1 // Copyright (C) 2007-2011 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_Constants.hxx"
51 #include "Batch_BatchManager_eLSF.hxx"
52 #include "Batch_JobInfo_eLSF.hxx"
58 BatchManager_eLSF::BatchManager_eLSF(const FactBatchManager * parent, const char * host,
59 const char * username,
60 CommunicationProtocolType protocolType, const char * mpiImpl)
61 : BatchManager(parent, host),
62 BatchManager_eClient(parent, host, username, protocolType, mpiImpl)
69 BatchManager_eLSF::~BatchManager_eLSF()
74 // Methode pour le controle des jobs : soumet un job au gestionnaire
75 const JobId BatchManager_eLSF::submitJob(const Job & job)
78 Parametre params = job.getParametre();
79 const std::string workDir = params[WORKDIR];
80 const string fileToExecute = params[EXECUTABLE];
81 string::size_type p1 = fileToExecute.find_last_of("/");
82 string::size_type p2 = fileToExecute.find_last_of(".");
83 std::string fileNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
85 // export input files on cluster
86 cerr << "Export des fichiers en entree" << endl;
87 exportInputFiles(job);
89 // build batch script for job
90 cerr << "Construction du script de batch" << endl;
91 buildBatchScript(job);
92 cerr << "Script envoye" << endl;
94 // define name of log file (local)
95 string logFile = generateTemporaryFileName("LSF-submitlog");
97 // define command to submit batch
98 string subCommand = string("cd ") + workDir + "; bsub < " + fileNameToExecute + "_Batch.sh";
99 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
103 cerr << command.c_str() << endl;
104 status = system(command.c_str());
107 ifstream error_message(logFile.c_str());
110 while(std::getline(error_message, temp))
112 error_message.close();
113 throw EmulationException("Error of connection on remote host, error was: " + mess);
116 // read id of submitted job in log file
118 FILE *fp = fopen(logFile.c_str(),"r");
119 fgets( line, 128, fp);
123 int p10 = sline.find("<");
124 int p20 = sline.find(">");
125 string strjob = sline.substr(p10+1,p20-p10-1);
127 JobId id(this, strjob);
131 // Ce manager permet de faire de la reprise
133 BatchManager_eLSF::addJob(const Batch::Job & job, const std::string reference)
135 return JobId(this, reference);
138 // Methode pour le controle des jobs : retire un job du gestionnaire
139 void BatchManager_eLSF::deleteJob(const JobId & jobid)
143 istringstream iss(jobid.getReference());
146 // define command to delete batch
147 string subCommand = string("bkill ") + iss.str();
148 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
149 cerr << command.c_str() << endl;
150 status = system(command.c_str());
152 throw EmulationException("Error of connection on remote host");
154 cerr << "jobId = " << ref << "killed" << endl;
157 // Methode pour le controle des jobs : suspend un job en file d'attente
158 void BatchManager_eLSF::holdJob(const JobId & jobid)
160 throw EmulationException("Not yet implemented");
163 // Methode pour le controle des jobs : relache un job suspendu
164 void BatchManager_eLSF::releaseJob(const JobId & jobid)
166 throw EmulationException("Not yet implemented");
170 // Methode pour le controle des jobs : modifie un job en file d'attente
171 void BatchManager_eLSF::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
173 throw EmulationException("Not yet implemented");
176 // Methode pour le controle des jobs : modifie un job en file d'attente
177 void BatchManager_eLSF::alterJob(const JobId & jobid, const Parametre & param)
179 alterJob(jobid, param, Environnement());
182 // Methode pour le controle des jobs : modifie un job en file d'attente
183 void BatchManager_eLSF::alterJob(const JobId & jobid, const Environnement & env)
185 alterJob(jobid, Parametre(), env);
188 // Methode pour le controle des jobs : renvoie l'etat du job
189 JobInfo BatchManager_eLSF::queryJob(const JobId & jobid)
192 istringstream iss(jobid.getReference());
195 // define name of log file (local)
196 string logFile = generateTemporaryFileName(string("LSF-querylog-id") + jobid.getReference());
198 // define command to query batch
199 string subCommand = string("bjobs ") + iss.str();
200 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
203 cerr << command.c_str() << endl;
204 int status = system(command.c_str());
206 throw EmulationException("Error of connection on remote host");
208 JobInfo_eLSF ji = JobInfo_eLSF(id,logFile);
