def getAttr( self, attrs, name ): |
---|
[354] | 550 | |
---|
[68] | 551 | """Return certain attribute from dictionary, if exists""" |
---|
[26] | 552 | |
---|
| 553 | if attrs.has_key( name ): |
---|
[354] | 554 | |
---|
| 555 | return attrs[ name ] |
---|
[26] | 556 | else: |
---|
| 557 | return '' |
---|
| 558 | |
---|
| 559 | def jobDataChanged( self, jobs, job_id, attrs ): |
---|
[354] | 560 | |
---|
[68] | 561 | """Check if job with attrs and job_id in jobs has changed""" |
---|
[26] | 562 | |
---|
| 563 | if jobs.has_key( job_id ): |
---|
[354] | 564 | |
---|
[26] | 565 | oldData = jobs[ job_id ] |
---|
| 566 | else: |
---|
| 567 | return 1 |
---|
| 568 | |
---|
| 569 | for name, val in attrs.items(): |
---|
| 570 | |
---|
| 571 | if oldData.has_key( name ): |
---|
| 572 | |
---|
| 573 | if oldData[ name ] != attrs[ name ]: |
---|
| 574 | |
---|
| 575 | return 1 |
---|
| 576 | |
---|
| 577 | else: |
---|
| 578 | return 1 |
---|
| 579 | |
---|
| 580 | return 0 |
---|
| 581 | |
---|
[348] | 582 | def getJobData( self ): |
---|
[354] | 583 | |
---|
[68] | 584 | """Gather all data on current jobs in Torque""" |
---|
[26] | 585 | |
---|
[354] | 586 | joblist = {} |
---|
[349] | 587 | |
---|
[354] | 588 | while len( joblist ) == 0: |
---|
| 589 | |
---|
[282] | 590 | try: |
---|
| 591 | joblist = self.pq.getjobs() |
---|
[354] | 592 | |
---|
[349] | 593 | except PBSError, detail: |
---|
[354] | 594 | |
---|
| 595 | debug_msg( 10, "Caught PBS unavailable, skipping until next polling interval: " + str( detail ) ) |
---|
[349] | 596 | return None |
---|
[26] | 597 | |
---|
[354] | 598 | self.cur_time = time.time() |
---|
[68] | 599 | |
---|
[354] | 600 | jobs_processed = [ ] |
---|
[26] | 601 | |
---|
| 602 | for name, attrs in joblist.items(): |
---|
| 603 | |
---|
[354] | 604 | job_id = name.split( '.' )[0] |
---|
[26] | 605 | |
---|
| 606 | jobs_processed.append( job_id ) |
---|
[61] | 607 | |
---|
[354] | 608 | name = self.getAttr( attrs, 'Job_Name' ) |
---|
| 609 | queue = self.getAttr( attrs, 'queue' ) |
---|
[317] | 610 | |
---|
| 611 | if QUEUE: |
---|
| 612 | |
---|
| 613 | if QUEUE != queue: |
---|
| 614 | |
---|
| 615 | continue |
---|
| 616 | |
---|
[354] | 617 | owner = self.getAttr( attrs, 'Job_Owner' ).split( '@' )[0] |
---|
| 618 | requested_time = self.getAttr( attrs, 'Resource_List.walltime' ) |
---|
| 619 | requested_memory = self.getAttr( attrs, 'Resource_List.mem' ) |
---|
[95] | 620 | |
---|
[354] | 621 | mynoderequest = self.getAttr( attrs, 'Resource_List.nodes' ) |
---|
[95] | 622 | |
---|
[354] | 623 | ppn = '' |
---|
[281] | 624 | |
---|
[26] | 625 | if mynoderequest.find( ':' ) != -1 and mynoderequest.find( 'ppn' ) != -1: |
---|
[95] | 626 | |
---|
[354] | 627 | mynoderequest_fields = mynoderequest.split( ':' ) |
---|
[281] | 628 | |
---|
| 629 | for mynoderequest_field in mynoderequest_fields: |
---|
| 630 | |
---|
| 631 | if mynoderequest_field.find( 'ppn' ) != -1: |
---|
| 632 | |
---|
[354] | 633 | ppn = mynoderequest_field.split( 'ppn=' )[1] |
---|
[281] | 634 | |
---|
[354] | 635 | status = self.getAttr( attrs, 'job_state' ) |
---|
[25] | 636 | |
---|
[354] | 637 | queued_timestamp = self.getAttr( attrs, 'ctime' ) |
---|
[243] | 638 | |
---|
[95] | 639 | if status == 'R': |
---|
