src/COORM/BatchManager_COORM.cxx

   1 // Copyright (C) 2012-2021  INRIA
   2 //
   3 // This library is free software; you can redistribute it and/or
   4 // modify it under the terms of the GNU Lesser General Public
   5 // License as published by the Free Software Foundation; either
   6 // version 2.1 of the License, or (at your option) any later version.
   7 //
   8 // This library is distributed in the hope that it will be useful,
   9 // but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11 // Lesser General Public License for more details.
  12 //
  13 // You should have received a copy of the GNU Lesser General Public
  14 // License along with this library; if not, write to the Free Software
  15 // Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  16 //
  17 // See http://www.salome-platform.org/ or email : webmaster.salome@opencascade.com
  18 //
  19
  20 #include <cstdlib>
  21 #include <fstream>
  22
  23 #include <Constants.hxx>
  24 #include <Utils.hxx>
  25 #include <Log.hxx>
  26
  27 #include "BatchManager_COORM.hxx"
  28 #include "JobInfo_COORM.hxx"
  29
  30 using namespace std;
  31
  32 namespace Batch
  33 {
  34       BatchManager_COORM::BatchManager_COORM(const FactBatchManager * parent, const char * host,
  35                       const char * username,
  36                       CommunicationProtocolType protocolType, const char * mpiImpl)
  37               : BatchManager(parent, host, username, protocolType, mpiImpl)
  38       {
  39       }
  40
  41       BatchManager_COORM::~BatchManager_COORM()
  42       {
  43       }
  44
  45       // Soumet un job au gestionnaire
  46       const JobId BatchManager_COORM::runJob(const Job & job)
  47       {
  48               Parametre params = job.getParametre();
  49               const string workDir = params[WORKDIR];
  50               const string fileToExecute = params[EXECUTABLE];
  51
  52               // For CooRM
  53               const string launcherArgs = params[LAUNCHER_ARGS];
  54               const string launcherFile  = params[LAUNCHER_FILE];
  55
  56               const string jobName = params[NAME];
  57
  58               string::size_type p1 = fileToExecute.find_last_of("/");
  59               string::size_type p2 = fileToExecute.find_last_of(".");
  60               std::string fileNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
  61
  62               // For CooRM
  63               p1 = launcherFile.find_last_of("/");
  64               std::string fileNameToLaunch = launcherFile.substr(p1+1);
  65
  66               // build batch script for job
  67               string scriptFile = buildBatchScript(job);
  68
  69               // Get REMOTE_COORM_PATH environment variable
  70               const char * remote_coorm_path = getenv("REMOTE_COORM_PATH");
  71               if (remote_coorm_path == NULL)
  72               {
  73                       throw RunTimeException("Unable to get REMOTE_COORM_PATH environment variable");
  74               }
  75
  76               // We need omniORB to execute launcher.py
  77               const string set_env_cmd = "source " + string(remote_coorm_path) + "/coorm_prerequis.sh;";
  78
  79
  80               // define command to submit batch
  81               string subCommand = set_env_cmd + "python " + workDir + "/" + fileNameToLaunch + " --name="+ jobName +
  82                                   " --workdir=" + workDir + " --outputs=" + workDir + "/logs/outputs.log" +
  83                                   " --errors=" + workDir + "/logs/errors.log" +
  84                                   " --executable=" + scriptFile + " " + launcherArgs;
  85               string command = _protocol.getExecCommand(subCommand, _hostname, _username);
  86               command += " 2>&1";
  87               LOG(command);
  88
  89               // submit job
  90               string output;
  91               int status = Utils::getCommandOutput(command, output);
  92               LOG(output);
  93               if (status != 0) throw RunTimeException("Can't submit job, error was: " + output);
  94
  95               // read id of submitted job in output
  96               istringstream logfile(output);
  97               string sline, idline, id;
  98
  99               if (logfile)
 100               {
 101                 while (getline(logfile, sline) && sline != "")
 102                 {
 103                         idline = sline;
 104                 }
 105
 106                 vector<string> tokens;
 107                 JobInfo::Tokenize(idline, tokens, "=");
 108                 id = tokens[1] ;
 109               }
 110               else
 111               {
 112                 throw RunTimeException("Error in the submission of the job on the remote host");
 113               }
 114
 115               JobId jobid(this, (string) id);
 116               return jobid;
 117               }
 118
 119               // retire un job du gestionnaire
 120               void BatchManager_COORM::deleteJob(const JobId & jobid)
 121               {
 122               // Get REMOTE_COORM_PATH environment variable
 123               const char * remote_coorm_path = getenv("REMOTE_COORM_PATH");
 124               if (remote_coorm_path == NULL)
 125               {
 126                 throw RunTimeException("Unable to get REMOTE_COORM_PATH environment variable");
 127               }
 128
 129               // We need omniORB to execute launcher.py
 130               const string set_env_cmd = "source " + string(remote_coorm_path) + "/coorm_prerequis.sh;";
 131
 132               // define command to delete job
 133               string subCommand = set_env_cmd + "python " + string(remote_coorm_path) + "/coormdel.py --jobID=" + jobid.getReference();
