EventProcessor.cc

Go to the documentation of this file.

#include "FWCore/Framework/interface/EventProcessor.h"

#include "DataFormats/Provenance/interface/BranchIDListHelper.h"
#include "DataFormats/Provenance/interface/EntryDescriptionRegistry.h"
#include "DataFormats/Provenance/interface/ModuleDescription.h"
#include "DataFormats/Provenance/interface/ParameterSetID.h"
#include "DataFormats/Provenance/interface/ParentageRegistry.h"
#include "DataFormats/Provenance/interface/ProcessConfigurationRegistry.h"
#include "DataFormats/Provenance/interface/ProcessHistoryRegistry.h"

#include "FWCore/Framework/interface/CommonParams.h"
#include "FWCore/Framework/interface/EDLooperBase.h"
#include "FWCore/Framework/interface/EventPrincipal.h"
#include "FWCore/Framework/interface/EventSetupProvider.h"
#include "FWCore/Framework/interface/EventSetupRecord.h"
#include "FWCore/Framework/interface/FileBlock.h"
#include "FWCore/Framework/interface/HistoryAppender.h"
#include "FWCore/Framework/interface/InputSourceDescription.h"
#include "FWCore/Framework/interface/IOVSyncValue.h"
#include "FWCore/Framework/interface/LooperFactory.h"
#include "FWCore/Framework/interface/LuminosityBlock.h"
#include "FWCore/Framework/interface/LuminosityBlockPrincipal.h"
#include "FWCore/Framework/interface/MessageReceiverForSource.h"
#include "FWCore/Framework/interface/ModuleChanger.h"
#include "FWCore/Framework/interface/OccurrenceTraits.h"
#include "FWCore/Framework/interface/ProcessingController.h"
#include "FWCore/Framework/interface/RunPrincipal.h"
#include "FWCore/Framework/interface/Schedule.h"
#include "FWCore/Framework/interface/ScheduleInfo.h"
#include "FWCore/Framework/interface/SubProcess.h"
#include "FWCore/Framework/src/Breakpoints.h"
#include "FWCore/Framework/src/EPStates.h"
#include "FWCore/Framework/src/EventSetupsController.h"
#include "FWCore/Framework/src/InputSourceFactory.h"

#include "FWCore/MessageLogger/interface/MessageLogger.h"

#include "FWCore/ParameterSet/interface/ConfigurationDescriptions.h"
#include "FWCore/ParameterSet/interface/IllegalParameters.h"
#include "FWCore/ParameterSet/interface/ParameterSetDescriptionFillerBase.h"
#include "FWCore/ParameterSet/interface/ParameterSetDescriptionFillerPluginFactory.h"
#include "FWCore/ParameterSet/interface/ProcessDesc.h"
#include "FWCore/ParameterSet/interface/Registry.h"
#include "FWCore/PythonParameterSet/interface/PythonProcessDesc.h"

#include "FWCore/ServiceRegistry/interface/ServiceRegistry.h"
#include "FWCore/ServiceRegistry/interface/Service.h"

#include "FWCore/Utilities/interface/DebugMacros.h"
#include "FWCore/Utilities/interface/EDMException.h"
#include "FWCore/Utilities/interface/Exception.h"
#include "FWCore/Utilities/interface/ConvertException.h"
#include "FWCore/Utilities/interface/RandomNumberGenerator.h"
#include "FWCore/Utilities/interface/UnixSignalHandlers.h"
#include "FWCore/Utilities/interface/ExceptionCollector.h"

#include "MessageForSource.h"
#include "MessageForParent.h"

#include "boost/bind.hpp"
#include "boost/thread/xtime.hpp"

#include <exception>
#include <iomanip>
#include <iostream>
#include <utility>
#include <sstream>

#include <sys/ipc.h>
#include <sys/msg.h>

//Used for forking
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/socket.h>
#include <sys/select.h>
#include <sys/fcntl.h>
#include <unistd.h>

//Used for CPU affinity
#ifndef __APPLE__
#include <sched.h>
#endif

//Needed for introspection
#include "Cintex/Cintex.h"

namespace edm {

  namespace event_processor {

    class StateSentry {
    public:
      StateSentry(EventProcessor* ep) : ep_(ep), success_(false) { }
      ~StateSentry() {if(!success_) ep_->changeState(mException);}
      void succeeded() {success_ = true;}

    private:
      EventProcessor* ep_;
      bool success_;
    };
  }

  using namespace event_processor;

  namespace {
    template <typename T>
    class ScheduleSignalSentry {
    public:
      ScheduleSignalSentry(ActivityRegistry* a, typename T::MyPrincipal* principal, EventSetup const* es) :
           a_(a), principal_(principal), es_(es), allowThrow_(false) {
        if (a_) T::preScheduleSignal(a_, principal_);
      }
      ~ScheduleSignalSentry() noexcept(false) {
        try {
          if (a_) if (principal_) T::postScheduleSignal(a_, principal_, es_);
        } catch(...) {
          if( allowThrow_) {throw;}
        }
      }

      void allowThrow() {
        allowThrow_ = true;
      }
    private:
      // We own none of these resources.
      ActivityRegistry* a_;
      typename T::MyPrincipal* principal_;
      EventSetup const* es_;
      bool allowThrow_ =true;
    };
  }

  namespace {

    // the next two tables must be kept in sync with the state and
    // message enums from the header

    char const* stateNames[] = {
      "Init",
      "JobReady",
      "RunGiven",
      "Running",
      "Stopping",
      "ShuttingDown",
      "Done",
      "JobEnded",
      "Error",
      "ErrorEnded",
      "End",
      "Invalid"
    };

    char const* msgNames[] = {
      "SetRun",
      "Skip",
      "RunAsync",
      "Run(ID)",
      "Run(count)",
      "BeginJob",
      "StopAsync",
      "ShutdownAsync",
      "EndJob",
      "CountComplete",
      "InputExhausted",
      "StopSignal",
      "ShutdownSignal",
      "Finished",
      "Any",
      "dtor",
      "Exception",
      "Rewind"
    };
  }
    // IMPORTANT NOTE:
    // the mAny messages are special, they must appear last in the
    // table if multiple entries for a CurrentState are present.
    // the changeState function does not use the mAny yet!!!

    struct TransEntry {
      State current;
      Msg   message;
      State final;
    };

    // we should use this information to initialize a two dimensional
    // table of t[CurrentState][Message] = FinalState

    /*
      the way this is current written, the async run can thread function
      can return in the "JobReady" state - but not yet cleaned up.  The
      problem is that only when stop/shutdown async is called is the
      thread cleaned up. But the stop/shudown async functions attempt
      first to change the state using messages that are not valid in
      "JobReady" state.

      I think most of the problems can be solved by using two states
      for "running": RunningS and RunningA (sync/async). The problems
      seems to be the all the transitions out of running for both
      modes of operation.  The other solution might be to only go to
      "Stopping" from Running, and use the return code from "run_p" to
      set the final state.  If this is used, then in the sync mode the
      "Stopping" state would be momentary.

     */

    TransEntry table[] = {
    // CurrentState   Message         FinalState
    // -----------------------------------------
      { sInit,          mException,      sError },
      { sInit,          mBeginJob,       sJobReady },
      { sJobReady,      mException,      sError },
      { sJobReady,      mSetRun,         sRunGiven },
      { sJobReady,      mInputRewind,    sRunning },
      { sJobReady,      mSkip,           sRunning },
      { sJobReady,      mRunID,          sRunning },
      { sJobReady,      mRunCount,       sRunning },
      { sJobReady,      mEndJob,         sJobEnded },
      { sJobReady,      mBeginJob,       sJobReady },
      { sJobReady,      mDtor,           sEnd },    // should this be allowed?

      { sJobReady,      mStopAsync,      sJobReady },
      { sJobReady,      mCountComplete,  sJobReady },
      { sJobReady,      mFinished,       sJobReady },

      { sRunGiven,      mException,      sError },
      { sRunGiven,      mRunAsync,       sRunning },
      { sRunGiven,      mBeginJob,       sRunGiven },
      { sRunGiven,      mShutdownAsync,  sShuttingDown },
      { sRunGiven,      mStopAsync,      sStopping },
      { sRunning,       mException,      sError },
      { sRunning,       mStopAsync,      sStopping },
      { sRunning,       mShutdownAsync,  sShuttingDown },
      { sRunning,       mShutdownSignal, sShuttingDown },
      { sRunning,       mCountComplete,  sStopping }, // sJobReady
      { sRunning,       mInputExhausted, sStopping }, // sJobReady