[133] | 640 | |
---|
[354] | 641 | start_timestamp = self.getAttr( attrs, 'mtime' ) |
---|
| 642 | nodes = self.getAttr( attrs, 'exec_host' ).split( '+' ) |
---|
[133] | 643 | |
---|
[354] | 644 | nodeslist = [ ] |
---|
| 645 | |
---|
[133] | 646 | for node in nodes: |
---|
| 647 | |
---|
[354] | 648 | host = node.split( '/' )[0] |
---|
| 649 | |
---|
[133] | 650 | if nodeslist.count( host ) == 0: |
---|
[215] | 651 | |
---|
| 652 | for translate_pattern in BATCH_HOST_TRANSLATE: |
---|
| 653 | |
---|
[220] | 654 | if translate_pattern.find( '/' ) != -1: |
---|
[215] | 655 | |
---|
[354] | 656 | translate_orig = translate_pattern.split( '/' )[1] |
---|
| 657 | translate_new = translate_pattern.split( '/' )[2] |
---|
[220] | 658 | |
---|
[354] | 659 | host = re.sub( translate_orig, translate_new, host ) |
---|
[216] | 660 | |
---|
[217] | 661 | if not host in nodeslist: |
---|
[216] | 662 | |
---|
| 663 | nodeslist.append( host ) |
---|
[133] | 664 | |
---|
[185] | 665 | if DETECT_TIME_DIFFS: |
---|
| 666 | |
---|
| 667 | # If a job start if later than our current date, |
---|
| 668 | # that must mean the Torque server's time is later |
---|
| 669 | # than our local time. |
---|
| 670 | |
---|
[354] | 671 | if int( start_timestamp ) > int( int( self.cur_time ) + int( self.timeoffset ) ): |
---|
[185] | 672 | |
---|
[354] | 673 | self.timeoffset = int( int(start_timestamp) - int(self.cur_time) ) |
---|
[185] | 674 | |
---|
[133] | 675 | elif status == 'Q': |
---|
[95] | 676 | |
---|
[354] | 677 | start_timestamp = '' |
---|
| 678 | count_mynodes = 0 |
---|
| 679 | numeric_node = 1 |
---|
| 680 | |
---|
[133] | 681 | for node in mynoderequest.split( '+' ): |
---|
[67] | 682 | |
---|
[354] | 683 | nodepart = node.split( ':' )[0] |
---|
[67] | 684 | |
---|
[133] | 685 | for letter in nodepart: |
---|
[67] | 686 | |
---|
[133] | 687 | if letter not in string.digits: |
---|
| 688 | |
---|
[354] | 689 | numeric_node = 0 |
---|
[133] | 690 | |
---|
| 691 | if not numeric_node: |
---|
[354] | 692 | |
---|
| 693 | count_mynodes = count_mynodes + 1 |
---|
[133] | 694 | else: |
---|
[327] | 695 | try: |
---|
[354] | 696 | count_mynodes = count_mynodes + int( nodepart ) |
---|
| 697 | |
---|
[327] | 698 | except ValueError, detail: |
---|
[354] | 699 | |
---|
[327] | 700 | debug_msg( 10, str( detail ) ) |
---|
| 701 | debug_msg( 10, "Encountered weird node in Resources_List?!" ) |
---|
| 702 | debug_msg( 10, 'nodepart = ' + str( nodepart ) ) |
---|
| 703 | debug_msg( 10, 'job = ' + str( name ) ) |
---|
| 704 | debug_msg( 10, 'attrs = ' + str( attrs ) ) |
---|
[133] | 705 | |
---|
[354] | 706 | nodeslist = str( count_mynodes ) |
---|
[172] | 707 | else: |
---|
[354] | 708 | start_timestamp = '' |
---|
| 709 | nodeslist = '' |
---|
[133] | 710 | |
---|
[354] | 711 | myAttrs = { } |
---|
[26] | 712 | |
---|
[354] | 713 | myAttrs[ 'name' ] = str( name ) |
---|
| 714 | myAttrs[ 'queue' ] = str( queue ) |
---|
| 715 | myAttrs[ 'owner' ] = str( owner ) |
---|
| 716 | myAttrs[ 'requested_time' ] = str( requested_time ) |
---|
| 717 | myAttrs[ 'requested_memory' ] = str( requested_memory ) |
---|
| 718 | myAttrs[ 'ppn' ] = str( ppn ) |
---|
| 719 | myAttrs[ 'status' ] = str( status ) |
---|
| 720 | myAttrs[ 'start_timestamp' ] = str( start_timestamp ) |
