[23] | 1 | #!/usr/bin/env python |
---|
[225] | 2 | # |
---|
| 3 | # This file is part of Jobmonarch |
---|
| 4 | # |
---|
[363] | 5 | # Copyright (C) 2006-2007 Ramon Bastiaans |
---|
[225] | 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 425 2007-07-06 18:41:09Z bastiaans $ |
---|
[227] | 22 | # |
---|
[23] | 23 | |
---|
[212] | 24 | import sys, getopt, ConfigParser |
---|
[354] | 25 | import time, os, socket, string, re |
---|
[373] | 26 | import xdrlib, socket, syslog |
---|
[318] | 27 | import xml, xml.sax |
---|
| 28 | from xml.sax import saxutils, make_parser |
---|
| 29 | from xml.sax import make_parser |
---|
| 30 | from xml.sax.handler import feature_namespaces |
---|
| 31 | |
---|
[307] | 32 | def usage(): |
---|
| 33 | |
---|
| 34 | print |
---|
| 35 | print 'usage: jobmond [options]' |
---|
| 36 | print 'options:' |
---|
| 37 | print ' --config, -c configuration file' |
---|
| 38 | print ' --pidfile, -p pid file' |
---|
| 39 | print ' --help, -h help' |
---|
| 40 | print |
---|
| 41 | |
---|
[212] | 42 | def processArgs( args ): |
---|
[26] | 43 | |
---|
[354] | 44 | SHORT_L = 'hc:' |
---|
| 45 | LONG_L = [ 'help', 'config=' ] |
---|
[165] | 46 | |
---|
[307] | 47 | global PIDFILE |
---|
[354] | 48 | PIDFILE = None |
---|
[61] | 49 | |
---|
[354] | 50 | config_filename = '/etc/jobmond.conf' |
---|
| 51 | |
---|
[212] | 52 | try: |
---|
[68] | 53 | |
---|
[354] | 54 | opts, args = getopt.getopt( args, SHORT_L, LONG_L ) |
---|
[185] | 55 | |
---|
[307] | 56 | except getopt.GetoptError, detail: |
---|
[212] | 57 | |
---|
| 58 | print detail |
---|
[307] | 59 | usage() |
---|
[354] | 60 | sys.exit( 1 ) |
---|
[212] | 61 | |
---|
| 62 | for opt, value in opts: |
---|
| 63 | |
---|
| 64 | if opt in [ '--config', '-c' ]: |
---|
| 65 | |
---|
[354] | 66 | config_filename = value |
---|
[212] | 67 | |
---|
[307] | 68 | if opt in [ '--pidfile', '-p' ]: |
---|
[212] | 69 | |
---|
[354] | 70 | PIDFILE = value |
---|
[307] | 71 | |
---|
| 72 | if opt in [ '--help', '-h' ]: |
---|
| 73 | |
---|
| 74 | usage() |
---|
[354] | 75 | sys.exit( 0 ) |
---|
[212] | 76 | |
---|
| 77 | return loadConfig( config_filename ) |
---|
| 78 | |
---|
| 79 | def loadConfig( filename ): |
---|
| 80 | |
---|
[215] | 81 | def getlist( cfg_string ): |
---|
| 82 | |
---|
| 83 | my_list = [ ] |
---|
| 84 | |
---|
| 85 | for item_txt in cfg_string.split( ',' ): |
---|
| 86 | |
---|
| 87 | sep_char = None |
---|
| 88 | |
---|
| 89 | item_txt = item_txt.strip() |
---|
| 90 | |
---|
| 91 | for s_char in [ "'", '"' ]: |
---|
| 92 | |
---|
| 93 | if item_txt.find( s_char ) != -1: |
---|
| 94 | |
---|
| 95 | if item_txt.count( s_char ) != 2: |
---|
| 96 | |
---|
| 97 | print 'Missing quote: %s' %item_txt |
---|
| 98 | sys.exit( 1 ) |
---|
| 99 | |
---|
| 100 | else: |
---|
| 101 | |
---|
| 102 | sep_char = s_char |
---|
| 103 | break |
---|
| 104 | |
---|
| 105 | if sep_char: |
---|
| 106 | |
---|
| 107 | item_txt = item_txt.split( sep_char )[1] |
---|
| 108 | |
---|
| 109 | my_list.append( item_txt ) |
---|
| 110 | |
---|
| 111 | return my_list |
---|
| 112 | |
---|
[354] | 113 | cfg = ConfigParser.ConfigParser() |
---|
[212] | 114 | |
---|
| 115 | cfg.read( filename ) |
---|
| 116 | |
---|
[354] | 117 | global DEBUG_LEVEL, DAEMONIZE, BATCH_SERVER, BATCH_POLL_INTERVAL |
---|
| 118 | global GMOND_CONF, DETECT_TIME_DIFFS, BATCH_HOST_TRANSLATE |
---|
[373] | 119 | global BATCH_API, QUEUE, GMETRIC_TARGET, USE_SYSLOG |
---|
| 120 | global SYSLOG_LEVEL, SYSLOG_FACILITY |
---|
[212] | 121 | |
---|
[354] | 122 | DEBUG_LEVEL = cfg.getint( 'DEFAULT', 'DEBUG_LEVEL' ) |
---|
[212] | 123 | |
---|
[354] | 124 | DAEMONIZE = cfg.getboolean( 'DEFAULT', 'DAEMONIZE' ) |
---|
[212] | 125 | |
---|
[377] | 126 | SYSLOG_LEVEL = -1 |
---|
| 127 | SYSLOG_FACILITY = None |
---|
| 128 | |
---|
[265] | 129 | try: |
---|
[373] | 130 | USE_SYSLOG = cfg.getboolean( 'DEFAULT', 'USE_SYSLOG' ) |
---|
[212] | 131 | |
---|
[373] | 132 | except ConfigParser.NoOptionError: |
---|
| 133 | |
---|
| 134 | USE_SYSLOG = True |
---|
| 135 | |
---|
| 136 | debug_msg( 0, 'ERROR: no option USE_SYSLOG found: assuming yes' ) |
---|
| 137 | |
---|
[377] | 138 | |
---|
[373] | 139 | if USE_SYSLOG: |
---|
| 140 | |
---|
| 141 | try: |
---|
| 142 | SYSLOG_LEVEL = cfg.getint( 'DEFAULT', 'SYSLOG_LEVEL' ) |
---|
| 143 | |
---|
| 144 | except ConfigParser.NoOptionError: |
---|
| 145 | |
---|
| 146 | debug_msg( 0, 'ERROR: no option SYSLOG_LEVEL found: assuming level 0' ) |
---|
| 147 | SYSLOG_LEVEL = 0 |
---|
| 148 | |
---|
| 149 | try: |
---|
| 150 | |
---|
| 151 | SYSLOG_FACILITY = eval( 'syslog.LOG_' + cfg.get( 'DEFAULT', 'SYSLOG_FACILITY' ) ) |
---|
| 152 | |
---|
[377] | 153 | except ConfigParser.NoOptionError: |
---|
[373] | 154 | |
---|
| 155 | SYSLOG_FACILITY = syslog.LOG_DAEMON |
---|
| 156 | |
---|