      { sStopping,      mInputRewind,    sRunning }, // The looper needs this
      { sStopping,      mException,      sError },
      { sStopping,      mFinished,       sJobReady },
      { sStopping,      mCountComplete,  sJobReady },
      { sStopping,      mShutdownSignal, sShuttingDown },
      { sStopping,      mStopAsync,      sStopping },     // stay
      { sStopping,      mInputExhausted, sStopping },     // stay
      //{ sStopping,      mAny,            sJobReady },     // <- ??????
      { sShuttingDown,  mException,      sError },
      { sShuttingDown,  mShutdownSignal, sShuttingDown },
      { sShuttingDown,  mCountComplete,  sDone }, // needed?
      { sShuttingDown,  mInputExhausted, sDone }, // needed?
      { sShuttingDown,  mFinished,       sDone },
      //{ sShuttingDown,  mShutdownAsync,  sShuttingDown }, // only one at
      //{ sShuttingDown,  mStopAsync,      sShuttingDown }, // a time
      //{ sShuttingDown,  mAny,            sDone },         // <- ??????
      { sDone,          mEndJob,         sJobEnded },
      { sDone,          mException,      sError },
      { sJobEnded,      mDtor,           sEnd },
      { sJobEnded,      mException,      sError },
      { sError,         mEndJob,         sError },   // funny one here
      { sError,         mDtor,           sError },   // funny one here
      { sInit,          mDtor,           sEnd },     // for StorM dummy EP
      { sStopping,      mShutdownAsync,  sShuttingDown }, // For FUEP tests
      { sInvalid,       mAny,            sInvalid }
    };


    // Note: many of the messages generate the mBeginJob message first
    //  mRunID, mRunCount, mSetRun

  // ---------------------------------------------------------------
  std::unique_ptr<InputSource>
  makeInput(ParameterSet& params,
            CommonParams const& common,
            ProductRegistry& preg,
            boost::shared_ptr<BranchIDListHelper> branchIDListHelper,
            boost::shared_ptr<ActivityRegistry> areg,
            boost::shared_ptr<ProcessConfiguration const> processConfiguration) {
    ParameterSet* main_input = params.getPSetForUpdate("@main_input");
    if(main_input == 0) {
      throw Exception(errors::Configuration)
        << "There must be exactly one source in the configuration.\n"
        << "It is missing (or there are sufficient syntax errors such that it is not recognized as the source)\n";
    }

    std::string modtype(main_input->getParameter<std::string>("@module_type"));

    std::auto_ptr<ParameterSetDescriptionFillerBase> filler(
      ParameterSetDescriptionFillerPluginFactory::get()->create(modtype));
    ConfigurationDescriptions descriptions(filler->baseType());
    filler->fill(descriptions);

    try {
      try {
        descriptions.validate(*main_input, std::string("source"));
      }
      catch (cms::Exception& e) { throw; }
      catch (std::bad_alloc& bda) { convertException::badAllocToEDM(); }
      catch (std::exception& e) { convertException::stdToEDM(e); }
      catch (std::string& s) { convertException::stringToEDM(s); }
      catch (char const* c) { convertException::charPtrToEDM(c); }
      catch (...) { convertException::unknownToEDM(); }
    }
    catch (cms::Exception & iException) {
      std::ostringstream ost;
      ost << "Validating configuration of input source of type " << modtype;
      iException.addContext(ost.str());
      throw;
    }

    main_input->registerIt();

    // Fill in "ModuleDescription", in case the input source produces
    // any EDproducts, which would be registered in the ProductRegistry.
    // Also fill in the process history item for this process.
    // There is no module label for the unnamed input source, so
    // just use "source".
    // Only the tracked parameters belong in the process configuration.
    ModuleDescription md(main_input->id(),
                         main_input->getParameter<std::string>("@module_type"),
                         "source",
                         processConfiguration.get());

    InputSourceDescription isdesc(md, preg, branchIDListHelper, areg, common.maxEventsInput_, common.maxLumisInput_);
    areg->preSourceConstructionSignal_(md);
    std::unique_ptr<InputSource> input;
    try {
      try {
        input = std::unique_ptr<InputSource>(InputSourceFactory::get()->makeInputSource(*main_input, isdesc).release());
      }
      catch (cms::Exception& e) { throw; }
      catch (std::bad_alloc& bda) { convertException::badAllocToEDM(); }
      catch (std::exception& e) { convertException::stdToEDM(e); }
      catch (std::string& s) { convertException::stringToEDM(s); }
      catch (char const* c) { convertException::charPtrToEDM(c); }
      catch (...) { convertException::unknownToEDM(); }
    }
    catch (cms::Exception& iException) {
      areg->postSourceConstructionSignal_(md);
      std::ostringstream ost;
      ost << "Constructing input source of type " << modtype;
      iException.addContext(ost.str());
      throw;
    }
    areg->postSourceConstructionSignal_(md);
    return input;
  }

  // ---------------------------------------------------------------
  boost::shared_ptr<EDLooperBase>
  fillLooper(eventsetup::EventSetupsController& esController,
             eventsetup::EventSetupProvider& cp,
                         ParameterSet& params) {
    boost::shared_ptr<EDLooperBase> vLooper;

    std::vector<std::string> loopers = params.getParameter<std::vector<std::string> >("@all_loopers");

    if(loopers.size() == 0) {
       return vLooper;
    }

    assert(1 == loopers.size());

    for(std::vector<std::string>::iterator itName = loopers.begin(), itNameEnd = loopers.end();
        itName != itNameEnd;
        ++itName) {

      ParameterSet* providerPSet = params.getPSetForUpdate(*itName);
      providerPSet->registerIt();
      vLooper = eventsetup::LooperFactory::get()->addTo(esController,
                                                        cp,
                                                        *providerPSet);
      }
      return vLooper;

  }

  // ---------------------------------------------------------------
  EventProcessor::EventProcessor(std::string const& config,
                                ServiceToken const& iToken,
                                serviceregistry::ServiceLegacy iLegacy,
                                std::vector<std::string> const& defaultServices,
                                std::vector<std::string> const& forcedServices) :
    preProcessEventSignal_(),
    postProcessEventSignal_(),
    actReg_(),
    preg_(),
    branchIDListHelper_(),
    serviceToken_(),
    input_(),
    espController_(new eventsetup::EventSetupsController),
    esp_(),
    act_table_(),
    processConfiguration_(),
    schedule_(),
    subProcess_(),
    historyAppender_(new HistoryAppender),
    state_(sInit),
    event_loop_(),
    state_lock_(),
    stop_lock_(),
    stopper_(),
    starter_(),
    stop_count_(-1),
    last_rc_(epSuccess),
    last_error_text_(),
    id_set_(false),
    event_loop_id_(),
    my_sig_num_(getSigNum()),
    fb_(),
    looper_(),
    principalCache_(),
    shouldWeStop_(false),
    stateMachineWasInErrorState_(false),
    fileMode_(),
    emptyRunLumiMode_(),
    exceptionMessageFiles_(),
    exceptionMessageRuns_(),
    exceptionMessageLumis_(),
    alreadyHandlingException_(false),
    forceLooperToEnd_(false),
    looperBeginJobRun_(false),
    forceESCacheClearOnNewRun_(false),
    numberOfForkedChildren_(0),
    numberOfSequentialEventsPerChild_(1),
    setCpuAffinity_(false),
    eventSetupDataToExcludeFromPrefetching_() {
    boost::shared_ptr<ParameterSet> parameterSet = PythonProcessDesc(config).parameterSet();
    boost::shared_ptr<ProcessDesc> processDesc(new ProcessDesc(parameterSet));
    processDesc->addServices(defaultServices, forcedServices);
    init(processDesc, iToken, iLegacy);
  }

  EventProcessor::EventProcessor(std::string const& config,
                                std::vector<std::string> const& defaultServices,
                                std::vector<std::string> const& forcedServices) :
    preProcessEventSignal_(),
    postProcessEventSignal_(),
    actReg_(),
    preg_(),
    branchIDListHelper_(),
    serviceToken_(),
    input_(),
    espController_(new eventsetup::EventSetupsController),
    esp_(),
    act_table_(),
    processConfiguration_(),
    schedule_(),
    subProcess_(),
    historyAppender_(new HistoryAppender),
    state_(sInit),
    event_loop_(),
    state_lock_(),
    stop_lock_(),
    stopper_(),
    starter_(),
    stop_count_(-1),
    last_rc_(epSuccess),
    last_error_text_(),
    id_set_(false),
    event_loop_id_(),
    my_sig_num_(getSigNum()),
    fb_(),
    looper_(),
    principalCache_(),
    shouldWeStop_(false),
    stateMachineWasInErrorState_(false),
    fileMode_(),
    emptyRunLumiMode_(),
    exceptionMessageFiles_(),
    exceptionMessageRuns_(),
    exceptionMessageLumis_(),
    alreadyHandlingException_(false),
    forceLooperToEnd_(false),
    looperBeginJobRun_(false),
    forceESCacheClearOnNewRun_(false),
    numberOfForkedChildren_(0),
    numberOfSequentialEventsPerChild_(1),
    setCpuAffinity_(false),
    eventSetupDataToExcludeFromPrefetching_() {
    boost::shared_ptr<ParameterSet> parameterSet = PythonProcessDesc(config).parameterSet();
    boost::shared_ptr<ProcessDesc> processDesc(new ProcessDesc(parameterSet));
    processDesc->addServices(defaultServices, forcedServices);
    init(processDesc, ServiceToken(), serviceregistry::kOverlapIsError);
  }