---|
| 721 | myAttrs[ 'queued_timestamp' ] = str( queued_timestamp ) |
---|
| 722 | myAttrs[ 'reported' ] = str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) |
---|
| 723 | myAttrs[ 'nodes' ] = nodeslist |
---|
| 724 | myAttrs[ 'domain' ] = string.join( socket.getfqdn().split( '.' )[1:], '.' ) |
---|
| 725 | myAttrs[ 'poll_interval' ] = str( BATCH_POLL_INTERVAL ) |
---|
| 726 | |
---|
[348] | 727 | if self.jobDataChanged( self.jobs, job_id, myAttrs ) and myAttrs['status'] in [ 'R', 'Q' ]: |
---|
[61] | 728 | |
---|
[354] | 729 | self.jobs[ job_id ] = myAttrs |
---|
[26] | 730 | |
---|
[348] | 731 | for id, attrs in self.jobs.items(): |
---|
[76] | 732 | |
---|
| 733 | if id not in jobs_processed: |
---|
| 734 | |
---|
| 735 | # This one isn't there anymore; toedeledoki! |
---|
| 736 | # |
---|
[348] | 737 | del self.jobs[ id ] |
---|
[76] | 738 | |
---|
[348] | 739 | def submitJobData( self ): |
---|
[354] | 740 | |
---|
[65] | 741 | """Submit job info list""" |
---|
| 742 | |
---|
[219] | 743 | self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) ) |
---|
[69] | 744 | |
---|
[61] | 745 | # Now let's spread the knowledge |
---|
| 746 | # |
---|
[348] | 747 | for jobid, jobattrs in self.jobs.items(): |
---|
[61] | 748 | |
---|
[354] | 749 | gmetric_val = self.compileGmetricVal( jobid, jobattrs ) |
---|
| 750 | metric_increment = 0 |
---|
[61] | 751 | |
---|
[354] | 752 | for val in gmetric_val: |
---|
[253] | 753 | |
---|
| 754 | self.dp.multicastGmetric( 'MONARCH-JOB-' + jobid + '-' + str(metric_increment), val ) |
---|
[61] | 755 | |
---|
[354] | 756 | metric_increment = metric_increment + 1 |
---|
| 757 | |
---|
[253] | 758 | def compileGmetricVal( self, jobid, jobattrs ): |
---|
[354] | 759 | |
---|
[253] | 760 | """Create a val string for gmetric of jobinfo""" |
---|
[67] | 761 | |
---|
[354] | 762 | gval_lists = [ ] |
---|
| 763 | mystr = None |
---|
| 764 | val_list = { } |
---|
[67] | 765 | |
---|
[253] | 766 | for val_name, val_value in jobattrs.items(): |
---|
[61] | 767 | |
---|
[253] | 768 | val_list_names_len = len( string.join( val_list.keys() ) ) + len(val_list.keys()) |
---|
| 769 | val_list_vals_len = len( string.join( val_list.values() ) ) + len(val_list.values()) |
---|
[95] | 770 | |
---|
[254] | 771 | if val_name == 'nodes' and jobattrs['status'] == 'R': |
---|
[95] | 772 | |
---|
[253] | 773 | node_str = None |
---|
[95] | 774 | |
---|
[253] | 775 | for node in val_value: |
---|
[65] | 776 | |
---|
[253] | 777 | if node_str: |
---|
[354] | 778 | |
---|
[253] | 779 | node_str = node_str + ';' + node |
---|
| 780 | else: |
---|
| 781 | node_str = node |
---|
[65] | 782 | |
---|
[253] | 783 | if (val_list_names_len + len(val_name) ) + (val_list_vals_len + len(node_str) ) > METRIC_MAX_VAL_LEN: |
---|
[65] | 784 | |
---|
[354] | 785 | val_list[ val_name ] = node_str |
---|
| 786 | |
---|
[253] | 787 | gval_lists.append( val_list ) |
---|
[65] | 788 | |
---|
[354] | 789 | val_list = { } |
---|
| 790 | node_str = None |
---|
| 791 | |
---|
| 792 | val_list[ val_name ] = node_str |
---|
| 793 | |
---|
[253] | 794 | gval_lists.append( val_list ) |
---|
[65] | 795 | |
---|
[354] | 796 | val_list = { } |
---|
| 797 | |
---|
[254] | 798 | elif val_value != '': |
---|
| 799 | |