 134               string command = _protocol.getExecCommand(subCommand, _hostname, _username);
 135               LOG(command);
 136
 137               int status = system(command.c_str());
 138               if (status)
 139                 throw RunTimeException("Can't delete job " + jobid.getReference());
 140
 141               LOG("job " << jobid.getReference() << " killed");
 142       }
 143
 144       // Renvoie l'etat du job
 145       JobInfo BatchManager_COORM::queryJob(const JobId & jobid)
 146       {
 147         // Get REMOTE_COORM_PATH environment variable
 148         const char * remote_coorm_path = getenv("REMOTE_COORM_PATH");
 149         if (remote_coorm_path == NULL)
 150         {
 151           throw RunTimeException("Unable to get REMOTE_COORM_PATH environment variable");
 152         }
 153
 154         // We need omniORB to execute launcher.py
 155         const string set_env_cmd = "source " + string(remote_coorm_path) + "/coorm_prerequis.sh;";
 156
 157         // define command to query batch
 158         string subCommand = set_env_cmd + "python " + string(remote_coorm_path) + "/coormstat.py --jobID=" + jobid.getReference();
 159         string command = _protocol.getExecCommand(subCommand, _hostname, _username);
 160         LOG(command);
 161         string output;
 162         int status = Utils::getCommandOutput(command, output);
 163         if(status && status != 153 && status != 256*153)
 164           throw RunTimeException("Error of connection on remote host");
 165
 166         JobInfo_COORM jobinfo = JobInfo_COORM(jobid.getReference(), output);
 167         return jobinfo;
 168       }
 169
 170       string BatchManager_COORM::buildBatchScript(const Job & job)
 171       {
 172         Parametre params = job.getParametre();
 173
 174         // Job Parameters
 175         string workDir       = "";
 176         string fileToExecute = "";
 177         string tmpDir = "";
 178         int nbproc               = 0;
 179         int edt          = 0;
 180         int mem              = 0;
 181         string queue         = "";
 182         LIBBATCH_UNUSED(nbproc);
 183         LIBBATCH_UNUSED(edt);
 184         LIBBATCH_UNUSED(mem);
 185
 186         // Mandatory parameters
 187         if (params.find(WORKDIR) != params.end())
 188           workDir = params[WORKDIR].str();
 189         else
 190           throw RunTimeException("params[WORKDIR] is not defined. Please define it, cannot submit this job.");
 191         if (params.find(EXECUTABLE) != params.end())
 192           fileToExecute = params[EXECUTABLE].str();
 193         else
 194           throw RunTimeException("params[EXECUTABLE] is not defined. Please define it, cannot submit this job.");
 195
 196         string::size_type p1 = fileToExecute.find_last_of("/");
 197         string::size_type p2 = fileToExecute.find_last_of(".");
 198         string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
 199         string fileNameToExecute = fileToExecute.substr(p1+1);
 200
 201         // Create batch submit file
 202         ofstream tempOutputFile;
 203         string tmpFileName = Utils::createAndOpenTemporaryFile("COORM-script", tempOutputFile);
 204
 205         tempOutputFile << "#!/bin/sh -f" << endl;
 206         tempOutputFile << "export LIBBATCH_NODEFILE=$COORM_NODEFILE" << endl;
 207         // Launch the executable
 208         tempOutputFile << "cd " << tmpDir << endl;
 209         tempOutputFile << "./" + fileNameToExecute << endl;
 210         tempOutputFile.flush();
 211         tempOutputFile.close();
 212
 213         Utils::chmod(tmpFileName.c_str(), 0x1ED);
 214         LOG("Batch script file generated is: " << tmpFileName);
 215
 216         string remoteFileName = rootNameToExecute + "_Batch.sh";
 217         int status = _protocol.copyFile(tmpFileName, "", "",
 218                                         workDir + "/" + remoteFileName,
 219                                         _hostname, _username);
 220         if (status)
 221           throw RunTimeException("Cannot copy batch submission file on host " + _hostname);
 222
 223         return remoteFileName;
 224       }
 225
 226       void BatchManager_COORM::exportInputFiles(const Job & job)
 227       {
 228         BatchManager::exportInputFiles(job);
 229
 230         int status;
 231         Parametre params = job.getParametre();
 232
 233         string launcherFile = params[LAUNCHER_FILE];
 234
 235         if (launcherFile.size() != 0)
 236         {
 237           // Copy launcherFile into batch working directory
 238           status = _protocol.copyFile(launcherFile, "", "",
 239                                       params[WORKDIR], _hostname, _username);
 240           if (status) {
 241             std::ostringstream oss;
 242             oss << "Cannot copy file " << launcherFile << " on host " << _hostname;
 243             oss << ". Return status is " << status;
 244             throw RunTimeException(oss.str());
 245           }
 246
 247           string remoteLauncher = launcherFile;
 248           remoteLauncher = remoteLauncher.substr(remoteLauncher.rfind("/") + 1, remoteLauncher.length());
 249           remoteLauncher = string(params[WORKDIR]) + "/" + remoteLauncher;
 250
 251           string subCommand = string("chmod u+x ") + remoteLauncher;
 252           string command = _protocol.getExecCommand(subCommand, _hostname, _username);
 253           LOG(command);
 254           status = system(command.c_str());
 255           if (status) {
 256             std::ostringstream oss;
 257             oss << "Cannot change permissions of file " << remoteLauncher << " on host " << _hostname;
 258             oss << ". Return status is " << status;
 259             throw RunTimeException(oss.str());
 260           }
 261         }
 262       }
 263 }