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 : // Force SH protocol for Vishnu
56 BatchManager(parent, host, username, SH, mpiImpl),
57 _nb_proc_per_node(nb_proc_per_node)
61 BatchManager_eVishnu::~BatchManager_eVishnu()
65 // Method to submit a job to the batch manager
66 const JobId BatchManager_eVishnu::submitJob(const Job & job)
68 // export input files on cluster
69 exportInputFiles(job);
71 // build command file to submit the job
72 string cmdFile = buildCommandFile(job);
74 // define extra parameters (that can not be defined in the command file)
75 Parametre params = job.getParametre();
76 ostringstream extraParams;
77 if (params.find(NBPROC) != params.end())
78 extraParams << "-P " << params[NBPROC] << " ";
79 if (params.find(MAXRAMSIZE) != params.end())
80 extraParams << "-m " << params[MAXRAMSIZE] << " ";
82 // define command to submit batch
83 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
84 subCommand += "vishnu_connect && ";
85 subCommand += "vishnu_submit_job " + extraParams.str() + _hostname + " " + cmdFile + " && ";
86 subCommand += "vishnu_close";
87 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
89 cerr << command.c_str() << endl;
93 int status = Utils::getCommandOutput(command, output);
95 if (status != 0) throw RunTimeException("Can't submit job, error was: " + output);
97 // find id of submitted job in output
98 string search = "Job Id : ";
99 string::size_type pos = output.find(search);
100 if (pos == string::npos)
101 throw RunTimeException("Error in the submission of the job on the remote host");
102 pos += search.size();
103 string::size_type endl_pos = output.find('\n', pos);
104 string::size_type count = (endl_pos == string::npos)? string::npos : endl_pos - pos;
105 string jobref = output.substr(pos, count);
106 if (jobref.size() == 0)
107 throw RunTimeException("Error in the submission of the job on the remote host");
109 JobId id(this, jobref);
114 void BatchManager_eVishnu::exportInputFiles(const Job& job)
116 Parametre params = job.getParametre();
117 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
118 subCommand += "vishnu_connect && ";
120 // create remote directories
121 subCommand += "vishnu_create_dir -p " + _hostname + ":" + params[WORKDIR].str() + "/logs && ";
124 string executeFile = params[EXECUTABLE];
125 if (executeFile.size() != 0) {
126 subCommand += "vishnu_copy_file " + executeFile + " " +
127 _hostname + ":" + params[WORKDIR].str() + "/ && ";
130 // copy filesToExportList
131 const Versatile & V = params[INFILE];
132 Versatile::const_iterator Vit;
133 for (Vit=V.begin(); Vit!=V.end(); Vit++) {
134 CoupleType cpt = *static_cast< CoupleType * >(*Vit);
135 Couple inputFile = cpt;
137 // Get absolute paths
138 char * buf = getcwd(NULL, 0);
142 string absremote = (Utils::isAbsolutePath(inputFile.getRemote()))?
143 inputFile.getRemote() :
144 params[WORKDIR].str() + "/" + inputFile.getRemote();
145 string abslocal = (Utils::isAbsolutePath(inputFile.getLocal()))?
146 inputFile.getLocal() :
147 cwd + "/" + inputFile.getLocal();
149 if (Vit != V.begin())
150 subCommand += " && ";
151 subCommand += "vishnu_copy_file " + abslocal + " " + _hostname + ":" + absremote;
153 subCommand += " && vishnu_close";
156 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
158 cerr << command.c_str() << endl;
160 int status = Utils::getCommandOutput(command, output);
163 throw RunTimeException("Can't copy input files, error was: " + output);
167 * Create Vishnu command file and copy it on the server.
168 * Return the name of the remote file.