| 157 | debug_msg( 0, 'ERROR: no option SYSLOG_FACILITY found: assuming facility DAEMON' ) |
---|
| 158 | |
---|
| 159 | try: |
---|
| 160 | |
---|
[354] | 161 | BATCH_SERVER = cfg.get( 'DEFAULT', 'BATCH_SERVER' ) |
---|
[212] | 162 | |
---|
[265] | 163 | except ConfigParser.NoOptionError: |
---|
| 164 | |
---|
| 165 | # Backwards compatibility for old configs |
---|
| 166 | # |
---|
| 167 | |
---|
[354] | 168 | BATCH_SERVER = cfg.get( 'DEFAULT', 'TORQUE_SERVER' ) |
---|
| 169 | api_guess = 'pbs' |
---|
[265] | 170 | |
---|
| 171 | try: |
---|
| 172 | |
---|
[354] | 173 | BATCH_POLL_INTERVAL = cfg.getint( 'DEFAULT', 'BATCH_POLL_INTERVAL' ) |
---|
[265] | 174 | |
---|
| 175 | except ConfigParser.NoOptionError: |
---|
| 176 | |
---|
| 177 | # Backwards compatibility for old configs |
---|
| 178 | # |
---|
| 179 | |
---|
[354] | 180 | BATCH_POLL_INTERVAL = cfg.getint( 'DEFAULT', 'TORQUE_POLL_INTERVAL' ) |
---|
| 181 | api_guess = 'pbs' |
---|
[353] | 182 | |
---|
| 183 | try: |
---|
[212] | 184 | |
---|
[354] | 185 | GMOND_CONF = cfg.get( 'DEFAULT', 'GMOND_CONF' ) |
---|
[353] | 186 | |
---|
| 187 | except ConfigParser.NoOptionError: |
---|
| 188 | |
---|
[354] | 189 | GMOND_CONF = None |
---|
[353] | 190 | |
---|
[354] | 191 | DETECT_TIME_DIFFS = cfg.getboolean( 'DEFAULT', 'DETECT_TIME_DIFFS' ) |
---|
[212] | 192 | |
---|
[354] | 193 | BATCH_HOST_TRANSLATE = getlist( cfg.get( 'DEFAULT', 'BATCH_HOST_TRANSLATE' ) ) |
---|
[215] | 194 | |
---|
[266] | 195 | try: |
---|
[256] | 196 | |
---|
[354] | 197 | BATCH_API = cfg.get( 'DEFAULT', 'BATCH_API' ) |
---|
[266] | 198 | |
---|
| 199 | except ConfigParser.NoOptionError, detail: |
---|
| 200 | |
---|
| 201 | if BATCH_SERVER and api_guess: |
---|
[354] | 202 | |
---|
| 203 | BATCH_API = api_guess |
---|
[266] | 204 | else: |
---|
[373] | 205 | debug_msg( 0, "FATAL ERROR: BATCH_API not set and can't make guess" ) |
---|
[266] | 206 | sys.exit( 1 ) |
---|
[317] | 207 | |
---|
| 208 | try: |
---|
| 209 | |
---|
[354] | 210 | QUEUE = getlist( cfg.get( 'DEFAULT', 'QUEUE' ) ) |
---|
[317] | 211 | |
---|
| 212 | except ConfigParser.NoOptionError, detail: |
---|
| 213 | |
---|
[354] | 214 | QUEUE = None |
---|
[353] | 215 | |
---|
| 216 | try: |
---|
| 217 | |
---|
[354] | 218 | GMETRIC_TARGET = cfg.get( 'DEFAULT', 'GMETRIC_TARGET' ) |
---|
[353] | 219 | |
---|
| 220 | except ConfigParser.NoOptionError: |
---|
| 221 | |
---|
[354] | 222 | GMETRIC_TARGET = None |
---|
[353] | 223 | |
---|
| 224 | if not GMOND_CONF: |
---|
| 225 | |
---|
[373] | 226 | debug_msg( 0, "FATAL ERROR: GMETRIC_TARGET and GMOND_CONF both not set! Set at least one!" ) |
---|
[353] | 227 | sys.exit( 1 ) |
---|
| 228 | else: |
---|
| 229 | |
---|
[373] | 230 | 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!" ) |
---|
[353] | 231 | |
---|
[212] | 232 | return True |
---|
| 233 | |
---|
[253] | 234 | METRIC_MAX_VAL_LEN = 900 |
---|
| 235 | |
---|
[61] | 236 | class DataProcessor: |
---|
[355] | 237 | |
---|
[68] | 238 | """Class for processing of data""" |
---|
[61] | 239 | |
---|
| 240 | binary = '/usr/bin/gmetric' |
---|
| 241 | |
---|
| 242 | def __init__( self, binary=None ): |
---|
[355] | 243 | |
---|
[68] | 244 | """Remember alternate binary location if supplied""" |
---|
[61] | 245 | |
---|
| 246 | if binary: |
---|
| 247 | self.binary = binary |
---|
| 248 | |
---|
[80] | 249 | # Timeout for XML |
---|
| 250 | # |
---|
| 251 | # From ganglia's documentation: |
---|
| 252 | # |
---|
| 253 | # 'A metric will be deleted DMAX seconds after it is received, and |
---|
| 254 | # DMAX=0 means eternal life.' |
---|
[61] | 255 | |
---|
[256] | 256 | self.dmax = str( int( int( BATCH_POLL_INTERVAL ) * 2 ) ) |
---|
[80] | 257 | |
---|
[353] | 258 | if GMOND_CONF: |
---|
[354] | 259 | |
---|
[353] | 260 | try: |
---|
| 261 | gmond_file = GMOND_CONF |
---|
[68] | 262 | |
---|
[353] | 263 | except NameError: |
---|
| 264 | gmond_file = '/etc/gmond.conf' |
---|
[68] | 265 | |
---|
[353] | 266 | if not os.path.exists( gmond_file ): |
---|
[373] | 267 | debug_msg( 0, 'FATAL ERROR: ' + gmond_file + ' does not exist' ) |
---|
[353] | 268 | sys.exit( 1 ) |
---|
[68] | 269 | |
---|
[353] | 270 | incompatible = self.checkGmetricVersion() |
---|
[61] | 271 | |
---|
[353] | 272 | if incompatible: |
---|
[355] | 273 | |
---|
| 274 | debug_msg( 0, 'Gmetric version not compatible, please upgrade to at least 3.0.1' ) |
---|
[353] | 275 | sys.exit( 1 ) |
---|
[65] | 276 | |
---|
| 277 | def checkGmetricVersion( self ): |
---|
[355] | 278 | |
---|
[68] | 279 | """ |
---|
| 280 | Check version of gmetric is at least 3.0.1 |
---|
| 281 | for the syntax we use |
---|
| 282 | """ |
---|
[65] | 283 | |
---|
[255] | 284 | global METRIC_MAX_VAL_LEN |
---|
| 285 | |
---|
[341] | 286 | incompatible = 0 |
---|
| 287 | |
---|
[355] | 288 | gfp = os.popen( self.binary + ' --version' ) |
---|
| 289 | lines = gfp.readlines() |
