[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 228 2006-03-10 16:26:43Z bastiaans $ |
---|
[227] | 22 | # |
---|
[23] | 23 | |
---|
[212] | 24 | import sys, getopt, ConfigParser |
---|
[26] | 25 | |
---|
[212] | 26 | def processArgs( args ): |
---|
[26] | 27 | |
---|
[212] | 28 | SHORT_L = 'c:' |
---|
| 29 | LONG_L = 'config=' |
---|
[165] | 30 | |
---|
[212] | 31 | config_filename = None |
---|
[61] | 32 | |
---|
[212] | 33 | try: |
---|
[68] | 34 | |
---|
[212] | 35 | opts, args = getopt.getopt( args, SHORT_L, LONG_L ) |
---|
[185] | 36 | |
---|
[212] | 37 | except getopt.error, detail: |
---|
| 38 | |
---|
| 39 | print detail |
---|
| 40 | sys.exit(1) |
---|
| 41 | |
---|
| 42 | for opt, value in opts: |
---|
| 43 | |
---|
| 44 | if opt in [ '--config', '-c' ]: |
---|
| 45 | |
---|
| 46 | config_filename = value |
---|
| 47 | |
---|
| 48 | if not config_filename: |
---|
| 49 | |
---|
| 50 | config_filename = '/etc/jobmond.conf' |
---|
| 51 | |
---|
| 52 | return loadConfig( config_filename ) |
---|
| 53 | |
---|
| 54 | def loadConfig( filename ): |
---|
| 55 | |
---|
[215] | 56 | def getlist( cfg_string ): |
---|
| 57 | |
---|
| 58 | my_list = [ ] |
---|
| 59 | |
---|
| 60 | for item_txt in cfg_string.split( ',' ): |
---|
| 61 | |
---|
| 62 | sep_char = None |
---|
| 63 | |
---|
| 64 | item_txt = item_txt.strip() |
---|
| 65 | |
---|
| 66 | for s_char in [ "'", '"' ]: |
---|
| 67 | |
---|
| 68 | if item_txt.find( s_char ) != -1: |
---|
| 69 | |
---|
| 70 | if item_txt.count( s_char ) != 2: |
---|
| 71 | |
---|
| 72 | print 'Missing quote: %s' %item_txt |
---|
| 73 | sys.exit( 1 ) |
---|
| 74 | |
---|
| 75 | else: |
---|
| 76 | |
---|
| 77 | sep_char = s_char |
---|
| 78 | break |
---|
| 79 | |
---|
| 80 | if sep_char: |
---|
| 81 | |
---|
| 82 | item_txt = item_txt.split( sep_char )[1] |
---|
| 83 | |
---|
| 84 | my_list.append( item_txt ) |
---|
| 85 | |
---|
| 86 | return my_list |
---|
| 87 | |
---|
[212] | 88 | cfg = ConfigParser.ConfigParser() |
---|
| 89 | |
---|
| 90 | cfg.read( filename ) |
---|
| 91 | |
---|
[215] | 92 | global DEBUG_LEVEL, DAEMONIZE, TORQUE_SERVER, TORQUE_POLL_INTERVAL, GMOND_CONF, DETECT_TIME_DIFFS, BATCH_HOST_TRANSLATE |
---|
[212] | 93 | |
---|
| 94 | DEBUG_LEVEL = cfg.getint( 'DEFAULT', 'DEBUG_LEVEL' ) |
---|
| 95 | |
---|
| 96 | DAEMONIZE = cfg.getboolean( 'DEFAULT', 'DAEMONIZE' ) |
---|
| 97 | |
---|
| 98 | TORQUE_SERVER = cfg.get( 'DEFAULT', 'TORQUE_SERVER' ) |
---|
| 99 | |
---|
| 100 | TORQUE_POLL_INTERVAL = cfg.getint( 'DEFAULT', 'TORQUE_POLL_INTERVAL' ) |
---|
| 101 | |
---|
| 102 | GMOND_CONF = cfg.get( 'DEFAULT', 'GMOND_CONF' ) |
---|
| 103 | |
---|
| 104 | DETECT_TIME_DIFFS = cfg.getboolean( 'DEFAULT', 'DETECT_TIME_DIFFS' ) |
---|
| 105 | |
---|
[215] | 106 | BATCH_HOST_TRANSLATE = getlist( cfg.get( 'DEFAULT', 'BATCH_HOST_TRANSLATE' ) ) |
---|
| 107 | |