170 string BatchManager_eVishnu::buildCommandFile(const Job & job)
172 Parametre params = job.getParametre();
176 string fileToExecute = "";
179 // Mandatory parameters
180 if (params.find(WORKDIR) != params.end())
181 workDir = params[WORKDIR].str();
183 throw RunTimeException("params[WORKDIR] is not defined. Please define it, cannot submit this job.");
184 if (params.find(EXECUTABLE) != params.end())
185 fileToExecute = params[EXECUTABLE].str();
187 throw RunTimeException("params[EXECUTABLE] is not defined. Please define it, cannot submit this job.");
189 string::size_type p1 = fileToExecute.find_last_of("/");
190 string::size_type p2 = fileToExecute.find_last_of(".");
191 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
192 string fileNameToExecute = fileToExecute.substr(p1+1);
194 // Create batch submit file
195 ofstream tempOutputFile;
196 string tmpFileName = Utils::createAndOpenTemporaryFile("vishnu-script", tempOutputFile);
198 tempOutputFile << "#!/bin/sh" << endl;
199 tempOutputFile << "#% vishnu_output=" << workDir << "/logs/output.log." << rootNameToExecute << endl;
200 tempOutputFile << "#% vishnu_error=" << workDir << "/logs/error.log." << rootNameToExecute << endl;
202 if (params.find(NAME) != params.end())
203 tempOutputFile << "#% vishnu_job_name=\"" << params[NAME] << "\"" << endl;
205 // Optional parameters
206 if (params.find(MAXWALLTIME) != params.end()) {
207 long totalMinutes = params[MAXWALLTIME];
208 long h = totalMinutes / 60;
209 long m = totalMinutes - h * 60;
210 tempOutputFile << "#% vishnu_wallclocklimit=" << h << ":";
212 tempOutputFile << "0";
213 tempOutputFile << m << ":00" << endl;
215 if (params.find(QUEUE) != params.end())
216 tempOutputFile << "#% vishnu_queue=" << params[QUEUE] << endl;
218 // Define environment for the job
219 Environnement env = job.getEnvironnement();
220 for (Environnement::const_iterator iter = env.begin() ; iter != env.end() ; ++iter) {
221 tempOutputFile << "export " << iter->first << "=" << iter->second << endl;
225 tempOutputFile << "export LIBBATCH_NODEFILE=$VISHNU_BATCHJOB_NODEFILE" << endl;
227 // Launch the executable
228 tempOutputFile << "cd " << workDir << endl;
229 tempOutputFile << "./" + fileNameToExecute << endl;
231 tempOutputFile.flush();
232 tempOutputFile.close();
234 cerr << "Batch script file generated is: " << tmpFileName << endl;
238 void BatchManager_eVishnu::deleteJob(const JobId & jobid)
240 // define command to delete job
241 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
242 subCommand += "vishnu_connect && ";
243 subCommand += "vishnu_cancel_job " + _hostname + " " + jobid.getReference() + " && ";
244 subCommand += "vishnu_close";
245 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
246 cerr << command.c_str() << endl;
248 int status = system(command.c_str());
250 throw RunTimeException("Can't delete job " + jobid.getReference());
252 cerr << "job " << jobid.getReference() << " killed" << endl;
255 void BatchManager_eVishnu::holdJob(const JobId & jobid)
257 throw NotYetImplementedException("BatchManager_eVishnu::holdJob");
260 void BatchManager_eVishnu::releaseJob(const JobId & jobid)
262 throw NotYetImplementedException("BatchManager_eVishnu::releaseJob");
265 void BatchManager_eVishnu::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
267 throw NotYetImplementedException("BatchManager_eVishnu::alterJob");
270 void BatchManager_eVishnu::alterJob(const JobId & jobid, const Parametre & param)
272 throw NotYetImplementedException("BatchManager_eVishnu::alterJob");
275 void BatchManager_eVishnu::alterJob(const JobId & jobid, const Environnement & env)
277 throw NotYetImplementedException("BatchManager_eVishnu::alterJob");
280 JobInfo BatchManager_eVishnu::queryJob(const JobId & jobid)
282 // define command to query batch
283 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
284 subCommand += "vishnu_connect && ";
285 subCommand += "vishnu_get_job_info " + _hostname + " " + jobid.getReference() + " && ";
286 subCommand += "vishnu_close";
287 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
288 cerr << command.c_str() << endl;
291 int status = Utils::getCommandOutput(command, output);
293 throw RunTimeException("Can't query job " + jobid.getReference());
294 JobInfo_eVishnu jobinfo = JobInfo_eVishnu(jobid.getReference(), output);
298 const JobId BatchManager_eVishnu::addJob(const Job & job, const string reference)
300 return JobId(this, reference);
303 void BatchManager_eVishnu::importOutputFiles(const Job & job, const std::string directory)
305 // Create local result directory
306 char * buf = getcwd(NULL, 0);
309 string absdir = (Utils::isAbsolutePath(directory))? directory : cwd + "/" + directory;
310 int status = CommunicationProtocol::getInstance(SH).makeDirectory(absdir, "", "");
312 throw RunTimeException("Can't create result directory");
315 string subCommand = string("export OMNIORB_CONFIG=$VISHNU_CONFIG_FILE; ");
316 subCommand += "vishnu_connect && ";
319 Parametre params = job.getParametre();
320 const Versatile & V = params[OUTFILE];
321 Versatile::const_iterator Vit;
322 for (Vit=V.begin(); Vit!=V.end(); Vit++) {
323 CoupleType cpt = *static_cast< CoupleType * >(*Vit);
324 Couple outputFile = cpt;
326 // Get absolute paths
327 string absremote = (Utils::isAbsolutePath(outputFile.getRemote()))?
328 outputFile.getRemote() :
329 params[WORKDIR].str() + "/" + outputFile.getRemote();
330 string abslocal = (Utils::isAbsolutePath(outputFile.getLocal()))?
331 outputFile.getLocal() :
332 absdir + "/" + outputFile.getLocal();
334 subCommand += "vishnu_copy_file " + _hostname + ":" + absremote + " " + abslocal + " && ";
338 subCommand += "vishnu_copy_file -r " +_hostname + ":" + params[WORKDIR].str() + "/logs" + " " + absdir + " && ";
339 subCommand += "vishnu_close";
342 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
344 cerr << command.c_str() << endl;
346 status = Utils::getCommandOutput(command, output);
349 throw RunTimeException("Can't import output files, error was: " + output);