---|
[65] | 290 | |
---|
[355] | 291 | gfp.close() |
---|
| 292 | |
---|
| 293 | for line in lines: |
---|
| 294 | |
---|
[65] | 295 | line = line.split( ' ' ) |
---|
| 296 | |
---|
[355] | 297 | if len( line ) == 2 and str( line ).find( 'gmetric' ) != -1: |
---|
[65] | 298 | |
---|
[355] | 299 | gmetric_version = line[1].split( '\n' )[0] |
---|
[65] | 300 | |
---|
[355] | 301 | version_major = int( gmetric_version.split( '.' )[0] ) |
---|
| 302 | version_minor = int( gmetric_version.split( '.' )[1] ) |
---|
| 303 | version_patch = int( gmetric_version.split( '.' )[2] ) |
---|
[65] | 304 | |
---|
[355] | 305 | incompatible = 0 |
---|
[65] | 306 | |
---|
| 307 | if version_major < 3: |
---|
| 308 | |
---|
| 309 | incompatible = 1 |
---|
| 310 | |
---|
| 311 | elif version_major == 3: |
---|
| 312 | |
---|
| 313 | if version_minor == 0: |
---|
| 314 | |
---|
| 315 | if version_patch < 1: |
---|
| 316 | |
---|
[91] | 317 | incompatible = 1 |
---|
[65] | 318 | |
---|
[255] | 319 | if version_patch < 3: |
---|
| 320 | |
---|
| 321 | METRIC_MAX_VAL_LEN = 900 |
---|
| 322 | |
---|
| 323 | elif version_patch >= 3: |
---|
| 324 | |
---|
| 325 | METRIC_MAX_VAL_LEN = 1400 |
---|
| 326 | |
---|
[65] | 327 | return incompatible |
---|
| 328 | |
---|
[409] | 329 | def multicastGmetric( self, metricname, metricval, valtype='string', units='' ): |
---|
[355] | 330 | |
---|
[68] | 331 | """Call gmetric binary and multicast""" |
---|
[65] | 332 | |
---|
| 333 | cmd = self.binary |
---|
| 334 | |
---|
[353] | 335 | if GMETRIC_TARGET: |
---|
[61] | 336 | |
---|
[353] | 337 | GMETRIC_TARGET_HOST = GMETRIC_TARGET.split( ':' )[0] |
---|
| 338 | GMETRIC_TARGET_PORT = GMETRIC_TARGET.split( ':' )[1] |
---|
| 339 | |
---|
| 340 | metric_debug = "[gmetric] name: %s - val: %s - dmax: %s" %( str( metricname ), str( metricval ), str( self.dmax ) ) |
---|
| 341 | |
---|
| 342 | debug_msg( 10, printTime() + ' ' + metric_debug) |
---|
| 343 | |
---|
| 344 | gm = Gmetric( GMETRIC_TARGET_HOST, GMETRIC_TARGET_PORT ) |
---|
| 345 | |
---|
[425] | 346 | gm.send( str( metricname ), str( metricval ), str( self.dmax ), valtype, units ) |
---|
[353] | 347 | |
---|
| 348 | else: |
---|
| 349 | try: |
---|
| 350 | cmd = cmd + ' -c' + GMOND_CONF |
---|
| 351 | |
---|
| 352 | except NameError: |
---|
| 353 | |
---|
| 354 | debug_msg( 10, 'Assuming /etc/gmond.conf for gmetric cmd (ommitting)' ) |
---|
| 355 | |
---|
| 356 | cmd = cmd + ' -n' + str( metricname )+ ' -v"' + str( metricval )+ '" -t' + str( valtype ) + ' -d' + str( self.dmax ) |
---|
| 357 | |
---|
[409] | 358 | if len( units ) > 0: |
---|
| 359 | |
---|
| 360 | cmd = cmd + ' -u"' + units + '"' |
---|
| 361 | |
---|
[353] | 362 | debug_msg( 10, printTime() + ' ' + cmd ) |
---|
| 363 | |
---|
| 364 | os.system( cmd ) |
---|
| 365 | |
---|
[318] | 366 | class DataGatherer: |
---|
[23] | 367 | |
---|
[318] | 368 | """Skeleton class for batch system DataGatherer""" |
---|
[256] | 369 | |
---|
[318] | 370 | def printJobs( self, jobs ): |
---|
[355] | 371 | |
---|
[318] | 372 | """Print a jobinfo overview""" |
---|
| 373 | |
---|
| 374 | for name, attrs in self.jobs.items(): |
---|
| 375 | |
---|
| 376 | print 'job %s' %(name) |
---|
| 377 | |
---|
| 378 | for name, val in attrs.items(): |
---|
| 379 | |
---|
| 380 | print '\t%s = %s' %( name, val ) |
---|
| 381 | |
---|
| 382 | def printJob( self, jobs, job_id ): |
---|
[355] | 383 | |
---|
[318] | 384 | """Print job with job_id from jobs""" |
---|
| 385 | |
---|
| 386 | print 'job %s' %(job_id) |
---|
| 387 | |
---|
| 388 | for name, val in jobs[ job_id ].items(): |
---|
| 389 | |
---|
| 390 | print '\t%s = %s' %( name, val ) |
---|
| 391 | |
---|
[256] | 392 | def daemon( self ): |
---|
[355] | 393 | |
---|
[318] | 394 | """Run as daemon forever""" |
---|
[256] | 395 | |
---|
[318] | 396 | # Fork the first child |
---|
| 397 | # |
---|
| 398 | pid = os.fork() |
---|
| 399 | if pid > 0: |
---|
| 400 | sys.exit(0) # end parent |
---|
[256] | 401 | |
---|
[318] | 402 | # creates a session and sets the process group ID |
---|
| 403 | # |
---|
| 404 | os.setsid() |
---|
| 405 | |
---|
| 406 | # Fork the second child |
---|
| 407 | # |
---|
| 408 | pid = os.fork() |
---|
| 409 | if pid > 0: |
---|
| 410 | sys.exit(0) # end parent |
---|
| 411 | |
---|
| 412 | write_pidfile() |
---|
| 413 | |
---|
| 414 | # Go to the root directory and set the umask |
---|
| 415 | # |
---|
| 416 | os.chdir('/') |
---|
| 417 | os.umask(0) |
---|
| 418 | |
---|
| 419 | sys.stdin.close() |
---|
| 420 | sys.stdout.close() |
---|
| 421 | sys.stderr.close() |
---|
| 422 | |
---|
| 423 | os.open('/dev/null', os.O_RDWR) |
---|
| 424 | os.dup2(0, 1) |
---|
| 425 | os.dup2(0, 2) |
---|
| 426 | |
---|
| 427 | self.run() |
---|
| 428 | |
---|
[256] | 429 | def run( self ): |
---|
[355] | 430 | |
---|
