[22755] | 1 | Index: ../trunk-jpl/src/m/classes/clusters/stallo.py
|
---|
| 2 | ===================================================================
|
---|
| 3 | --- ../trunk-jpl/src/m/classes/clusters/stallo.py (nonexistent)
|
---|
| 4 | +++ ../trunk-jpl/src/m/classes/clusters/stallo.py (revision 21987)
|
---|
| 5 | @@ -0,0 +1,171 @@
|
---|
| 6 | +import subprocess
|
---|
| 7 | +from fielddisplay import fielddisplay
|
---|
| 8 | +from pairoptions import pairoptions
|
---|
| 9 | +from issmssh import issmssh
|
---|
| 10 | +from issmscpin import issmscpin
|
---|
| 11 | +from issmscpout import issmscpout
|
---|
| 12 | +from QueueRequirements import QueueRequirements
|
---|
| 13 | +import datetime
|
---|
| 14 | +try:
|
---|
| 15 | + from stallo_settings import stallo_settings
|
---|
| 16 | +except ImportError:
|
---|
| 17 | + print 'You need stallo_settings.py to proceed, check presence and sys.path'
|
---|
| 18 | +
|
---|
| 19 | +class stallo(object):
|
---|
| 20 | + """
|
---|
| 21 | + Stallo cluster class definition
|
---|
| 22 | + This is a SLURM queue
|
---|
| 23 | + The priorities are given to:
|
---|
| 24 | + - Large jobs
|
---|
| 25 | + - Short jobs
|
---|
| 26 | + - small number of job per user
|
---|
| 27 | +
|
---|
| 28 | + There are some 20cpu nodes and 16cpu nodes, with 32GB (a few with 128GB) mem per node, you can ask for part of a node if you need more memory.(1 node, 2 CPUS and 10GB per cpu for example)
|
---|
| 29 | +
|
---|
| 30 | +
|
---|
| 31 | + Usage:
|
---|
| 32 | + cluster=stallo();
|
---|
| 33 | + """
|
---|
| 34 | +
|
---|
| 35 | + def __init__(self,*args):
|
---|
| 36 | + # {{{
|
---|
| 37 | + self.name = 'stallo'
|
---|
| 38 | + self.login = ''
|
---|
| 39 | + self.numnodes = 2
|
---|
| 40 | + self.cpuspernode = 20
|
---|
| 41 | + self.mem = 1.6
|
---|
| 42 | + self.queue = 'normal'
|
---|
| 43 | + self.time = 2*60
|
---|
| 44 | + self.codepath = ''
|
---|
| 45 | + self.executionpath = ''
|
---|
| 46 | + self.interactive = 0
|
---|
| 47 | + self.port = []
|
---|
| 48 | + self.accountname = ''
|
---|
| 49 | + self.profiling = 0
|
---|
| 50 | + #use provided options to change fields
|
---|
| 51 | + options=pairoptions(*args)
|
---|
| 52 | +
|
---|
| 53 | + #initialize cluster using user settings if provided
|
---|
| 54 | + self=vilje_settings(self)
|
---|
| 55 | + #OK get other fields
|
---|
| 56 | + self=options.AssignObjectFields(self)
|
---|
| 57 | + self.np=self.numnodes*self.procspernodes
|
---|
| 58 | + # }}}
|
---|
| 59 | +
|
---|
| 60 | + def __repr__(self):
|
---|
| 61 | + # {{{
|
---|
| 62 | + # display the object
|
---|
| 63 | + s = "class vilje object:"
|
---|
| 64 | + s = "%s\n%s"%(s,fielddisplay(self,'name','name of the cluster'))
|
---|
| 65 | + s = "%s\n%s"%(s,fielddisplay(self,'login','login'))
|
---|
| 66 | + s = "%s\n%s"%(s,fielddisplay(self,'numnodes','number of nodes'))
|
---|
| 67 | + s = "%s\n%s"%(s,fielddisplay(self,'cpuspernode','number of nodes per CPUs'))
|
---|
| 68 | + s = "%s\n%s"%(s,fielddisplay(self,'mem','memory per CPU'))
|
---|
