[23] | 1 | #!/usr/bin/env python |
---|
| 2 | |
---|
[26] | 3 | # Specify debugging level here; |
---|
| 4 | # |
---|
| 5 | DEBUG_LEVEL = 10 |
---|
| 6 | |
---|
| 7 | # Wether or not to run as a daemon in background |
---|
| 8 | # |
---|
| 9 | DAEMONIZE = 0 |
---|
| 10 | |
---|
[64] | 11 | # How many seconds interval for polling of jobs |
---|
[61] | 12 | # |
---|
[64] | 13 | # this will effect directly how accurate the |
---|
| 14 | # end time of a job can be determined |
---|
| 15 | # |
---|
| 16 | TORQUE_POLL_INTERVAL = 10 |
---|
[61] | 17 | |
---|
[23] | 18 | from PBSQuery import PBSQuery |
---|
[26] | 19 | import sys |
---|
| 20 | import time |
---|
[65] | 21 | import os |
---|
[67] | 22 | import socket |
---|
[65] | 23 | import string |
---|
[23] | 24 | |
---|
[61] | 25 | class DataProcessor: |
---|
| 26 | |
---|
| 27 | binary = '/usr/bin/gmetric' |
---|
| 28 | |
---|
| 29 | def __init__( self, binary=None ): |
---|
| 30 | |
---|
| 31 | if binary: |
---|
| 32 | self.binary = binary |
---|
| 33 | |
---|
[65] | 34 | self.dmax = TORQUE_POLL_INTERVAL |
---|
[61] | 35 | |
---|
[65] | 36 | #incompatible = self.checkGmetricVersion |
---|
| 37 | incompatible = 0 |
---|
[61] | 38 | |
---|
[65] | 39 | if incompatible: |
---|
| 40 | debug_msg( 0, 'Gmetric version not compatible, pls upgrade to at least 3.0.1' ) |
---|
| 41 | sys.exit( 1 ) |
---|
| 42 | |
---|
| 43 | def checkGmetricVersion( self ): |
---|
| 44 | |
---|
| 45 | for line in os.popen( self.binary + ' --version' ).readlines(): |
---|
| 46 | |
---|
| 47 | line = line.split( ' ' ) |
---|
| 48 | |
---|
| 49 | if len( line ) == 2 and line.find( 'gmetric' ) != -1: |
---|
| 50 | |
---|
| 51 | gmetric_version = line[1] |
---|
| 52 | |
---|
| 53 | version_major = int( gemtric_version.split( '.' )[0] ) |
---|
| 54 | version_minor = int( gemtric_version.split( '.' )[1] ) |
---|
| 55 | version_patch = int( gemtric_version.split( '.' )[2] ) |
---|
| 56 | |
---|
| 57 | incompatible = 0 |
---|
| 58 | |
---|
| 59 | if version_major < 3: |
---|
| 60 | |
---|
| 61 | incompatible = 1 |
---|
| 62 | |
---|
| 63 | elif version_major == 3: |
---|
| 64 | |
---|
| 65 | if version_minor == 0: |
---|
| 66 | |
---|
| 67 | if version_patch < 1: |
---|
| 68 | |
---|
| 69 | incompatbiel = 1 |
---|
| 70 | |
---|
| 71 | return incompatible |
---|
| 72 | |
---|
[66] | 73 | def multicastGmetric( self, metricname, metricval, valtype='string', tmax='15' ): |
---|
[65] | 74 | |
---|
| 75 | cmd = self.binary |
---|
| 76 | |
---|
[61] | 77 | try: |
---|
| 78 | cmd = cmd + ' -c' + GMOND_CONF |
---|
| 79 | except NameError: |
---|
[64] | 80 | debug_msg( 10, 'Assuming /etc/gmond.conf for gmetric cmd (ommitting)' ) |
---|
[61] | 81 | |
---|
[66] | 82 | cmd = cmd + ' -n' + metricname + ' -v"' + metricval + '" -t' + valtype + ' -x' + tmax + ' -d' + str( self.dmax ) |
---|
[61] | 83 | |
---|
| 84 | print cmd |
---|
| 85 | #os.system( cmd ) |
---|
| 86 | |
---|
[23] | 87 | class PBSDataGatherer: |
---|
| 88 | |
---|
[61] | 89 | jobs = { } |
---|
| 90 | |
---|
[23] | 91 | def __init__( self ): |
---|
| 92 | |
---|