  EventProcessor::EventProcessor(boost::shared_ptr<ProcessDesc>& processDesc,
                 ServiceToken const& token,
                 serviceregistry::ServiceLegacy legacy) :
    preProcessEventSignal_(),
    postProcessEventSignal_(),
    actReg_(),
    preg_(),
    branchIDListHelper_(),
    serviceToken_(),
    input_(),
    espController_(new eventsetup::EventSetupsController),
    esp_(),
    act_table_(),
    processConfiguration_(),
    schedule_(),
    subProcess_(),
    historyAppender_(new HistoryAppender),
    state_(sInit),
    event_loop_(),
    state_lock_(),
    stop_lock_(),
    stopper_(),
    starter_(),
    stop_count_(-1),
    last_rc_(epSuccess),
    last_error_text_(),
    id_set_(false),
    event_loop_id_(),
    my_sig_num_(getSigNum()),
    fb_(),
    looper_(),
    principalCache_(),
    shouldWeStop_(false),
    stateMachineWasInErrorState_(false),
    fileMode_(),
    emptyRunLumiMode_(),
    exceptionMessageFiles_(),
    exceptionMessageRuns_(),
    exceptionMessageLumis_(),
    alreadyHandlingException_(false),
    forceLooperToEnd_(false),
    looperBeginJobRun_(false),
    forceESCacheClearOnNewRun_(false),
    numberOfForkedChildren_(0),
    numberOfSequentialEventsPerChild_(1),
    setCpuAffinity_(false),
    eventSetupDataToExcludeFromPrefetching_() {
    init(processDesc, token, legacy);
  }


  EventProcessor::EventProcessor(std::string const& config, bool isPython):
    preProcessEventSignal_(),
    postProcessEventSignal_(),
    actReg_(),
    preg_(),
    branchIDListHelper_(),
    serviceToken_(),
    input_(),
    espController_(new eventsetup::EventSetupsController),
    esp_(),
    act_table_(),
    processConfiguration_(),
    schedule_(),
    subProcess_(),
    historyAppender_(new HistoryAppender),
    state_(sInit),
    event_loop_(),
    state_lock_(),
    stop_lock_(),
    stopper_(),
    starter_(),
    stop_count_(-1),
    last_rc_(epSuccess),
    last_error_text_(),
    id_set_(false),
    event_loop_id_(),
    my_sig_num_(getSigNum()),
    fb_(),
    looper_(),
    principalCache_(),
    shouldWeStop_(false),
    stateMachineWasInErrorState_(false),
    fileMode_(),
    emptyRunLumiMode_(),
    exceptionMessageFiles_(),
    exceptionMessageRuns_(),
    exceptionMessageLumis_(),
    alreadyHandlingException_(false),
    forceLooperToEnd_(false),
    looperBeginJobRun_(false),
    forceESCacheClearOnNewRun_(false),
    numberOfForkedChildren_(0),
    numberOfSequentialEventsPerChild_(1),
    setCpuAffinity_(false),
    eventSetupDataToExcludeFromPrefetching_() {
    if(isPython) {
      boost::shared_ptr<ParameterSet> parameterSet = PythonProcessDesc(config).parameterSet();
      boost::shared_ptr<ProcessDesc> processDesc(new ProcessDesc(parameterSet));
      init(processDesc, ServiceToken(), serviceregistry::kOverlapIsError);
    }
    else {
      boost::shared_ptr<ProcessDesc> processDesc(new ProcessDesc(config));
      init(processDesc, ServiceToken(), serviceregistry::kOverlapIsError);
    }
  }

  void
  EventProcessor::init(boost::shared_ptr<ProcessDesc>& processDesc,
                        ServiceToken const& iToken,
                        serviceregistry::ServiceLegacy iLegacy) {

    //std::cerr << processDesc->dump() << std::endl;
   
    ROOT::Cintex::Cintex::Enable();

    boost::shared_ptr<ParameterSet> parameterSet = processDesc->getProcessPSet();
    //std::cerr << parameterSet->dump() << std::endl;

    // If there is a subprocess, pop the subprocess parameter set out of the process parameter set
    boost::shared_ptr<ParameterSet> subProcessParameterSet(popSubProcessParameterSet(*parameterSet).release());

    // Now set some parameters specific to the main process.
    ParameterSet const& optionsPset(parameterSet->getUntrackedParameterSet("options", ParameterSet()));
    fileMode_ = optionsPset.getUntrackedParameter<std::string>("fileMode", "");
    emptyRunLumiMode_ = optionsPset.getUntrackedParameter<std::string>("emptyRunLumiMode", "");
    forceESCacheClearOnNewRun_ = optionsPset.getUntrackedParameter<bool>("forceEventSetupCacheClearOnNewRun", false);
    ParameterSet const& forking = optionsPset.getUntrackedParameterSet("multiProcesses", ParameterSet());
    numberOfForkedChildren_ = forking.getUntrackedParameter<int>("maxChildProcesses", 0);
    numberOfSequentialEventsPerChild_ = forking.getUntrackedParameter<unsigned int>("maxSequentialEventsPerChild", 1);
    setCpuAffinity_ = forking.getUntrackedParameter<bool>("setCpuAffinity", false);
    continueAfterChildFailure_ = forking.getUntrackedParameter<bool>("continueAfterChildFailure",false);
    std::vector<ParameterSet> const& excluded = forking.getUntrackedParameterSetVector("eventSetupDataToExcludeFromPrefetching", std::vector<ParameterSet>());
    for(std::vector<ParameterSet>::const_iterator itPS = excluded.begin(), itPSEnd = excluded.end();
        itPS != itPSEnd;
        ++itPS) {
      eventSetupDataToExcludeFromPrefetching_[itPS->getUntrackedParameter<std::string>("record")].insert(
                                                std::make_pair(itPS->getUntrackedParameter<std::string>("type", "*"),
                                                               itPS->getUntrackedParameter<std::string>("label", "")));
    }
    IllegalParameters::setThrowAnException(optionsPset.getUntrackedParameter<bool>("throwIfIllegalParameter", true));

    // Now do general initialization
    ScheduleItems items;

    //initialize the services
    boost::shared_ptr<std::vector<ParameterSet> > pServiceSets = processDesc->getServicesPSets();
    ServiceToken token = items.initServices(*pServiceSets, *parameterSet, iToken, iLegacy, true);
    serviceToken_ = items.addCPRandTNS(*parameterSet, token);

    //make the services available
    ServiceRegistry::Operate operate(serviceToken_);

    // intialize miscellaneous items
    boost::shared_ptr<CommonParams> common(items.initMisc(*parameterSet));

    // intialize the event setup provider
    esp_ = espController_->makeProvider(*parameterSet);

    // initialize the looper, if any
    looper_ = fillLooper(*espController_, *esp_, *parameterSet);
    if(looper_) {
      looper_->setActionTable(items.act_table_.get());
      looper_->attachTo(*items.actReg_);
    }

    // initialize the input source
    input_ = makeInput(*parameterSet, *common, *items.preg_, items.branchIDListHelper_, items.actReg_, items.processConfiguration_);

    // intialize the Schedule
    schedule_ = items.initSchedule(*parameterSet,subProcessParameterSet.get());

    // set the data members
    act_table_ = std::move(items.act_table_);
    actReg_ = items.actReg_;
    preg_.reset(items.preg_.release());
    branchIDListHelper_ = items.branchIDListHelper_;
    processConfiguration_ = items.processConfiguration_;

    FDEBUG(2) << parameterSet << std::endl;
    connectSigs(this);

    // Reusable event principal
    boost::shared_ptr<EventPrincipal> ep(new EventPrincipal(preg_, branchIDListHelper_, *processConfiguration_, historyAppender_.get()));
    principalCache_.insert(ep);
      
    // initialize the subprocess, if there is one
    if(subProcessParameterSet) {
      subProcess_.reset(new SubProcess(*subProcessParameterSet, *parameterSet, preg_, branchIDListHelper_, *espController_, *actReg_, token, serviceregistry::kConfigurationOverrides));
    }
  }

  EventProcessor::~EventProcessor() {
    // Make the services available while everything is being deleted.
    ServiceToken token = getToken();
    ServiceRegistry::Operate op(token);
    try {
      changeState(mDtor);
    }
    catch(cms::Exception& e) {
      LogError("System")
        << e.explainSelf() << "\n";
    }

    // manually destroy all these thing that may need the services around
    espController_.reset();
    subProcess_.reset();
    esp_.reset();
    schedule_.reset();
    input_.reset();
    looper_.reset();
    actReg_.reset();

    pset::Registry* psetRegistry = pset::Registry::instance();
    psetRegistry->data().clear();
    psetRegistry->extra().setID(ParameterSetID());

    EntryDescriptionRegistry::instance()->data().clear();
    ParentageRegistry::instance()->data().clear();
    ProcessConfigurationRegistry::instance()->data().clear();
    ProcessHistoryRegistry::instance()->data().clear();
  }

  void
  EventProcessor::beginJob() {
    if(state_ != sInit) return;
    bk::beginJob();
    // can only be run if in the initial state
    changeState(mBeginJob);

    // StateSentry toerror(this); // should we add this ?
    //make the services available
    ServiceRegistry::Operate operate(serviceToken_);

    //NOTE:  This implementation assumes 'Job' means one call
    // the EventProcessor::run
    // If it really means once per 'application' then this code will
    // have to be changed.
    // Also have to deal with case where have 'run' then new Module
    // added and do 'run'
    // again.  In that case the newly added Module needs its 'beginJob'
    // to be called.

