| 1 |
spiga |
1.6 |
#!/usr/bin/env python
|
| 2 |
|
|
"""
|
| 3 |
|
|
basic SGE CLI interaction class
|
| 4 |
|
|
"""
|
| 5 |
|
|
|
| 6 |
|
|
__revision__ = "$Id: SchedulerSge.py,v 1.5 2009/01/20 18:49:45 gcodispo Exp $"
|
| 7 |
|
|
__version__ = "$Revision: 1.5 $"
|
| 8 |
|
|
|
| 9 |
|
|
import re, os
|
| 10 |
|
|
|
| 11 |
|
|
from ProdCommon.BossLite.Scheduler.SchedulerInterface import SchedulerInterface
|
| 12 |
|
|
from ProdCommon.BossLite.Common.Exceptions import SchedulerError
|
| 13 |
|
|
from ProdCommon.BossLite.DbObjects.Job import Job
|
| 14 |
|
|
from ProdCommon.BossLite.DbObjects.Task import Task
|
| 15 |
|
|
from ProdCommon.BossLite.DbObjects.RunningJob import RunningJob
|
| 16 |
|
|
|
| 17 |
|
|
class SchedulerSge (SchedulerInterface) :
|
| 18 |
|
|
"""
|
| 19 |
|
|
basic class to handle sge jobs
|
| 20 |
|
|
"""
|
| 21 |
|
|
def __init__( self, **args):
|
| 22 |
|
|
|
| 23 |
|
|
# call super class init method
|
| 24 |
|
|
super(SchedulerSge, self).__init__(**args)
|
| 25 |
|
|
self.statusMap = {
|
| 26 |
|
|
'd':'K',
|
| 27 |
|
|
'E':'DA',
|
| 28 |
|
|
'h':'R',
|
| 29 |
|
|
'r':'R',
|
| 30 |
|
|
'R':'R',
|
| 31 |
|
|
's':'R',
|
| 32 |
|
|
'S':'R',
|
| 33 |
|
|
't':'SS',
|
| 34 |
|
|
'T':'R',
|
| 35 |
|
|
'w':'R',
|
| 36 |
|
|
'qw':'R',
|
| 37 |
|
|
'Eqw':'K',
|
| 38 |
|
|
'DONE':'SD'
|
| 39 |
|
|
}
|
| 40 |
|
|
|
| 41 |
|
|
|
| 42 |
|
|
def checkUserProxy( self, cert='' ):
|
| 43 |
|
|
return
|
| 44 |
slacapra |
1.1 |
|
| 45 |
spiga |
1.6 |
def jobDescription ( self, obj, requirements='', config='', service = '' ):
|
| 46 |
|
|
"""
|
| 47 |
|
|
retrieve scheduler specific job description
|
| 48 |
|
|
return it as a string
|
| 49 |
|
|
"""
|
| 50 |
|
|
args=''
|
| 51 |
|
|
if type(obj) == RunningJob or type(obj) == Job:
|
| 52 |
|
|
return self.decode(obj, requirements)
|
| 53 |
|
|
elif type(obj) == Task :
|
| 54 |
|
|
task = obj
|
| 55 |
|
|
for job in task.getJobs() :
|
| 56 |
|
|
args += self.decode(job, task, requirements)+' \n'
|
| 57 |
|
|
return args
|
| 58 |
slacapra |
1.1 |
|
| 59 |
|
|
|
| 60 |
spiga |
1.6 |
def submit ( self, obj, requirements='', config='', service = '' ) :
|
| 61 |
|
|
"""
|
| 62 |
|
|
set up submission parameters and submit
|
| 63 |
slacapra |
1.1 |
|
| 64 |
spiga |
1.6 |
return jobAttributes, bulkId, service
|
| 65 |
slacapra |
1.1 |
|
| 66 |
spiga |
1.6 |
- jobAttributs is a map of the format
|
| 67 |
|
|
jobAttributes[ 'name' : 'schedulerId' ]
|
| 68 |
|
|
- bulkId is an eventual bulk submission identifier
|
| 69 |
|
|
- service is a endpoit to connect withs (such as the WMS)
|
| 70 |
|
|
"""
|
| 71 |
ewv |
1.5 |
|
| 72 |
spiga |
1.6 |
#print "config"+config
|
| 73 |
|
|
if type(obj) == RunningJob or type(obj) == Job:
|
| 74 |
|
|
return self.submitJob(obj, requirements)
|
| 75 |
|
|
elif type(obj) == Task :
|
| 76 |
|
|
return self.submitTask (obj, requirements )
|
| 77 |
|
|
|
| 78 |
|
|
def submitTask ( self, task, requirements=''):
|
| 79 |
|
|
ret_map={}
|
| 80 |
|
|
for job in task.getJobs() :
