-// Copyright (C) 2007-2008 CEA/DEN, EDF R&D, OPEN CASCADE
+// Copyright (C) 2007-2010 CEA/DEN, EDF R&D, OPEN CASCADE
//
// Copyright (C) 2003-2007 OPEN CASCADE, EADS/CCR, LIP6, CEA/DEN,
// CEDRAT, EDF R&D, LEG, PRINCIPIA R&D, BUREAU VERITAS
*
*/
+#include <stdlib.h>
+#include <string.h>
+
#include <iostream>
#include <fstream>
#include <sstream>
+#include <string>
#include <sys/stat.h>
-#include <string.h>
+
#include <stdlib.h>
+#include <string.h>
-#include "Batch_BatchManager_eLSF.hxx"
#ifdef WIN32
-# include <time.h>
-# include <io.h>
+#include <io.h>
#else
-# include <libgen.h>
+#include <libgen.h>
#endif
+#include "Batch_BatchManager_eLSF.hxx"
+#include "Batch_JobInfo_eLSF.hxx"
+
using namespace std;
namespace Batch {
- BatchManager_eLSF::BatchManager_eLSF(const FactBatchManager * parent, const char * host, const char * protocol, const char * mpiImpl) throw(InvalidArgumentException,ConnexionFailureException) : BatchManager_eClient(parent,host,protocol,mpiImpl)
+ BatchManager_eLSF::BatchManager_eLSF(const FactBatchManager * parent, const char * host,
+ const char * username,
+ CommunicationProtocolType protocolType, const char * mpiImpl)
+ : BatchManager_eClient(parent, host, username, protocolType, mpiImpl),
+ BatchManager(parent, host)
{
// Nothing to do
}
{
int status;
Parametre params = job.getParametre();
- const std::string dirForTmpFiles = params[TMPDIR];
+ const std::string workDir = params[WORKDIR];
const string fileToExecute = params[EXECUTABLE];
- std::string fileNameToExecute;
- if( fileToExecute.size() > 0 ){
- string::size_type p1 = fileToExecute.find_last_of("/");
- string::size_type p2 = fileToExecute.find_last_of(".");
- fileNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
- }
- else
- fileNameToExecute = "command";
+ string::size_type p1 = fileToExecute.find_last_of("/");
+ string::size_type p2 = fileToExecute.find_last_of(".");
+ std::string fileNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
// export input files on cluster
+ cerr << "Export des fichiers en entree" << endl;
exportInputFiles(job);
// build batch script for job
+ cerr << "Construction du script de batch" << endl;
buildBatchScript(job);
+ cerr << "Script envoye" << endl;
- // define name of log file
- string logFile="/tmp/logs/";
- logFile += getenv("USER");
- logFile += "/batchSalome_";
- srand ( time(NULL) );
- int ir = rand();
- ostringstream oss;
- oss << ir;
- logFile += oss.str();
- logFile += ".log";
-
- string command;
+ // define name of log file (local)
+ string logFile = generateTemporaryFileName("LSF-submitlog");
// define command to submit batch
- command = _protocol;
- command += " ";
-
- if(_username != ""){
- command += _username;
- command += "@";
- }
-
- command += _hostname;
- command += " \"cd " ;
- command += dirForTmpFiles ;
- command += "; bsub < " ;
- command += fileNameToExecute ;
- command += "_Batch.sh\" > ";
+ string subCommand = string("cd ") + workDir + "; bsub < " + fileNameToExecute + "_Batch.sh";
+ string command = _protocol.getExecCommand(subCommand, _hostname, _username);
+ command += " > ";
command += logFile;
+ command += " 2>&1";
cerr << command.c_str() << endl;
status = system(command.c_str());
if(status)
- throw EmulationException("Error of connection on remote host");
+ {
+ ifstream error_message(logFile.c_str());
+ std::string mess;
+ std::string temp;
+ while(std::getline(error_message, temp))
+ mess += temp;
+ error_message.close();
+ throw EmulationException("Error of connection on remote host, error was: " + mess);
+ }
// read id of submitted job in log file
char line[128];
return id;
}
+ // Ce manager permet de faire de la reprise
+ const Batch::JobId
+ BatchManager_eLSF::addJob(const Batch::Job & job, const std::string reference)
+ {
+ return JobId(this, reference);
+ }
+
// Methode pour le controle des jobs : retire un job du gestionnaire
void BatchManager_eLSF::deleteJob(const JobId & jobid)
{
istringstream iss(jobid.getReference());
iss >> ref;
- // define command to submit batch
- string command;
- command = _protocol;
- command += " ";
-
- if (_username != ""){
- command += _username;
- command += "@";
- }
-
- command += _hostname;