    //NOTE: in future we should have a beginOfJob for looper that takes no arguments
    //  For now we delay calling beginOfJob until first beginOfRun
    //if(looper_) {
    //   looper_->beginOfJob(es);
    //}
    try {
      try {
        input_->doBeginJob();
      }
      catch (cms::Exception& e) { throw; }
      catch(std::bad_alloc& bda) { convertException::badAllocToEDM(); }
      catch (std::exception& e) { convertException::stdToEDM(e); }
      catch(std::string& s) { convertException::stringToEDM(s); }
      catch(char const* c) { convertException::charPtrToEDM(c); }
      catch (...) { convertException::unknownToEDM(); }
    }
    catch(cms::Exception& ex) {
      ex.addContext("Calling beginJob for the source");
      throw;
    }
    schedule_->beginJob(*preg_);
    // toerror.succeeded(); // should we add this?
    if(hasSubProcess()) subProcess_->doBeginJob();
    actReg_->postBeginJobSignal_();
  }

  void
  EventProcessor::endJob() {
    // Collects exceptions, so we don't throw before all operations are performed.
    ExceptionCollector c("Multiple exceptions were thrown while executing endJob. An exception message follows for each.\n");

    // only allowed to run if state is sIdle, sJobReady, sRunGiven
    c.call(boost::bind(&EventProcessor::changeState, this, mEndJob));

    //make the services available
    ServiceRegistry::Operate operate(serviceToken_);

    schedule_->endJob(c);
    if(hasSubProcess()) {
      c.call(boost::bind(&SubProcess::doEndJob, subProcess_.get()));
    }
    c.call(boost::bind(&InputSource::doEndJob, input_.get()));
    if(looper_) {
      c.call(boost::bind(&EDLooperBase::endOfJob, looper_));
    }
    auto actReg = actReg_.get();
    c.call([actReg](){actReg->postEndJobSignal_();});
    if(c.hasThrown()) {
      c.rethrow();
    }
  }

  ServiceToken
  EventProcessor::getToken() {
    return serviceToken_;
  }

  //Setup signal handler to listen for when forked children stop
  namespace {
    //These are volatile since the compiler can not be allowed to optimize them
    // since they can be modified in the signaller handler
    volatile bool child_failed = false;
    volatile unsigned int num_children_done = 0;
    volatile int child_fail_exit_status = 0;
    volatile int child_fail_signal = 0;
    
    //NOTE: We setup the signal handler to run in the main thread which
    // is also the same thread that then reads the above values

    extern "C" {
      void ep_sigchld(int, siginfo_t*, void*) {
        //printf("in sigchld\n");
        //FDEBUG(1) << "in sigchld handler\n";
        int stat_loc;
        pid_t p = waitpid(-1, &stat_loc, WNOHANG);
        while(0<p) {
          //printf("  looping\n");
          if(WIFEXITED(stat_loc)) {
            ++num_children_done;
            if(0 != WEXITSTATUS(stat_loc)) {
              child_fail_exit_status = WEXITSTATUS(stat_loc);
              child_failed = true;
            }
          }
          if(WIFSIGNALED(stat_loc)) {
            ++num_children_done;
            child_fail_signal = WTERMSIG(stat_loc);
            child_failed = true;
          }
          p = waitpid(-1, &stat_loc, WNOHANG);
        }
      }
    }

  }

  enum {
    kChildSucceed,
    kChildExitBadly,
    kChildSegv,
    kMaxChildAction
  };

  namespace {
    unsigned int numberOfDigitsInChildIndex(unsigned int numberOfChildren) {
      unsigned int n = 0;
      while(numberOfChildren != 0) {
        ++n;
        numberOfChildren /= 10;
      }
      if(n == 0) {
        n = 3; // Protect against zero numberOfChildren
      }
      return n;
    }
    
    /*This class embodied the thread which is used to listen to the forked children and
     then tell them which events they should process */
    class MessageSenderToSource {
    public:
      MessageSenderToSource(std::vector<int> const& childrenSockets, std::vector<int> const& childrenPipes, long iNEventsToProcess);
      void operator()();

    private:
      const std::vector<int>& m_childrenPipes;
      long const m_nEventsToProcess;
      fd_set m_socketSet;
      unsigned int m_aliveChildren;
      int m_maxFd;
    };
    
    MessageSenderToSource::MessageSenderToSource(std::vector<int> const& childrenSockets,
                                                 std::vector<int> const& childrenPipes,
                                                 long iNEventsToProcess):
    m_childrenPipes(childrenPipes),
    m_nEventsToProcess(iNEventsToProcess),
    m_aliveChildren(childrenSockets.size()),
    m_maxFd(0)
    {
      FD_ZERO(&m_socketSet);
      for (std::vector<int>::const_iterator it = childrenSockets.begin(), itEnd = childrenSockets.end();
           it != itEnd; it++) {
        FD_SET(*it, &m_socketSet);
        if (*it > m_maxFd) {
          m_maxFd = *it;
        }
      }
      for (std::vector<int>::const_iterator it = childrenPipes.begin(), itEnd = childrenPipes.end();
           it != itEnd; ++it) {
        FD_SET(*it, &m_socketSet);
        if (*it > m_maxFd) {
          m_maxFd = *it;
        }
      }
      m_maxFd++; // select reads [0,m_maxFd).
    }
   
    /* This function is the heart of the communication between parent and child.
     * When ready for more data, the child (see MessageReceiverForSource) requests
     * data through a AF_UNIX socket message.  The parent will then assign the next
     * chunk of data by sending a message back.
     *
     * Additionally, this function also monitors the read-side of the pipe fd from the child.
     * If the child dies unexpectedly, the pipe will be selected as ready for read and
     * will return EPIPE when read from.  Further, if the child thinks the parent has died
     * (defined as waiting more than 1s for a response), it will write a single byte to
     * the pipe.  If the parent has died, the child will get a EPIPE and throw an exception.
     * If still alive, the parent will read the byte and ignore it.
     *
     * Note this function is complemented by the SIGCHLD handler above as currently only the SIGCHLD
     * handler can distinguish between success and failure cases.
     */
 
    void
    MessageSenderToSource::operator()() {
      multicore::MessageForParent childMsg;
      LogInfo("ForkingController") << "I am controller";
      //this is the master and therefore the controller
      
      multicore::MessageForSource sndmsg;
      sndmsg.startIndex = 0;
      sndmsg.nIndices = m_nEventsToProcess;
      do {
        
        fd_set readSockets, errorSockets;
        // Wait for a request from a child for events.
        memcpy(&readSockets, &m_socketSet, sizeof(m_socketSet));
        memcpy(&errorSockets, &m_socketSet, sizeof(m_socketSet));
        // Note that we don't timeout; may be reconsidered in the future.
        ssize_t rc;
        while (((rc = select(m_maxFd, &readSockets, NULL, &errorSockets, NULL)) < 0) && (errno == EINTR)) {}
        if (rc < 0) {
          std::cerr << "select failed; should be impossible due to preconditions.\n";
          abort();
          break;
        }

        // Read the message from the child.
        for (int idx=0; idx<m_maxFd; idx++) {

          // Handle errors
          if (FD_ISSET(idx, &errorSockets)) {
            LogInfo("ForkingController") << "Error on socket " << idx;
            FD_CLR(idx, &m_socketSet);
            close(idx);
            // See if it was the watchdog pipe that died.
            for (std::vector<int>::const_iterator it = m_childrenPipes.begin(); it != m_childrenPipes.end(); it++) {
              if (*it == idx) {
                m_aliveChildren--;
              }
            }
            continue;
          }
          
          if (!FD_ISSET(idx, &readSockets)) {
            continue;
          }

          // See if this FD is a child watchdog pipe.  If so, read from it to prevent
          // writes from blocking.
          bool is_pipe = false;
          for (std::vector<int>::const_iterator it = m_childrenPipes.begin(), itEnd = m_childrenPipes.end(); it != itEnd; it++) {
              if (*it == idx) {
                is_pipe = true;
                char buf;
                while (((rc = read(idx, &buf, 1)) < 0) && (errno == EINTR)) {}
                if (rc <= 0) {
                  m_aliveChildren--;
                  FD_CLR(idx, &m_socketSet);
                  close(idx);
                }
              }
          }

          // Only execute this block if the FD is a socket for sending the child work.
          if (!is_pipe) {
            while (((rc = recv(idx, reinterpret_cast<char*>(&childMsg),childMsg.sizeForBuffer() , 0)) < 0) && (errno == EINTR)) {}
            if (rc < 0) {
              FD_CLR(idx, &m_socketSet);
              close(idx);
              continue;
            }
          