214 // Methode pour le controle des jobs : teste si un job est present en machine
215 bool BatchManager_eLSF::isRunning(const JobId & jobid)
217 throw EmulationException("Not yet implemented");
220 void BatchManager_eLSF::buildBatchScript(const Job & job)
222 #ifndef WIN32 //TODO: need for porting on Windows
223 Parametre params = job.getParametre();
227 string fileToExecute = "";
233 // Mandatory parameters
234 if (params.find(WORKDIR) != params.end())
235 workDir = params[WORKDIR].str();
237 throw EmulationException("params[WORKDIR] is not defined ! Please defined it, cannot submit this job");
238 if (params.find(EXECUTABLE) != params.end())
239 fileToExecute = params[EXECUTABLE].str();
241 throw EmulationException("params[EXECUTABLE] is not defined ! Please defined it, cannot submit this job");
243 // Optional parameters
244 if (params.find(NBPROC) != params.end())
245 nbproc = params[NBPROC];
246 if (params.find(MAXWALLTIME) != params.end())
247 edt = params[MAXWALLTIME];
248 if (params.find(MAXRAMSIZE) != params.end())
249 mem = params[MAXRAMSIZE];
250 if (params.find(QUEUE) != params.end())
251 queue = params[QUEUE].str();
253 string::size_type p1 = fileToExecute.find_last_of("/");
254 string::size_type p2 = fileToExecute.find_last_of(".");
255 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
256 string fileNameToExecute = fileToExecute.substr(p1+1);
258 // Create batch submit file
259 ofstream tempOutputFile;
260 std::string TmpFileName = createAndOpenTemporaryFile("LSF-script", tempOutputFile);
262 tempOutputFile << "#! /bin/sh -f" << endl ;
264 tempOutputFile << "#BSUB -q " << queue << endl;
266 tempOutputFile << "#BSUB -W " << getWallTime(edt) << endl ;
268 tempOutputFile << "#BSUB -M " << mem*1024 << endl ;
269 tempOutputFile << "#BSUB -n " << nbproc << endl ;
270 size_t pos = workDir.find("$HOME");
272 if( pos != string::npos )
273 baseDir = getHomeDir(workDir) + workDir.substr(pos+5,workDir.length()-5);
275 pos = workDir.find("~");
276 if( pos != string::npos )
277 baseDir = getHomeDir(workDir) + workDir.substr(pos+1,workDir.length()-1);
281 tempOutputFile << "#BSUB -o " << baseDir << "/logs/output.log." << rootNameToExecute << endl ;
282 tempOutputFile << "#BSUB -e " << baseDir << "/logs/error.log." << rootNameToExecute << endl ;
284 tempOutputFile << "cd " << workDir << endl ;
286 // generate nodes file
287 tempOutputFile << "NODEFILE=`mktemp nodefile-XXXXXXXXXX` || exit 1" << endl;
288 tempOutputFile << "bool=0" << endl;
289 tempOutputFile << "for i in $LSB_MCPU_HOSTS; do" << endl;
290 tempOutputFile << " if test $bool = 0; then" << endl;
291 tempOutputFile << " n=$i" << endl;
292 tempOutputFile << " bool=1" << endl;
293 tempOutputFile << " else" << endl;
294 tempOutputFile << " for ((j=0;j<$i;j++)); do" << endl;
295 tempOutputFile << " echo $n >> $NODEFILE" << endl;
296 tempOutputFile << " done" << endl;
297 tempOutputFile << " bool=0" << endl;
298 tempOutputFile << " fi" << endl;
299 tempOutputFile << "done" << endl;
301 // Abstraction of PBS_NODEFILE - TODO
302 tempOutputFile << "export LIBBATCH_NODEFILE=$NODEFILE" << endl;
304 // Launch the executable
305 tempOutputFile << "./" + fileNameToExecute << endl;
307 // Remove the node file
308 tempOutputFile << "rm $NODEFILE" << endl;
310 tempOutputFile.flush();
311 tempOutputFile.close();
313 BATCH_CHMOD(TmpFileName.c_str(), 0x1ED);
314 cerr << "Batch script file generated is: " << TmpFileName.c_str() << endl;
316 int status = _protocol.copyFile(TmpFileName, "", "",
317 workDir + "/" + rootNameToExecute + "_Batch.sh",
318 _hostname, _username);
320 throw EmulationException("Error of connection on remote host");
326 std::string BatchManager_eLSF::getWallTime(const long edt)
333 oss << h << ":" << m;
335 oss << h << ":0" << m;
339 std::string BatchManager_eLSF::getHomeDir(std::string tmpdir)
342 int idx = tmpdir.find("Batch/");
343 std::string filelogtemp = tmpdir.substr(idx+6, tmpdir.length());
344 filelogtemp = "/tmp/logs" + filelogtemp + "_home";
346 string subCommand = string("echo $HOME");
347 string command = _protocol.getExecCommand(subCommand, _hostname, _username) + " > " + filelogtemp;
348 cerr << command.c_str() << endl;
349 int status = system(command.c_str());
351 throw EmulationException("Error of launching home command on remote host");
353 std::ifstream file_home(filelogtemp.c_str());
354 std::getline(file_home, home);