[318] | 431 | """Main thread""" |
---|
[256] | 432 | |
---|
[318] | 433 | while ( 1 ): |
---|
| 434 | |
---|
[348] | 435 | self.getJobData() |
---|
| 436 | self.submitJobData() |
---|
[318] | 437 | time.sleep( BATCH_POLL_INTERVAL ) |
---|
[256] | 438 | |
---|
[318] | 439 | class SgeQstatXMLParser(xml.sax.handler.ContentHandler): |
---|
[256] | 440 | |
---|
[318] | 441 | """Babu Sundaram's experimental SGE qstat XML parser""" |
---|
[256] | 442 | |
---|
[318] | 443 | def __init__(self, qstatinxml): |
---|
| 444 | |
---|
| 445 | self.qstatfile = qstatinxml |
---|
| 446 | self.attribs = {} |
---|
| 447 | self.value = '' |
---|
| 448 | self.jobID = '' |
---|
| 449 | self.currentJobInfo = '' |
---|
| 450 | self.job_list = [] |
---|
| 451 | self.EOFFlag = 0 |
---|
| 452 | self.jobinfoCount = 0 |
---|
| 453 | |
---|
| 454 | |
---|
| 455 | def startElement(self, name, attrs): |
---|
| 456 | |
---|
| 457 | if name == 'job_list': |
---|
| 458 | self.currentJobInfo = 'Status=' + attrs.get('state', None) + ' ' |
---|
| 459 | elif name == 'job_info': |
---|
| 460 | self.job_list = [] |
---|
| 461 | self.jobinfoCount += 1 |
---|
| 462 | |
---|
| 463 | def characters(self, ch): |
---|
| 464 | |
---|
| 465 | self.value = self.value + ch |
---|
| 466 | |
---|
| 467 | def endElement(self, name): |
---|
| 468 | |
---|
| 469 | if len(self.value.strip()) > 0 : |
---|
| 470 | |
---|
| 471 | self.currentJobInfo += name + '=' + self.value.strip() + ' ' |
---|
| 472 | elif name != 'job_list': |
---|
| 473 | |
---|
| 474 | self.currentJobInfo += name + '=Unknown ' |
---|
| 475 | |
---|
| 476 | if name == 'JB_job_number': |
---|
| 477 | |
---|
| 478 | self.jobID = self.value.strip() |
---|
| 479 | self.job_list.append(self.jobID) |
---|
| 480 | |
---|
| 481 | if name == 'job_list': |
---|
| 482 | |
---|
| 483 | if self.attribs.has_key(self.jobID) == False: |
---|
| 484 | self.attribs[self.jobID] = self.currentJobInfo |
---|
| 485 | elif self.attribs.has_key(self.jobID) and self.attribs[self.jobID] != self.currentJobInfo: |
---|
| 486 | self.attribs[self.jobID] = self.currentJobInfo |
---|
| 487 | self.currentJobInfo = '' |
---|
| 488 | self.jobID = '' |
---|
| 489 | |
---|
| 490 | elif name == 'job_info' and self.jobinfoCount == 2: |
---|
| 491 | |
---|
| 492 | deljobs = [] |
---|
| 493 | for id in self.attribs: |
---|
| 494 | try: |
---|
| 495 | self.job_list.index(str(id)) |
---|
| 496 | except ValueError: |
---|
| 497 | deljobs.append(id) |
---|
| 498 | for i in deljobs: |
---|
| 499 | del self.attribs[i] |
---|
| 500 | deljobs = [] |
---|
| 501 | self.jobinfoCount = 0 |
---|
| 502 | |
---|
| 503 | self.value = '' |
---|
| 504 | |
---|
| 505 | class SgeDataGatherer(DataGatherer): |
---|
| 506 | |
---|
[61] | 507 | jobs = { } |
---|
[347] | 508 | SGE_QSTAT_XML_FILE = '/tmp/.jobmonarch.sge.qstat' |
---|
[61] | 509 | |
---|
[318] | 510 | def __init__( self ): |
---|
| 511 | """Setup appropriate variables""" |
---|
| 512 | |
---|
| 513 | self.jobs = { } |
---|
| 514 | self.timeoffset = 0 |
---|
| 515 | self.dp = DataProcessor() |
---|
| 516 | self.initSgeJobInfo() |
---|
| 517 | |
---|
| 518 | def initSgeJobInfo( self ): |
---|
| 519 | """This is outside the scope of DRMAA; Get the current jobs in SGE""" |
---|
| 520 | """This is a hack because we cant get info about jobs beyond""" |
---|
| 521 | """those in the current DRMAA session""" |
---|
| 522 | |
---|
[347] | 523 | self.qstatparser = SgeQstatXMLParser( self.SGE_QSTAT_XML_FILE ) |
---|
[318] | 524 | |
---|
| 525 | # Obtain the qstat information from SGE in XML format |
---|
| 526 | # This would change to DRMAA-specific calls from 6.0u9 |
---|
| 527 | |
---|
| 528 | def getJobData(self): |
---|
| 529 | """Gather all data on current jobs in SGE""" |
---|
| 530 | |
---|
| 531 | # Get the information about the current jobs in the SGE queue |
---|
| 532 | info = os.popen("qstat -ext -xml").readlines() |
---|
[347] | 533 | f = open(self.SGE_QSTAT_XML_FILE,'w') |
---|
[318] | 534 | for lines in info: |
---|
| 535 | f.write(lines) |
---|
| 536 | f.close() |
---|
| 537 | |
---|
| 538 | # Parse the input |
---|
| 539 | f = open(self.qstatparser.qstatfile, 'r') |
---|
| 540 | xml.sax.parse(f, self.qstatparser) |
---|
| 541 | f.close() |
---|
| 542 | |
---|
| 543 | self.cur_time = time.time() |
---|
| 544 | |
---|
| 545 | return self.qstatparser.attribs |
---|
| 546 | |
---|
| 547 | def submitJobData(self): |
---|
| 548 | """Submit job info list""" |
---|
| 549 | |
---|
| 550 | self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) ) |
---|
| 551 | # Now let's spread the knowledge |
---|
| 552 | # |
---|
| 553 | metric_increment = 0 |
---|
| 554 | for jobid, jobattrs in self.qstatparser.attribs.items(): |
---|
| 555 | |
---|
| 556 | self.dp.multicastGmetric( 'MONARCH-JOB-' + jobid + '-' + str(metric_increment), jobattrs) |
---|
| 557 | |
---|