            // Tell the child what events to process.
            // If 'send' fails, then the child process has failed (any other possibilities are
            // eliminated because we are using fixed-size messages with Unix datagram sockets).
            // Thus, the SIGCHLD handler will fire and set child_fail = true.
            while (((rc = send(idx, (char *)(&sndmsg), multicore::MessageForSource::sizeForBuffer(), 0)) < 0) && (errno == EINTR)) {}
            if (rc < 0) {
              FD_CLR(idx, &m_socketSet);
              close(idx);
              continue;
            }
            //std::cout << "Sent chunk starting at " << sndmsg.startIndex << " to child, length " << sndmsg.nIndices << std::endl;
            sndmsg.startIndex += sndmsg.nIndices;
          }
        }
      
      } while (m_aliveChildren > 0);
      
      return;
    }

  }

  
  void EventProcessor::possiblyContinueAfterForkChildFailure() {
    if(child_failed && continueAfterChildFailure_) {
      if (child_fail_signal) {
        LogSystem("ForkedChildFailed") << "child process ended abnormally with signal " << child_fail_signal;
        child_fail_signal=0;
      } else if (child_fail_exit_status) {
        LogSystem("ForkedChildFailed") << "child process ended abnormally with exit code " << child_fail_exit_status;
        child_fail_exit_status=0;
      } else {
        LogSystem("ForkedChildFailed") << "child process ended abnormally for unknown reason";
      }
      child_failed =false;
    }
  }

  bool
  EventProcessor::forkProcess(std::string const& jobReportFile) {

    if(0 == numberOfForkedChildren_) {return true;}
    assert(0<numberOfForkedChildren_);
    //do what we want done in common
    {
      beginJob(); //make sure this was run
      // make the services available
      ServiceRegistry::Operate operate(serviceToken_);

      InputSource::ItemType itemType;
      itemType = input_->nextItemType();

      assert(itemType == InputSource::IsFile);
      {
        readFile();
      }
      itemType = input_->nextItemType();
      assert(itemType == InputSource::IsRun);

      LogSystem("ForkingEventSetupPreFetching") << " prefetching for run " << input_->runAuxiliary()->run();
      IOVSyncValue ts(EventID(input_->runAuxiliary()->run(), 0, 0),
                      input_->runAuxiliary()->beginTime());
      espController_->eventSetupForInstance(ts);
      EventSetup const& es = esp_->eventSetup();

      //now get all the data available in the EventSetup
      std::vector<eventsetup::EventSetupRecordKey> recordKeys;
      es.fillAvailableRecordKeys(recordKeys);
      std::vector<eventsetup::DataKey> dataKeys;
      for(std::vector<eventsetup::EventSetupRecordKey>::const_iterator itKey = recordKeys.begin(), itEnd = recordKeys.end();
          itKey != itEnd;
          ++itKey) {
        eventsetup::EventSetupRecord const* recordPtr = es.find(*itKey);
        //see if this is on our exclusion list
        ExcludedDataMap::const_iterator itExcludeRec = eventSetupDataToExcludeFromPrefetching_.find(itKey->type().name());
        ExcludedData const* excludedData(nullptr);
        if(itExcludeRec != eventSetupDataToExcludeFromPrefetching_.end()) {
          excludedData = &(itExcludeRec->second);
          if(excludedData->size() == 0 || excludedData->begin()->first == "*") {
            //skip all items in this record
            continue;
          }
        }
        if(0 != recordPtr) {
          dataKeys.clear();
          recordPtr->fillRegisteredDataKeys(dataKeys);
          for(std::vector<eventsetup::DataKey>::const_iterator itDataKey = dataKeys.begin(), itDataKeyEnd = dataKeys.end();
              itDataKey != itDataKeyEnd;
              ++itDataKey) {
            //std::cout << "  " << itDataKey->type().name() << " " << itDataKey->name().value() << std::endl;
            if(0 != excludedData && excludedData->find(std::make_pair(itDataKey->type().name(), itDataKey->name().value())) != excludedData->end()) {
              LogInfo("ForkingEventSetupPreFetching") << "   excluding:" << itDataKey->type().name() << " " << itDataKey->name().value() << std::endl;
              continue;
            }
            try {
              recordPtr->doGet(*itDataKey);
            } catch(cms::Exception& e) {
             LogWarning("ForkingEventSetupPreFetching") << e.what();
            }
          }
        }
      }
    }
    LogSystem("ForkingEventSetupPreFetching") <<"  done prefetching";
    {
      // make the services available
      ServiceRegistry::Operate operate(serviceToken_);
      Service<JobReport> jobReport;
      jobReport->parentBeforeFork(jobReportFile, numberOfForkedChildren_);

      //Now actually do the forking
      actReg_->preForkReleaseResourcesSignal_();
      input_->doPreForkReleaseResources();
      schedule_->preForkReleaseResources();
    }
    installCustomHandler(SIGCHLD, ep_sigchld);


    unsigned int childIndex = 0;
    unsigned int const kMaxChildren = numberOfForkedChildren_;
    unsigned int const numberOfDigitsInIndex = numberOfDigitsInChildIndex(kMaxChildren);
    std::vector<pid_t> childrenIds;
    childrenIds.reserve(kMaxChildren);
    std::vector<int> childrenSockets;
    childrenSockets.reserve(kMaxChildren);
    std::vector<int> childrenPipes;
    childrenPipes.reserve(kMaxChildren);
    std::vector<int> childrenSocketsCopy;
    childrenSocketsCopy.reserve(kMaxChildren);
    std::vector<int> childrenPipesCopy;
    childrenPipesCopy.reserve(kMaxChildren);
    int pipes[] {0, 0};

    {
      // make the services available
      ServiceRegistry::Operate operate(serviceToken_);
      Service<JobReport> jobReport;
      int sockets[2], fd_flags;
      for(; childIndex < kMaxChildren; ++childIndex) {
        // Create a UNIX_DGRAM socket pair
        if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sockets)) {
          printf("Error creating communication socket (errno=%d, %s)\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }
        if (pipe(pipes)) {
          printf("Error creating communication pipes (errno=%d, %s)\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }
        // set CLOEXEC so the socket/pipe doesn't get leaked if the child exec's.
        if ((fd_flags = fcntl(sockets[1], F_GETFD, NULL)) == -1) {
          printf("Failed to get fd flags: %d %s\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }
        // Mark socket as non-block.  Child must be careful to do select prior
        // to reading from socket.
        if (fcntl(sockets[1], F_SETFD, fd_flags | FD_CLOEXEC | O_NONBLOCK) == -1) {
          printf("Failed to set new fd flags: %d %s\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }
        if ((fd_flags = fcntl(pipes[1], F_GETFD, NULL)) == -1) {
          printf("Failed to get fd flags: %d %s\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }
        if (fcntl(pipes[1], F_SETFD, fd_flags | FD_CLOEXEC) == -1) {
          printf("Failed to set new fd flags: %d %s\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }
        // Linux man page notes there are some edge cases where reading from a
        // fd can block, even after a select.
        if ((fd_flags = fcntl(pipes[0], F_GETFD, NULL)) == -1) {
          printf("Failed to get fd flags: %d %s\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }
        if (fcntl(pipes[0], F_SETFD, fd_flags | O_NONBLOCK) == -1) {
          printf("Failed to set new fd flags: %d %s\n", errno, strerror(errno));
          exit(EXIT_FAILURE);
        }

        childrenPipesCopy = childrenPipes;
        childrenSocketsCopy = childrenSockets;

        pid_t value = fork();
        if(value == 0) {
          // Close the parent's side of the socket and pipe which will talk to us.
          close(pipes[0]);
          close(sockets[0]);
          // Close our copies of the parent's other communication pipes.
          for(std::vector<int>::const_iterator it=childrenPipesCopy.begin(); it != childrenPipesCopy.end(); it++) {
            close(*it);
          }
          for(std::vector<int>::const_iterator it=childrenSocketsCopy.begin(); it != childrenSocketsCopy.end(); it++) {
            close(*it);
          }

          // this is the child process, redirect stdout and stderr to a log file
          fflush(stdout);
          fflush(stderr);
          std::stringstream stout;
          stout << "redirectout_" << getpgrp() << "_" << std::setw(numberOfDigitsInIndex) << std::setfill('0') << childIndex << ".log";
          if(0 == freopen(stout.str().c_str(), "w", stdout)) {
            LogError("ForkingStdOutRedirect") << "Error during freopen of child process "<< childIndex;
          }
          if(dup2(fileno(stdout), fileno(stderr)) < 0) {
            LogError("ForkingStdOutRedirect") << "Error during dup2 of child process"<< childIndex;
          }

