source: src/Fragmentation/Automation/FragmentScheduler.cpp@ a8f54b6

/*
 * Project: MoleCuilder
 * Description: creates and alters molecular systems
 * Copyright (C)  2011 University of Bonn. All rights reserved.
 * Please see the LICENSE file or "Copyright notice" in builder.cpp for details.
 */

/*
 * \file FragmentScheduler.cpp
 *
 * This file strongly follows the Serialization example from the boost::asio
 * library (see server.cpp)
 *
 *  Created on: Oct 19, 2011
 *      Author: heber
 */

// include config.h
#ifdef HAVE_CONFIG_H
#include <config.h>
#endif

// boost asio needs specific operator new
#include <boost/asio.hpp>

#include "CodePatterns/MemDebug.hpp"

#include <boost/bind.hpp>
#include <boost/lexical_cast.hpp>
#include <iostream>
#include <vector>
#include "Connection.hpp" // Must come before boost/serialization headers.
#include <boost/serialization/vector.hpp>
#include "CodePatterns/Info.hpp"
#include "CodePatterns/Log.hpp"
#include "Controller/Commands/EnrollInPoolOperation.hpp"
#include "Jobs/MPQCCommandJob.hpp"
#include "Jobs/SystemCommandJob.hpp"
#include "JobId.hpp"

#include "FragmentScheduler.hpp"

FragmentJob::ptr FragmentScheduler::WorkerListener_t::NoJob(new SystemCommandJob(std::string("/bin/true"), std::string("dosomething"), JobId::NoJob));

/** Helper function to enforce binding of FragmentWorker to possible derived
 * FragmentJob classes.
 */
void dummyInit() {
  SystemCommandJob("/bin/false", "something", JobId::IllegalJob);
  MPQCCommandJob("nofile", JobId::IllegalJob);
}

/** Constructor of class FragmentScheduler.
 *
 * We setup both acceptors to accept connections from workers and Controller.
 *
 * \param io_service io_service of the asynchronous communications
 * \param workerport port to listen for worker connections
 * \param controllerport port to listen for controller connections.
 */
FragmentScheduler::FragmentScheduler(boost::asio::io_service& io_service, unsigned short workerport, unsigned short controllerport) :
    WorkerListener(io_service, workerport, JobsQueue, pool,
        boost::bind(&FragmentScheduler::sendJobToWorker, boost::ref(*this), _1, _2)),
    ControllerListener(io_service, controllerport, JobsQueue,
        boost::bind(&Listener::initiateSocket, boost::ref(WorkerListener))),
    connection(io_service),
    sendJobOp(connection)
{
  Info info(__FUNCTION__);

  // listen for controller
  ControllerListener.initiateSocket();

  // only initiate socket if jobs are already present
  if (JobsQueue.isJobPresent()) {
    WorkerListener.initiateSocket();
  }
}

/** Handle a new worker connection.
 *
 * We store the given address in the pool.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::WorkerListener_t::handle_Accept(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  if (!e)
  {
    // Successfully accepted a new connection.
    // read address
    conn->async_read(address,
      boost::bind(&FragmentScheduler::WorkerListener_t::handle_checkAddress, this,
      boost::asio::placeholders::error, conn));
  }  else {
    // An error occurred. Log it and return. Since we are not starting a new
    // accept operation the io_service will run out of work to do and the
    // server will exit.
    Exitflag = ErrorFlag;
    ELOG(0, e.message());
  }
}

/** Callback function when worker address has been received.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::WorkerListener_t::handle_checkAddress(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  if (!e)
  {
    if (pool.presentInPool(address)) {
      // check whether its priority is busy_priority
      if (pool.isWorkerBusy(address)) {
        conn->async_read(result,
          boost::bind(&FragmentScheduler::WorkerListener_t::handle_ReceiveResultFromWorker, this,
          boost::asio::placeholders::error, conn));
      } else {
        ELOG(1, "INFO: Idle worker "+toString(address)+" has logged in again.");
      }
    } else {
      // insert as its new worker
      pool.addWorker(address);
      enum EnrollInPoolOperation::EnrollFlag flag = EnrollInPoolOperation::Success;
      conn->async_write(flag,
        boost::bind(&FragmentScheduler::WorkerListener_t::handle_enrolled, this,
        boost::asio::placeholders::error, conn));
    }
  }
  else
  {
    // An error occurred. Log it and return. Since we are not starting a new
    // accept operation the io_service will run out of work to do and the
    // server will exit.
    Exitflag = ErrorFlag;
    ELOG(0, e.message());
  }

  if (JobsQueue.isJobPresent()) {
    // Start an accept operation for a new Connection only when there
    // are still jobs present
    initiateSocket();
  }
}