|
| 81 |
|
|
map, taskId, queue = self.submitJob(job, task, requirements)
|
| 82 |
|
|
ret_map.update(map)
|
| 83 |
|
|
|
| 84 |
|
|
return ret_map, taskId, queue
|
| 85 |
|
|
|
| 86 |
|
|
def submitJob ( self, job, task=None, requirements=''):
|
| 87 |
|
|
"""Need to copy the inputsandbox to WN before submitting a job"""
|
| 88 |
|
|
|
| 89 |
|
|
arg = self.decode(job, task, requirements )
|
| 90 |
|
|
|
| 91 |
|
|
command = "qsub " + arg
|
| 92 |
|
|
#print command + "\n"
|
| 93 |
|
|
out, ret = self.ExecuteCommand(command)
|
| 94 |
|
|
#print "out:" + out + "\n"
|
| 95 |
|
|
r = re.compile("Your job (\d+) .* has been submitted")
|
| 96 |
|
|
|
| 97 |
|
|
m= r.search(out)
|
| 98 |
|
|
if m is not None:
|
| 99 |
|
|
jobId =m.group(1)
|
| 100 |
|
|
command = "qstat -j " + jobId
|
| 101 |
|
|
#out, ret = self.ExecuteCommand(command)
|
| 102 |
|
|
#print "out:" + out + "\n"
|
| 103 |
|
|
#queue = m.group(2)
|
| 104 |
|
|
queue = "all"
|
| 105 |
|
|
else:
|
| 106 |
|
|
#rNot = re.compile("Job not submitted.*<(\w+)>")
|
| 107 |
|
|
#m= rNot.search(out)
|
| 108 |
|
|
#if m is not None:
|
| 109 |
|
|
# print m
|
| 110 |
|
|
# print "Job NOT submitted"
|
| 111 |
|
|
# print out
|
| 112 |
|
|
raise SchedulerError('error', out)
|
| 113 |
|
|
taskId = None
|
| 114 |
|
|
#print "Your job identifier is: ", taskId, queue
|
| 115 |
|
|
map={ job[ 'name' ] : jobId }
|
| 116 |
|
|
return map, taskId, queue
|
| 117 |
slacapra |
1.1 |
|
| 118 |
spiga |
1.4 |
|
| 119 |
spiga |
1.6 |
def decode (self, job, task=None, requirements='' , config ='', service='' ):
|
| 120 |
slacapra |
1.1 |
"""
|
| 121 |
spiga |
1.6 |
prepare file for submission
|
| 122 |
slacapra |
1.1 |
"""
|
| 123 |
|
|
|
| 124 |
spiga |
1.6 |
txt = "#batch script for SGE jobs\n"
|
| 125 |
|
|
txt += "MYTMPDIR=$TMP/$JOB_NAME\n"
|
| 126 |
|
|
txt += "mkdir -p $MYTMPDIR \n"
|
| 127 |
|
|
txt += "cd $MYTMPDIR\n"
|
| 128 |
|
|
# Need to copy InputSandBox to WN
|
| 129 |
|
|
if task:
|
| 130 |
|
|
subDir=task[ 'startDirectory' ]
|
| 131 |
|
|
for inpFile in task[ 'globalSandbox' ].split(','):
|
| 132 |
|
|
#txt += "cp "+subDir+inpFile+" . \n"
|
| 133 |
|
|
txt += "cp "+inpFile+" . \n"
|
| 134 |
|
|
## Job specific ISB
|
| 135 |
|
|
#for inpFile in job[ 'inputFiles' ]:
|
| 136 |
|
|
# if inpFile != '': txt += self.cpCmd+" "+self.rfioSer+"/"+inpFile+" . \n"
|
| 137 |
|
|
|
| 138 |
|
|
## Now the actual wrapper
|
| 139 |
|
|
args = job[ 'arguments' ]
|
| 140 |
|
|
exe = job[ 'executable' ]
|
| 141 |
|
|
txt += "./"+os.path.basename(exe)+" "+args+"\n"
|
| 142 |
|
|
|
| 143 |
|
|
## And finally copy back the output
|
| 144 |
|
|
outputDir=task['outputDirectory']
|
| 145 |
|
|
for outFile in job['outputFiles']:
|
| 146 |
|
|
#print "outputFile:"+outFile
|
| 147 |
|
|
txt += "cp "+outFile+" "+outputDir+"/. \n"
|
| 148 |
|
|
|
| 149 |
|
|
txt += "cd $SGE_O_HOME\n"
|
| 150 |
|
|
txt += "rm -rf $MYTMPDIR\n"
|
| 151 |
|
|
arg = ""
|
| 152 |
|
|
if job[ 'standardInput' ] != '':
|
| 153 |
|
|
arg += ' -i %s ' % job[ 'standardInput' ]