---|
[212] | 108 | return True |
---|
| 109 | |
---|
[23] | 110 | from PBSQuery import PBSQuery |
---|
| 111 | |
---|
[215] | 112 | import time, os, socket, string, re |
---|
[212] | 113 | |
---|
[61] | 114 | class DataProcessor: |
---|
[68] | 115 | """Class for processing of data""" |
---|
[61] | 116 | |
---|
| 117 | binary = '/usr/bin/gmetric' |
---|
| 118 | |
---|
| 119 | def __init__( self, binary=None ): |
---|
[68] | 120 | """Remember alternate binary location if supplied""" |
---|
[61] | 121 | |
---|
| 122 | if binary: |
---|
| 123 | self.binary = binary |
---|
| 124 | |
---|
[80] | 125 | # Timeout for XML |
---|
| 126 | # |
---|
| 127 | # From ganglia's documentation: |
---|
| 128 | # |
---|
| 129 | # 'A metric will be deleted DMAX seconds after it is received, and |
---|
| 130 | # DMAX=0 means eternal life.' |
---|
[61] | 131 | |
---|
[194] | 132 | self.dmax = str( int( int( TORQUE_POLL_INTERVAL ) + 2 ) ) |
---|
[80] | 133 | |
---|
[68] | 134 | try: |
---|
| 135 | gmond_file = GMOND_CONF |
---|
| 136 | |
---|
| 137 | except NameError: |
---|
| 138 | gmond_file = '/etc/gmond.conf' |
---|
| 139 | |
---|
| 140 | if not os.path.exists( gmond_file ): |
---|
| 141 | debug_msg( 0, gmond_file + ' does not exist' ) |
---|
| 142 | sys.exit( 1 ) |
---|
| 143 | |
---|
[69] | 144 | incompatible = self.checkGmetricVersion() |
---|
[61] | 145 | |
---|
[65] | 146 | if incompatible: |
---|
| 147 | debug_msg( 0, 'Gmetric version not compatible, pls upgrade to at least 3.0.1' ) |
---|
| 148 | sys.exit( 1 ) |
---|
| 149 | |
---|
| 150 | def checkGmetricVersion( self ): |
---|
[68] | 151 | """ |
---|
| 152 | Check version of gmetric is at least 3.0.1 |
---|
| 153 | for the syntax we use |
---|
| 154 | """ |
---|
[65] | 155 | |
---|
| 156 | for line in os.popen( self.binary + ' --version' ).readlines(): |
---|
| 157 | |
---|
| 158 | line = line.split( ' ' ) |
---|
| 159 | |
---|
[69] | 160 | if len( line ) == 2 and str(line).find( 'gmetric' ) != -1: |
---|
[65] | 161 | |
---|
[69] | 162 | gmetric_version = line[1].split( '\n' )[0] |
---|
[65] | 163 | |
---|
[69] | 164 | version_major = int( gmetric_version.split( '.' )[0] ) |
---|
| 165 | version_minor = int( gmetric_version.split( '.' )[1] ) |
---|
| 166 | version_patch = int( gmetric_version.split( '.' )[2] ) |
---|
[65] | 167 | |
---|
| 168 | incompatible = 0 |
---|
| 169 | |
---|
| 170 | if version_major < 3: |
---|
| 171 | |
---|
| 172 | incompatible = 1 |
---|
| 173 | |
---|
| 174 | elif version_major == 3: |
---|
| 175 | |
---|
| 176 | if version_minor == 0: |
---|
| 177 | |
---|
| 178 | if version_patch < 1: |
---|
| 179 | |
---|
[91] | 180 | incompatible = 1 |
---|
[65] | 181 | |
---|
| 182 | return incompatible |
---|
| 183 | |
---|
[75] | 184 | def multicastGmetric( self, metricname, metricval, valtype='string' ): |
---|
[68] | 185 | """Call gmetric binary and multicast""" |