---|
| 800 | if (val_list_names_len + len(val_name) ) + (val_list_vals_len + len(str(val_value)) ) > METRIC_MAX_VAL_LEN: |
---|
| 801 | |
---|
| 802 | gval_lists.append( val_list ) |
---|
| 803 | |
---|
[354] | 804 | val_list = { } |
---|
[254] | 805 | |
---|
[354] | 806 | val_list[ val_name ] = val_value |
---|
| 807 | |
---|
| 808 | if len( val_list ) > 0: |
---|
| 809 | |
---|
[254] | 810 | gval_lists.append( val_list ) |
---|
| 811 | |
---|
[354] | 812 | str_list = [ ] |
---|
[65] | 813 | |
---|
[253] | 814 | for val_list in gval_lists: |
---|
[65] | 815 | |
---|
[354] | 816 | my_val_str = None |
---|
[65] | 817 | |
---|
[253] | 818 | for val_name, val_value in val_list.items(): |
---|
[65] | 819 | |
---|
[253] | 820 | if my_val_str: |
---|
[65] | 821 | |
---|
[253] | 822 | my_val_str = my_val_str + ' ' + val_name + '=' + val_value |
---|
| 823 | else: |
---|
| 824 | my_val_str = val_name + '=' + val_value |
---|
| 825 | |
---|
| 826 | str_list.append( my_val_str ) |
---|
| 827 | |
---|
| 828 | return str_list |
---|
| 829 | |
---|
[26] | 830 | def printTime( ): |
---|
[354] | 831 | |
---|
[65] | 832 | """Print current time/date in human readable format for log/debug""" |
---|
[26] | 833 | |
---|
| 834 | return time.strftime("%a, %d %b %Y %H:%M:%S") |
---|
| 835 | |
---|
| 836 | def debug_msg( level, msg ): |
---|
[354] | 837 | |
---|
[65] | 838 | """Print msg if at or above current debug level""" |
---|
[26] | 839 | |
---|
| 840 | if (DEBUG_LEVEL >= level): |
---|
| 841 | sys.stderr.write( msg + '\n' ) |
---|
| 842 | |
---|
[307] | 843 | def write_pidfile(): |
---|
| 844 | |
---|
| 845 | # Write pidfile if PIDFILE exists |
---|
| 846 | if PIDFILE: |
---|
| 847 | |
---|
[354] | 848 | pid = os.getpid() |
---|
| 849 | |
---|
| 850 | pidfile = open(PIDFILE, 'w') |
---|
| 851 | |
---|
| 852 | pidfile.write( str( pid ) ) |
---|
[307] | 853 | pidfile.close() |
---|
| 854 | |
---|
[23] | 855 | def main(): |
---|
[354] | 856 | |
---|
[65] | 857 | """Application start""" |
---|
[23] | 858 | |
---|
[352] | 859 | global PBSQuery, PBSError |
---|
[256] | 860 | |
---|
[212] | 861 | if not processArgs( sys.argv[1:] ): |
---|
[354] | 862 | |
---|
[212] | 863 | sys.exit( 1 ) |
---|
| 864 | |
---|
[256] | 865 | if BATCH_API == 'pbs': |
---|
| 866 | |
---|
| 867 | try: |
---|
[282] | 868 | from PBSQuery import PBSQuery, PBSError |
---|
[256] | 869 | |
---|
| 870 | except ImportError: |
---|
| 871 | |
---|
| 872 | debug_msg( 0, "fatal error: BATCH_API set to 'pbs' but python module 'pbs_python' is not installed" ) |
---|
| 873 | sys.exit( 1 ) |
---|
| 874 | |
---|
| 875 | gather = PbsDataGatherer() |
---|
| 876 | |
---|
| 877 | elif BATCH_API == 'sge': |
---|
| 878 | |
---|
[347] | 879 | gather = SgeDataGatherer() |
---|
[256] | 880 | |
---|
| 881 | else: |
---|
| 882 | debug_msg( 0, "fatal error: unknown BATCH_API '" + BATCH_API + "' is not supported" ) |
---|
[354] | 883 | |
---|
[256] | 884 | sys.exit( 1 ) |
---|
| 885 | |
---|
[26] | 886 | if DAEMONIZE: |
---|
[354] | 887 | |
---|
[26] | 888 | gather.daemon() |
---|
| 889 | else: |
---|
| 890 | gather.run() |
---|
[23] | 891 | |
---|
[256] | 892 | # wh00t? someone started me! :) |
---|
[65] | 893 | # |
---|
[23] | 894 | if __name__ == '__main__': |
---|
| 895 | main() |
---|