|
| 154 |
|
|
|
| 155 |
|
|
#delete old log files as SGE will append to the file
|
| 156 |
|
|
if os.path.exists(job[ 'standardOutput' ]):
|
| 157 |
|
|
os.remove(job[ 'standardOutput' ])
|
| 158 |
|
|
|
| 159 |
|
|
if os.path.exists(job[ 'standardError' ]):
|
| 160 |
|
|
os.remove(job[ 'standardError' ])
|
| 161 |
|
|
|
| 162 |
|
|
arg += ' -o %s ' % job[ 'standardOutput' ]
|
| 163 |
|
|
arg += ' -e %s ' % job[ 'standardError' ]
|
| 164 |
|
|
|
| 165 |
|
|
# jobrundir = outputDir
|
| 166 |
|
|
# jobrundir += "/%s" % job['id']
|
| 167 |
|
|
# if not os.path.exists(jobrundir):
|
| 168 |
|
|
# os.mkdir(jobrundir)
|
| 169 |
|
|
|
| 170 |
|
|
arg +='-wd '+outputDir+ ' '
|
| 171 |
|
|
# txt += "rm -rf "+jobrundir+"\n"
|
| 172 |
|
|
# blindly append user requirements
|
| 173 |
|
|
arg += requirements
|
| 174 |
|
|
arg += '-N '+task[ 'name' ]+' '
|
| 175 |
|
|
#create job script
|
| 176 |
|
|
f = open(outputDir+'/batchscript', 'wc')
|
| 177 |
|
|
f.write("#!/bin/sh\n")
|
| 178 |
|
|
f.write(txt)
|
| 179 |
|
|
f.close()
|
| 180 |
|
|
|
| 181 |
|
|
# and finally the wrapper
|
| 182 |
|
|
#arg += ' %s ' % txt
|
| 183 |
|
|
arg += outputDir+"/batchscript"
|
| 184 |
|
|
return arg
|
| 185 |
|
|
|
| 186 |
|
|
|
| 187 |
|
|
##########################################################################
|
| 188 |
|
|
def query(self, obj, service='', objType='node') :
|
| 189 |
slacapra |
1.1 |
"""
|
| 190 |
spiga |
1.6 |
query status and eventually other scheduler related information
|
| 191 |
slacapra |
1.1 |
"""
|
| 192 |
|
|
|
| 193 |
spiga |
1.6 |
# the object passed is a Task:
|
| 194 |
|
|
# check whether parent id are provided, make a list of ids
|
| 195 |
|
|
# and check the status
|
| 196 |
|
|
if type(obj) == Task :
|
| 197 |
|
|
schedIds = []
|
| 198 |
|
|
|
| 199 |
|
|
# query performed through single job ids
|
| 200 |
|
|
if objType == 'node' :
|
| 201 |
|
|
for job in obj.jobs :
|
| 202 |
|
|
if self.valid( job.runningJob ) and \
|
| 203 |
|
|
job.runningJob['status'] != 'SD':
|
| 204 |
|
|
schedIds.append( job.runningJob['schedulerId'] )
|
| 205 |
|
|
jobAttributes = self.queryLocal( schedIds, objType )
|
| 206 |
|
|
|
| 207 |
|
|
# query performed through a bulk id
|
| 208 |
|
|
elif objType == 'parent' :
|
| 209 |
|
|
for job in obj.jobs :
|
| 210 |
|
|
if self.valid( job.runningJob ) \
|
| 211 |
|
|
and job.runningJob['status'] != 'SD' \
|
| 212 |
|
|
and job.runningJob['schedulerParentId'] not in schedIds:
|
| 213 |
|
|
schedIds.append( job.runningJob['schedulerParentId'] )
|
| 214 |
|
|
jobAttributes = self.queryLocal( schedIds, objType )
|
| 215 |
|
|
|
| 216 |
|
|
# status association
|
| 217 |
|
|
for job in obj.jobs :
|
| 218 |
|
|
try:
|
| 219 |
|
|
valuesMap = jobAttributes[ job.runningJob['schedulerId'] ]
|
| 220 |
|
|
except:
|
| 221 |
|
|
continue
|
| 222 |
|
|
for key, value in valuesMap.iteritems() :
|
| 223 |
|
|
job.runningJob[key] = value
|
| 224 |
|
|
|
| 225 |
|
|
# unknown object type
|
| 226 |
|
|
else:
|
