1 // Copyright (C) 2007-2010 CEA/DEN, EDF R&D, OPEN CASCADE
3 // Copyright (C) 2003-2007 OPEN CASCADE, EADS/CCR, LIP6, CEA/DEN,
4 // CEDRAT, EDF R&D, LEG, PRINCIPIA R&D, BUREAU VERITAS
6 // This library is free software; you can redistribute it and/or
7 // modify it under the terms of the GNU Lesser General Public
8 // License as published by the Free Software Foundation; either
9 // version 2.1 of the License.
11 // This library is distributed in the hope that it will be useful,
12 // but WITHOUT ANY WARRANTY; without even the implied warranty of
13 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 // Lesser General Public License for more details.
16 // You should have received a copy of the GNU Lesser General Public
17 // License along with this library; if not, write to the Free Software
18 // Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 // See http://www.salome-platform.org/ or email : webmaster.salome@opencascade.com
23 * BatchManager_eLSF.cxx : emulation of LSF client
25 * Auteur : Bernard SECHER - CEA DEN
26 * Mail : mailto:bernard.secher@cea.fr
27 * Date : Thu Apr 24 10:17:22 2008
50 #include "Batch_BatchManager_eLSF.hxx"
51 #include "Batch_JobInfo_eLSF.hxx"
57 BatchManager_eLSF::BatchManager_eLSF(const FactBatchManager * parent, const char * host,
58 CommunicationProtocolType protocolType, const char * mpiImpl)
59 : BatchManager_eClient(parent, host, protocolType, mpiImpl),
60 BatchManager(parent, host)
66 BatchManager_eLSF::~BatchManager_eLSF()
71 // Methode pour le controle des jobs : soumet un job au gestionnaire
72 const JobId BatchManager_eLSF::submitJob(const Job & job)
75 Parametre params = job.getParametre();
76 const std::string workDir = params[WORKDIR];
77 const string fileToExecute = params[EXECUTABLE];
78 string::size_type p1 = fileToExecute.find_last_of("/");
79 string::size_type p2 = fileToExecute.find_last_of(".");
80 std::string fileNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
82 // export input files on cluster
83 cerr << "Export des fichiers en entree" << endl;
84 exportInputFiles(job);
86 // build batch script for job
87 cerr << "Construction du script de batch" << endl;
88 buildBatchScript(job);
89 cerr << "Script envoye" << endl;
91 // define name of log file (local)
92 string logFile = generateTemporaryFileName("LSF-submitlog");
94 // define command to submit batch
95 string subCommand = string("cd ") + workDir + "; bsub < " + fileNameToExecute + "_Batch.sh";
96 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
100 cerr << command.c_str() << endl;
101 status = system(command.c_str());
104 ifstream error_message(logFile.c_str());
107 while(std::getline(error_message, temp))
109 error_message.close();
110 throw EmulationException("Error of connection on remote host, error was: " + mess);
113 // read id of submitted job in log file
115 FILE *fp = fopen(logFile.c_str(),"r");
116 fgets( line, 128, fp);
120 int p10 = sline.find("<");
121 int p20 = sline.find(">");
122 string strjob = sline.substr(p10+1,p20-p10-1);
124 JobId id(this, strjob);
128 // Ce manager permet de faire de la reprise
130 BatchManager_eLSF::addJob(const Batch::Job & job, const std::string reference)
132 return JobId(this, reference);
135 // Methode pour le controle des jobs : retire un job du gestionnaire
136 void BatchManager_eLSF::deleteJob(const JobId & jobid)
140 istringstream iss(jobid.getReference());
143 // define command to delete batch
144 string subCommand = string("bkill ") + iss.str();
145 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
146 cerr << command.c_str() << endl;
147 status = system(command.c_str());
149 throw EmulationException("Error of connection on remote host");
151 cerr << "jobId = " << ref << "killed" << endl;
154 // Methode pour le controle des jobs : suspend un job en file d'attente
155 void BatchManager_eLSF::holdJob(const JobId & jobid)
157 throw EmulationException("Not yet implemented");
160 // Methode pour le controle des jobs : relache un job suspendu
161 void BatchManager_eLSF::releaseJob(const JobId & jobid)
163 throw EmulationException("Not yet implemented");
167 // Methode pour le controle des jobs : modifie un job en file d'attente
168 void BatchManager_eLSF::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
170 throw EmulationException("Not yet implemented");
173 // Methode pour le controle des jobs : modifie un job en file d'attente
174 void BatchManager_eLSF::alterJob(const JobId & jobid, const Parametre & param)
176 alterJob(jobid, param, Environnement());
179 // Methode pour le controle des jobs : modifie un job en file d'attente
180 void BatchManager_eLSF::alterJob(const JobId & jobid, const Environnement & env)
182 alterJob(jobid, Parametre(), env);
185 // Methode pour le controle des jobs : renvoie l'etat du job
186 JobInfo BatchManager_eLSF::queryJob(const JobId & jobid)
189 istringstream iss(jobid.getReference());
