* Author : Renaud BARATE - EDF R&D
*/
+#include <cstdlib>
#include <iostream>
#include <fstream>
#include <Batch_NotYetImplementedException.hxx>
#include "Batch_BatchManager_eLL.hxx"
+#include "Batch_JobInfo_eLL.hxx"
using namespace std;
CommunicationProtocolType protocolType, const char * mpiImpl,
int nb_proc_per_node)
: BatchManager(parent, host),
- BatchManager_eClient(parent, host, username, protocolType, mpiImpl)
+ BatchManager_eClient(parent, host, username, protocolType, mpiImpl),
+ _nb_proc_per_node(nb_proc_per_node)
{
// Nothing to do
}
command += logFile;
cerr << command.c_str() << endl;
status = system(command.c_str());
- if(status)
+ if (status)
{
ifstream error_message(logFile.c_str());
string mess;
string temp;
- while(std::getline(error_message, temp))
- mess += temp;
+ while(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
string jobref;
ifstream idfile(logFile.c_str());
- unsigned int linebufsize = 1024;
- char linebuf[linebufsize];
- idfile.getline(linebuf, linebufsize);
- while (!idfile.eof() && strncmp(linebuf, "llsubmit:", 9) != 0)
- idfile.getline(linebuf, linebufsize);
+ string line;
+ while (idfile && line.compare(0, 9, "llsubmit:") != 0)
+ getline(idfile, line);
idfile.close();
- if (strncmp(linebuf, "llsubmit:", 9) == 0)
+ if (line.compare(0, 9, "llsubmit:") == 0)
{
- string line(linebuf);
string::size_type p1 = line.find_first_of("\"");
string::size_type p2 = line.find_last_of("\"");
if (p1 != p2)
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");
+ throw EmulationException("params[WORKDIR] is not defined. Please define 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(QUEUE) != params.end())
- queue = params[QUEUE].str();
+ throw EmulationException("params[EXECUTABLE] is not defined. Please define it, cannot submit this job.");
string::size_type p1 = fileToExecute.find_last_of("/");
string::size_type p2 = fileToExecute.find_last_of(".");
// Create batch submit file
ofstream tempOutputFile;
- std::string tmpFileName = createAndOpenTemporaryFile("LL-script", tempOutputFile);
+ string tmpFileName = createAndOpenTemporaryFile("LL-script", tempOutputFile);
- tempOutputFile << "# @ executable = " << fileNameToExecute << endl;
+ tempOutputFile << "#!/bin/bash" << endl;
tempOutputFile << "# @ output = " << workDir << "/logs/output.log." << rootNameToExecute << endl;
tempOutputFile << "# @ error = " << workDir << "/logs/error.log." << rootNameToExecute << endl;
- if (queue != "")
- tempOutputFile << "# @ class = " << queue << endl;
- tempOutputFile << "# @ job_type = bluegene" << endl;
+ tempOutputFile << "# @ node_usage = not_shared" << endl;
+
+ if (params.find(NAME) != params.end())
+ tempOutputFile << "# @ job_name = " << params[NAME] << endl;
+
+ // Optional parameters
+ int nbproc = 1;
+ if (params.find(NBPROC) != params.end())
+ nbproc = params[NBPROC];
+ //if (nbproc == 1)
+ // tempOutputFile << "# @ job_type = serial" << endl;
+ //else {
+ // tempOutputFile << "# @ job_type = parallel" << endl;
+ int nodes_requested = (nbproc + _nb_proc_per_node -1) / _nb_proc_per_node;
+ tempOutputFile << "# @ job_type = mpich" << endl;
+ tempOutputFile << "# @ node = " << nodes_requested << endl;
+ tempOutputFile << "# @ tasks_per_node = " << _nb_proc_per_node << endl;
+ //}
+ //tempOutputFile << "# @ job_type = bluegene" << endl;
+
+ if (params.find(MAXWALLTIME) != params.end())
+ tempOutputFile << "# @ wall_clock_limit = " << params[MAXWALLTIME] << ":00" << endl;
+ if (params.find(MAXRAMSIZE) != params.end())
+ tempOutputFile << "# @ as_limit = " << params[MAXRAMSIZE] << "mb" << endl;
+ if (params.find(QUEUE) != params.end())
+ tempOutputFile << "# @ class = " << params[QUEUE] << endl;
+
+ // Define environment for the job
+ Environnement env = job.getEnvironnement();
+ if (!env.empty()) {
+ tempOutputFile << "# @ environment = ";
+ Environnement::const_iterator iter;
+ for (iter = env.begin() ; iter != env.end() ; ++iter) {
+ tempOutputFile << iter->first << "=" << iter->second << "; ";
+ }
+ tempOutputFile << endl;
+ }
+
tempOutputFile << "# @ queue" << endl;
+ // generate nodes file
+ tempOutputFile << "NODEFILE=`mktemp nodefile-XXXXXXXXXX` || exit 1" << endl;
+ tempOutputFile << "for node in $LOADL_PROCESSOR_LIST; do" << endl;
+ tempOutputFile << " echo $node >> $NODEFILE" << endl;
+ tempOutputFile << "done" << endl;
+ tempOutputFile << "export LIBBATCH_NODEFILE=$NODEFILE" << endl;
+
+ // Launch the executable
+ tempOutputFile << "cd " << workDir << endl;
+ tempOutputFile << "./" + fileNameToExecute << endl;
+
tempOutputFile.flush();
tempOutputFile.close();
- cerr << "Batch script file generated is: " << tmpFileName.c_str() << endl;
+ cerr << "Batch script file generated is: " << tmpFileName << endl;
string remoteFileName = rootNameToExecute + "_LL.cmd";
int status = _protocol.copyFile(tmpFileName, "", "",
workDir + "/" + remoteFileName,
_hostname, _username);
if (status)
- throw EmulationException("Error of connection on remote host, cannot copy batch submission file");
+ throw EmulationException("Cannot copy command file on host " + _hostname);
return remoteFileName;
}
void BatchManager_eLL::deleteJob(const JobId & jobid)
{
- throw NotYetImplementedException("BatchManager_eLL::deleteJob");
+ // define command to delete job
+ string subCommand = "llcancel " + jobid.getReference();
+ string command = _protocol.getExecCommand(subCommand, _hostname, _username);
+ cerr << command.c_str() << endl;
+
+ int status = system(command.c_str());
+ if (status)
+ throw EmulationException("Can't delete job " + jobid.getReference());
+
+ cerr << "job " << jobid.getReference() << " killed" << endl;
}
void BatchManager_eLL::holdJob(const JobId & jobid)
JobInfo BatchManager_eLL::queryJob(const JobId & jobid)
{
- throw NotYetImplementedException("BatchManager_eLL::queryJob");
+ // define name of log file (local)
+ string logFile = generateTemporaryFileName("LL-querylog-" + jobid.getReference());
+
+ // define command to query batch
+ string subCommand = "llq -f %st " + jobid.getReference();
+ string command = _protocol.getExecCommand(subCommand, _hostname, _username);
+ command += " > ";
+ command += logFile;
+ cerr << command.c_str() << endl;
+ int status = system(command.c_str());
+ if (status != 0)
+ throw EmulationException("Can't query job " + jobid.getReference());
+
+ JobInfo_eLL jobinfo = JobInfo_eLL(jobid.getReference(), logFile);
+ return jobinfo;
}
const JobId BatchManager_eLL::addJob(const Job & job, const string reference)