| 69 | + s = "%s\n%s"%(s,fielddisplay(self,'queue','name of the queue (normal (D), short,singlenode,multinode,devel)'))
|
---|
| 70 | + s = "%s\n%s"%(s,fielddisplay(self,'time','walltime requested in minutes'))
|
---|
| 71 | + s = "%s\n%s"%(s,fielddisplay(self,'codepath','code path on the cluster'))
|
---|
| 72 | + s = "%s\n%s"%(s,fielddisplay(self,'executionpath','execution path on the cluster'))
|
---|
| 73 | + s = "%s\n%s"%(s,fielddisplay(self,'interactive',''))
|
---|
| 74 | + s = "%s\n%s"%(s,fielddisplay(self,'accountname','your cluster account'))
|
---|
| 75 | + s = "%s\n%s"%(s,fielddisplay(self,'profiling','enable profiling if 1 default is 0'))
|
---|
| 76 | + return s
|
---|
| 77 | + # }}}
|
---|
| 78 | + def checkconsistency(self,md,solution,analyses):
|
---|
| 79 | + # {{{
|
---|
| 80 | + #Queue dictionarry gives queue name as key and max walltime and cpus as var
|
---|
| 81 | + queuedict = {'short':[60, 2048],
|
---|
| 82 | + 'normal':[2*24*60,2048],
|
---|
| 83 | + 'singlenode':[28*24*60,20],
|
---|
| 84 | + 'multinode':[28*24*60,2048],
|
---|
| 85 | + 'devel':[4*60,2048]}
|
---|
| 86 | + QueueRequirements(queuedict,self.queue,self.np,self.time)
|
---|
| 87 | +
|
---|
| 88 | + #Miscelaneous
|
---|
| 89 | + if not self.login:
|
---|
| 90 | + md = md.checkmessage('login empty')
|
---|
| 91 | + if not self.codepath:
|
---|
| 92 | + md = md.checkmessage('codepath empty')
|
---|
| 93 | + if not self.executionpath:
|
---|
| 94 | + md = md.checkmessage('executionpath empty')
|
---|
| 95 | + if self.interactive==1:
|
---|
| 96 | + md = md.checkmessage('interactive mode not implemented')
|
---|
| 97 | + return self
|
---|
| 98 | + # }}}
|
---|
| 99 | + def BuildQueueScript(self,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling):
|
---|
| 100 | + # {{{
|
---|
| 101 | +
|
---|
| 102 | + executable='issm.exe'
|
---|
| 103 | + if isdakota:
|
---|
| 104 | + version=IssmConfig('_DAKOTA_VERSION_')[0:2]
|
---|
| 105 | + version=float(version)
|
---|
| 106 | + if version>=6:
|
---|
| 107 | + executable='issm_dakota.exe'
|
---|
| 108 | + if isoceancoupling:
|
---|
| 109 | + executable='issm_ocean.exe'
|
---|
| 110 | + #write queuing script
|
---|
| 111 | + shortname=modelname[0:min(12,len(modelname))]
|
---|
| 112 | + fid=open(modelname+'.queue','w')
|
---|
| 113 | +
|
---|
| 114 | + fid.write('#!/bin/bash -l\n')
|
---|
| 115 | + fid.write('#SBATCH --job-name=%s \n' % shortname)
|
---|
| 116 | + fid.write('#SBATCH --partition %s \n' % self.queue)
|
---|
| 117 | + fid.write('#SBATCH --nodes=%i' % self.numnodes)
|
---|
| 118 | + fid.write('#SBATCH --ntasks-per-nodes==%i' % self.cpuspernode)
|
---|
| 119 | + fid.write('#SBATCH --time=%s\n' % timestring) #walltime is minutes
|
---|
| 120 | + fid.write('#SBATCH --mem-per-cpu=%iGB\n' % self.mem)# mem is in GB
|
---|
| 121 | + if (mod(self.np,16)+mod(self.np,20))==0:
|
---|
| 122 | + fid.write('#SBATCH --ntask=%i\n' % self.np)
|
---|
| 123 | + fid.write('#SBATCH --account=%s\n' % self.accountname)
|
---|
| 124 | + fid.write('#SBATCH --output %s/%s/%s.outlog \n' % (self.executionpath,dirname,modelname))