          LogInfo("ForkingChild") << "I am child " << childIndex << " with pgid " << getpgrp();
          if(setCpuAffinity_) {
            // CPU affinity is handled differently on macosx.
            // We disable it and print a message until someone reads:
            //
            // http://developer.apple.com/mac/library/releasenotes/Performance/RN-AffinityAPI/index.html
            //
            // and implements it.
#ifdef __APPLE__
            LogInfo("ForkingChildAffinity") << "Architecture support for CPU affinity not implemented.";
#else
            LogInfo("ForkingChildAffinity") << "Setting CPU affinity, setting this child to cpu " << childIndex;
            cpu_set_t mask;
            CPU_ZERO(&mask);
            CPU_SET(childIndex, &mask);
            if(sched_setaffinity(0, sizeof(mask), &mask) != 0) {
              LogError("ForkingChildAffinity") << "Failed to set the cpu affinity, errno " << errno;
              exit(-1);
            }
#endif
          }
          break;
        } else {
          //this is the parent
          close(pipes[1]);
          close(sockets[1]);
        }
        if(value < 0) {
          LogError("ForkingChild") << "failed to create a child";
          exit(-1);
        }
        childrenIds.push_back(value);
        childrenSockets.push_back(sockets[0]);
        childrenPipes.push_back(pipes[0]);
      }

      if(childIndex < kMaxChildren) {
        jobReport->childAfterFork(jobReportFile, childIndex, kMaxChildren);
        actReg_->postForkReacquireResourcesSignal_(childIndex, kMaxChildren);

        boost::shared_ptr<multicore::MessageReceiverForSource> receiver(new multicore::MessageReceiverForSource(sockets[1], pipes[1]));
        input_->doPostForkReacquireResources(receiver);
        schedule_->postForkReacquireResources(childIndex, kMaxChildren);
        //NOTE: sources have to reset themselves by listening to the post fork message
        //rewindInput();
        return true;
      }
      jobReport->parentAfterFork(jobReportFile);
    }

    //this is the original, which is now the master for all the children

    //Need to wait for signals from the children or externally
    // To wait we must
    // 1) block the signals we want to wait on so we do not have a race condition
    // 2) check that we haven't already meet our ending criteria
    // 3) call sigsuspend, which unblocks the signals and waits until a signal is caught
    sigset_t blockingSigSet;
    sigset_t unblockingSigSet;
    sigset_t oldSigSet;
    pthread_sigmask(SIG_SETMASK, NULL, &unblockingSigSet);
    pthread_sigmask(SIG_SETMASK, NULL, &blockingSigSet);
    sigaddset(&blockingSigSet, SIGCHLD);
    sigaddset(&blockingSigSet, SIGUSR2);
    sigaddset(&blockingSigSet, SIGINT);
    sigdelset(&unblockingSigSet, SIGCHLD);
    sigdelset(&unblockingSigSet, SIGUSR2);
    sigdelset(&unblockingSigSet, SIGINT);
    pthread_sigmask(SIG_BLOCK, &blockingSigSet, &oldSigSet);

    // If there are too many fd's (unlikely, but possible) for select, denote this 
    // because the sender will fail.
    bool too_many_fds = false;
    if (pipes[1]+1 > FD_SETSIZE) {
      LogError("ForkingFileDescriptors") << "too many file descriptors for multicore job";
      too_many_fds = true;
    }

    //create a thread that sends the units of work to workers
    // we create it after all signals were blocked so that this
    // thread is never interupted by a signal
    MessageSenderToSource sender(childrenSockets, childrenPipes, numberOfSequentialEventsPerChild_);
    boost::thread senderThread(sender);

    if(not too_many_fds) {
      //NOTE: a child could have failed before we got here and even after this call
      // which is why the 'if' is conditional on continueAfterChildFailure_
      possiblyContinueAfterForkChildFailure();
      while(!shutdown_flag && (!child_failed or continueAfterChildFailure_) && (childrenIds.size() != num_children_done)) {
        sigsuspend(&unblockingSigSet);
        possiblyContinueAfterForkChildFailure();
        LogInfo("ForkingAwake") << "woke from sigwait" << std::endl;
      }
    }
    pthread_sigmask(SIG_SETMASK, &oldSigSet, NULL);

    LogInfo("ForkingStopping") << "num children who have already stopped " << num_children_done;
    if(child_failed) {
      LogError("ForkingStopping") << "child failed";
    }
    if(shutdown_flag) {
      LogSystem("ForkingStopping") << "asked to shutdown";
    }

    if(too_many_fds || shutdown_flag || (child_failed && (num_children_done != childrenIds.size()))) {
      LogInfo("ForkingStopping") << "must stop children" << std::endl;
      for(std::vector<pid_t>::iterator it = childrenIds.begin(), itEnd = childrenIds.end();
          it != itEnd; ++it) {
        /* int result = */ kill(*it, SIGUSR2);
      }
      pthread_sigmask(SIG_BLOCK, &blockingSigSet, &oldSigSet);
      while(num_children_done != kMaxChildren) {
        sigsuspend(&unblockingSigSet);
      }
      pthread_sigmask(SIG_SETMASK, &oldSigSet, NULL);
    }
    // The senderThread will notice the pipes die off, one by one.  Once all children are gone, it will exit.
    senderThread.join();
    if(child_failed && !continueAfterChildFailure_) {
      if (child_fail_signal) {
        throw cms::Exception("ForkedChildFailed") << "child process ended abnormally with signal " << child_fail_signal;
      } else if (child_fail_exit_status) {
        throw cms::Exception("ForkedChildFailed") << "child process ended abnormally with exit code " << child_fail_exit_status;
      } else {
        throw cms::Exception("ForkedChildFailed") << "child process ended abnormally for unknown reason";
      }
    }
    if(too_many_fds) {
      throw cms::Exception("ForkedParentFailed") << "hit select limit for number of fds";
    }
    return false;
  }

  void
  EventProcessor::connectSigs(EventProcessor* ep) {
    // When the FwkImpl signals are given, pass them to the
    // appropriate EventProcessor signals so that the outside world
    // can see the signal.
    actReg_->preProcessEventSignal_.connect(std::cref(ep->preProcessEventSignal_));
    actReg_->postProcessEventSignal_.connect(std::cref(ep->postProcessEventSignal_));
  }

  std::vector<ModuleDescription const*>
  EventProcessor::getAllModuleDescriptions() const {
    return schedule_->getAllModuleDescriptions();
  }

  int
  EventProcessor::totalEvents() const {
    return schedule_->totalEvents();
  }

  int
  EventProcessor::totalEventsPassed() const {
    return schedule_->totalEventsPassed();
  }

  int
  EventProcessor::totalEventsFailed() const {
    return schedule_->totalEventsFailed();
  }

  void
  EventProcessor::enableEndPaths(bool active) {
    schedule_->enableEndPaths(active);
  }

  bool
  EventProcessor::endPathsEnabled() const {
    return schedule_->endPathsEnabled();
  }

  void
  EventProcessor::getTriggerReport(TriggerReport& rep) const {
    schedule_->getTriggerReport(rep);
  }

  void
  EventProcessor::clearCounters() {
    schedule_->clearCounters();
  }

  char const* EventProcessor::currentStateName() const {
    return stateName(getState());
  }

  char const* EventProcessor::stateName(State s) const {
    return stateNames[s];
  }

  char const* EventProcessor::msgName(Msg m) const {
    return msgNames[m];
  }

  State EventProcessor::getState() const {
    return state_;
  }

  EventProcessor::StatusCode EventProcessor::statusAsync() const {
    // the thread will record exception/error status in the event processor
    // for us to look at and report here
    return last_rc_;
  }

  void
  EventProcessor::setRunNumber(RunNumber_t runNumber) {
    if(runNumber == 0) {
      runNumber = 1;
      LogWarning("Invalid Run")
        << "EventProcessor::setRunNumber was called with an invalid run number (nullptr)\n"
        << "Run number was set to 1 instead\n";
    }

    // inside of beginJob there is a check to see if it has been called before
    beginJob();
    changeState(mSetRun);

    // interface not correct yet
    input_->setRunNumber(runNumber);
  }

  void
  EventProcessor::declareRunNumber(RunNumber_t /*runNumber*/) {
    // inside of beginJob there is a check to see if it has been called before
    beginJob();
    changeState(mSetRun);

    // interface not correct yet - wait for Bill to be done with run/lumi loop stuff 21-Jun-2007
    //input_->declareRunNumber(runNumber);
  }

  EventProcessor::StatusCode
  EventProcessor::waitForAsyncCompletion(unsigned int timeout_seconds) {
    bool rc = true;
    boost::xtime timeout;

#if BOOST_VERSION >= 105000
    boost::xtime_get(&timeout, boost::TIME_UTC_);
#else
    boost::xtime_get(&timeout, boost::TIME_UTC);
#endif
    timeout.sec += timeout_seconds;

    // make sure to include a timeout here so we don't wait forever
    // I suspect there are still timing issues with thread startup
    // and the setting of the various control variables (stop_count, id_set)
    {
      boost::mutex::scoped_lock sl(stop_lock_);

      // look here - if runAsync not active, just return the last return code
      if(stop_count_ < 0) return last_rc_;

      if(timeout_seconds == 0) {
        while(stop_count_ == 0) stopper_.wait(sl);
      } else {
        while(stop_count_ == 0 && (rc = stopper_.timed_wait(sl, timeout)) == true);
      }

      if(rc == false) {
          // timeout occurred
          // if(id_set_) pthread_kill(event_loop_id_, my_sig_num_);
          // this is a temporary hack until we get the input source
          // upgraded to allow blocking input sources to be unblocked

          // the next line is dangerous and causes all sorts of trouble
          if(id_set_) pthread_cancel(event_loop_id_);