/** Callback function when new worker has enrolled.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::WorkerListener_t::handle_enrolled(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  if (!e)
  {
    FragmentJob::ptr job;
    if (JobsQueue.isJobPresent()) {
      job = JobsQueue.popJob();
    } else {
      job = NoJob;
    }
    callback_sendJobToWorker(address, job);
  }
  else
  {
    // An error occurred. Log it and return. Since we are not starting a new
    // accept operation the io_service will run out of work to do and the
    // server will exit.
    Exitflag = ErrorFlag;
    ELOG(0, e.message());
  }
}

/** Callback function when result has been received.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::WorkerListener_t::handle_ReceiveResultFromWorker(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  LOG(1, "INFO: Received result for job #" << result->getId() << " ...");

  // and push into queue
  ASSERT(result->getId() != (JobId_t)JobId::NoJob,
      "WorkerListener_t::handle_ReceiveResultFromWorker() - result received has NoJob id.");
  ASSERT(result->getId() != (JobId_t)JobId::IllegalJob,
      "WorkerListener_t::handle_ReceiveResultFromWorker() - result received has IllegalJob id.");
  // place id into expected
  if ((result->getId() != (JobId_t)JobId::NoJob) && (result->getId() != (JobId_t)JobId::IllegalJob))
    JobsQueue.pushResult(result);

  // mark as idle
  pool.unmarkWorkerBusy(address);
  // for now remove worker again from pool such that other may connect
  pool.removeWorker(address);

  // erase result
  result.reset();
  LOG(1, "INFO: JobsQueue has " << JobsQueue.getDoneJobs() << " results.");
}


/** Handle a new controller connection.
 *
 * \sa handle_ReceiveJobs()
 * \sa handle_CheckResultState()
 * \sa handle_SendResults()
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::ControllerListener_t::handle_Accept(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  if (!e)
  {
    conn->async_read(choice,
      boost::bind(&FragmentScheduler::ControllerListener_t::handle_ReadChoice, this,
      boost::asio::placeholders::error, conn));
  }
  else
  {
    // An error occurred. Log it and return. Since we are not starting a new
    // accept operation the io_service will run out of work to do and the
    // server will exit.
    Exitflag = ErrorFlag;
    ELOG(0, e.message());
  }
}

/** Controller callback function to read the choice for next operation.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::ControllerListener_t::handle_ReadChoice(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  if (!e)
  {
    bool LaunchNewAcceptor = true;
    LOG(1, "INFO: Received request for operation " << choice << ".");
    // switch over the desired choice read previously
    switch(choice) {
    case NoControllerOperation:
    {
      ELOG(1, "FragmentScheduler::handle_ReadChoice() - called with NoOperation.");
      break;
    }
    case GetNextJobId:
    {
      const JobId_t nextid = globalId.getNextId();
      LOG(1, "INFO: Sending next available job id " << nextid << " to controller ...");
      conn->async_write(nextid,
        boost::bind(&FragmentScheduler::ControllerListener_t::handle_GetNextJobIdState, this,
        boost::asio::placeholders::error, conn));
      break;
    }
    case ReceiveJobs:
    {
      // The connection::async_write() function will automatically
      // serialize the data structure for us.
      LOG(1, "INFO: Receiving bunch of jobs from a controller ...");
      conn->async_read(jobs,
        boost::bind(&FragmentScheduler::ControllerListener_t::handle_ReceiveJobs, this,
        boost::asio::placeholders::error, conn));
      break;
    }
    case CheckState:
    {
      // first update number
      jobInfo[0] = JobsQueue.getPresentJobs();
      jobInfo[1] = JobsQueue.getDoneJobs();
      // now we accept connections to check for state of calculations
      LOG(1, "INFO: Sending state that "+toString(jobInfo[0])+" jobs are present and "+toString(jobInfo[1])+" jobs are done to controller ...");
      conn->async_write(jobInfo,
        boost::bind(&FragmentScheduler::ControllerListener_t::handle_CheckResultState, this,
        boost::asio::placeholders::error, conn));
      break;
    }
    case SendResults:
    {
      const std::vector<FragmentResult::ptr> results = JobsQueue.getAllResults();
      // ... or we give the results
      LOG(1, "INFO: Sending "+toString(results.size())+" results to controller ...");
      conn->async_write(results,
        boost::bind(&FragmentScheduler::ControllerListener_t::handle_SendResults, this,
        boost::asio::placeholders::error, conn));
      break;
    }
    case ShutdownControllerSocket:
    {
      LaunchNewAcceptor = false;
      break;
    }
    default:
      Exitflag = ErrorFlag;
      ELOG(1, "FragmentScheduler::handle_ReadChoice() - called with no valid choice.");
      break;
    }
    // restore NoControllerOperation choice such that choice is not read twice
    choice = NoControllerOperation;