|
---|
| 125 | + fid.write('#SBATCH --error %s/%s/%s.errlog \n\n' % (self.executionpath,dirname,modelname))
|
---|
| 126 | +
|
---|
| 127 | + fid.write('export ISSM_DIR="%s/../"\n' % self.codepath)
|
---|
| 128 | + fid.write('module load Automake/1.15-intel-2016a\n')
|
---|
| 129 | + fid.write('module load libtool/2.4.6-intel-2016a\n')
|
---|
| 130 | + fid.write('module load CMake/3.5.2-intel-2016a\n')
|
---|
| 131 | + fid.write('module load intel/2016a\n')
|
---|
| 132 | + fid.write('module load ParMETIS/4.0.3-intel-2016a\n')
|
---|
| 133 | + fid.write('module load MUMPS/5.1.1-intel-2016a-parmetis\n')
|
---|
| 134 | + fid.write('module load PETSc/3.7.2-intel-2016a-Python-2.7.11\n')
|
---|
| 135 | + fid.write('module load FFTW/3.3.4-intel-2016a\n')
|
---|
| 136 | + fid.write('module load OpenSSL/1.0.1s-intel-2016a\n')
|
---|
| 137 | +
|
---|
| 138 | + fid.write('cd %s/%s/\n\n' % (self.executionpath,dirname))
|
---|
| 139 | + if self.profiling==1:
|
---|
| 140 | + fid.write('module load perf-report\n')
|
---|
| 141 | + fid.write('perf-report mpirun -np %i %s/%s %s %s/%s %s\n' % (self.np,self.codepath,executable,str(solution),self.executionpath,dirname,modelname))
|
---|
| 142 | + else:
|
---|
| 143 | + fid.write('mpirun -np %i %s/%s %s %s/%s %s\n' % (self.np,self.codepath,executable,str(solution),self.executionpath,dirname,modelname))
|
---|
| 144 | + fid.close()
|
---|
| 145 | +
|
---|
| 146 | + # }}}
|
---|
| 147 | + def UploadQueueJob(self,modelname,dirname,filelist):
|
---|
| 148 | + # {{{
|
---|
| 149 | + #compress the files into one zip.
|
---|
| 150 | + compressstring='tar -zcf %s.tar.gz ' % dirname
|
---|
| 151 | + for file in filelist:
|
---|
| 152 | + compressstring += ' %s' % file
|
---|
| 153 | + subprocess.call(compressstring,shell=True)
|
---|
| 154 | +
|
---|
| 155 | + print 'uploading input file and queueing script'
|
---|
| 156 | + issmscpout(self.name,self.executionpath,self.login,self.port,[dirname+'.tar.gz'])
|
---|
| 157 | +
|
---|
| 158 | + # }}}
|
---|
| 159 | + def LaunchQueueJob(self,modelname,dirname,filelist,restart,batch):
|
---|
| 160 | + # {{{
|
---|
| 161 | +
|
---|
| 162 | + print 'launching solution sequence on remote cluster'
|
---|
| 163 | + if restart:
|
---|
| 164 | + launchcommand='cd %s && cd %s && sbatch %s.queue' % (self.executionpath,dirname,modelname)
|
---|
| 165 | + else:
|
---|
| 166 | + launchcommand='cd %s && rm -rf ./%s && mkdir %s && cd %s && mv ../%s.tar.gz ./ && tar -zxf %s.tar.gz && sbatch %s.queue' % (self.executionpath,dirname,dirname,dirname,dirname,dirname,modelname)
|
---|
| 167 | + issmssh(self.name,self.login,self.port,launchcommand)
|
---|
| 168 | +
|
---|
| 169 | + # }}}
|
---|
| 170 | + def Download(self,dirname,filelist):
|
---|
| 171 | + # {{{
|
---|
| 172 | +
|
---|
| 173 | + #copy files from cluster to current directory
|
---|
| 174 | + directory='%s/%s/' % (self.executionpath,dirname)
|
---|
| 175 | + issmscpin(self.name,self.login,self.port,directory,filelist)
|
---|
| 176 | + # }}}
|
---|