[355] | 558 | class PbsDataGatherer( DataGatherer ): |
---|
[318] | 559 | |
---|
| 560 | """This is the DataGatherer for PBS and Torque""" |
---|
| 561 | |
---|
[256] | 562 | global PBSQuery |
---|
| 563 | |
---|
[23] | 564 | def __init__( self ): |
---|
[354] | 565 | |
---|
[68] | 566 | """Setup appropriate variables""" |
---|
[23] | 567 | |
---|
[354] | 568 | self.jobs = { } |
---|
| 569 | self.timeoffset = 0 |
---|
| 570 | self.dp = DataProcessor() |
---|
| 571 | |
---|
[91] | 572 | self.initPbsQuery() |
---|
[23] | 573 | |
---|
[91] | 574 | def initPbsQuery( self ): |
---|
| 575 | |
---|
[354] | 576 | self.pq = None |
---|
| 577 | |
---|
[256] | 578 | if( BATCH_SERVER ): |
---|
[354] | 579 | |
---|
| 580 | self.pq = PBSQuery( BATCH_SERVER ) |
---|
[174] | 581 | else: |
---|
[354] | 582 | self.pq = PBSQuery() |
---|
[91] | 583 | |
---|
[26] | 584 | def getAttr( self, attrs, name ): |
---|
[354] | 585 | |
---|
[68] | 586 | """Return certain attribute from dictionary, if exists""" |
---|
[26] | 587 | |
---|
| 588 | if attrs.has_key( name ): |
---|
[354] | 589 | |
---|
| 590 | return attrs[ name ] |
---|
[26] | 591 | else: |
---|
| 592 | return '' |
---|
| 593 | |
---|
| 594 | def jobDataChanged( self, jobs, job_id, attrs ): |
---|
[354] | 595 | |
---|
[68] | 596 | """Check if job with attrs and job_id in jobs has changed""" |
---|
[26] | 597 | |
---|
| 598 | if jobs.has_key( job_id ): |
---|
[354] | 599 | |
---|
[26] | 600 | oldData = jobs[ job_id ] |
---|
| 601 | else: |
---|
| 602 | return 1 |
---|
| 603 | |
---|
| 604 | for name, val in attrs.items(): |
---|
| 605 | |
---|
| 606 | if oldData.has_key( name ): |
---|
| 607 | |
---|
| 608 | if oldData[ name ] != attrs[ name ]: |
---|
| 609 | |
---|
| 610 | return 1 |
---|
| 611 | |
---|
| 612 | else: |
---|
| 613 | return 1 |
---|
| 614 | |
---|
| 615 | return 0 |
---|
| 616 | |
---|
[348] | 617 | def getJobData( self ): |
---|
[354] | 618 | |
---|
[68] | 619 | """Gather all data on current jobs in Torque""" |
---|
[26] | 620 | |
---|
[354] | 621 | joblist = {} |
---|
[359] | 622 | self.cur_time = 0 |
---|
[349] | 623 | |
---|
[359] | 624 | try: |
---|
| 625 | joblist = self.pq.getjobs() |
---|
| 626 | self.cur_time = time.time() |
---|
[354] | 627 | |
---|
[359] | 628 | except PBSError, detail: |
---|
[354] | 629 | |
---|
[359] | 630 | debug_msg( 10, "Caught PBS unavailable, skipping until next polling interval: " + str( detail ) ) |
---|
| 631 | return None |
---|
[354] | 632 | |
---|
| 633 | jobs_processed = [ ] |
---|
[26] | 634 | |
---|
| 635 | for name, attrs in joblist.items(): |
---|
| 636 | |
---|
[354] | 637 | job_id = name.split( '.' )[0] |
---|
[26] | 638 | |
---|
| 639 | jobs_processed.append( job_id ) |
---|
[61] | 640 | |
---|
[354] | 641 | name = self.getAttr( attrs, 'Job_Name' ) |
---|
| 642 | queue = self.getAttr( attrs, 'queue' ) |
---|
[317] | 643 | |
---|
| 644 | if QUEUE: |
---|
| 645 | |
---|
| 646 | if QUEUE != queue: |
---|
| 647 | |
---|
| 648 | continue |
---|
| 649 | |
---|
[354] | 650 | owner = self.getAttr( attrs, 'Job_Owner' ).split( '@' )[0] |
---|
| 651 | requested_time = self.getAttr( attrs, 'Resource_List.walltime' ) |
---|
| 652 | requested_memory = self.getAttr( attrs, 'Resource_List.mem' ) |
---|
[95] | 653 | |
---|
[354] | 654 | mynoderequest = self.getAttr( attrs, 'Resource_List.nodes' ) |
---|
[95] | 655 | |
---|
[354] | 656 | ppn = '' |
---|
[281] | 657 | |
---|
[26] | 658 | if mynoderequest.find( ':' ) != -1 and mynoderequest.find( 'ppn' ) != -1: |
---|
[95] | 659 | |
---|
[354] | 660 | mynoderequest_fields = mynoderequest.split( ':' ) |
---|
[281] | 661 | |
---|
| 662 | for mynoderequest_field in mynoderequest_fields: |
---|
| 663 | |
---|
| 664 | if mynoderequest_field.find( 'ppn' ) != -1: |
---|
| 665 | |
---|
[354] | 666 | ppn = mynoderequest_field.split( 'ppn=' )[1] |
---|
[281] | 667 | |
---|
[354] | 668 | status = self.getAttr( attrs, 'job_state' ) |
---|
[25] | 669 | |
---|
[354] | 670 | queued_timestamp = self.getAttr( attrs, 'ctime' ) |
---|
[243] | 671 | |
---|
[95] | 672 | if status == 'R': |
---|
[133] | 673 | |
---|
[354] | 674 | start_timestamp = self.getAttr( attrs, 'mtime' ) |
---|
| 675 | nodes = self.getAttr( attrs, 'exec_host' ).split( '+' ) |
---|
[133] | 676 | |
---|
[354] | 677 | nodeslist = [ ] |
---|
| 678 | |
---|
[133] | 679 | for node in nodes: |
---|
| 680 | |
---|
[354] | 681 | host = node.split( '/' )[0] |
---|
| 682 | |
---|
[133] | 683 | if nodeslist.count( host ) == 0: |
---|
[215] | 684 | |
---|
| 685 | for translate_pattern in BATCH_HOST_TRANSLATE: |
---|
| 686 | |
---|
[220] | 687 | if translate_pattern.find( '/' ) != -1: |
---|
[215] | 688 | |
---|
[354] | 689 | translate_orig = translate_pattern.split( '/' )[1] |
---|
| 690 | translate_new = translate_pattern.split( '/' )[2] |
---|
[220] | 691 | |
---|