- command += " \"bkill " ;
- command += iss.str();
- command += "\"";
+ // define command to delete batch
+ string subCommand = string("bkill ") + iss.str();
+ string command = _protocol.getExecCommand(subCommand, _hostname, _username);
cerr << command.c_str() << endl;
status = system(command.c_str());
- if(status)
+ if (status)
throw EmulationException("Error of connection on remote host");
cerr << "jobId = " << ref << "killed" << endl;
istringstream iss(jobid.getReference());
iss >> id;
- // define name of log file
- string logFile="/tmp/logs/";
- logFile += getenv("USER");
- logFile += "/batchSalome_";
-
- srand ( time(NULL) );
- int ir = rand();
- ostringstream oss;
- oss << ir;
- logFile += oss.str();
- logFile += ".log";
-
- string command;
- int status;
-
- // define command to submit batch
- command = _protocol;
- command += " ";
-
- if (_username != ""){
- command += _username;
- command += "@";
- }
+ // define name of log file (local)
+ string logFile = generateTemporaryFileName(string("LSF-querylog-id") + jobid.getReference());
- command += _hostname;
- command += " \"bjobs " ;
- command += iss.str();
- command += "\" > ";
+ // define command to query batch
+ string subCommand = string("bjobs ") + iss.str();
+ string command = _protocol.getExecCommand(subCommand, _hostname, _username);
+ command += " > ";
command += logFile;
cerr << command.c_str() << endl;
- status = system(command.c_str());
- if(status)
+ int status = system(command.c_str());
+ if (status)
throw EmulationException("Error of connection on remote host");
JobInfo_eLSF ji = JobInfo_eLSF(id,logFile);
void BatchManager_eLSF::buildBatchScript(const Job & job)
{
#ifndef WIN32 //TODO: need for porting on Windows
- int status;
Parametre params = job.getParametre();
- Environnement env = job.getEnvironnement();
- const int nbproc = params[NBPROC];
- const long edt = params[MAXWALLTIME];
- const long mem = params[MAXRAMSIZE];
- const string workDir = params[WORKDIR];
- const std::string dirForTmpFiles = params[TMPDIR];
- const string fileToExecute = params[EXECUTABLE];
- const string home = params[HOMEDIR];
- const std::string queue = params[QUEUE];
- std::string rootNameToExecute;
- std::string fileNameToExecute;
- std::string filelogtemp;
- if( fileToExecute.size() > 0 ){
- string::size_type p1 = fileToExecute.find_last_of("/");
- string::size_type p2 = fileToExecute.find_last_of(".");
- rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
- char* basec=strdup(fileToExecute.c_str());
- fileNameToExecute = "~/" + dirForTmpFiles + "/" + string(basename(basec));
- free(basec);
-
- int idx = dirForTmpFiles.find("Batch/");
- filelogtemp = dirForTmpFiles.substr(idx+6, dirForTmpFiles.length());
- }
- else{
- rootNameToExecute = "command";
- }
+ // Job Parameters
+ string workDir = "";
+ string fileToExecute = "";
+ int nbproc = 0;
+ int edt = 0;
+ int mem = 0;
+ string queue = "";
+
+ // Mandatory parameters
+ if (params.find(WORKDIR) != params.end())
+ workDir = params[WORKDIR].str();
+ else
+ throw EmulationException("params[WORKDIR] is not defined ! Please defined it, cannot submit this job");
+ if (params.find(EXECUTABLE) != params.end())
+ fileToExecute = params[EXECUTABLE].str();
+ else
+ throw EmulationException("params[EXECUTABLE] is not defined ! Please defined it, cannot submit this job");
+
+ // Optional parameters
+ if (params.find(NBPROC) != params.end())
+ nbproc = params[NBPROC];
+ if (params.find(MAXWALLTIME) != params.end())
+ edt = params[MAXWALLTIME];
+ if (params.find(MAXRAMSIZE) != params.end())
+ mem = params[MAXRAMSIZE];
+ if (params.find(QUEUE) != params.end())
+ queue = params[QUEUE].str();
+
+ string::size_type p1 = fileToExecute.find_last_of("/");
+ string::size_type p2 = fileToExecute.find_last_of(".");
+ string rootNameToExecute = fileToExecute.substr(p1+1,p2-p1-1);
+ string fileNameToExecute = fileToExecute.substr(p1+1);
+
+ // Create batch submit file
ofstream tempOutputFile;
- std::string TmpFileName = createAndOpenTemporaryFile(tempOutputFile);
+ std::string TmpFileName = createAndOpenTemporaryFile("LSF-script", tempOutputFile);
tempOutputFile << "#! /bin/sh -f" << endl ;
if (queue != "")
