1 // Copyright (C) 2007-2012 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 * Batch_BatchManager_eVishnu.cxx :
25 * Created on: 24 june 2011
26 * Author : Renaud BARATE - EDF R&D
38 #include <Batch_NotYetImplementedException.hxx>
39 #include <Batch_Constants.hxx>
40 #include <Batch_Utils.hxx>
42 #include "Batch_BatchManager_eVishnu.hxx"
43 #include "Batch_JobInfo_eVishnu.hxx"
49 BatchManager_eVishnu::BatchManager_eVishnu(const FactBatchManager * parent,
51 const char * username,
52 CommunicationProtocolType protocolType,
55 : BatchManager(parent, host),
56 // Force SH protocol for Vishnu
57 BatchManager_eClient(parent, host, username, SH, mpiImpl),
58 _nb_proc_per_node(nb_proc_per_node)
62 BatchManager_eVishnu::~BatchManager_eVishnu()
66 // Method to submit a job to the batch manager
67 const JobId BatchManager_eVishnu::submitJob(const Job & job)
69 // export input files on cluster
70 exportInputFiles(job);
72 // build command file to submit the job
73 string cmdFile = buildCommandFile(job);
75 // define extra parameters (that can not be defined in the command file)
76 Parametre params = job.getParametre();
77 ostringstream extraParams;
78 if (params.find(NBPROC) != params.end())
79 extraParams << "-P " << params[NBPROC] << " ";
80 if (params.find(MAXRAMSIZE) != params.end())
81 extraParams << "-m " << params[MAXRAMSIZE] << " ";
83 // define command to submit batch
84 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
85 subCommand += "vishnu_connect && ";
86 subCommand += "vishnu_submit_job " + extraParams.str() + _hostname + " " + cmdFile + " && ";
87 subCommand += "vishnu_close";
88 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
90 cerr << command.c_str() << endl;
94 int status = Utils::getCommandOutput(command, output);
96 if (status != 0) throw EmulationException("Can't submit job, error was: " + output);
98 // find id of submitted job in output
99 string search = "Job Id : ";
100 string::size_type pos = output.find(search);
101 if (pos == string::npos)
102 throw EmulationException("Error in the submission of the job on the remote host");
103 pos += search.size();
104 string::size_type endl_pos = output.find('\n', pos);
105 string::size_type count = (endl_pos == string::npos)? string::npos : endl_pos - pos;
106 string jobref = output.substr(pos, count);
107 if (jobref.size() == 0)
108 throw EmulationException("Error in the submission of the job on the remote host");
110 JobId id(this, jobref);
115 void BatchManager_eVishnu::exportInputFiles(const Job& job)
117 Parametre params = job.getParametre();
118 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
119 subCommand += "vishnu_connect && ";
121 // create remote directories
122 subCommand += "vishnu_create_dir -p " + _hostname + ":" + params[WORKDIR].str() + "/logs && ";
125 string executeFile = params[EXECUTABLE];
126 if (executeFile.size() != 0) {
127 subCommand += "vishnu_copy_file " + executeFile + " " +
128 _hostname + ":" + params[WORKDIR].str() + "/ && ";
131 // copy filesToExportList
132 const Versatile & V = params[INFILE];
133 Versatile::const_iterator Vit;
134 for (Vit=V.begin(); Vit!=V.end(); Vit++) {
135 CoupleType cpt = *static_cast< CoupleType * >(*Vit);
136 Couple inputFile = cpt;
138 // Get absolute paths
139 char * buf = getcwd(NULL, 0);
143 string absremote = (Utils::isAbsolutePath(inputFile.getRemote()))?
144 inputFile.getRemote() :
145 params[WORKDIR].str() + "/" + inputFile.getRemote();
146 string abslocal = (Utils::isAbsolutePath(inputFile.getLocal()))?
147 inputFile.getLocal() :
148 cwd + "/" + inputFile.getLocal();
150 if (Vit != V.begin())
151 subCommand += " && ";
152 subCommand += "vishnu_copy_file " + abslocal + " " + _hostname + ":" + absremote;
154 subCommand += " && vishnu_close";
157 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
159 cerr << command.c_str() << endl;
161 int status = Utils::getCommandOutput(command, output);
164 throw EmulationException("Can't copy input files, error was: " + output);
168 * Create Vishnu command file and copy it on the server.
169 * Return the name of the remote file.