[354] | 692 | host = re.sub( translate_orig, translate_new, host ) |
---|
[216] | 693 | |
---|
[217] | 694 | if not host in nodeslist: |
---|
[216] | 695 | |
---|
| 696 | nodeslist.append( host ) |
---|
[133] | 697 | |
---|
[185] | 698 | if DETECT_TIME_DIFFS: |
---|
| 699 | |
---|
| 700 | # If a job start if later than our current date, |
---|
| 701 | # that must mean the Torque server's time is later |
---|
| 702 | # than our local time. |
---|
| 703 | |
---|
[354] | 704 | if int( start_timestamp ) > int( int( self.cur_time ) + int( self.timeoffset ) ): |
---|
[185] | 705 | |
---|
[354] | 706 | self.timeoffset = int( int(start_timestamp) - int(self.cur_time) ) |
---|
[185] | 707 | |
---|
[133] | 708 | elif status == 'Q': |
---|
[95] | 709 | |
---|
[354] | 710 | start_timestamp = '' |
---|
| 711 | count_mynodes = 0 |
---|
| 712 | numeric_node = 1 |
---|
| 713 | |
---|
[133] | 714 | for node in mynoderequest.split( '+' ): |
---|
[67] | 715 | |
---|
[354] | 716 | nodepart = node.split( ':' )[0] |
---|
[67] | 717 | |
---|
[133] | 718 | for letter in nodepart: |
---|
[67] | 719 | |
---|
[133] | 720 | if letter not in string.digits: |
---|
| 721 | |
---|
[354] | 722 | numeric_node = 0 |
---|
[133] | 723 | |
---|
| 724 | if not numeric_node: |
---|
[354] | 725 | |
---|
| 726 | count_mynodes = count_mynodes + 1 |
---|
[133] | 727 | else: |
---|
[327] | 728 | try: |
---|
[354] | 729 | count_mynodes = count_mynodes + int( nodepart ) |
---|
| 730 | |
---|
[327] | 731 | except ValueError, detail: |
---|
[354] | 732 | |
---|
[327] | 733 | debug_msg( 10, str( detail ) ) |
---|
| 734 | debug_msg( 10, "Encountered weird node in Resources_List?!" ) |
---|
| 735 | debug_msg( 10, 'nodepart = ' + str( nodepart ) ) |
---|
| 736 | debug_msg( 10, 'job = ' + str( name ) ) |
---|
| 737 | debug_msg( 10, 'attrs = ' + str( attrs ) ) |
---|
[133] | 738 | |
---|
[354] | 739 | nodeslist = str( count_mynodes ) |
---|
[172] | 740 | else: |
---|
[354] | 741 | start_timestamp = '' |
---|
| 742 | nodeslist = '' |
---|
[133] | 743 | |
---|
[354] | 744 | myAttrs = { } |
---|
[26] | 745 | |
---|
[354] | 746 | myAttrs[ 'name' ] = str( name ) |
---|
| 747 | myAttrs[ 'queue' ] = str( queue ) |
---|
| 748 | myAttrs[ 'owner' ] = str( owner ) |
---|
| 749 | myAttrs[ 'requested_time' ] = str( requested_time ) |
---|
| 750 | myAttrs[ 'requested_memory' ] = str( requested_memory ) |
---|
| 751 | myAttrs[ 'ppn' ] = str( ppn ) |
---|
| 752 | myAttrs[ 'status' ] = str( status ) |
---|
| 753 | myAttrs[ 'start_timestamp' ] = str( start_timestamp ) |
---|
| 754 | myAttrs[ 'queued_timestamp' ] = str( queued_timestamp ) |
---|
| 755 | myAttrs[ 'reported' ] = str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) |
---|
| 756 | myAttrs[ 'nodes' ] = nodeslist |
---|
| 757 | myAttrs[ 'domain' ] = string.join( socket.getfqdn().split( '.' )[1:], '.' ) |
---|
| 758 | myAttrs[ 'poll_interval' ] = str( BATCH_POLL_INTERVAL ) |
---|
| 759 | |
---|
[348] | 760 | if self.jobDataChanged( self.jobs, job_id, myAttrs ) and myAttrs['status'] in [ 'R', 'Q' ]: |
---|
[61] | 761 | |
---|
[354] | 762 | self.jobs[ job_id ] = myAttrs |
---|
[26] | 763 | |
---|
[348] | 764 | for id, attrs in self.jobs.items(): |
---|
[76] | 765 | |
---|
| 766 | if id not in jobs_processed: |
---|
| 767 | |
---|
| 768 | # This one isn't there anymore; toedeledoki! |
---|
| 769 | # |
---|
[348] | 770 | del self.jobs[ id ] |
---|
[76] | 771 | |
---|
[348] | 772 | def submitJobData( self ): |
---|
[354] | 773 | |
---|
[65] | 774 | """Submit job info list""" |
---|
| 775 | |
---|
[219] | 776 | self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) ) |
---|
[69] | 777 | |
---|
[409] | 778 | running_jobs = 0 |
---|
| 779 | queued_jobs = 0 |
---|
| 780 | |
---|
| 781 | for jobid, jobattrs in self.jobs.items(): |
---|
| 782 | |
---|
| 783 | if jobattrs[ 'status' ] == 'Q': |
---|
| 784 | |
---|
| 785 | queued_jobs += 1 |
---|
| 786 | |
---|
| 787 | elif jobattrs[ 'status' ] == 'R': |
---|
| 788 | |
---|
| 789 | running_jobs += 1 |
---|
| 790 | |
---|
| 791 | self.dp.multicastGmetric( 'MONARCH-RJ', str( running_jobs ), 'uint32', 'jobs' ) |
---|
| 792 | self.dp.multicastGmetric( 'MONARCH-QJ', str( queued_jobs ), 'uint32', 'jobs' ) |
---|
| 793 | |
---|
[61] | 794 | # Now let's spread the knowledge |
---|
| 795 | # |
---|
[348] | 796 | for jobid, jobattrs in self.jobs.items(): |
---|
[61] | 797 | |
---|
[354] | 798 | gmetric_val = self.compileGmetricVal( jobid, jobattrs ) |
---|
| 799 | metric_increment = 0 |
---|
[61] | 800 | |
---|
[354] | 801 | for val in gmetric_val: |
---|
[253] | 802 | |
---|
| 803 | self.dp.multicastGmetric( 'MONARCH-JOB-' + jobid + '-' + str(metric_increment), val ) |
---|
[61] | 804 | |
---|
[354] | 805 | metric_increment = metric_increment + 1 |
---|
| 806 | |
---|
[253] | 807 | def compileGmetricVal( self, jobid, jobattrs ): |
---|
[354] | 808 | |
---|