          // we will not do anything yet
          LogWarning("timeout")
            << "An asynchronous request was made to shut down "
            << "the event loop "
            << "and the event loop did not shutdown after "
            << timeout_seconds << " seconds\n";
      } else {
          event_loop_->join();
          event_loop_.reset();
          id_set_ = false;
          stop_count_ = -1;
      }
    }
    return rc == false ? epTimedOut : last_rc_;
  }

  EventProcessor::StatusCode
  EventProcessor::waitTillDoneAsync(unsigned int timeout_value_secs) {
    StatusCode rc = waitForAsyncCompletion(timeout_value_secs);
    if(rc != epTimedOut) changeState(mCountComplete);
    else errorState();
    return rc;
  }


  EventProcessor::StatusCode EventProcessor::stopAsync(unsigned int secs) {
    changeState(mStopAsync);
    StatusCode rc = waitForAsyncCompletion(secs);
    if(rc != epTimedOut) changeState(mFinished);
    else errorState();
    return rc;
  }

  EventProcessor::StatusCode EventProcessor::shutdownAsync(unsigned int secs) {
    changeState(mShutdownAsync);
    StatusCode rc = waitForAsyncCompletion(secs);
    if(rc != epTimedOut) changeState(mFinished);
    else errorState();
    return rc;
  }

  void EventProcessor::errorState() {
    state_ = sError;
  }

  // next function irrelevant now
  EventProcessor::StatusCode EventProcessor::doneAsync(Msg m) {
    // make sure to include a timeout here so we don't wait forever
    // I suspect there are still timing issues with thread startup
    // and the setting of the various control variables (stop_count, id_set)
    changeState(m);
    return waitForAsyncCompletion(60*2);
  }

  void EventProcessor::changeState(Msg msg) {
    // most likely need to serialize access to this routine

    boost::mutex::scoped_lock sl(state_lock_);
    State curr = state_;
    int rc;
    // found if(not end of table) and
    // (state == table.state && (msg == table.message || msg == any))
    for(rc = 0;
        table[rc].current != sInvalid &&
          (curr != table[rc].current ||
           (curr == table[rc].current &&
             msg != table[rc].message && table[rc].message != mAny));
        ++rc);

    if(table[rc].current == sInvalid)
      throw cms::Exception("BadState")
        << "A member function of EventProcessor has been called in an"
        << " inappropriate order.\n"
        << "Bad transition from " << stateName(curr) << " "
        << "using message " << msgName(msg) << "\n"
        << "No where to go from here.\n";

    FDEBUG(1) << "changeState: current=" << stateName(curr)
              << ", message=" << msgName(msg)
              << " -> new=" << stateName(table[rc].final) << "\n";

    state_ = table[rc].final;
  }

  void EventProcessor::runAsync() {
    beginJob();
    {
      boost::mutex::scoped_lock sl(stop_lock_);
      if(id_set_ == true) {
          std::string err("runAsync called while async event loop already running\n");
          LogError("FwkJob") << err;
          throw cms::Exception("BadState") << err;
      }

      changeState(mRunAsync);

      stop_count_ = 0;
      last_rc_ = epSuccess; // forget the last value!
      event_loop_.reset(new boost::thread(boost::bind(EventProcessor::asyncRun, this)));
      boost::xtime timeout;
#if BOOST_VERSION >= 105000
      boost::xtime_get(&timeout, boost::TIME_UTC_);
#else
      boost::xtime_get(&timeout, boost::TIME_UTC);
#endif
      timeout.sec += 60; // 60 seconds to start!!!!
      if(starter_.timed_wait(sl, timeout) == false) {
          // yikes - the thread did not start
          throw cms::Exception("BadState")
            << "Async run thread did not start in 60 seconds\n";
      }
    }
  }

  void EventProcessor::asyncRun(EventProcessor* me) {
    // set up signals to allow for interruptions
    // ignore all other signals
    // make sure no exceptions escape out

    // temporary hack until we modify the input source to allow
    // wakeup calls from other threads.  This mimics the solution
    // in EventFilter/Processor, which I do not like.
    // allowing cancels means that the thread just disappears at
    // certain points.  This is bad for C++ stack variables.
    pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, 0);
    //pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, 0);
    pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, 0);
    pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, 0);

    {
      boost::mutex::scoped_lock sl(me->stop_lock_);
      me->event_loop_id_ = pthread_self();
      me->id_set_ = true;
      me->starter_.notify_all();
    }

    Status rc = epException;
    FDEBUG(2) << "asyncRun starting ......................\n";

    try {
      bool onlineStateTransitions = true;
      rc = me->runToCompletion(onlineStateTransitions);
    }
    catch (cms::Exception& e) {
      LogError("FwkJob") << "cms::Exception caught in "
                         << "EventProcessor::asyncRun"
                         << "\n"
                         << e.explainSelf();
      me->last_error_text_ = e.explainSelf();
    }
    catch (std::exception& e) {
      LogError("FwkJob") << "Standard library exception caught in "
                         << "EventProcessor::asyncRun"
                         << "\n"
                         << e.what();
      me->last_error_text_ = e.what();
    }
    catch (...) {
      LogError("FwkJob") << "Unknown exception caught in "
                         << "EventProcessor::asyncRun"
                         << "\n";
      me->last_error_text_ = "Unknown exception caught";
      rc = epOther;
    }

    me->last_rc_ = rc;

    {
      // notify anyone waiting for exit that we are doing so now
      boost::mutex::scoped_lock sl(me->stop_lock_);
      ++me->stop_count_;
      me->stopper_.notify_all();
    }
    FDEBUG(2) << "asyncRun ending ......................\n";
  }

  std::auto_ptr<statemachine::Machine>
  EventProcessor::createStateMachine() {
    statemachine::FileMode fileMode;
    if(fileMode_.empty()) fileMode = statemachine::FULLMERGE;
    else if(fileMode_ == std::string("NOMERGE")) fileMode = statemachine::NOMERGE;
    else if(fileMode_ == std::string("FULLMERGE")) fileMode = statemachine::FULLMERGE;
    else {
      throw Exception(errors::Configuration, "Illegal fileMode parameter value: ")
      << fileMode_ << ".\n"
      << "Legal values are 'NOMERGE' and 'FULLMERGE'.\n";
    }
    
    statemachine::EmptyRunLumiMode emptyRunLumiMode;
    if(emptyRunLumiMode_.empty()) emptyRunLumiMode = statemachine::handleEmptyRunsAndLumis;
    else if(emptyRunLumiMode_ == std::string("handleEmptyRunsAndLumis")) emptyRunLumiMode = statemachine::handleEmptyRunsAndLumis;
    else if(emptyRunLumiMode_ == std::string("handleEmptyRuns")) emptyRunLumiMode = statemachine::handleEmptyRuns;
    else if(emptyRunLumiMode_ == std::string("doNotHandleEmptyRunsAndLumis")) emptyRunLumiMode = statemachine::doNotHandleEmptyRunsAndLumis;
    else {
      throw Exception(errors::Configuration, "Illegal emptyMode parameter value: ")
      << emptyRunLumiMode_ << ".\n"
      << "Legal values are 'handleEmptyRunsAndLumis', 'handleEmptyRuns', and 'doNotHandleEmptyRunsAndLumis'.\n";
    }
    
    std::auto_ptr<statemachine::Machine> machine(new statemachine::Machine(this,
                                             fileMode,
                                             emptyRunLumiMode));
    
    machine->initiate();
    return machine;
  }


  EventProcessor::StatusCode
  EventProcessor::runToCompletion(bool onlineStateTransitions) {

    StateSentry toerror(this);

    StatusCode returnCode=epSuccess;
    std::auto_ptr<statemachine::Machine> machine;
    {
      beginJob(); //make sure this was called
      
      if(!onlineStateTransitions) changeState(mRunCount);
      
      //StatusCode returnCode = epSuccess;
      stateMachineWasInErrorState_ = false;
      
      // make the services available
      ServiceRegistry::Operate operate(serviceToken_);

      machine = createStateMachine();
      try {
        try {
          
          InputSource::ItemType itemType;
          
          while(true) {
            
            bool more = true;
            if(numberOfForkedChildren_ > 0) {
              size_t size = preg_->size();
              more = input_->skipForForking();
              if(more) {
                if(size < preg_->size()) {
                  principalCache_.adjustIndexesAfterProductRegistryAddition();
                }
                principalCache_.adjustEventToNewProductRegistry(preg_);
              }
            } 
            itemType = (more ? input_->nextItemType() : InputSource::IsStop);
            
            FDEBUG(1) << "itemType = " << itemType << "\n";
            
            // These are used for asynchronous running only and
            // and are checking to see if stopAsync or shutdownAsync
            // were called from another thread.  In the future, we
            // may need to do something better than polling the state.
            // With the current code this is the simplest thing and
            // it should always work.  If the interaction between
            // threads becomes more complex this may cause problems.
            if(state_ == sStopping) {
              FDEBUG(1) << "In main processing loop, encountered sStopping state\n";
              forceLooperToEnd_ = true;
              machine->process_event(statemachine::Stop());
              forceLooperToEnd_ = false;
              break;
            }
            else if(state_ == sShuttingDown) {
              FDEBUG(1) << "In main processing loop, encountered sShuttingDown state\n";
              forceLooperToEnd_ = true;
              machine->process_event(statemachine::Stop());
              forceLooperToEnd_ = false;
              break;
            }
            