---|
[65] | 186 | |
---|
| 187 | cmd = self.binary |
---|
| 188 | |
---|
[61] | 189 | try: |
---|
| 190 | cmd = cmd + ' -c' + GMOND_CONF |
---|
| 191 | except NameError: |
---|
[64] | 192 | debug_msg( 10, 'Assuming /etc/gmond.conf for gmetric cmd (ommitting)' ) |
---|
[61] | 193 | |
---|
[168] | 194 | cmd = cmd + ' -n' + str( metricname )+ ' -v"' + str( metricval )+ '" -t' + str( valtype ) + ' -d' + str( self.dmax ) |
---|
[61] | 195 | |
---|
[101] | 196 | debug_msg( 10, printTime() + ' ' + cmd ) |
---|
[69] | 197 | os.system( cmd ) |
---|
[61] | 198 | |
---|
[174] | 199 | class DataGatherer: |
---|
[23] | 200 | |
---|
[61] | 201 | jobs = { } |
---|
| 202 | |
---|
[23] | 203 | def __init__( self ): |
---|
[68] | 204 | """Setup appropriate variables""" |
---|
[23] | 205 | |
---|
[26] | 206 | self.jobs = { } |
---|
[185] | 207 | self.timeoffset = 0 |
---|
[61] | 208 | self.dp = DataProcessor() |
---|
[91] | 209 | self.initPbsQuery() |
---|
[23] | 210 | |
---|
[91] | 211 | def initPbsQuery( self ): |
---|
| 212 | |
---|
| 213 | self.pq = None |
---|
[165] | 214 | if( TORQUE_SERVER ): |
---|
| 215 | self.pq = PBSQuery( TORQUE_SERVER ) |
---|
[174] | 216 | else: |
---|
[165] | 217 | self.pq = PBSQuery() |
---|
[91] | 218 | |
---|
[26] | 219 | def getAttr( self, attrs, name ): |
---|
[68] | 220 | """Return certain attribute from dictionary, if exists""" |
---|
[26] | 221 | |
---|
| 222 | if attrs.has_key( name ): |
---|
| 223 | return attrs[name] |
---|
| 224 | else: |
---|
| 225 | return '' |
---|
| 226 | |
---|
| 227 | def jobDataChanged( self, jobs, job_id, attrs ): |
---|
[68] | 228 | """Check if job with attrs and job_id in jobs has changed""" |
---|
[26] | 229 | |
---|
| 230 | if jobs.has_key( job_id ): |
---|
| 231 | oldData = jobs[ job_id ] |
---|
| 232 | else: |
---|
| 233 | return 1 |
---|
| 234 | |
---|
| 235 | for name, val in attrs.items(): |
---|
| 236 | |
---|
| 237 | if oldData.has_key( name ): |
---|
| 238 | |
---|
| 239 | if oldData[ name ] != attrs[ name ]: |
---|
| 240 | |
---|
| 241 | return 1 |
---|
| 242 | |
---|
| 243 | else: |
---|
| 244 | return 1 |
---|
| 245 | |
---|
| 246 | return 0 |
---|
| 247 | |
---|
[65] | 248 | def getJobData( self, known_jobs ): |
---|
[68] | 249 | """Gather all data on current jobs in Torque""" |
---|
[26] | 250 | |
---|
[65] | 251 | if len( known_jobs ) > 0: |
---|
| 252 | jobs = known_jobs |
---|
| 253 | else: |
---|
| 254 | jobs = { } |
---|
[26] | 255 | |
---|
[101] | 256 | #self.initPbsQuery() |
---|
[125] | 257 | |
---|
| 258 | #print self.pq.getnodes() |
---|
| 259 | |
---|
[26] | 260 | joblist = self.pq.getjobs() |
---|
| 261 | |
---|
[69] | 262 | self.cur_time = time.time() |
---|
[68] | 263 | |
---|
[26] | 264 | jobs_processed = [ ] |
---|
| 265 | |
---|
[125] | 266 | #self.printJobs( joblist ) |
---|
| 267 | |
---|
