1 // Copyright (C) 2007-2019 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::runJob(const Job & job)
60 Parametre params = job.getParametre();
61 const string workDir = params[WORKDIR];
63 // build command file to submit the job and copy it on the server
64 string cmdFile = buildCommandFile(job);
66 // define command to submit batch
67 string subCommand = string("cd ") + workDir + "; sbatch " + cmdFile;
68 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
74 int status = Utils::getCommandOutput(command, output);
76 if (status != 0) throw RunTimeException("Can't submit job, error was: " + output);
78 // find id of submitted job in output
79 string search = "Submitted batch job ";
80 string::size_type pos = output.find(search);
81 if (pos == string::npos)
82 throw RunTimeException("Error in the submission of the job on the remote host");
84 string::size_type endl_pos = output.find('\n', pos);
85 string::size_type count = (endl_pos == string::npos)? string::npos : endl_pos - pos;
86 string jobref = output.substr(pos, count);
88 JobId id(this, jobref);
93 * Create Slurm command file and copy it on the server.
94 * Return the name of the remote file.
96 string BatchManager_Slurm::buildCommandFile(const Job & job)
98 Parametre params = job.getParametre();
102 string fileToExecute = "";
105 // Mandatory parameters
106 if (params.find(WORKDIR) != params.end())
107 workDir = params[WORKDIR].str();
109 throw RunTimeException("params[WORKDIR] is not defined. Please define it, cannot submit this job.");
110 if (params.find(EXECUTABLE) != params.end())
111 fileToExecute = params[EXECUTABLE].str();
113 throw RunTimeException("params[EXECUTABLE] is not defined. Please define it, cannot submit this job.");
115 string::size_type p1 = fileToExecute.find_last_of("/");
116 string::size_type p2 = fileToExecute.find_last_of(".");
117 string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
118 string fileNameToExecute = fileToExecute.substr(p1+1);
120 // Create batch submit file
121 ofstream tempOutputFile;
122 string tmpFileName = Utils::createAndOpenTemporaryFile("slurm-script", tempOutputFile);
124 tempOutputFile << "#!/bin/bash -l" << endl;
125 tempOutputFile << "#SBATCH --output=" << workDir << "/logs/output.log." << rootNameToExecute << endl;
126 tempOutputFile << "#SBATCH --error=" << workDir << "/logs/error.log." << rootNameToExecute << endl;
128 if (params.find(NAME) != params.end())
129 tempOutputFile << "#SBATCH --job-name=\"" << params[NAME] << "\"" << endl;
131 // Optional parameters
133 if (params.find(NBPROC) != params.end())
134 nbproc = params[NBPROC];
135 tempOutputFile << "#SBATCH --ntasks=" << nbproc << endl;
137 if (params.find(EXCLUSIVE) != params.end()) {
138 if (params[EXCLUSIVE])
139 tempOutputFile << "#SBATCH --exclusive" << endl;
141 tempOutputFile << "#SBATCH --share" << endl;
144 if (params.find(MAXWALLTIME) != params.end())
145 tempOutputFile << "#SBATCH --time=" << params[MAXWALLTIME] << endl;
146 if (params.find(MAXRAMSIZE) != params.end())
147 tempOutputFile << "#SBATCH --mem=" << params[MAXRAMSIZE] << endl;
148 else if (params.find(MEMPERCPU) != params.end())
149 tempOutputFile << "#SBATCH --mem-per-cpu=" << params[MEMPERCPU] << endl;
150 if (params.find(QUEUE) != params.end())
151 tempOutputFile << "#SBATCH --qos=" << params[QUEUE] << endl;
152 if (params.find(PARTITION) != params.end())
153 tempOutputFile << "#SBATCH --partition=" << params[PARTITION] << endl;
154 if (params.find(WCKEY) != params.end())
155 tempOutputFile << "#SBATCH --wckey=" << params[WCKEY] << endl;
156 if (params.find(NBNODE) != params.end())
157 tempOutputFile << "#SBATCH --nodes=" << params[NBNODE] << endl;
158 if (params.find(EXTRAPARAMS) != params.end())
159 tempOutputFile << params[EXTRAPARAMS] << endl;
161 // Define environment for the job
162 Environnement env = job.getEnvironnement();
163 for (Environnement::const_iterator iter = env.begin() ; iter != env.end() ; ++iter) {
164 tempOutputFile << "export " << iter->first << "=" << iter->second << endl;
167 // generate nodes file
168 tempOutputFile << "LIBBATCH_NODEFILE=$(mktemp nodefile-XXXXXXXXXX)" << endl;
169 tempOutputFile << "srun hostname > \"$LIBBATCH_NODEFILE\"" << endl;
170 tempOutputFile << "export LIBBATCH_NODEFILE" << endl;
172 // Launch the executable
173 tempOutputFile << "cd " << workDir << endl;
174 tempOutputFile << "./" + fileNameToExecute;
175 if (params.find(ARGUMENTS) != params.end()) {
176 Versatile V = params[ARGUMENTS];
177 for(Versatile::const_iterator it=V.begin(); it!=V.end(); it++) {
178 StringType argt = * static_cast<StringType *>(*it);
180 tempOutputFile << " " << arg;
183 tempOutputFile << endl;
185 // Remove the node file
186 tempOutputFile << "rm \"$LIBBATCH_NODEFILE\"" << endl;
188 tempOutputFile.flush();
189 tempOutputFile.close();
191 LOG("Batch script file generated is: " << tmpFileName);
193 string remoteFileName = rootNameToExecute + "_slurm.cmd";
194 int status = _protocol.copyFile(tmpFileName, "", "",
195 workDir + "/" + remoteFileName,
196 _hostname, _username);
198 throw RunTimeException("Cannot copy command file on host " + _hostname);
200 return remoteFileName;
203 void BatchManager_Slurm::deleteJob(const JobId & jobid)
205 // define command to delete job
206 string subCommand = "scancel " + jobid.getReference();
207 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
210 int status = system(command.c_str());
212 throw RunTimeException("Can't delete job " + jobid.getReference());
214 LOG("job " << jobid.getReference() << " killed");
217 JobInfo BatchManager_Slurm::queryJob(const JobId & jobid)
219 // First try to query the job with "squeue" command
220 string subCommand = "squeue -h -o %T -j " + jobid.getReference() + " 2>/dev/null";
221 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
224 int status = Utils::getCommandOutput(command, output);
225 LOG("status: " << status << ", output: " << output);
230 jobinfo = JobInfo_Slurm(jobid.getReference(), output);
232 } catch (const RunTimeException & exc) {
237 // If "squeue" failed, the job may be finished. In this case, try to query the job with
240 string subCommand = "sacct -X -o State%-10 -n -j " + jobid.getReference();
241 string command = _protocol.getExecCommand(subCommand, _hostname, _username);
244 int status = Utils::getCommandOutput(command, output);
245 LOG("status: " << status << ", output: " << output);
248 jobinfo = JobInfo_Slurm(jobid.getReference(), output);
249 } catch (const RunTimeException & exc) {
254 throw RunTimeException("sacct command failed with return code: " + status);