171 string BatchManager_eVishnu::buildCommandFile(const Job & job)
173 Parametre params = job.getParametre();
177 string fileToExecute = "";
180 // Mandatory parameters
181 if (params.find(WORKDIR) != params.end())
182 workDir = params[WORKDIR].str();
184 throw EmulationException("params[WORKDIR] is not defined. Please define it, cannot submit this job.");
185 if (params.find(EXECUTABLE) != params.end())
186 fileToExecute = params[EXECUTABLE].str();
188 throw EmulationException("params[EXECUTABLE] is not defined. Please define it, cannot submit this job.");
190 string::size_type p1 = fileToExecute.find_last_of("/");
191 string::size_type p2 = fileToExecute.find_last_of(".");
192 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
193 string fileNameToExecute = fileToExecute.substr(p1+1);
195 // Create batch submit file
196 ofstream tempOutputFile;
197 string tmpFileName = createAndOpenTemporaryFile("vishnu-script", tempOutputFile);
199 tempOutputFile << "#!/bin/sh" << endl;
200 tempOutputFile << "#% vishnu_output=" << workDir << "/logs/output.log." << rootNameToExecute << endl;
201 tempOutputFile << "#% vishnu_error=" << workDir << "/logs/error.log." << rootNameToExecute << endl;
203 if (params.find(NAME) != params.end())
204 tempOutputFile << "#% vishnu_job_name=\"" << params[NAME] << "\"" << endl;
206 // Optional parameters
207 if (params.find(MAXWALLTIME) != params.end()) {
208 long totalMinutes = params[MAXWALLTIME];
209 long h = totalMinutes / 60;
210 long m = totalMinutes - h * 60;
211 tempOutputFile << "#% vishnu_wallclocklimit=" << h << ":";
213 tempOutputFile << "0";
214 tempOutputFile << m << ":00" << endl;
216 if (params.find(QUEUE) != params.end())
217 tempOutputFile << "#% vishnu_queue=" << params[QUEUE] << endl;
219 // Define environment for the job
220 Environnement env = job.getEnvironnement();
221 for (Environnement::const_iterator iter = env.begin() ; iter != env.end() ; ++iter) {
222 tempOutputFile << "export " << iter->first << "=" << iter->second << endl;
226 tempOutputFile << "export LIBBATCH_NODEFILE=$VISHNU_BATCHJOB_NODEFILE" << endl;
228 // Launch the executable
229 tempOutputFile << "cd " << workDir << endl;
230 tempOutputFile << "./" + fileNameToExecute << endl;
232 tempOutputFile.flush();
233 tempOutputFile.close();
235 cerr << "Batch script file generated is: " << tmpFileName << endl;
239 void BatchManager_eVishnu::deleteJob(const JobId & jobid)
241 // define command to delete job
242 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
243 subCommand += "vishnu_connect && ";
244 subCommand += "vishnu_cancel_job " + _hostname + " " + jobid.getReference() + " && ";
245 subCommand += "vishnu_close";
246 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
247 cerr << command.c_str() << endl;
249 int status = system(command.c_str());
251 throw EmulationException("Can't delete job " + jobid.getReference());
253 cerr << "job " << jobid.getReference() << " killed" << endl;
256 void BatchManager_eVishnu::holdJob(const JobId & jobid)
258 throw NotYetImplementedException("BatchManager_eVishnu::holdJob");
261 void BatchManager_eVishnu::releaseJob(const JobId & jobid)
263 throw NotYetImplementedException("BatchManager_eVishnu::releaseJob");
266 void BatchManager_eVishnu::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
268 throw NotYetImplementedException("BatchManager_eVishnu::alterJob");
271 void BatchManager_eVishnu::alterJob(const JobId & jobid, const Parametre & param)
273 throw NotYetImplementedException("BatchManager_eVishnu::alterJob");
276 void BatchManager_eVishnu::alterJob(const JobId & jobid, const Environnement & env)
278 throw NotYetImplementedException("BatchManager_eVishnu::alterJob");
281 JobInfo BatchManager_eVishnu::queryJob(const JobId & jobid)
283 // define command to query batch
284 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
285 subCommand += "vishnu_connect && ";
286 subCommand += "vishnu_get_job_info " + _hostname + " " + jobid.getReference() + " && ";
287 subCommand += "vishnu_close";
288 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
289 cerr << command.c_str() << endl;
292 int status = Utils::getCommandOutput(command, output);
294 throw EmulationException("Can't query job " + jobid.getReference());
295 JobInfo_eVishnu jobinfo = JobInfo_eVishnu(jobid.getReference(), output);
299 const JobId BatchManager_eVishnu::addJob(const Job & job, const string reference)
301 return JobId(this, reference);
304 void BatchManager_eVishnu::importOutputFiles(const Job & job, const std::string directory)
306 // Create local result directory
307 char * buf = getcwd(NULL, 0);
310 string absdir = (Utils::isAbsolutePath(directory))? directory : cwd + "/" + directory;
311 int status = CommunicationProtocol::getInstance(SH).makeDirectory(absdir, "", "");
313 throw EmulationException("Can't create result directory");
316 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
317 subCommand += "vishnu_connect && ";
320 Parametre params = job.getParametre();
321 const Versatile & V = params[OUTFILE];
322 Versatile::const_iterator Vit;
323 for (Vit=V.begin(); Vit!=V.end(); Vit++) {
324 CoupleType cpt = *static_cast< CoupleType * >(*Vit);
325 Couple outputFile = cpt;
327 // Get absolute paths
328 string absremote = (Utils::isAbsolutePath(outputFile.getRemote()))?
329 outputFile.getRemote() :
330 params[WORKDIR].str() + "/" + outputFile.getRemote();
331 string abslocal = (Utils::isAbsolutePath(outputFile.getLocal()))?
332 outputFile.getLocal() :
333 absdir + "/" + outputFile.getLocal();
335 subCommand += "vishnu_copy_file " + _hostname + ":" + absremote + " " + abslocal + " && ";
339 subCommand += "vishnu_copy_file -r " +_hostname + ":" + params[WORKDIR].str() + "/logs" + " " + absdir + " && ";
340 subCommand += "vishnu_close";
343 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
345 cerr << command.c_str() << endl;
347 status = Utils::getCommandOutput(command, output);
350 throw EmulationException("Can't import output files, error was: " + output);