[26] | 268 | for name, attrs in joblist.items(): |
---|
| 269 | |
---|
| 270 | job_id = name.split( '.' )[0] |
---|
| 271 | |
---|
| 272 | jobs_processed.append( job_id ) |
---|
[61] | 273 | |
---|
[26] | 274 | name = self.getAttr( attrs, 'Job_Name' ) |
---|
| 275 | queue = self.getAttr( attrs, 'queue' ) |
---|
| 276 | owner = self.getAttr( attrs, 'Job_Owner' ).split( '@' )[0] |
---|
| 277 | requested_time = self.getAttr( attrs, 'Resource_List.walltime' ) |
---|
| 278 | requested_memory = self.getAttr( attrs, 'Resource_List.mem' ) |
---|
[95] | 279 | |
---|
[26] | 280 | mynoderequest = self.getAttr( attrs, 'Resource_List.nodes' ) |
---|
[95] | 281 | |
---|
[26] | 282 | if mynoderequest.find( ':' ) != -1 and mynoderequest.find( 'ppn' ) != -1: |
---|
| 283 | ppn = mynoderequest.split( ':' )[1].split( 'ppn=' )[1] |
---|
| 284 | else: |
---|
| 285 | ppn = '' |
---|
[95] | 286 | |
---|
[26] | 287 | status = self.getAttr( attrs, 'job_state' ) |
---|
[25] | 288 | |
---|
[95] | 289 | if status == 'R': |
---|
| 290 | start_timestamp = self.getAttr( attrs, 'mtime' ) |
---|
| 291 | nodes = self.getAttr( attrs, 'exec_host' ).split( '+' ) |
---|
[133] | 292 | |
---|
| 293 | nodeslist = [ ] |
---|
| 294 | |
---|
| 295 | for node in nodes: |
---|
| 296 | host = node.split( '/' )[0] |
---|
| 297 | |
---|
| 298 | if nodeslist.count( host ) == 0: |
---|
[215] | 299 | |
---|
| 300 | for translate_pattern in BATCH_HOST_TRANSLATE: |
---|
| 301 | |
---|
[220] | 302 | if translate_pattern.find( '/' ) != -1: |
---|
[215] | 303 | |
---|
[220] | 304 | translate_orig = translate_pattern.split( '/' )[1] |
---|
| 305 | translate_new = translate_pattern.split( '/' )[2] |
---|
| 306 | |
---|
| 307 | host = re.sub( translate_orig, translate_new, host ) |
---|
[216] | 308 | |
---|
[217] | 309 | if not host in nodeslist: |
---|
[216] | 310 | |
---|
| 311 | nodeslist.append( host ) |
---|
[133] | 312 | |
---|
[185] | 313 | if DETECT_TIME_DIFFS: |
---|
| 314 | |
---|
| 315 | # If a job start if later than our current date, |
---|
| 316 | # that must mean the Torque server's time is later |
---|
| 317 | # than our local time. |
---|
| 318 | |
---|
| 319 | if int(start_timestamp) > int( int(self.cur_time) + int(self.timeoffset) ): |
---|
| 320 | |
---|
| 321 | self.timeoffset = int( int(start_timestamp) - int(self.cur_time) ) |
---|
| 322 | |
---|
[133] | 323 | elif status == 'Q': |
---|
[95] | 324 | start_timestamp = '' |
---|
[133] | 325 | count_mynodes = 0 |
---|
| 326 | numeric_node = 1 |
---|
[95] | 327 | |
---|
[133] | 328 | for node in mynoderequest.split( '+' ): |
---|
[67] | 329 | |
---|
[133] | 330 | nodepart = node.split( ':' )[0] |
---|
[67] | 331 | |
---|
[133] | 332 | for letter in nodepart: |
---|
[67] | 333 | |
---|
[133] | 334 | if letter not in string.digits: |
---|
| 335 | |
---|
| 336 | numeric_node = 0 |
---|
| 337 | |
---|
