Files
ResInsight/ApplicationLibCode/ProjectDataModel/Batch/RimBatchQueueSlurm.cpp
T
JJ 6c39729e77 Support submitting opm flow jobs to Slurm or LSF (#14943)
Add preferences for HPC/Batch scheduler
Add Slurm and LSF support
Add batch job monitor
Support launcher specific user settings
Use internal timer to refresh job state
Add some more options to batch monitor
Improve naming
Add tests for job lists
Change WSL handling for processes.
2026-10-21 18:14:13 +02:00

204 lines
7.0 KiB
C++

/////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2026 Equinor ASA
//
// ResInsight is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// ResInsight is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or
// FITNESS FOR A PARTICULAR PURPOSE.
//
// See the GNU General Public License at <http://www.gnu.org/licenses/gpl.html>
// for more details.
//
/////////////////////////////////////////////////////////////////////////////////
#include "RimBatchQueueSlurm.h"
#include "RiaHpcTools.h"
#include "RiaLogging.h"
#include "RiaPreferencesBatch.h"
#include "RiaPreferencesOpm.h"
#include "RiaWslTools.h"
#include "ProcessControl/RimProcess.h"
#include "RimBatchProcessMonitor.h"
CAF_PDM_SOURCE_INIT( RimBatchQueueSlurm, "BatchQueueSlurm" );
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
RimBatchQueueSlurm::RimBatchQueueSlurm( RiaDefines::BatchSchedulerType schedulerType )
: RimBatchQueue( schedulerType )
{
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
RimBatchQueueSlurm::~RimBatchQueueSlurm()
{
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
void RimBatchQueueSlurm::queueProcess( std::shared_ptr<RimProcess> process, int nProcessesPerTask )
{
if ( process == nullptr ) return;
m_process = process;
m_monitor = std::make_shared<RimBatchProcessMonitor>( this );
auto useWsl = RiaPreferencesOpm::current()->useWsl();
auto prefs = RiaPreferencesBatch::current();
QString workDir = m_process->workingDirectory();
// build launch script
auto [builtOk, script] = buildLaunchScript( workDir );
if ( !builtOk )
{
m_monitor->finished( 1, QProcess::ExitStatus::NormalExit );
RiaLogging::warning( QString( script ).toStdString() );
return;
}
QString jobName = generateJobName();
m_stdOutFileName = QString( "%1/%2.out" ).arg( workDir ).arg( jobName );
m_stdErrFileName = QString( "%1/%2.err" ).arg( workDir ).arg( jobName );
cleanUpOldLogFiles();
QString stdOut = m_stdOutFileName;
QString stdErr = m_stdErrFileName;
if ( useWsl )
{
workDir = RiaWslTools::convertToWslPath( workDir );
script = RiaWslTools::convertToWslPath( script );
stdOut = RiaWslTools::convertToWslPath( stdOut );
stdErr = RiaWslTools::convertToWslPath( stdErr );
}
QStringList arguments =
generateCommand( script, jobName, workDir, stdOut, stdErr, prefs->queueName(), prefs->exclusiveJob(), nProcessesPerTask );
auto [batchProcess, output] = runCommand( arguments, m_monitor );
if ( !batchProcess )
{
m_monitor->finished( 1, QProcess::ExitStatus::NormalExit );
RiaLogging::warning( output.toStdString() );
}
else
{
if ( m_process->monitor() ) m_process->monitor()->started();
m_batchProcess = std::move( batchProcess );
}
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
QStringList RimBatchQueueSlurm::generateCommand( QString script,
QString jobName,
QString workDir,
QString stdOutLog,
QString stdErrLog,
QString queueName,
bool exclusive,
int numberOfProcesses )
{
QStringList arguments;
arguments << "sbatch";
arguments << "-p";
arguments << queueName;
arguments << "-J";
arguments << jobName;
if ( exclusive )
{
arguments << "--exclusive";
}
if ( !workDir.isEmpty() )
{
arguments << "-D";
arguments << workDir;
arguments << "-o";
arguments << stdOutLog;
arguments << "-e";
arguments << stdErrLog;
}
arguments << "-n";
arguments << QString( "%1" ).arg( numberOfProcesses );
arguments << "--wait";
arguments << script;
return arguments;
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
QString RimBatchQueueSlurm::jobId() const
{
if ( m_batchProcess )
{
return RiaHpcTools::findJobId( schedulerType(), m_batchProcess->stdOut() );
}
return QString();
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
void RimBatchQueueSlurm::stopProcess()
{
if ( m_batchProcess && m_batchProcess->isRunning() )
{
auto id = jobId();
if ( !id.isEmpty() )
{
RiaLogging::info( QString( "Stopping batch job %1" ).arg( id ).toStdString() );
RiaHpcTools::stopJob( schedulerType(), id, RiaPreferencesOpm::current()->wslOptions() );
}
}
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
void RimBatchQueueSlurm::setFinished( bool runOk )
{
if ( m_batchProcess && m_process && m_process->monitor() )
{
auto jobId = RiaHpcTools::findJobId( schedulerType(), m_batchProcess->stdOut() );
readStdOutErrIntoProcessLog();
if ( runOk )
{
RiaLogging::info( QString( "Batch job %1 completed successfully" ).arg( jobId ).toStdString() );
m_process->monitor()->finished( 0, QProcess::ExitStatus::NormalExit );
}
else
{
RiaLogging::warning( QString( "Batch job %1 failed." ).arg( jobId ).toStdString() );
m_process->monitor()->finished( 1, QProcess::ExitStatus::CrashExit );
}
}
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
QString RimBatchQueueSlurm::launchScriptDirective() const
{
return "#SBATCH";
}