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_eSlurm.cxx :
25 * Created on: 12 may 2011
26 * Author : Renaud BARATE - EDF R&D
32 #include <Batch_NotYetImplementedException.hxx>
33 #include <Batch_Constants.hxx>
34 #include <Batch_Utils.hxx>
36 #include "Batch_BatchManager_eSlurm.hxx"
37 #include "Batch_JobInfo_eSlurm.hxx"
43 BatchManager_eSlurm::BatchManager_eSlurm(const FactBatchManager * parent,
45 const char * username,
46 CommunicationProtocolType protocolType,
49 : BatchManager(parent, host, username, protocolType, mpiImpl)
53 BatchManager_eSlurm::~BatchManager_eSlurm()
57 // Method to submit a job to the batch manager
58 const JobId BatchManager_eSlurm::submitJob(const Job & job)
60 Parametre params = job.getParametre();
61 const string workDir = params[WORKDIR];
63 // export input files on cluster
64 exportInputFiles(job);
66 // build command file to submit the job and copy it on the server
67 string cmdFile = buildCommandFile(job);
69 // define command to submit batch
70 string subCommand = string("cd ") + workDir + "; sbatch " + cmdFile;
71 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
73 cout << command.c_str() << endl;
77 int status = Utils::getCommandOutput(command, output);
79 if (status != 0) throw RunTimeException("Can't submit job, error was: " + output);
81 // find id of submitted job in output
82 string search = "Submitted batch job ";
83 string::size_type pos = output.find(search);
84 if (pos == string::npos)
85 throw RunTimeException("Error in the submission of the job on the remote host");
87 string::size_type endl_pos = output.find('\n', pos);
88 string::size_type count = (endl_pos == string::npos)? string::npos : endl_pos - pos;
89 string jobref = output.substr(pos, count);
91 JobId id(this, jobref);
96 * Create Slurm command file and copy it on the server.
97 * Return the name of the remote file.
99 string BatchManager_eSlurm::buildCommandFile(const Job & job)
101 Parametre params = job.getParametre();
105 string fileToExecute = "";
108 // Mandatory parameters
109 if (params.find(WORKDIR) != params.end())
110 workDir = params[WORKDIR].str();
112 throw RunTimeException("params[WORKDIR] is not defined. Please define it, cannot submit this job.");
113 if (params.find(EXECUTABLE) != params.end())
114 fileToExecute = params[EXECUTABLE].str();
116 throw RunTimeException("params[EXECUTABLE] is not defined. Please define it, cannot submit this job.");
118 string::size_type p1 = fileToExecute.find_last_of("/");
119 string::size_type p2 = fileToExecute.find_last_of(".");
120 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
121 string fileNameToExecute = fileToExecute.substr(p1+1);
123 // Create batch submit file
124 ofstream tempOutputFile;
125 string tmpFileName = Utils::createAndOpenTemporaryFile("slurm-script", tempOutputFile);
127 tempOutputFile << "#!/bin/bash" << endl;
128 tempOutputFile << "#SBATCH --output=" << workDir << "/logs/output.log." << rootNameToExecute << endl;
129 tempOutputFile << "#SBATCH --error=" << workDir << "/logs/error.log." << rootNameToExecute << endl;
131 if (params.find(NAME) != params.end())
132 tempOutputFile << "#SBATCH --job-name=\"" << params[NAME] << "\"" << endl;
134 // Optional parameters
136 if (params.find(NBPROC) != params.end())
137 nbproc = params[NBPROC];
138 tempOutputFile << "#SBATCH --ntasks=" << nbproc << endl;
140 if (params.find(EXCLUSIVE) != params.end()) {
141 if (params[EXCLUSIVE])
142 tempOutputFile << "#SBATCH --exclusive" << endl;
144 tempOutputFile << "#SBATCH --share" << endl;
147 if (params.find(MAXWALLTIME) != params.end())
148 tempOutputFile << "#SBATCH --time=" << params[MAXWALLTIME] << endl;
149 if (params.find(MAXRAMSIZE) != params.end())
150 tempOutputFile << "#SBATCH --mem=" << params[MAXRAMSIZE] << endl;
151 if (params.find(QUEUE) != params.end())
152 tempOutputFile << "#SBATCH --partition=" << params[QUEUE] << endl;
154 // Define environment for the job
155 Environnement env = job.getEnvironnement();
156 for (Environnement::const_iterator iter = env.begin() ; iter != env.end() ; ++iter) {
157 tempOutputFile << "export " << iter->first << "=" << iter->second << endl;
160 // generate nodes file
161 tempOutputFile << "LIBBATCH_NODEFILE=`mktemp nodefile-XXXXXXXXXX`" << endl;
162 tempOutputFile << "srun hostname > $LIBBATCH_NODEFILE" << endl;
163 tempOutputFile << "export LIBBATCH_NODEFILE" << endl;
165 // Launch the executable
166 tempOutputFile << "cd " << workDir << endl;
167 tempOutputFile << "./" + fileNameToExecute << endl;
169 // Remove the node file
170 tempOutputFile << "rm $LIBBATCH_NODEFILE" << endl;
172 tempOutputFile.flush();
173 tempOutputFile.close();
175 cerr << "Batch script file generated is: " << tmpFileName << endl;
177 string remoteFileName = rootNameToExecute + "_slurm.cmd";
178 int status = _protocol.copyFile(tmpFileName, "", "",
179 workDir + "/" + remoteFileName,
180 _hostname, _username);
182 throw RunTimeException("Cannot copy command file on host " + _hostname);
184 return remoteFileName;
187 void BatchManager_eSlurm::deleteJob(const JobId & jobid)
189 // define command to delete job
190 string subCommand = "scancel " + jobid.getReference();
191 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
192 cerr << command.c_str() << endl;
194 int status = system(command.c_str());
196 throw RunTimeException("Can't delete job " + jobid.getReference());
198 cerr << "job " << jobid.getReference() << " killed" << endl;
201 void BatchManager_eSlurm::holdJob(const JobId & jobid)
203 throw NotYetImplementedException("BatchManager_eSlurm::holdJob");
206 void BatchManager_eSlurm::releaseJob(const JobId & jobid)
208 throw NotYetImplementedException("BatchManager_eSlurm::releaseJob");
211 void BatchManager_eSlurm::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
213 throw NotYetImplementedException("BatchManager_eSlurm::alterJob");
216 void BatchManager_eSlurm::alterJob(const JobId & jobid, const Parametre & param)
218 throw NotYetImplementedException("BatchManager_eSlurm::alterJob");
221 void BatchManager_eSlurm::alterJob(const JobId & jobid, const Environnement & env)
223 throw NotYetImplementedException("BatchManager_eSlurm::alterJob");
226 JobInfo BatchManager_eSlurm::queryJob(const JobId & jobid)
228 // define command to query batch
229 string subCommand = "squeue -o %t -j " + jobid.getReference();
230 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
231 cerr << command.c_str() << endl;
233 Utils::getCommandOutput(command, output);
234 // We don't test the return code here because with jobs finished since a long time Slurm
235 // returns an error and a message like "slurm_load_jobs error: Invalid job id specified".
236 // So we consider that the job is finished when we get an error.
238 JobInfo_eSlurm jobinfo = JobInfo_eSlurm(jobid.getReference(), output);
242 const JobId BatchManager_eSlurm::addJob(const Job & job, const string reference)
244 return JobId(this, reference);