192 // define name of log file (local)
193 string logFile = generateTemporaryFileName(string("LSF-querylog-id") + jobid.getReference());
195 // define command to query batch
196 string subCommand = string("bjobs ") + iss.str();
197 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
200 cerr << command.c_str() << endl;
201 int status = system(command.c_str());
203 throw EmulationException("Error of connection on remote host");
205 JobInfo_eLSF ji = JobInfo_eLSF(id,logFile);
211 // Methode pour le controle des jobs : teste si un job est present en machine
212 bool BatchManager_eLSF::isRunning(const JobId & jobid)
214 throw EmulationException("Not yet implemented");
217 void BatchManager_eLSF::buildBatchScript(const Job & job)
219 #ifndef WIN32 //TODO: need for porting on Windows
220 Parametre params = job.getParametre();
224 string fileToExecute = "";
230 // Mandatory parameters
231 if (params.find(WORKDIR) != params.end())
232 workDir = params[WORKDIR].str();
234 throw EmulationException("params[WORKDIR] is not defined ! Please defined it, cannot submit this job");
235 if (params.find(EXECUTABLE) != params.end())
236 fileToExecute = params[EXECUTABLE].str();
238 throw EmulationException("params[EXECUTABLE] is not defined ! Please defined it, cannot submit this job");
240 // Optional parameters
241 if (params.find(NBPROC) != params.end())
242 nbproc = params[NBPROC];
243 if (params.find(MAXWALLTIME) != params.end())
244 edt = params[MAXWALLTIME];
245 if (params.find(MAXRAMSIZE) != params.end())
246 mem = params[MAXRAMSIZE];
247 if (params.find(QUEUE) != params.end())
248 queue = params[QUEUE].str();
250 string::size_type p1 = fileToExecute.find_last_of("/");
251 string::size_type p2 = fileToExecute.find_last_of(".");
252 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
253 string fileNameToExecute = fileToExecute.substr(p1+1);
255 // Create batch submit file
256 ofstream tempOutputFile;
257 std::string TmpFileName = createAndOpenTemporaryFile("LSF-script", tempOutputFile);
259 tempOutputFile << "#! /bin/sh -f" << endl ;
261 tempOutputFile << "#BSUB -q " << queue << endl;
263 tempOutputFile << "#BSUB -W " << getWallTime(edt) << endl ;
265 tempOutputFile << "#BSUB -M " << mem*1024 << endl ;
266 tempOutputFile << "#BSUB -n " << nbproc << endl ;
267 size_t pos = workDir.find("$HOME");
269 if( pos != string::npos )
270 baseDir = getHomeDir(workDir) + workDir.substr(pos+5,workDir.length()-5);
272 pos = workDir.find("~");
273 if( pos != string::npos )
274 baseDir = getHomeDir(workDir) + workDir.substr(pos+1,workDir.length()-1);
278 tempOutputFile << "#BSUB -o " << baseDir << "/logs/output.log." << rootNameToExecute << endl ;
279 tempOutputFile << "#BSUB -e " << baseDir << "/logs/error.log." << rootNameToExecute << endl ;
281 tempOutputFile << "cd " << workDir << endl ;
283 // generate nodes file
284 tempOutputFile << "NODEFILE=`mktemp nodefile-XXXXXXXXXX` || exit 1" << endl;
285 tempOutputFile << "bool=0" << endl;
286 tempOutputFile << "for i in $LSB_MCPU_HOSTS; do" << endl;
287 tempOutputFile << " if test $bool = 0; then" << endl;
288 tempOutputFile << " n=$i" << endl;
289 tempOutputFile << " bool=1" << endl;
290 tempOutputFile << " else" << endl;
291 tempOutputFile << " for ((j=0;j<$i;j++)); do" << endl;
292 tempOutputFile << " echo $n >> $NODEFILE" << endl;
293 tempOutputFile << " done" << endl;
294 tempOutputFile << " bool=0" << endl;
295 tempOutputFile << " fi" << endl;
296 tempOutputFile << "done" << endl;
298 // Abstraction of PBS_NODEFILE - TODO
299 tempOutputFile << "export LIBBATCH_NODEFILE=$NODEFILE" << endl;
301 // Launch the executable
302 tempOutputFile << "./" + fileNameToExecute << endl;
304 // Remove the node file
305 tempOutputFile << "rm $NODEFILE" << endl;
307 tempOutputFile.flush();
308 tempOutputFile.close();
310 BATCH_CHMOD(TmpFileName.c_str(), 0x1ED);
311 cerr << "Batch script file generated is: " << TmpFileName.c_str() << endl;
313 int status = _protocol.copyFile(TmpFileName, "", "",
314 workDir + "/" + rootNameToExecute + "_Batch.sh",
315 _hostname, _username);
317 throw EmulationException("Error of connection on remote host");
323 std::string BatchManager_eLSF::getWallTime(const long edt)
330 oss << h << ":" << m;
332 oss << h << ":0" << m;
336 std::string BatchManager_eLSF::getHomeDir(std::string tmpdir)
339 int idx = tmpdir.find("Batch/");
340 std::string filelogtemp = tmpdir.substr(idx+6, tmpdir.length());
341 filelogtemp = "/tmp/logs" + filelogtemp + "_home";
343 string subCommand = string("echo $HOME");
344 string command = _protocol.getExecCommand(subCommand, _hostname, _username) + " > " + filelogtemp;
345 cerr << command.c_str() << endl;
346 int status = system(command.c_str());
348 throw EmulationException("Error of launching home command on remote host");
350 std::ifstream file_home(filelogtemp.c_str());
351 std::getline(file_home, home);