| 338 | if not numeric_node: |
---|
| 339 | count_mynodes = count_mynodes + 1 |
---|
| 340 | else: |
---|
| 341 | count_mynodes = count_mynodes + int( nodepart ) |
---|
| 342 | |
---|
[134] | 343 | nodeslist = count_mynodes |
---|
[172] | 344 | else: |
---|
| 345 | start_timestamp = '' |
---|
[173] | 346 | nodeslist = '' |
---|
[133] | 347 | |
---|
[26] | 348 | myAttrs = { } |
---|
| 349 | myAttrs['name'] = name |
---|
| 350 | myAttrs['queue'] = queue |
---|
| 351 | myAttrs['owner'] = owner |
---|
| 352 | myAttrs['requested_time'] = requested_time |
---|
| 353 | myAttrs['requested_memory'] = requested_memory |
---|
| 354 | myAttrs['ppn'] = ppn |
---|
| 355 | myAttrs['status'] = status |
---|
| 356 | myAttrs['start_timestamp'] = start_timestamp |
---|
[185] | 357 | myAttrs['reported'] = str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) |
---|
[67] | 358 | myAttrs['nodes'] = nodeslist |
---|
| 359 | myAttrs['domain'] = string.join( socket.getfqdn().split( '.' )[1:], '.' ) |
---|
[80] | 360 | myAttrs['poll_interval'] = TORQUE_POLL_INTERVAL |
---|
[26] | 361 | |
---|
[184] | 362 | if self.jobDataChanged( jobs, job_id, myAttrs ) and myAttrs['status'] in [ 'R', 'Q' ]: |
---|
[26] | 363 | jobs[ job_id ] = myAttrs |
---|
[61] | 364 | |
---|
[101] | 365 | #debug_msg( 10, printTime() + ' job %s state changed' %(job_id) ) |
---|
[26] | 366 | |
---|
[76] | 367 | for id, attrs in jobs.items(): |
---|
| 368 | |
---|
| 369 | if id not in jobs_processed: |
---|
| 370 | |
---|
| 371 | # This one isn't there anymore; toedeledoki! |
---|
| 372 | # |
---|
| 373 | del jobs[ id ] |
---|
| 374 | |
---|
[65] | 375 | return jobs |
---|
| 376 | |
---|
| 377 | def submitJobData( self, jobs ): |
---|
| 378 | """Submit job info list""" |
---|
| 379 | |
---|
[219] | 380 | self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) ) |
---|
[69] | 381 | |
---|
[61] | 382 | # Now let's spread the knowledge |
---|
| 383 | # |
---|
| 384 | for jobid, jobattrs in jobs.items(): |
---|
| 385 | |
---|
[95] | 386 | gmetric_val = self.compileGmetricVal( jobid, jobattrs ) |
---|
[61] | 387 | |
---|
[95] | 388 | for val in gmetric_val: |
---|
[219] | 389 | self.dp.multicastGmetric( 'MONARCH-JOB-' + jobid, val ) |
---|
[61] | 390 | |
---|
[67] | 391 | def makeNodeString( self, nodelist ): |
---|
[68] | 392 | """Make one big string of all hosts""" |
---|
[67] | 393 | |
---|
| 394 | node_str = None |
---|
| 395 | |
---|
| 396 | for node in nodelist: |
---|
| 397 | if not node_str: |
---|
| 398 | node_str = node |
---|
| 399 | else: |
---|
| 400 | node_str = node_str + ';' + node |
---|
| 401 | |
---|
| 402 | return node_str |
---|
| 403 | |
---|
[65] | 404 | def compileGmetricVal( self, jobid, jobattrs ): |
---|
| 405 | """Create a val string for gmetric of jobinfo""" |
---|
[61] | 406 | |
---|
[80] | 407 | appendList = [ ] |
---|