    if (LaunchNewAcceptor) {
      LOG(1, "Launching new acceptor on socket.");
      // Start an accept operation for a new Connection.
      initiateSocket();
    }
  }
  else
  {
    // An error occurred. Log it and return. Since we are not starting a new
    // accept operation the io_service will run out of work to do and the
    // server will exit.
    Exitflag = ErrorFlag;
    ELOG(0, e.message());
  }
}

/** Controller callback function when job has been sent.
 *
 * We check here whether the worker socket is accepting, if there
 * have been no jobs we re-activate it, as it is shut down after
 * last job.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::ControllerListener_t::handle_ReceiveJobs(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  bool need_initiateSocket = !JobsQueue.isJobPresent();

  // jobs are received, hence place in JobsQueue
  if (!jobs.empty()) {
    LOG(1, "INFO: Pushing " << jobs.size() << " jobs into queue.");
    JobsQueue.pushJobs(jobs);
  }

  jobs.clear();

  // initiate socket if we had no jobs before
  if (need_initiateSocket)
    initiateWorkerSocket();
}

/** Controller callback function when checking on state of results.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::ControllerListener_t::handle_CheckResultState(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  // do nothing
  LOG(1, "INFO: Sent that " << jobInfo << " jobs are (scheduled, done).");
}

/** Controller callback function when checking on state of results.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::ControllerListener_t::handle_GetNextJobIdState(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  // do nothing
  LOG(1, "INFO: Sent next available job id.");
}

/** Controller callback function when result has been received.
 *
 * \param e error code if something went wrong
 * \param conn reference with the connection
 */
void FragmentScheduler::ControllerListener_t::handle_SendResults(const boost::system::error_code& e, connection_ptr conn)
{
  Info info(__FUNCTION__);
  // do nothing
  LOG(1, "INFO: Results have been sent.");
}


/** Helper function to send a job to worker.
 *
 * @param address address of worker
 * @param job job to send
 */
void FragmentScheduler::sendJobToWorker(const WorkerAddress &address, FragmentJob::ptr &job)
{
  ASSERT( !pool.isWorkerBusy(address),
      "FragmentScheduler::sendJobToWorker() - Worker "+toString(address)+" is already busy.");
  LOG(1, "INFO: Sending job " << job->getId() << " to worker " << address << ".");
  sendJobOp.setJob(job);
  sendJobOp(address.host, address.service);
  // set worker as busy (assuming it has been removed from idle queue already)
  WorkerPool::Idle_Queue_t::iterator iter = pool.getIdleWorker(address);
  ASSERT( iter != pool.getIdleEnd(),
      "FragmentScheduler::sendJobToWorker() - cannot find worker "
      +toString(address)+" in idle queue.");
  pool.markWorkerBusy(iter);
}

///** Helper function to shutdown a single worker.
// *
// * We send NoJob to indicate shutdown
// *
// * @param address of worker to shutdown
// */
//void FragmentScheduler::shutdownWorker(const WorkerAddress &address)
//{
//  sendJobToWorker(address, NoJob);
//}
//
///** Sends shutdown to all current workers in the pool.
// *
// */
//void FragmentScheduler::removeAllWorkers()
//{
//  // give all workers shutdown signal
//  while (pool.presentIdleWorkers()) {
//    const WorkerAddress address = pool.getNextIdleWorker();
//    shutdownWorker(address);
//  }
//}