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_eSGE.cxx : emulation of SGE client
25 * Auteur : Bernard SECHER - CEA DEN
26 * Mail : mailto:bernard.secher@cea.fr
27 * Date : Thu Apr 24 10:17:22 2008
49 #include "Batch_BatchManager_eSGE.hxx"
50 #include "Batch_JobInfo_eSGE.hxx"
56 BatchManager_eSGE::BatchManager_eSGE(const FactBatchManager * parent, const char * host,
57 const char * username,
58 CommunicationProtocolType protocolType, const char * mpiImpl)
59 : BatchManager_eClient(parent, host, username, protocolType, mpiImpl),
60 BatchManager(parent, host)
66 BatchManager_eSGE::~BatchManager_eSGE()
71 // Methode pour le controle des jobs : soumet un job au gestionnaire
72 const JobId BatchManager_eSGE::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 exportInputFiles(job);
85 // build batch script for job
86 buildBatchScript(job);
88 // define name of log file (local)
89 string logFile = generateTemporaryFileName("SGE-submitlog");
91 // define command to submit batch
92 string subCommand = string("cd ") + workDir + "; qsub " + fileNameToExecute + "_Batch.sh";
93 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
97 cerr << command.c_str() << endl;
98 status = system(command.c_str());
101 ifstream error_message(logFile.c_str());
104 while(std::getline(error_message, temp))
106 error_message.close();
107 throw EmulationException("Error of connection on remote host, error was: " + mess);
110 // read id of submitted job in log file
112 FILE *fp = fopen(logFile.c_str(),"r");
113 fgets( line, 128, fp);
117 istringstream iss(line);
118 iss >> strjob >> strjob >> strjob;
120 JobId id(this, strjob);
124 // Ce manager permet de faire de la reprise
126 BatchManager_eSGE::addJob(const Batch::Job & job, const std::string reference)
128 return JobId(this, reference);
131 // Methode pour le controle des jobs : retire un job du gestionnaire
132 void BatchManager_eSGE::deleteJob(const JobId & jobid)
136 istringstream iss(jobid.getReference());
139 // define command to delete batch
140 string subCommand = string("qdel ") + iss.str();
141 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
142 cerr << command.c_str() << endl;
143 status = system(command.c_str());
145 throw EmulationException("Error of connection on remote host");
147 cerr << "jobId = " << ref << "killed" << endl;
150 // Methode pour le controle des jobs : suspend un job en file d'attente
151 void BatchManager_eSGE::holdJob(const JobId & jobid)
153 throw EmulationException("Not yet implemented");
156 // Methode pour le controle des jobs : relache un job suspendu
157 void BatchManager_eSGE::releaseJob(const JobId & jobid)
159 throw EmulationException("Not yet implemented");
163 // Methode pour le controle des jobs : modifie un job en file d'attente
164 void BatchManager_eSGE::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
166 throw EmulationException("Not yet implemented");
169 // Methode pour le controle des jobs : modifie un job en file d'attente
170 void BatchManager_eSGE::alterJob(const JobId & jobid, const Parametre & param)
172 alterJob(jobid, param, Environnement());
175 // Methode pour le controle des jobs : modifie un job en file d'attente
176 void BatchManager_eSGE::alterJob(const JobId & jobid, const Environnement & env)
178 alterJob(jobid, Parametre(), env);
181 // Methode pour le controle des jobs : renvoie l'etat du job
182 JobInfo BatchManager_eSGE::queryJob(const JobId & jobid)
185 istringstream iss(jobid.getReference());
188 // define name of log file (local)
189 string logFile = generateTemporaryFileName(string("SGE-querylog-id") + jobid.getReference());
191 // define command to query batch
192 string subCommand = string("qstat | grep ") + iss.str();
193 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
196 cerr << command.c_str() << endl;
197 int status = system(command.c_str());
198 if (status && status != 256)
199 throw EmulationException("Error of connection on remote host");
201 JobInfo_eSGE ji = JobInfo_eSGE(id,logFile);
205 // Methode pour le controle des jobs : teste si un job est present en machine
206 bool BatchManager_eSGE::isRunning(const JobId & jobid)
208 throw EmulationException("Not yet implemented");
211 void BatchManager_eSGE::buildBatchScript(const Job & job)
214 //TODO porting on Win32 platform
215 std::cerr << "BuildBatchScript" << std::endl;
216 Parametre params = job.getParametre();
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("SGE-script", tempOutputFile);
255 tempOutputFile << "#! /bin/sh -f" << endl;
257 tempOutputFile << "#$ -q " << queue << endl;
258 tempOutputFile << "#$ -pe mpich " << nbproc << endl;
260 tempOutputFile << "#$ -l h_rt=" << getWallTime(edt) << endl ;
262 tempOutputFile << "#$ -l h_vmem=" << mem << "M" << endl ;
263 tempOutputFile << "#$ -o " << workDir << "/logs/output.log." << rootNameToExecute << endl ;
264 tempOutputFile << "#$ -e " << workDir << "/logs/error.log." << rootNameToExecute << endl ;
266 // Abstraction of PBS_NODEFILE - TODO
267 tempOutputFile << "export LIBBATCH_NODEFILE=$TMPDIR/machines" << endl;
269 // Launch the executable
270 tempOutputFile << "cd " << workDir << endl ;
271 tempOutputFile << "./" + fileNameToExecute << endl;
272 tempOutputFile.flush();
273 tempOutputFile.close();
275 BATCH_CHMOD(TmpFileName.c_str(), 0x1ED);
276 cerr << "Batch script file generated is: " << TmpFileName.c_str() << endl;
278 int status = _protocol.copyFile(TmpFileName, "", "",
279 workDir + "/" + rootNameToExecute + "_Batch.sh",
280 _hostname, _username);
282 throw EmulationException("Error of connection on remote host");
287 std::string BatchManager_eSGE::getWallTime(const long edt)
294 oss << h << ":" << m;
296 oss << h << ":0" << m;