1 // Copyright (C) 2007-2008 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 // Methode pour le controle des jobs : retire un job du gestionnaire
129 void BatchManager_eLSF::deleteJob(const JobId & jobid)
133 istringstream iss(jobid.getReference());
136 // define command to delete batch
137 string subCommand = string("bkill ") + iss.str();
138 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
139 cerr << command.c_str() << endl;
140 status = system(command.c_str());
142 throw EmulationException("Error of connection on remote host");
144 cerr << "jobId = " << ref << "killed" << endl;
147 // Methode pour le controle des jobs : suspend un job en file d'attente
148 void BatchManager_eLSF::holdJob(const JobId & jobid)
150 throw EmulationException("Not yet implemented");
153 // Methode pour le controle des jobs : relache un job suspendu
154 void BatchManager_eLSF::releaseJob(const JobId & jobid)
156 throw EmulationException("Not yet implemented");
160 // Methode pour le controle des jobs : modifie un job en file d'attente
161 void BatchManager_eLSF::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
163 throw EmulationException("Not yet implemented");
166 // Methode pour le controle des jobs : modifie un job en file d'attente
167 void BatchManager_eLSF::alterJob(const JobId & jobid, const Parametre & param)
169 alterJob(jobid, param, Environnement());
172 // Methode pour le controle des jobs : modifie un job en file d'attente
173 void BatchManager_eLSF::alterJob(const JobId & jobid, const Environnement & env)
175 alterJob(jobid, Parametre(), env);
178 // Methode pour le controle des jobs : renvoie l'etat du job
179 JobInfo BatchManager_eLSF::queryJob(const JobId & jobid)
182 istringstream iss(jobid.getReference());
185 // define name of log file (local)
186 string logFile = generateTemporaryFileName(string("LSF-querylog-id") + jobid.getReference());
188 // define command to query batch
189 string subCommand = string("bjobs ") + iss.str();
190 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
193 cerr << command.c_str() << endl;
194 int status = system(command.c_str());
196 throw EmulationException("Error of connection on remote host");
198 JobInfo_eLSF ji = JobInfo_eLSF(id,logFile);
204 // Methode pour le controle des jobs : teste si un job est present en machine
205 bool BatchManager_eLSF::isRunning(const JobId & jobid)
207 throw EmulationException("Not yet implemented");
210 void BatchManager_eLSF::buildBatchScript(const Job & job)
212 #ifndef WIN32 //TODO: need for porting on Windows
213 Parametre params = job.getParametre();
217 string fileToExecute = "";
223 // Mandatory parameters
224 if (params.find(WORKDIR) != params.end())
225 workDir = params[WORKDIR].str();
227 throw EmulationException("params[WORKDIR] is not defined ! Please defined it, cannot submit this job");
228 if (params.find(EXECUTABLE) != params.end())
229 fileToExecute = params[EXECUTABLE].str();
231 throw EmulationException("params[EXECUTABLE] is not defined ! Please defined it, cannot submit this job");
233 // Optional parameters
234 if (params.find(NBPROC) != params.end())
235 nbproc = params[NBPROC];
236 if (params.find(MAXWALLTIME) != params.end())
237 edt = params[MAXWALLTIME];
238 if (params.find(MAXRAMSIZE) != params.end())
239 mem = params[MAXRAMSIZE];
240 if (params.find(QUEUE) != params.end())
241 queue = params[QUEUE].str();
243 string::size_type p1 = fileToExecute.find_last_of("/");
244 string::size_type p2 = fileToExecute.find_last_of(".");
245 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
246 string fileNameToExecute = fileToExecute.substr(p1+1);
248 // Create batch submit file
249 ofstream tempOutputFile;
250 std::string TmpFileName = createAndOpenTemporaryFile("LSF-script", tempOutputFile);
252 tempOutputFile << "#! /bin/sh -f" << endl ;
254 tempOutputFile << "#BSUB -q " << queue << endl;
256 tempOutputFile << "#BSUB -W " << getWallTime(edt) << endl ;
258 tempOutputFile << "#BSUB -M " << mem*1024 << endl ;
259 tempOutputFile << "#BSUB -n " << nbproc << endl ;
260 size_t pos = workDir.find("$HOME");
262 if( pos != string::npos )
263 baseDir = getHomeDir(workDir) + workDir.substr(pos+5,workDir.length()-5);
265 pos = workDir.find("~");
266 if( pos != string::npos )
267 baseDir = getHomeDir(workDir) + workDir.substr(pos+1,workDir.length()-1);
271 tempOutputFile << "#BSUB -o " << baseDir << "/logs/output.log." << rootNameToExecute << endl ;
272 tempOutputFile << "#BSUB -e " << baseDir << "/logs/error.log." << rootNameToExecute << endl ;
274 tempOutputFile << "cd " << workDir << endl ;
276 // generate nodes file
277 tempOutputFile << "bool=0" << endl;
278 tempOutputFile << "for i in $LSB_MCPU_HOSTS; do" << endl;
279 tempOutputFile << " if test $bool = 0; then" << endl;
280 tempOutputFile << " n=$i" << endl;
281 tempOutputFile << " bool=1" << endl;
282 tempOutputFile << " else" << endl;
283 tempOutputFile << " for ((j=0;j<$i;j++)); do" << endl;
284 tempOutputFile << " echo $n >> nodesFile" << endl;
285 tempOutputFile << " done" << endl;
286 tempOutputFile << " bool=0" << endl;
287 tempOutputFile << " fi" << endl;
288 tempOutputFile << "done" << endl;
290 // Abstraction of PBS_NODEFILE - TODO
291 tempOutputFile << "export LIBBATCH_NODEFILE=nodesFile" << endl;
293 // Launch the executable
294 tempOutputFile << "./" + fileNameToExecute << endl;
295 tempOutputFile.flush();
296 tempOutputFile.close();
298 BATCH_CHMOD(TmpFileName.c_str(), 0x1ED);
299 cerr << "Batch script file generated is: " << TmpFileName.c_str() << endl;
301 int status = _protocol.copyFile(TmpFileName, "", "",
302 workDir + "/" + rootNameToExecute + "_Batch.sh",
303 _hostname, _username);
305 throw EmulationException("Error of connection on remote host");
311 std::string BatchManager_eLSF::getWallTime(const long edt)
318 oss << h << ":" << m;
320 oss << h << ":0" << m;
324 std::string BatchManager_eLSF::getHomeDir(std::string tmpdir)
327 int idx = tmpdir.find("Batch/");
328 std::string filelogtemp = tmpdir.substr(idx+6, tmpdir.length());
329 filelogtemp = "/tmp/logs" + filelogtemp + "_home";
331 string subCommand = string("echo $HOME");
332 string command = _protocol.getExecCommand(subCommand, _hostname, _username) + " > " + filelogtemp;
333 cerr << command.c_str() << endl;
334 int status = system(command.c_str());
336 throw EmulationException("Error of launching home command on remote host");
338 std::ifstream file_home(filelogtemp.c_str());
339 std::getline(file_home, home);