| 227 |
|
|
raise SchedulerError('wrong argument type', str( type(obj) ))
|
| 228 |
slacapra |
1.1 |
|
| 229 |
spiga |
1.3 |
|
| 230 |
slacapra |
1.1 |
|
| 231 |
spiga |
1.6 |
def queryLocal(self, schedIdList, objType='node' ) :
|
| 232 |
slacapra |
1.1 |
|
| 233 |
|
|
"""
|
| 234 |
spiga |
1.6 |
query status and eventually other scheduler related information
|
| 235 |
|
|
It may use single 'node' scheduler id or bulk id for association
|
| 236 |
|
|
|
| 237 |
|
|
return jobAttributes
|
| 238 |
|
|
|
| 239 |
|
|
where jobAttributes is a map of the format:
|
| 240 |
|
|
jobAttributes[ schedId :
|
| 241 |
|
|
[ key : val ]
|
| 242 |
|
|
]
|
| 243 |
|
|
where key can be any parameter of the Job object and at least status
|
| 244 |
|
|
|
| 245 |
slacapra |
1.1 |
"""
|
| 246 |
spiga |
1.6 |
ret_map={}
|
| 247 |
|
|
#print schedIdList, service, objType
|
| 248 |
|
|
r = re.compile("(\d+) .* "+os.getlogin()+" \W+(\w+) .* (\S+)@(\w+)")
|
| 249 |
|
|
rnohost = re.compile("(\d+) .* "+os.getlogin()+" \W+(\w+) ")
|
| 250 |
|
|
cmd='qstat -u '+os.getlogin()
|
| 251 |
|
|
#print cmd
|
| 252 |
|
|
out, ret = self.ExecuteCommand(cmd)
|
| 253 |
|
|
#print "<"+out+">"
|
| 254 |
|
|
for line in out.split('\n'):
|
| 255 |
|
|
#print line
|
| 256 |
|
|
queue=None
|
| 257 |
|
|
host=None
|
| 258 |
|
|
id=None
|
| 259 |
|
|
st=None
|
| 260 |
|
|
mfull= r.search(line)
|
| 261 |
|
|
if (mfull):
|
| 262 |
|
|
#print "matched"
|
| 263 |
|
|
id,st,queue,host=mfull.groups()
|
| 264 |
|
|
else:
|
| 265 |
|
|
mnohost = rnohost.search(line)
|
| 266 |
|
|
if (mnohost):
|
| 267 |
|
|
id,st = mnohost.groups()
|
| 268 |
|
|
pass
|
| 269 |
|
|
pass
|
| 270 |
|
|
#print "got id %s" % id
|
| 271 |
|
|
if(id) and (id in schedIdList):
|
| 272 |
|
|
map={}
|
| 273 |
|
|
map[ 'status' ] = self.statusMap[st]
|
| 274 |
|
|
map[ 'statusScheduler' ] = st
|
| 275 |
|
|
if (host): map[ 'destination' ] = host
|
| 276 |
|
|
ret_map[id]=map
|
| 277 |
|
|
#print "set state to "+map['statusScheduler']
|
| 278 |
|
|
pass
|
| 279 |
|
|
pass
|
| 280 |
|
|
|
| 281 |
|
|
#set all missing jobs to done
|
| 282 |
|
|
for jobid in schedIdList:
|
| 283 |
|
|
jobid = jobid.strip()
|
| 284 |
|
|
if not jobid in ret_map:
|
| 285 |
|
|
#print "job "+jobid+" not found in qstat list setting to DONE"
|
| 286 |
|
|
id = jobid
|
| 287 |
|
|
st = "DONE"
|
| 288 |
|
|
map={}
|
| 289 |
|
|
map[ 'status' ] = self.statusMap[st]
|
| 290 |
|
|
map[ 'statusScheduler' ]= st
|
| 291 |
|
|
ret_map[id]=map
|
| 292 |
|
|
pass
|
| 293 |
|
|
pass
|
| 294 |
|
|
|
| 295 |
|
|
return ret_map
|
| 296 |
|
|
|
| 297 |
slacapra |
1.1 |
|
| 298 |
spiga |
1.6 |
def getOutput( self, obj, outdir):
|
| 299 |
|
|
"""
|
| 300 |
|
|
retrieve output or just put it in the destination directory
|
| 301 |
slacapra |
1.1 |
|
| 302 |
spiga |
1.6 |
does not return
|
| 303 |
|
|
"""
|
| 304 |
|
|
#output ends up in the wrong location with a user defined
|
| 305 |
|
|
#output directory...Thus we have to move it to the correct
|
| 306 |
|
|
#directory here....