| 408 | appendList.append( 'name=' + jobattrs['name'] ) |
---|
| 409 | appendList.append( 'queue=' + jobattrs['queue'] ) |
---|
| 410 | appendList.append( 'owner=' + jobattrs['owner'] ) |
---|
| 411 | appendList.append( 'requested_time=' + jobattrs['requested_time'] ) |
---|
[95] | 412 | |
---|
| 413 | if jobattrs['requested_memory'] != '': |
---|
| 414 | appendList.append( 'requested_memory=' + jobattrs['requested_memory'] ) |
---|
| 415 | |
---|
| 416 | if jobattrs['ppn'] != '': |
---|
| 417 | appendList.append( 'ppn=' + jobattrs['ppn'] ) |
---|
| 418 | |
---|
[80] | 419 | appendList.append( 'status=' + jobattrs['status'] ) |
---|
[95] | 420 | |
---|
| 421 | if jobattrs['start_timestamp'] != '': |
---|
| 422 | appendList.append( 'start_timestamp=' + jobattrs['start_timestamp'] ) |
---|
| 423 | |
---|
[80] | 424 | appendList.append( 'reported=' + jobattrs['reported'] ) |
---|
[85] | 425 | appendList.append( 'poll_interval=' + str( jobattrs['poll_interval'] ) ) |
---|
[80] | 426 | appendList.append( 'domain=' + jobattrs['domain'] ) |
---|
[26] | 427 | |
---|
[134] | 428 | if jobattrs['status'] == 'R': |
---|
| 429 | if len( jobattrs['nodes'] ) > 0: |
---|
| 430 | appendList.append( 'nodes=' + self.makeNodeString( jobattrs['nodes'] ) ) |
---|
| 431 | elif jobattrs['status'] == 'Q': |
---|
| 432 | appendList.append( 'nodes=' + str(jobattrs['nodes']) ) |
---|
[95] | 433 | |
---|
[65] | 434 | return self.makeAppendLists( appendList ) |
---|
| 435 | |
---|
| 436 | def makeAppendLists( self, append_list ): |
---|
[68] | 437 | """ |
---|
| 438 | Divide all values from append_list over strings with a maximum |
---|
| 439 | size of 1400 |
---|
| 440 | """ |
---|
[65] | 441 | |
---|
| 442 | app_lists = [ ] |
---|
| 443 | |
---|
| 444 | mystr = None |
---|
| 445 | |
---|
| 446 | for val in append_list: |
---|
| 447 | |
---|
| 448 | if not mystr: |
---|
| 449 | mystr = val |
---|
| 450 | else: |
---|
| 451 | if not self.checkValAppendMaxSize( mystr, val ): |
---|
| 452 | mystr = mystr + ' ' + val |
---|
| 453 | else: |
---|
| 454 | # Too big, new appenlist |
---|
| 455 | app_lists.append( mystr ) |
---|
| 456 | mystr = val |
---|
| 457 | |
---|
| 458 | app_lists.append( mystr ) |
---|
| 459 | |
---|
| 460 | return app_lists |
---|
| 461 | |
---|
| 462 | def checkValAppendMaxSize( self, val, text ): |
---|
| 463 | """Check if val + text size is not above 1400 (max msg size)""" |
---|
| 464 | |
---|
[69] | 465 | # Max frame size of a udp datagram is 1500 bytes |
---|
[216] | 466 | # removing misc header and gmetric stuff leaves about 1300 bytes |
---|
[69] | 467 | # |
---|
[216] | 468 | if len( val + text ) > 900: |
---|
[65] | 469 | return 1 |
---|
| 470 | else: |
---|
| 471 | return 0 |
---|
| 472 | |
---|
[61] | 473 | def printJobs( self, jobs ): |
---|
[65] | 474 | """Print a jobinfo overview""" |
---|
| 475 | |
---|