[253] | 809 | """Create a val string for gmetric of jobinfo""" |
---|
[67] | 810 | |
---|
[354] | 811 | gval_lists = [ ] |
---|
| 812 | mystr = None |
---|
| 813 | val_list = { } |
---|
[67] | 814 | |
---|
[253] | 815 | for val_name, val_value in jobattrs.items(): |
---|
[61] | 816 | |
---|
[253] | 817 | val_list_names_len = len( string.join( val_list.keys() ) ) + len(val_list.keys()) |
---|
| 818 | val_list_vals_len = len( string.join( val_list.values() ) ) + len(val_list.values()) |
---|
[95] | 819 | |
---|
[254] | 820 | if val_name == 'nodes' and jobattrs['status'] == 'R': |
---|
[95] | 821 | |
---|
[253] | 822 | node_str = None |
---|
[95] | 823 | |
---|
[253] | 824 | for node in val_value: |
---|
[65] | 825 | |
---|
[253] | 826 | if node_str: |
---|
[354] | 827 | |
---|
[253] | 828 | node_str = node_str + ';' + node |
---|
| 829 | else: |
---|
| 830 | node_str = node |
---|
[65] | 831 | |
---|
[253] | 832 | if (val_list_names_len + len(val_name) ) + (val_list_vals_len + len(node_str) ) > METRIC_MAX_VAL_LEN: |
---|
[65] | 833 | |
---|
[354] | 834 | val_list[ val_name ] = node_str |
---|
| 835 | |
---|
[253] | 836 | gval_lists.append( val_list ) |
---|
[65] | 837 | |
---|
[354] | 838 | val_list = { } |
---|
| 839 | node_str = None |
---|
| 840 | |
---|
| 841 | val_list[ val_name ] = node_str |
---|
| 842 | |
---|
[253] | 843 | gval_lists.append( val_list ) |
---|
[65] | 844 | |
---|
[354] | 845 | val_list = { } |
---|
| 846 | |
---|
[254] | 847 | elif val_value != '': |
---|
| 848 | |
---|
| 849 | if (val_list_names_len + len(val_name) ) + (val_list_vals_len + len(str(val_value)) ) > METRIC_MAX_VAL_LEN: |
---|
| 850 | |
---|
| 851 | gval_lists.append( val_list ) |
---|
| 852 | |
---|
[354] | 853 | val_list = { } |
---|
[254] | 854 | |
---|
[354] | 855 | val_list[ val_name ] = val_value |
---|
| 856 | |
---|
| 857 | if len( val_list ) > 0: |
---|
| 858 | |
---|
[254] | 859 | gval_lists.append( val_list ) |
---|
| 860 | |
---|
[354] | 861 | str_list = [ ] |
---|
[65] | 862 | |
---|
[253] | 863 | for val_list in gval_lists: |
---|
[65] | 864 | |
---|
[354] | 865 | my_val_str = None |
---|
[65] | 866 | |
---|
[253] | 867 | for val_name, val_value in val_list.items(): |
---|
[65] | 868 | |
---|
[253] | 869 | if my_val_str: |
---|
[65] | 870 | |
---|
[253] | 871 | my_val_str = my_val_str + ' ' + val_name + '=' + val_value |
---|
| 872 | else: |
---|
| 873 | my_val_str = val_name + '=' + val_value |
---|
| 874 | |
---|
| 875 | str_list.append( my_val_str ) |
---|
| 876 | |
---|
| 877 | return str_list |
---|
| 878 | |
---|
[363] | 879 | # |
---|
| 880 | # Gmetric by Nick Galbreath - nickg(a.t)modp(d.o.t)com |
---|
| 881 | # Version 1.0 - 21-April2-2007 |
---|
| 882 | # http://code.google.com/p/embeddedgmetric/ |
---|
| 883 | # |
---|
| 884 | # Modified by: Ramon Bastiaans |
---|
| 885 | # For the Job Monarch Project, see: https://subtrac.sara.nl/oss/jobmonarch/ |
---|
| 886 | # |
---|
| 887 | # added: DEFAULT_TYPE for Gmetric's |
---|
| 888 | # added: checkHostProtocol to determine if target is multicast or not |
---|
| 889 | # changed: allow default for Gmetric constructor |
---|
| 890 | # changed: allow defaults for all send() values except dmax |
---|
| 891 | # |
---|
| 892 | |
---|
[362] | 893 | GMETRIC_DEFAULT_TYPE = 'string' |
---|
| 894 | GMETRIC_DEFAULT_HOST = '127.0.0.1' |
---|
| 895 | GMETRIC_DEFAULT_PORT = '8649' |
---|
| 896 | |
---|
| 897 | class Gmetric: |
---|
| 898 | |
---|
| 899 | global GMETRIC_DEFAULT_HOST, GMETRIC_DEFAULT_PORT |
---|
| 900 | |
---|
| 901 | slope = { 'zero' : 0, 'positive' : 1, 'negative' : 2, 'both' : 3, 'unspecified' : 4 } |
---|
| 902 | type = ( '', 'string', 'uint16', 'int16', 'uint32', 'int32', 'float', 'double', 'timestamp' ) |
---|
| 903 | protocol = ( 'udp', 'multicast' ) |
---|
| 904 | |
---|
| 905 | def __init__( self, host=GMETRIC_DEFAULT_HOST, port=GMETRIC_DEFAULT_PORT ): |
---|
| 906 | |
---|
| 907 | global GMETRIC_DEFAULT_TYPE |
---|
| 908 | |
---|
| 909 | self.prot = self.checkHostProtocol( host ) |
---|
| 910 | self.msg = xdrlib.Packer() |
---|
| 911 | self.socket = socket.socket( socket.AF_INET, socket.SOCK_DGRAM ) |
---|
| 912 | |
---|
| 913 | if self.prot not in self.protocol: |
---|
| 914 | |
---|
| 915 | raise ValueError( "Protocol must be one of: " + str( self.protocol ) ) |
---|
| 916 | |
---|
| 917 | if self.prot == 'multicast': |
---|
| 918 | |
---|
| 919 | self.socket.setsockopt( socket.IPPROTO_IP, socket.IP_MULTICAST_TTL, 20 ) |
---|
| 920 | |
---|
| 921 | self.hostport = ( host, int( port ) ) |
---|
| 922 | self.type = GMETRIC_DEFAULT_TYPE |
---|
| 923 | self.unitstr = '' |
---|
| 924 | self.slopestr = 'both' |
---|
| 925 | self.tmax = 60 |
---|
| 926 | |
---|
| 927 | def checkHostProtocol( self, ip ): |
---|
| 928 | |
---|
| 929 | MULTICAST_ADDRESS_MIN = ( "224", "0", "0", "0" ) |
---|