if( mem > 0 )
tempOutputFile << "#BSUB -M " << mem*1024 << endl ;
tempOutputFile << "#BSUB -n " << nbproc << endl ;
- if( fileToExecute.size() > 0 ){
- tempOutputFile << "#BSUB -o " << home << "/" << dirForTmpFiles << "/output.log." << filelogtemp << endl ;
- tempOutputFile << "#BSUB -e " << home << "/" << dirForTmpFiles << "/error.log." << filelogtemp << endl ;
- }
- else{
- tempOutputFile << "#BSUB -o " << dirForTmpFiles << "/" << env["LOGFILE"] << ".output.log" << endl ;
- tempOutputFile << "#BSUB -e " << dirForTmpFiles << "/" << env["LOGFILE"] << ".error.log" << endl ;
- }
- if( workDir.size() > 0 )
- tempOutputFile << "cd " << workDir << endl ;
- if( fileToExecute.size() > 0 ){
- tempOutputFile << _mpiImpl->boot("",nbproc);
- tempOutputFile << _mpiImpl->run("",nbproc,fileNameToExecute);
- tempOutputFile << _mpiImpl->halt();
- }
+ size_t pos = workDir.find("$HOME");
+ string baseDir;
+ if( pos != string::npos )
+ baseDir = getHomeDir(workDir) + workDir.substr(pos+5,workDir.length()-5);
else{
- tempOutputFile << "source " << env["SOURCEFILE"] << endl ;
- tempOutputFile << env["COMMAND"];
+ pos = workDir.find("~");
+ if( pos != string::npos )
+ baseDir = getHomeDir(workDir) + workDir.substr(pos+1,workDir.length()-1);
+ else
+ baseDir = workDir;
}
+ tempOutputFile << "#BSUB -o " << baseDir << "/logs/output.log." << rootNameToExecute << endl ;
+ tempOutputFile << "#BSUB -e " << baseDir << "/logs/error.log." << rootNameToExecute << endl ;
+
+ tempOutputFile << "cd " << workDir << endl ;
+
+ // generate nodes file
+ tempOutputFile << "NODEFILE=`mktemp nodefile-XXXXXXXXXX` || exit 1" << endl;
+ tempOutputFile << "bool=0" << endl;
+ tempOutputFile << "for i in $LSB_MCPU_HOSTS; do" << endl;
+ tempOutputFile << " if test $bool = 0; then" << endl;
+ tempOutputFile << " n=$i" << endl;
+ tempOutputFile << " bool=1" << endl;
+ tempOutputFile << " else" << endl;
+ tempOutputFile << " for ((j=0;j<$i;j++)); do" << endl;
+ tempOutputFile << " echo $n >> $NODEFILE" << endl;
+ tempOutputFile << " done" << endl;
+ tempOutputFile << " bool=0" << endl;
+ tempOutputFile << " fi" << endl;
+ tempOutputFile << "done" << endl;
+
+ // Abstraction of PBS_NODEFILE - TODO
+ tempOutputFile << "export LIBBATCH_NODEFILE=$NODEFILE" << endl;
+
+ // Launch the executable
+ tempOutputFile << "./" + fileNameToExecute << endl;
+
+ // Remove the node file
+ tempOutputFile << "rm $NODEFILE" << endl;
tempOutputFile.flush();
tempOutputFile.close();
-#ifdef WIN32
- _chmod(
-#else
- chmod(
-#endif
- TmpFileName.c_str(), 0x1ED);
- cerr << TmpFileName.c_str() << endl;
-
- string command;
- if( _protocol == "rsh" )
- command = "rcp ";
- else if( _protocol == "ssh" )
- command = "scp ";
- else
- throw EmulationException("Unknown protocol");
- command += TmpFileName;
- command += " ";
- if(_username != ""){
- command += _username;
- command += "@";
- }
- command += _hostname;
- command += ":";
- command += dirForTmpFiles ;
- command += "/" ;
- command += rootNameToExecute ;
- command += "_Batch.sh" ;
- cerr << command.c_str() << endl;
- status = system(command.c_str());
- if(status)
+
+ BATCH_CHMOD(TmpFileName.c_str(), 0x1ED);
+ cerr << "Batch script file generated is: " << TmpFileName.c_str() << endl;
+
+ int status = _protocol.copyFile(TmpFileName, "", "",
+ workDir + "/" + rootNameToExecute + "_Batch.sh",
+ _hostname, _username);
+ if (status)
throw EmulationException("Error of connection on remote host");
- remove(TmpFileName.c_str());
#endif
}
return oss.str();
}
+ std::string BatchManager_eLSF::getHomeDir(std::string tmpdir)
+ {
+ std::string home;
+ int idx = tmpdir.find("Batch/");
+ std::string filelogtemp = tmpdir.substr(idx+6, tmpdir.length());
+ filelogtemp = "/tmp/logs" + filelogtemp + "_home";
+
+ string subCommand = string("echo $HOME");
+ string command = _protocol.getExecCommand(subCommand, _hostname, _username) + " > " + filelogtemp;
+ cerr << command.c_str() << endl;
+ int status = system(command.c_str());
+ if (status)
+ throw EmulationException("Error of launching home command on remote host");
+
+ std::ifstream file_home(filelogtemp.c_str());
+ std::getline(file_home, home);
+ file_home.close();
+ return home;
+ }
+
}
