1 // Copyright (C) 2007-2015 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, or (at your option) any later version.
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_Slurm.cxx :
25 * Created on: 12 may 2011
26 * Author : Renaud BARATE - EDF R&D
32 #include <NotYetImplementedException.hxx>
33 #include <Constants.hxx>
37 #include "BatchManager_Slurm.hxx"
38 #include "JobInfo_Slurm.hxx"
44 BatchManager_Slurm::BatchManager_Slurm(const FactBatchManager * parent,
46 const char * username,
47 CommunicationProtocolType protocolType,
49 : BatchManager(parent, host, username, protocolType, mpiImpl)
53 BatchManager_Slurm::~BatchManager_Slurm()
57 // Method to submit a job to the batch manager
58 const JobId BatchManager_Slurm::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);
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_Slurm::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 -l" << 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 else if (params.find(MEMPERCPU) != params.end())
152 tempOutputFile << "#SBATCH --mem-per-cpu=" << params[MEMPERCPU] << endl;
153 if (params.find(QUEUE) != params.end())
154 tempOutputFile << "#SBATCH --qos=" << params[QUEUE] << endl;
155 if (params.find(WCKEY) != params.end())
156 tempOutputFile << "#SBATCH --wckey=" << params[WCKEY] << endl;
157 if (params.find(EXTRAPARAMS) != params.end())
158 tempOutputFile << params[EXTRAPARAMS] << endl;
160 // Define environment for the job
161 Environnement env = job.getEnvironnement();
162 for (Environnement::const_iterator iter = env.begin() ; iter != env.end() ; ++iter) {
163 tempOutputFile << "export " << iter->first << "=" << iter->second << endl;
166 // generate nodes file
167 tempOutputFile << "LIBBATCH_NODEFILE=$(mktemp nodefile-XXXXXXXXXX)" << endl;
168 tempOutputFile << "srun hostname > \"$LIBBATCH_NODEFILE\"" << endl;
169 tempOutputFile << "export LIBBATCH_NODEFILE" << endl;
171 // Launch the executable
172 tempOutputFile << "cd " << workDir << endl;
173 tempOutputFile << "./" + fileNameToExecute;
174 if (params.find(ARGUMENTS) != params.end()) {
175 Versatile V = params[ARGUMENTS];
176 for(Versatile::const_iterator it=V.begin(); it!=V.end(); it++) {
177 StringType argt = * static_cast<StringType *>(*it);
179 tempOutputFile << " " << arg;
182 tempOutputFile << endl;
184 // Remove the node file
185 tempOutputFile << "rm \"$LIBBATCH_NODEFILE\"" << endl;
187 tempOutputFile.flush();
188 tempOutputFile.close();
190 LOG("Batch script file generated is: " << tmpFileName);
192 string remoteFileName = rootNameToExecute + "_slurm.cmd";
193 int status = _protocol.copyFile(tmpFileName, "", "",
194 workDir + "/" + remoteFileName,
195 _hostname, _username);
197 throw RunTimeException("Cannot copy command file on host " + _hostname);
199 return remoteFileName;
202 void BatchManager_Slurm::deleteJob(const JobId & jobid)
204 // define command to delete job
205 string subCommand = "scancel " + jobid.getReference();
206 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
209 int status = system(command.c_str());
211 throw RunTimeException("Can't delete job " + jobid.getReference());
213 LOG("job " << jobid.getReference() << " killed");
216 JobInfo BatchManager_Slurm::queryJob(const JobId & jobid)
218 // define command to query batch
219 string subCommand = "squeue -o %t -j " + jobid.getReference();
220 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
223 Utils::getCommandOutput(command, output);
224 // We don't test the return code here because with jobs finished since a long time Slurm
225 // returns an error and a message like "slurm_load_jobs error: Invalid job id specified".
226 // So we consider that the job is finished when we get an error.
228 JobInfo_Slurm jobinfo = JobInfo_Slurm(jobid.getReference(), output);