| 93 | self.pq = PBSQuery() |
---|
[26] | 94 | self.jobs = { } |
---|
[61] | 95 | self.dp = DataProcessor() |
---|
[23] | 96 | |
---|
[26] | 97 | def getAttr( self, attrs, name ): |
---|
| 98 | |
---|
| 99 | if attrs.has_key( name ): |
---|
| 100 | return attrs[name] |
---|
| 101 | else: |
---|
| 102 | return '' |
---|
| 103 | |
---|
| 104 | def jobDataChanged( self, jobs, job_id, attrs ): |
---|
| 105 | |
---|
| 106 | if jobs.has_key( job_id ): |
---|
| 107 | oldData = jobs[ job_id ] |
---|
| 108 | else: |
---|
| 109 | return 1 |
---|
| 110 | |
---|
| 111 | for name, val in attrs.items(): |
---|
| 112 | |
---|
| 113 | if oldData.has_key( name ): |
---|
| 114 | |
---|
| 115 | if oldData[ name ] != attrs[ name ]: |
---|
| 116 | |
---|
| 117 | return 1 |
---|
| 118 | |
---|
| 119 | else: |
---|
| 120 | return 1 |
---|
| 121 | |
---|
| 122 | return 0 |
---|
| 123 | |
---|
[65] | 124 | def getJobData( self, known_jobs ): |
---|
[26] | 125 | |
---|
[65] | 126 | if len( known_jobs ) > 0: |
---|
| 127 | jobs = known_jobs |
---|
| 128 | else: |
---|
| 129 | jobs = { } |
---|
[26] | 130 | |
---|
| 131 | joblist = self.pq.getjobs() |
---|
| 132 | |
---|
| 133 | jobs_processed = [ ] |
---|
| 134 | |
---|
| 135 | for name, attrs in joblist.items(): |
---|
| 136 | |
---|
[67] | 137 | print attrs |
---|
| 138 | |
---|
[26] | 139 | job_id = name.split( '.' )[0] |
---|
| 140 | |
---|
| 141 | jobs_processed.append( job_id ) |
---|
[61] | 142 | |
---|
[26] | 143 | name = self.getAttr( attrs, 'Job_Name' ) |
---|
| 144 | queue = self.getAttr( attrs, 'queue' ) |
---|
| 145 | owner = self.getAttr( attrs, 'Job_Owner' ).split( '@' )[0] |
---|
| 146 | requested_time = self.getAttr( attrs, 'Resource_List.walltime' ) |
---|
| 147 | requested_memory = self.getAttr( attrs, 'Resource_List.mem' ) |
---|
| 148 | mynoderequest = self.getAttr( attrs, 'Resource_List.nodes' ) |
---|
| 149 | if mynoderequest.find( ':' ) != -1 and mynoderequest.find( 'ppn' ) != -1: |
---|
| 150 | ppn = mynoderequest.split( ':' )[1].split( 'ppn=' )[1] |
---|
| 151 | else: |
---|
| 152 | ppn = '' |
---|
| 153 | status = self.getAttr( attrs, 'job_state' ) |
---|
| 154 | start_timestamp = self.getAttr( attrs, 'mtime' ) |
---|
[25] | 155 | |
---|
[67] | 156 | nodes = self.getAttr( attrs, 'exec_host' ).split( '+' ) |
---|
| 157 | nodeslist = [ ] |
---|
| 158 | |
---|
| 159 | for node in nodes: |
---|
| 160 | host = node.split( '/' )[0] |
---|
| 161 | |
---|
| 162 | if nodeslist.count( host ) == 0: |
---|
| 163 | nodeslist.append( node.split( '/' )[0] ) |
---|
| 164 | |
---|
[26] | 165 | myAttrs = { } |
---|
| 166 | myAttrs['name'] = name |
---|
| 167 | myAttrs['queue'] = queue |
---|
| 168 | myAttrs['owner'] = owner |
---|
| 169 | myAttrs['requested_time'] = requested_time |
---|
| 170 | myAttrs['requested_memory'] = requested_memory |
---|
| 171 | myAttrs['ppn'] = ppn |
---|
| 172 | myAttrs['status'] = status |
---|
| 173 | myAttrs['start_timestamp'] = start_timestamp |
---|
[67] | 174 | myAttrs['nodes'] = nodeslist |
---|
| 175 | myAttrs['domain'] = string.join( socket.getfqdn().split( '.' )[1:], '.' ) |