[26] | 476 | for name, attrs in self.jobs.items(): |
---|
| 477 | |
---|
| 478 | print 'job %s' %(name) |
---|
| 479 | |
---|
| 480 | for name, val in attrs.items(): |
---|
| 481 | |
---|
| 482 | print '\t%s = %s' %( name, val ) |
---|
| 483 | |
---|
[61] | 484 | def printJob( self, jobs, job_id ): |
---|
[65] | 485 | """Print job with job_id from jobs""" |
---|
[26] | 486 | |
---|
| 487 | print 'job %s' %(job_id) |
---|
| 488 | |
---|
[65] | 489 | for name, val in jobs[ job_id ].items(): |
---|
[26] | 490 | |
---|
| 491 | print '\t%s = %s' %( name, val ) |
---|
| 492 | |
---|
| 493 | def daemon( self ): |
---|
[65] | 494 | """Run as daemon forever""" |
---|
[26] | 495 | |
---|
| 496 | # Fork the first child |
---|
| 497 | # |
---|
| 498 | pid = os.fork() |
---|
| 499 | if pid > 0: |
---|
[212] | 500 | sys.exit(0) # end parent |
---|
[26] | 501 | |
---|
| 502 | # creates a session and sets the process group ID |
---|
| 503 | # |
---|
| 504 | os.setsid() |
---|
| 505 | |
---|
| 506 | # Fork the second child |
---|
| 507 | # |
---|
| 508 | pid = os.fork() |
---|
| 509 | if pid > 0: |
---|
[212] | 510 | sys.exit(0) # end parent |
---|
[26] | 511 | |
---|
| 512 | # Go to the root directory and set the umask |
---|
| 513 | # |
---|
| 514 | os.chdir('/') |
---|
| 515 | os.umask(0) |
---|
| 516 | |
---|
| 517 | sys.stdin.close() |
---|
| 518 | sys.stdout.close() |
---|
| 519 | sys.stderr.close() |
---|
| 520 | |
---|
| 521 | os.open('/dev/null', 0) |
---|
| 522 | os.dup(0) |
---|
| 523 | os.dup(0) |
---|
| 524 | |
---|
| 525 | self.run() |
---|
| 526 | |
---|
| 527 | def run( self ): |
---|
[65] | 528 | """Main thread""" |
---|
[26] | 529 | |
---|
| 530 | while ( 1 ): |
---|
| 531 | |
---|
[65] | 532 | self.jobs = self.getJobData( self.jobs ) |
---|
| 533 | self.submitJobData( self.jobs ) |
---|
[64] | 534 | time.sleep( TORQUE_POLL_INTERVAL ) |
---|
[26] | 535 | |
---|
| 536 | def printTime( ): |
---|
[65] | 537 | """Print current time/date in human readable format for log/debug""" |
---|
[26] | 538 | |
---|
| 539 | return time.strftime("%a, %d %b %Y %H:%M:%S") |
---|
| 540 | |
---|
| 541 | def debug_msg( level, msg ): |
---|
[65] | 542 | """Print msg if at or above current debug level""" |
---|
[26] | 543 | |
---|
| 544 | if (DEBUG_LEVEL >= level): |
---|
| 545 | sys.stderr.write( msg + '\n' ) |
---|
| 546 | |
---|
[23] | 547 | def main(): |
---|
[65] | 548 | """Application start""" |
---|
[23] | 549 | |
---|
[212] | 550 | if not processArgs( sys.argv[1:] ): |
---|
| 551 | sys.exit( 1 ) |
---|
| 552 | |
---|
[174] | 553 | gather = DataGatherer() |
---|
[26] | 554 | if DAEMONIZE: |
---|
| 555 | gather.daemon() |
---|
| 556 | else: |
---|
| 557 | gather.run() |
---|
[23] | 558 | |
---|
[65] | 559 | # w00t someone started me |
---|
| 560 | # |
---|
[23] | 561 | if __name__ == '__main__': |
---|
| 562 | main() |
---|