SlurmDriver.cxx File Reference

#include <EventLoop/SlurmDriver.h>
#include <AsgMessaging/StatusCode.h>
#include <EventLoop/BatchJob.h>
#include <EventLoop/Job.h>
#include <EventLoop/ManagerData.h>
#include <EventLoop/MessageCheck.h>
#include <RootCoreUtils/Assert.h>
#include <RootCoreUtils/ShellExec.h>
#include <TSystem.h>
#include <fstream>
#include <memory>
#include <sstream>

Go to the source code of this file.

Functions
	ClassImp (EL::SlurmDriver) namespace EL

Function Documentation

ClassImp()

ClassImp

(

EL::SlurmDriver

)

Author

Nils Krumnack

Definition at line 30 of file SlurmDriver.cxx.

{
  //****************************************************
  void SlurmDriver :: testInvariant () const
  {}
  //****************************************************
  SlurmDriver :: SlurmDriver ()
  {
    m_b_job_name = false;
    m_b_account = false;
    m_b_run_time = false;
 
    RCU_NEW_INVARIANT (this);
  }
  //****************************************************
  ::StatusCode SlurmDriver ::
  doManagerStep (Detail::ManagerData& data) const
  {
    RCU_READ_INVARIANT (this);
    using namespace msgEventLoop;
    ANA_CHECK (BatchDriver::doManagerStep (data));
    switch (data.step)
    {
    case Detail::ManagerStep::batchScriptVar:
      {
        data.batchInit = "export PATH LD_LIBRARY_PATH PYTHONPATH";
      }
      break;
 
    case Detail::ManagerStep::submitJob:
    case Detail::ManagerStep::doResubmit:
      {
        auto all_set = m_b_job_name && m_b_account && m_b_run_time;
        if (!all_set)
        {
          ANA_MSG_INFO ("Job Name " << m_job_name);
          ANA_MSG_INFO ("Account " << m_account);
          ANA_MSG_INFO ("Run Time " << m_run_time);
 
          ANA_MSG_ERROR("All parameters need to be set before job can be submitted");
          return ::StatusCode::FAILURE;
        }
 
        if (data.resubmit)
        {
          ANA_MSG_ERROR ("resubmission not supported for the Slurm driver");
          return StatusCode::FAILURE;
        }
 
        if (data.batchJobIndices.empty())
        {
          ANA_MSG_ERROR ("no job indices to submit");
          return ::StatusCode::FAILURE;
        }
        if (data.batchJobIndices.back() + 1 != data.batchJobIndices.size())
        {
          ANA_MSG_ERROR ("submitting a non-contiguous set of job indices is not supported");
          return ::StatusCode::FAILURE;
        }
        const std::size_t njob = data.batchJobIndices.size();
 
        if(!data.sharedFileSystem)
        {
          int status=gSystem->CopyFile("RootCore.par",(data.submitDir+"/submit/RootCore.par").c_str());
          if(status != 0)
          {
            ANA_MSG_ERROR ("failed to copy RootCore.par");
            return StatusCode::FAILURE;
          }
        }
 
        {
          std::ofstream file ((data.submitDir + "/submit/submit").c_str());
 
          file << "#!/bin/bash \n";
          file << "\n";
          file << "#SBATCH --job-name=" << m_job_name << "\n";
          file << "#SBATCH --output=slurm-%j.out\n";
          file << "#SBATCH --error=slurm-%j.err\n";
          file << "#SBATCH --account=" << m_account << "\n";
          if(!m_partition .empty()) file << "#SBATCH --partition=" << m_partition << "\n";
          file << "#SBATCH --time=" << m_run_time << "\n";
          if(!m_memory    .empty()) file << "#SBATCH --mem=" << m_memory << "\n";
          if(!m_constraint.empty()) file << "#SBATCH --constraint=" << m_constraint << "\n";
          file << "\n";
          file << data.options.castString(Job::optBatchSlurmExtraConfigLines) << "\n";
          file << "\n";
          //note: no "\n" at the of this string since this goes as pre-command to the execution of the next line
          file << data.options.castString(Job::optBatchSlurmWrapperExec);
          file << "./run ${SLURM_ARRAY_TASK_ID}\n";
        }
 
        {
          std::ostringstream cmd;
          cmd << "cd " << RCU::Shell::quote (data.submitDir) << "/submit && sbatch --array=0-" << njob-1 << " " << data.options.castString (Job::optSubmitFlags) << " submit";
          if (gSystem->Exec (cmd.str().c_str()) != 0)
          {
            ANA_MSG_ERROR ("failed to execute: " << cmd.str());
            return StatusCode::FAILURE;
          }
        }
        data.submitted = true;
      }
      break;
 
    default:
      break;
    }
    return ::StatusCode::SUCCESS;
  }
 
  //****************************************************
  void SlurmDriver :: SetJobName(std::string job_name)
  {
    m_b_job_name = true;
    m_job_name = job_name;
  }
  void SlurmDriver :: SetAccount(std::string account)
  {
    m_b_account = true;
    m_account = account;
  }
  void SlurmDriver :: SetPartition(std::string partition)
  {
    m_partition = partition;
  }
  void SlurmDriver :: SetRunTime(std::string run_time)
  {
    m_b_run_time = true;
    m_run_time = run_time;
  }
  void SlurmDriver :: SetMemory(std::string memory)
  {
    m_memory = memory;
  }
  void SlurmDriver :: SetConstrain(std::string constraint)
  {
    m_constraint = constraint;
  }
  //****************************************************
}