---|
[26] | 176 | |
---|
| 177 | if self.jobDataChanged( jobs, job_id, myAttrs ): |
---|
| 178 | jobs[ job_id ] = myAttrs |
---|
[61] | 179 | |
---|
| 180 | self.printJob( jobs, job_id ) |
---|
| 181 | |
---|
[26] | 182 | debug_msg( 10, printTime() + ' job %s state changed' %(job_id) ) |
---|
| 183 | |
---|
[65] | 184 | return jobs |
---|
| 185 | |
---|
| 186 | def submitJobData( self, jobs ): |
---|
| 187 | """Submit job info list""" |
---|
| 188 | |
---|
| 189 | time_now = time.time() |
---|
| 190 | |
---|
[66] | 191 | self.dp.multicastGmetric( 'TOGA-HEARTBEAT', str( time_now ), 'float' ) |
---|
[65] | 192 | |
---|
[61] | 193 | # Now let's spread the knowledge |
---|
| 194 | # |
---|
| 195 | for jobid, jobattrs in jobs.items(): |
---|
| 196 | |
---|
[65] | 197 | gmetric_val = self.compileGmetricVal( jobid, jobattrs ) |
---|
[61] | 198 | |
---|
[65] | 199 | for val in gmetric_val: |
---|
| 200 | self.dp.multicastGmetric( 'TOGA-JOB-' + jobid, val ) |
---|
[61] | 201 | |
---|
[67] | 202 | def makeNodeString( self, nodelist ): |
---|
| 203 | |
---|
| 204 | node_str = None |
---|
| 205 | |
---|
| 206 | for node in nodelist: |
---|
| 207 | if not node_str: |
---|
| 208 | node_str = node |
---|
| 209 | else: |
---|
| 210 | node_str = node_str + ';' + node |
---|
| 211 | |
---|
| 212 | return node_str |
---|
| 213 | |
---|
[65] | 214 | def compileGmetricVal( self, jobid, jobattrs ): |
---|
| 215 | """Create a val string for gmetric of jobinfo""" |
---|
[61] | 216 | |
---|
[65] | 217 | name_str = 'name=' + jobattrs['name'] |
---|
| 218 | queue_str = 'queue=' + jobattrs['queue'] |
---|
| 219 | owner_str = 'owner=' + jobattrs['owner'] |
---|
| 220 | rtime_str = 'rtime=' + jobattrs['requested_time'] |
---|
| 221 | rmem_str = 'rmem=' + jobattrs['requested_memory'] |
---|
| 222 | ppn_str = 'ppn=' + jobattrs['ppn'] |
---|
| 223 | status_str = 'status=' + jobattrs['status'] |
---|
| 224 | stime_str = 'stime=' + jobattrs['start_timestamp'] |
---|
[67] | 225 | domain_str = 'domain=' + jobattrs['domain'] |
---|
| 226 | node_str = 'nodes=' + self.makeNodeString( jobattrs['nodes'] ) |
---|
[26] | 227 | |
---|
[67] | 228 | appendList = [ name_str, queue_str, owner_str, rtime_str, rmem_str, ppn_str, status_str, stime_str, domain_str, node_str ] |
---|
[65] | 229 | |
---|
| 230 | return self.makeAppendLists( appendList ) |
---|
| 231 | |
---|
| 232 | def makeAppendLists( self, append_list ): |
---|
| 233 | |
---|
| 234 | app_lists = [ ] |
---|
| 235 | |
---|
| 236 | mystr = None |
---|
| 237 | |
---|
| 238 | for val in append_list: |
---|
| 239 | |
---|
| 240 | if not mystr: |
---|
| 241 | mystr = val |
---|
| 242 | else: |
---|
| 243 | if not self.checkValAppendMaxSize( mystr, val ): |
---|
| 244 | mystr = mystr + ' ' + val |
---|
| 245 | else: |
---|
| 246 | # Too big, new appenlist |
---|
| 247 | app_lists.append( mystr ) |
---|
| 248 | mystr = val |
---|
| 249 | |
---|
| 250 | app_lists.append( mystr ) |
---|
| 251 | |
---|
| 252 | return app_lists |
---|
| 253 | |
---|
| 254 | def checkValAppendMaxSize( self, val, text ): |
