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_ePBS.cxx : emulation of PBS client
25 * Auteur : Bernard SECHER - CEA DEN, André RIBES - EDF R&D
26 * Mail : mailto:bernard.secher@cea.fr
27 * Date : Thu Apr 24 10:17:22 2008
42 #include <Batch_config.h>
50 #include "Batch_BatchManager_ePBS.hxx"
51 #include "Batch_JobInfo_ePBS.hxx"
57 BatchManager_ePBS::BatchManager_ePBS(const FactBatchManager * parent, const char * host,
58 const char * username,
59 CommunicationProtocolType protocolType, const char * mpiImpl,
61 : BatchManager_eClient(parent, host, username, protocolType, mpiImpl),
62 BatchManager(parent, host)
65 _nb_proc_per_node = nb_proc_per_node;
69 BatchManager_ePBS::~BatchManager_ePBS()
74 // Methode pour le controle des jobs : soumet un job au gestionnaire
75 const JobId BatchManager_ePBS::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 exportInputFiles(job);
88 // build batch script for job
89 buildBatchScript(job);
91 // define name of log file (local)
92 string logFile = generateTemporaryFileName("PBS-submitlog");
94 // define command to submit batch
95 string subCommand = string("cd ") + workDir + "; qsub " + 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
114 ifstream idfile(logFile.c_str());
118 if (sline.size() == 0)
119 throw EmulationException("Error in the submission of the job on the remote host");
121 JobId id(this, sline);
125 // Ce manager permet de faire de la reprise
127 BatchManager_ePBS::addJob(const Batch::Job & job, const std::string reference)
129 return JobId(this, reference);
132 // Methode pour le controle des jobs : retire un job du gestionnaire
133 void BatchManager_ePBS::deleteJob(const JobId & jobid)
137 istringstream iss(jobid.getReference());
140 // define command to delete batch
141 string subCommand = string("qdel ") + iss.str();
142 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
143 cerr << command.c_str() << endl;
144 status = system(command.c_str());
146 throw EmulationException("Error of connection on remote host");
148 cerr << "jobId = " << ref << "killed" << endl;
151 // Methode pour le controle des jobs : suspend un job en file d'attente
152 void BatchManager_ePBS::holdJob(const JobId & jobid)
154 throw EmulationException("Not yet implemented");
157 // Methode pour le controle des jobs : relache un job suspendu
158 void BatchManager_ePBS::releaseJob(const JobId & jobid)
160 throw EmulationException("Not yet implemented");
164 // Methode pour le controle des jobs : modifie un job en file d'attente
165 void BatchManager_ePBS::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
167 throw EmulationException("Not yet implemented");
170 // Methode pour le controle des jobs : modifie un job en file d'attente
171 void BatchManager_ePBS::alterJob(const JobId & jobid, const Parametre & param)
173 alterJob(jobid, param, Environnement());
176 // Methode pour le controle des jobs : modifie un job en file d'attente
177 void BatchManager_ePBS::alterJob(const JobId & jobid, const Environnement & env)
179 alterJob(jobid, Parametre(), env);
182 // Methode pour le controle des jobs : renvoie l'etat du job
183 JobInfo BatchManager_ePBS::queryJob(const JobId & jobid)
186 istringstream iss(jobid.getReference());
189 // define name of log file (local)
190 string logFile = generateTemporaryFileName(string("PBS-querylog-id") + jobid.getReference());
192 // define command to query batch
193 string subCommand = string("qstat -f ") + iss.str();
194 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
197 cerr << command.c_str() << endl;
198 int status = system(command.c_str());
199 if(status && status != 153 && status != 256*153)
200 throw EmulationException("Error of connection on remote host");
202 JobInfo_ePBS ji = JobInfo_ePBS(id,logFile);
206 // Methode pour le controle des jobs : teste si un job est present en machine
207 bool BatchManager_ePBS::isRunning(const JobId & jobid)
209 throw EmulationException("Not yet implemented");
212 void BatchManager_ePBS::buildBatchScript(const Job & job)
214 std::cerr << "BuildBatchScript" << std::endl;
215 Parametre params = job.getParametre();
216 Environnement env = job.getEnvironnement();
220 string fileToExecute = "";
226 // Mandatory parameters
227 if (params.find(WORKDIR) != params.end())
228 workDir = params[WORKDIR].str();
230 throw EmulationException("params[WORKDIR] is not defined ! Please defined it, cannot submit this job");
231 if (params.find(EXECUTABLE) != params.end())
232 fileToExecute = params[EXECUTABLE].str();
234 throw EmulationException("params[EXECUTABLE] is not defined ! Please defined it, cannot submit this job");
236 // Optional parameters
237 if (params.find(NBPROC) != params.end())
238 nbproc = params[NBPROC];
239 if (params.find(MAXWALLTIME) != params.end())
240 edt = params[MAXWALLTIME];
241 if (params.find(MAXRAMSIZE) != params.end())
242 mem = params[MAXRAMSIZE];
243 if (params.find(QUEUE) != params.end())
244 queue = params[QUEUE].str();
246 string::size_type p1 = fileToExecute.find_last_of("/");
247 string::size_type p2 = fileToExecute.find_last_of(".");
248 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
249 string fileNameToExecute = fileToExecute.substr(p1+1);
251 // Create batch submit file
252 ofstream tempOutputFile;
253 std::string TmpFileName = createAndOpenTemporaryFile("PBS-script", tempOutputFile);
255 tempOutputFile << "#! /bin/sh -f" << endl;
258 // Division - arrondi supérieur
259 int nodes_requested = (nbproc + _nb_proc_per_node -1) / _nb_proc_per_node;
260 tempOutputFile << "#PBS -l nodes=" << nodes_requested << ":ppn=" << _nb_proc_per_node << endl;
263 tempOutputFile << "#PBS -q " << queue << endl;
265 tempOutputFile << "#PBS -l walltime=" << edt*60 << endl;
267 tempOutputFile << "#PBS -l mem=" << mem << "MB" << endl;
268 tempOutputFile << "#PBS -o " << workDir << "/logs/output.log." << rootNameToExecute << endl;
269 tempOutputFile << "#PBS -e " << workDir << "/logs/error.log." << rootNameToExecute << endl;
271 // Define environment for the job
273 tempOutputFile << "#PBS -v ";
274 Environnement::const_iterator iter;
275 for (iter = env.begin() ; iter != env.end() ; ++iter) {
276 tempOutputFile << iter->first << "=" << iter->second << ",";
278 tempOutputFile << endl;
281 // Abstraction of PBS_NODEFILE - TODO
282 tempOutputFile << "export LIBBATCH_NODEFILE=$PBS_NODEFILE" << endl;
284 // Launch the executable
285 tempOutputFile << "cd " << workDir << endl;
286 tempOutputFile << "./" + fileNameToExecute << endl;
287 tempOutputFile.flush();
288 tempOutputFile.close();
290 BATCH_CHMOD(TmpFileName.c_str(), 0x1ED);
291 cerr << "Batch script file generated is: " << TmpFileName.c_str() << endl;
293 int status = _protocol.copyFile(TmpFileName, "", "",
294 workDir + "/" + rootNameToExecute + "_Batch.sh",
295 _hostname, _username);
297 throw EmulationException("Error of connection on remote host, cannot copy batch submission file");