|
| 307 |
|
|
#print "SchedulerSGE:getOutput called!"
|
| 308 |
|
|
|
| 309 |
|
|
if type(obj) == Task :
|
| 310 |
|
|
oldoutdir=obj[ 'outputDirectory' ]
|
| 311 |
|
|
if(outdir != oldoutdir):
|
| 312 |
|
|
for job in obj.jobs:
|
| 313 |
|
|
jobid = job[ 'id' ];
|
| 314 |
|
|
#print "job:"+str(jobid)
|
| 315 |
|
|
if self.valid( job.runningJob ):
|
| 316 |
|
|
#print "is valid"
|
| 317 |
|
|
for outFile in job['outputFiles']:
|
| 318 |
|
|
#print "outputFile:"+outFile
|
| 319 |
|
|
command = "mv "+oldoutdir+"/"+outFile+" "+outdir+"/. \n"
|
| 320 |
|
|
#print command
|
| 321 |
|
|
out, ret = self.ExecuteCommand(command)
|
| 322 |
|
|
if (out!=""):
|
| 323 |
|
|
raise SchedulerError('unable to move file', out)
|
| 324 |
|
|
#raise SchedulerError("unable to move file "+oldoutdir+"/"+outFile+" ",out)
|
| 325 |
|
|
pass
|
| 326 |
|
|
pass
|
| 327 |
|
|
pass
|
| 328 |
|
|
pass
|
| 329 |
|
|
pass
|
| 330 |
|
|
|
| 331 |
spiga |
1.2 |
|
| 332 |
spiga |
1.6 |
def kill( self, obj ):
|
| 333 |
|
|
"""
|
| 334 |
|
|
kill the job instance
|
| 335 |
slacapra |
1.1 |
|
| 336 |
spiga |
1.6 |
does not return
|
| 337 |
|
|
"""
|
| 338 |
|
|
r = re.compile("has registered the job (\d+) for deletion")
|
| 339 |
|
|
rFinished = re.compile("Job <(\d+)>: Job has already finished")
|
| 340 |
|
|
# for jobid in schedIdList:
|
| 341 |
|
|
for job in obj.jobs:
|
| 342 |
|
|
if not self.valid( job.runningJob ):
|
| 343 |
|
|
continue
|
| 344 |
|
|
jobid = str( job.runningJob[ 'schedulerId' ] ).strip()
|
| 345 |
|
|
cmd='qdel '+str(jobid)
|
| 346 |
|
|
out, ret = self.ExecuteCommand(cmd)
|
| 347 |
|
|
#print "kill:"+out
|
| 348 |
|
|
mKilled= r.search(out)
|
| 349 |
|
|
if not mKilled:
|
| 350 |
|
|
raise SchedulerError ( "Unable to kill job "+jobid+" . Reason: ", out )
|
| 351 |
|
|
pass
|
| 352 |
|
|
pass
|
| 353 |
slacapra |
1.1 |
|
| 354 |
spiga |
1.6 |
def postMortem ( self, schedIdList, outfile, service ) :
|
| 355 |
spiga |
1.3 |
"""
|
| 356 |
spiga |
1.6 |
execute any post mortem command such as logging-info
|
| 357 |
|
|
and write it in outfile
|
| 358 |
ewv |
1.5 |
"""
|
| 359 |
|
|
|