---|
| 255 | """Check if val + text size is not above 1400 (max msg size)""" |
---|
| 256 | |
---|
| 257 | if len( val + text ) > 1400: |
---|
| 258 | return 1 |
---|
| 259 | else: |
---|
| 260 | return 0 |
---|
| 261 | |
---|
[61] | 262 | def printJobs( self, jobs ): |
---|
[65] | 263 | """Print a jobinfo overview""" |
---|
| 264 | |
---|
[26] | 265 | for name, attrs in self.jobs.items(): |
---|
| 266 | |
---|
| 267 | print 'job %s' %(name) |
---|
| 268 | |
---|
| 269 | for name, val in attrs.items(): |
---|
| 270 | |
---|
| 271 | print '\t%s = %s' %( name, val ) |
---|
| 272 | |
---|
[61] | 273 | def printJob( self, jobs, job_id ): |
---|
[65] | 274 | """Print job with job_id from jobs""" |
---|
[26] | 275 | |
---|
| 276 | print 'job %s' %(job_id) |
---|
| 277 | |
---|
[65] | 278 | for name, val in jobs[ job_id ].items(): |
---|
[26] | 279 | |
---|
| 280 | print '\t%s = %s' %( name, val ) |
---|
| 281 | |
---|
| 282 | def daemon( self ): |
---|
[65] | 283 | """Run as daemon forever""" |
---|
[26] | 284 | |
---|
| 285 | # Fork the first child |
---|
| 286 | # |
---|
| 287 | pid = os.fork() |
---|
| 288 | if pid > 0: |
---|
| 289 | sys.exit(0) # end parrent |
---|
| 290 | |
---|
| 291 | # creates a session and sets the process group ID |
---|
| 292 | # |
---|
| 293 | os.setsid() |
---|
| 294 | |
---|
| 295 | # Fork the second child |
---|
| 296 | # |
---|
| 297 | pid = os.fork() |
---|
| 298 | if pid > 0: |
---|
| 299 | sys.exit(0) # end parrent |
---|
| 300 | |
---|
| 301 | # Go to the root directory and set the umask |
---|
| 302 | # |
---|
| 303 | os.chdir('/') |
---|
| 304 | os.umask(0) |
---|
| 305 | |
---|
| 306 | sys.stdin.close() |
---|
| 307 | sys.stdout.close() |
---|
| 308 | sys.stderr.close() |
---|
| 309 | |
---|
| 310 | os.open('/dev/null', 0) |
---|
| 311 | os.dup(0) |
---|
| 312 | os.dup(0) |
---|
| 313 | |
---|
| 314 | self.run() |
---|
| 315 | |
---|
| 316 | def run( self ): |
---|
[65] | 317 | """Main thread""" |
---|
[26] | 318 | |
---|
| 319 | while ( 1 ): |
---|
| 320 | |
---|
[65] | 321 | self.jobs = self.getJobData( self.jobs ) |
---|
| 322 | self.submitJobData( self.jobs ) |
---|
[64] | 323 | time.sleep( TORQUE_POLL_INTERVAL ) |
---|
[26] | 324 | |
---|
| 325 | def printTime( ): |
---|
[65] | 326 | """Print current time/date in human readable format for log/debug""" |
---|
[26] | 327 | |
---|
| 328 | return time.strftime("%a, %d %b %Y %H:%M:%S") |
---|
| 329 | |
---|
| 330 | def debug_msg( level, msg ): |
---|
[65] | 331 | """Print msg if at or above current debug level""" |
---|
[26] | 332 | |
---|
| 333 | if (DEBUG_LEVEL >= level): |
---|
| 334 | sys.stderr.write( msg + '\n' ) |
---|
| 335 | |
---|
[23] | 336 | def main(): |
---|
[65] | 337 | """Application start""" |
---|
[23] | 338 | |
---|
| 339 | gather = PBSDataGatherer() |
---|
[26] | 340 | if DAEMONIZE: |
---|
| 341 | gather.daemon() |
---|
| 342 | else: |
---|
| 343 | gather.run() |
---|
[23] | 344 | |
---|
[65] | 345 | # w00t someone started me |
---|
| 346 | # |
---|
[23] | 347 | if __name__ == '__main__': |
---|
| 348 | main() |
---|