| 930 | MULTICAST_ADDRESS_MAX = ( "239", "255", "255", "255" ) |
---|
| 931 | |
---|
| 932 | ip_fields = ip.split( '.' ) |
---|
| 933 | |
---|
| 934 | if ip_fields >= MULTICAST_ADDRESS_MIN and ip_fields <= MULTICAST_ADDRESS_MAX: |
---|
| 935 | |
---|
| 936 | return 'multicast' |
---|
| 937 | else: |
---|
| 938 | return 'udp' |
---|
| 939 | |
---|
[424] | 940 | def send( self, name, value, dmax, type = '', units = '' ): |
---|
[362] | 941 | |
---|
[409] | 942 | if len( units ) == 0: |
---|
[421] | 943 | units = self.unitstr |
---|
[424] | 944 | if len( type ) == 0: |
---|
| 945 | typestr = self.type |
---|
[362] | 946 | |
---|
[424] | 947 | msg = self.makexdr( name, value, typestr, units, self.slopestr, self.tmax, dmax ) |
---|
[409] | 948 | |
---|
[362] | 949 | return self.socket.sendto( msg, self.hostport ) |
---|
| 950 | |
---|
| 951 | def makexdr( self, name, value, typestr, unitstr, slopestr, tmax, dmax ): |
---|
| 952 | |
---|
| 953 | if slopestr not in self.slope: |
---|
| 954 | |
---|
| 955 | raise ValueError( "Slope must be one of: " + str( self.slope.keys() ) ) |
---|
| 956 | |
---|
| 957 | if typestr not in self.type: |
---|
| 958 | |
---|
| 959 | raise ValueError( "Type must be one of: " + str( self.type ) ) |
---|
| 960 | |
---|
| 961 | if len( name ) == 0: |
---|
| 962 | |
---|
| 963 | raise ValueError( "Name must be non-empty" ) |
---|
| 964 | |
---|
| 965 | self.msg.reset() |
---|
| 966 | self.msg.pack_int( 0 ) |
---|
| 967 | self.msg.pack_string( typestr ) |
---|
| 968 | self.msg.pack_string( name ) |
---|
| 969 | self.msg.pack_string( str( value ) ) |
---|
| 970 | self.msg.pack_string( unitstr ) |
---|
| 971 | self.msg.pack_int( self.slope[ slopestr ] ) |
---|
| 972 | self.msg.pack_uint( int( tmax ) ) |
---|
| 973 | self.msg.pack_uint( int( dmax ) ) |
---|
| 974 | |
---|
| 975 | return self.msg.get_buffer() |
---|
| 976 | |
---|
[26] | 977 | def printTime( ): |
---|
[354] | 978 | |
---|
[65] | 979 | """Print current time/date in human readable format for log/debug""" |
---|
[26] | 980 | |
---|
| 981 | return time.strftime("%a, %d %b %Y %H:%M:%S") |
---|
| 982 | |
---|
| 983 | def debug_msg( level, msg ): |
---|
[354] | 984 | |
---|
[65] | 985 | """Print msg if at or above current debug level""" |
---|
[26] | 986 | |
---|
[377] | 987 | global DAEMONIZE, DEBUG_LEVEL, SYSLOG_LEVEL |
---|
| 988 | |
---|
[373] | 989 | if (not DAEMONIZE and DEBUG_LEVEL >= level): |
---|
| 990 | sys.stderr.write( msg + '\n' ) |
---|
[26] | 991 | |
---|
[373] | 992 | if (DAEMONIZE and USE_SYSLOG and SYSLOG_LEVEL >= level): |
---|
| 993 | syslog.syslog( msg ) |
---|
| 994 | |
---|
[307] | 995 | def write_pidfile(): |
---|
| 996 | |
---|
| 997 | # Write pidfile if PIDFILE exists |
---|
| 998 | if PIDFILE: |
---|
| 999 | |
---|
[354] | 1000 | pid = os.getpid() |
---|
| 1001 | |
---|
| 1002 | pidfile = open(PIDFILE, 'w') |
---|
| 1003 | |
---|
| 1004 | pidfile.write( str( pid ) ) |
---|
[307] | 1005 | pidfile.close() |
---|
| 1006 | |
---|
[23] | 1007 | def main(): |
---|
[354] | 1008 | |
---|
[65] | 1009 | """Application start""" |
---|
[23] | 1010 | |
---|
[352] | 1011 | global PBSQuery, PBSError |
---|
[373] | 1012 | global SYSLOG_FACILITY, USE_SYSLOG, BATCH_API, DAEMONIZE |
---|
[256] | 1013 | |
---|
[212] | 1014 | if not processArgs( sys.argv[1:] ): |
---|
[354] | 1015 | |
---|
[212] | 1016 | sys.exit( 1 ) |
---|
| 1017 | |
---|
[256] | 1018 | if BATCH_API == 'pbs': |
---|
| 1019 | |
---|
| 1020 | try: |
---|
[282] | 1021 | from PBSQuery import PBSQuery, PBSError |
---|
[256] | 1022 | |
---|
| 1023 | except ImportError: |
---|
| 1024 | |
---|
[373] | 1025 | debug_msg( 0, "FATAL ERROR: BATCH_API set to 'pbs' but python module 'pbs_python' is not installed" ) |
---|
[256] | 1026 | sys.exit( 1 ) |
---|
| 1027 | |
---|
| 1028 | gather = PbsDataGatherer() |
---|
| 1029 | |
---|
| 1030 | elif BATCH_API == 'sge': |
---|
| 1031 | |
---|
[373] | 1032 | debug_msg( 0, "FATAL ERROR: BATCH_API 'sge' implementation is currently broken, check future releases" ) |
---|
[368] | 1033 | |
---|
| 1034 | sys.exit( 1 ) |
---|
| 1035 | |
---|
[347] | 1036 | gather = SgeDataGatherer() |
---|
[256] | 1037 | |
---|
| 1038 | else: |
---|
[373] | 1039 | debug_msg( 0, "FATAL ERROR: unknown BATCH_API '" + BATCH_API + "' is not supported" ) |
---|
[354] | 1040 | |
---|
[256] | 1041 | sys.exit( 1 ) |
---|
| 1042 | |
---|
[373] | 1043 | if( DAEMONIZE and USE_SYSLOG ): |
---|
| 1044 | |
---|
| 1045 | syslog.openlog( 'jobmond', syslog.LOG_NOWAIT, SYSLOG_FACILITY ) |
---|
| 1046 | |
---|
| 1047 | |
---|
[26] | 1048 | if DAEMONIZE: |
---|
[354] | 1049 | |
---|
[26] | 1050 | gather.daemon() |
---|
| 1051 | else: |
---|
| 1052 | gather.run() |
---|
[23] | 1053 | |
---|
[256] | 1054 | # wh00t? someone started me! :) |
---|
[65] | 1055 | # |
---|
[23] | 1056 | if __name__ == '__main__': |
---|
| 1057 | main() |
---|