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,
48 : BatchManager(parent, host, username, protocolType, mpiImpl)
52 BatchManager_eSlurm::~BatchManager_eSlurm()
56 // Method to submit a job to the batch manager
57 const JobId BatchManager_eSlurm::submitJob(const Job & job)
59 Parametre params = job.getParametre();
60 const string workDir = params[WORKDIR];
62 // export input files on cluster
63 exportInputFiles(job);
65 // build command file to submit the job and copy it on the server
66 string cmdFile = buildCommandFile(job);
68 // define command to submit batch
69 string subCommand = string("cd ") + workDir + "; sbatch " + cmdFile;
70 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
72 cout << command.c_str() << endl;
76 int status = Utils::getCommandOutput(command, output);
78 if (status != 0) throw RunTimeException("Can't submit job, error was: " + output);
80 // find id of submitted job in output
81 string search = "Submitted batch job ";
82 string::size_type pos = output.find(search);
83 if (pos == string::npos)
84 throw RunTimeException("Error in the submission of the job on the remote host");
86 string::size_type endl_pos = output.find('\n', pos);
87 string::size_type count = (endl_pos == string::npos)? string::npos : endl_pos - pos;
88 string jobref = output.substr(pos, count);
90 JobId id(this, jobref);
95 * Create Slurm command file and copy it on the server.
96 * Return the name of the remote file.
98 string BatchManager_eSlurm::buildCommandFile(const Job & job)
100 Parametre params = job.getParametre();
104 string fileToExecute = "";
107 // Mandatory parameters
108 if (params.find(WORKDIR) != params.end())
109 workDir = params[WORKDIR].str();
111 throw RunTimeException("params[WORKDIR] is not defined. Please define it, cannot submit this job.");
112 if (params.find(EXECUTABLE) != params.end())
113 fileToExecute = params[EXECUTABLE].str();
115 throw RunTimeException("params[EXECUTABLE] is not defined. Please define it, cannot submit this job.");
117 string::size_type p1 = fileToExecute.find_last_of("/");
118 string::size_type p2 = fileToExecute.find_last_of(".");
119 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
120 string fileNameToExecute = fileToExecute.substr(p1+1);
122 // Create batch submit file
123 ofstream tempOutputFile;
124 string tmpFileName = Utils::createAndOpenTemporaryFile("slurm-script", tempOutputFile);
126 tempOutputFile << "#!/bin/bash" << endl;
127 tempOutputFile << "#SBATCH --output=" << workDir << "/logs/output.log." << rootNameToExecute << endl;
128 tempOutputFile << "#SBATCH --error=" << workDir << "/logs/error.log." << rootNameToExecute << endl;
130 if (params.find(NAME) != params.end())
131 tempOutputFile << "#SBATCH --job-name=\"" << params[NAME] << "\"" << endl;
133 // Optional parameters
135 if (params.find(NBPROC) != params.end())
136 nbproc = params[NBPROC];
137 tempOutputFile << "#SBATCH --ntasks=" << nbproc << endl;
139 if (params.find(EXCLUSIVE) != params.end()) {
140 if (params[EXCLUSIVE])
141 tempOutputFile << "#SBATCH --exclusive" << endl;
143 tempOutputFile << "#SBATCH --share" << endl;
146 if (params.find(MAXWALLTIME) != params.end())
147 tempOutputFile << "#SBATCH --time=" << params[MAXWALLTIME] << endl;
148 if (params.find(MAXRAMSIZE) != params.end())
149 tempOutputFile << "#SBATCH --mem=" << params[MAXRAMSIZE] << endl;
150 if (params.find(QUEUE) != params.end())
151 tempOutputFile << "#SBATCH --partition=" << params[QUEUE] << endl;
153 // Define environment for the job
154 Environnement env = job.getEnvironnement();
155 for (Environnement::const_iterator iter = env.begin() ; iter != env.end() ; ++iter) {
156 tempOutputFile << "export " << iter->first << "=" << iter->second << endl;
159 // generate nodes file
160 tempOutputFile << "LIBBATCH_NODEFILE=`mktemp nodefile-XXXXXXXXXX`" << endl;
161 tempOutputFile << "srun hostname > $LIBBATCH_NODEFILE" << endl;
162 tempOutputFile << "export LIBBATCH_NODEFILE" << endl;
164 // Launch the executable
165 tempOutputFile << "cd " << workDir << endl;
166 tempOutputFile << "./" + fileNameToExecute << endl;
168 // Remove the node file
169 tempOutputFile << "rm $LIBBATCH_NODEFILE" << endl;
171 tempOutputFile.flush();
172 tempOutputFile.close();
174 cerr << "Batch script file generated is: " << tmpFileName << endl;
176 string remoteFileName = rootNameToExecute + "_slurm.cmd";
177 int status = _protocol.copyFile(tmpFileName, "", "",
178 workDir + "/" + remoteFileName,
179 _hostname, _username);
181 throw RunTimeException("Cannot copy command file on host " + _hostname);
183 return remoteFileName;
186 void BatchManager_eSlurm::deleteJob(const JobId & jobid)
188 // define command to delete job
189 string subCommand = "scancel " + jobid.getReference();
190 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
191 cerr << command.c_str() << endl;
193 int status = system(command.c_str());
195 throw RunTimeException("Can't delete job " + jobid.getReference());
197 cerr << "job " << jobid.getReference() << " killed" << endl;
200 void BatchManager_eSlurm::holdJob(const JobId & jobid)
202 throw NotYetImplementedException("BatchManager_eSlurm::holdJob");
205 void BatchManager_eSlurm::releaseJob(const JobId & jobid)
207 throw NotYetImplementedException("BatchManager_eSlurm::releaseJob");
210 void BatchManager_eSlurm::alterJob(const JobId & jobid, const Parametre & param, const Environnement & env)
212 throw NotYetImplementedException("BatchManager_eSlurm::alterJob");
215 void BatchManager_eSlurm::alterJob(const JobId & jobid, const Parametre & param)
217 throw NotYetImplementedException("BatchManager_eSlurm::alterJob");
220 void BatchManager_eSlurm::alterJob(const JobId & jobid, const Environnement & env)
222 throw NotYetImplementedException("BatchManager_eSlurm::alterJob");
225 JobInfo BatchManager_eSlurm::queryJob(const JobId & jobid)
227 // define command to query batch
228 string subCommand = "squeue -o %t -j " + jobid.getReference();
229 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
230 cerr << command.c_str() << endl;
232 Utils::getCommandOutput(command, output);
233 // We don't test the return code here because with jobs finished since a long time Slurm
234 // returns an error and a message like "slurm_load_jobs error: Invalid job id specified".
235 // So we consider that the job is finished when we get an error.
237 JobInfo_eSlurm jobinfo = JobInfo_eSlurm(jobid.getReference(), output);
241 const JobId BatchManager_eSlurm::addJob(const Job & job, const string reference)
243 return JobId(this, reference);