            // Look for a shutdown signal
            {
              boost::mutex::scoped_lock sl(usr2_lock);
              if(shutdown_flag) {
                changeState(mShutdownSignal);
                returnCode = epSignal;
                forceLooperToEnd_ = true;
                machine->process_event(statemachine::Stop());
                forceLooperToEnd_ = false;
                break;
              }
            }
            
            if(itemType == InputSource::IsStop) {
              machine->process_event(statemachine::Stop());
            }
            else if(itemType == InputSource::IsFile) {
              machine->process_event(statemachine::File());
            }
            else if(itemType == InputSource::IsRun) {
              machine->process_event(statemachine::Run(input_->reducedProcessHistoryID(), input_->run()));
            }
            else if(itemType == InputSource::IsLumi) {
              machine->process_event(statemachine::Lumi(input_->luminosityBlock()));
            }
            else if(itemType == InputSource::IsEvent) {
              machine->process_event(statemachine::Event());
            }
            // This should be impossible
            else {
              throw Exception(errors::LogicError)
              << "Unknown next item type passed to EventProcessor\n"
              << "Please report this error to the Framework group\n";
            }
            
            if(machine->terminated()) {
              changeState(mInputExhausted);
              break;
            }
          }  // End of loop over state machine events
        } // Try block
        catch (cms::Exception& e) { throw; }
        catch(std::bad_alloc& bda) { convertException::badAllocToEDM(); }
        catch (std::exception& e) { convertException::stdToEDM(e); }
        catch(std::string& s) { convertException::stringToEDM(s); }
        catch(char const* c) { convertException::charPtrToEDM(c); }
        catch (...) { convertException::unknownToEDM(); }
      } // Try block
      // Some comments on exception handling related to the boost state machine:
      //
      // Some states used in the machine are special because they
      // perform actions while the machine is being terminated, actions
      // such as close files, call endRun, call endLumi etc ...  Each of these
      // states has two functions that perform these actions.  The functions
      // are almost identical.  The major difference is that one version
      // catches all exceptions and the other lets exceptions pass through.
      // The destructor catches them and the other function named "exit" lets
      // them pass through.  On a normal termination, boost will always call
      // "exit" and then the state destructor.  In our state classes, the
      // the destructors do nothing if the exit function already took
      // care of things.  Here's the interesting part.  When boost is
      // handling an exception the "exit" function is not called (a boost
      // feature).
      //
      // If an exception occurs while the boost machine is in control
      // (which usually means inside a process_event call), then
      // the boost state machine destroys its states and "terminates" itself.
      // This already done before we hit the catch blocks below. In this case
      // the call to terminateMachine below only destroys an already
      // terminated state machine.  Because exit is not called, the state destructors
      // handle cleaning up lumis, runs, and files.  The destructors swallow
      // all exceptions and only pass through the exceptions messages, which
      // are tacked onto the original exception below.
      //
      // If an exception occurs when the boost state machine is not
      // in control (outside the process_event functions), then boost
      // cannot destroy its own states.  The terminateMachine function
      // below takes care of that.  The flag "alreadyHandlingException"
      // is set true so that the state exit functions do nothing (and
      // cannot throw more exceptions while handling the first).  Then the
      // state destructors take care of this because exit did nothing.
      //
      // In both cases above, the EventProcessor::endOfLoop function is
      // not called because it can throw exceptions.
      //
      // One tricky aspect of the state machine is that things that can
      // throw should not be invoked by the state machine while another
      // exception is being handled.
      // Another tricky aspect is that it appears to be important to
      // terminate the state machine before invoking its destructor.
      // We've seen crashes that are not understood when that is not
      // done.  Maintainers of this code should be careful about this.
      
      catch (cms::Exception & e) {
        alreadyHandlingException_ = true;
        terminateMachine(machine);
        alreadyHandlingException_ = false;
        if (!exceptionMessageLumis_.empty()) {
          e.addAdditionalInfo(exceptionMessageLumis_);
          if (e.alreadyPrinted()) {
            LogAbsolute("Additional Exceptions") << exceptionMessageLumis_;
          }
        }
        if (!exceptionMessageRuns_.empty()) {
          e.addAdditionalInfo(exceptionMessageRuns_);
          if (e.alreadyPrinted()) {
            LogAbsolute("Additional Exceptions") << exceptionMessageRuns_;
          }
        }
        if (!exceptionMessageFiles_.empty()) {
          e.addAdditionalInfo(exceptionMessageFiles_);
          if (e.alreadyPrinted()) {
            LogAbsolute("Additional Exceptions") << exceptionMessageFiles_;
          }
        }
        throw;
      }
      
      if(machine->terminated()) {
        FDEBUG(1) << "The state machine reports it has been terminated\n";
        machine.reset();
      }
      
      if(!onlineStateTransitions) changeState(mFinished);
      
      if(stateMachineWasInErrorState_) {
        throw cms::Exception("BadState")
        << "The boost state machine in the EventProcessor exited after\n"
        << "entering the Error state.\n";
      }
      
    }
    if(machine.get() != 0) {
      terminateMachine(machine);
      throw Exception(errors::LogicError)
        << "State machine not destroyed on exit from EventProcessor::runToCompletion\n"
        << "Please report this error to the Framework group\n";
    }

    toerror.succeeded();

    return returnCode;
  }

  void EventProcessor::readFile() {
    FDEBUG(1) << " \treadFile\n";
    size_t size = preg_->size();
    fb_ = input_->readFile();
    if(size < preg_->size()) {
      principalCache_.adjustIndexesAfterProductRegistryAddition();
    }
    principalCache_.adjustEventToNewProductRegistry(preg_);
    if(numberOfForkedChildren_ > 0) {
        fb_->setNotFastClonable(FileBlock::ParallelProcesses);
    }
  }

  void EventProcessor::closeInputFile(bool cleaningUpAfterException) {
    if (fb_.get() != nullptr) {
      input_->closeFile(fb_.get(), cleaningUpAfterException);
    }
    FDEBUG(1) << "\tcloseInputFile\n";
  }

  void EventProcessor::openOutputFiles() {
    if (fb_.get() != nullptr) {
      schedule_->openOutputFiles(*fb_);
      if(hasSubProcess()) subProcess_->openOutputFiles(*fb_);
    }
    FDEBUG(1) << "\topenOutputFiles\n";
  }

  void EventProcessor::closeOutputFiles() {
    if (fb_.get() != nullptr) {
      schedule_->closeOutputFiles();
      if(hasSubProcess()) subProcess_->closeOutputFiles();
    }
    FDEBUG(1) << "\tcloseOutputFiles\n";
  }

  void EventProcessor::respondToOpenInputFile() {
    if (fb_.get() != nullptr) {
      schedule_->respondToOpenInputFile(*fb_);
      if(hasSubProcess()) subProcess_->respondToOpenInputFile(*fb_);
    }
    FDEBUG(1) << "\trespondToOpenInputFile\n";
  }

  void EventProcessor::respondToCloseInputFile() {
    if (fb_.get() != nullptr) {
      schedule_->respondToCloseInputFile(*fb_);
      if(hasSubProcess()) subProcess_->respondToCloseInputFile(*fb_);
    }
    FDEBUG(1) << "\trespondToCloseInputFile\n";
  }

  void EventProcessor::respondToOpenOutputFiles() {
    if (fb_.get() != nullptr) {
      schedule_->respondToOpenOutputFiles(*fb_);
      if(hasSubProcess()) subProcess_->respondToOpenOutputFiles(*fb_);
    }
    FDEBUG(1) << "\trespondToOpenOutputFiles\n";
  }

  void EventProcessor::respondToCloseOutputFiles() {
    if (fb_.get() != nullptr) {
      schedule_->respondToCloseOutputFiles(*fb_);
      if(hasSubProcess()) subProcess_->respondToCloseOutputFiles(*fb_);
    }
    FDEBUG(1) << "\trespondToCloseOutputFiles\n";
  }

  void EventProcessor::startingNewLoop() {
    shouldWeStop_ = false;
    //NOTE: for first loop, need to delay calling 'doStartingNewLoop'
    // until after we've called beginOfJob
    if(looper_ && looperBeginJobRun_) {
      looper_->doStartingNewLoop();
    }
    FDEBUG(1) << "\tstartingNewLoop\n";
  }

  bool EventProcessor::endOfLoop() {
    if(looper_) {
      ModuleChanger changer(schedule_.get());
      looper_->setModuleChanger(&changer);
      EDLooperBase::Status status = looper_->doEndOfLoop(esp_->eventSetup());
      looper_->setModuleChanger(nullptr);
      if(status != EDLooperBase::kContinue || forceLooperToEnd_) return true;
      else return false;
    }
    FDEBUG(1) << "\tendOfLoop\n";
    return true;
  }

  void EventProcessor::rewindInput() {
    input_->repeat();
    input_->rewind();
    FDEBUG(1) << "\trewind\n";
  }

  void EventProcessor::prepareForNextLoop() {
    looper_->prepareForNextLoop(esp_.get());
    FDEBUG(1) << "\tprepareForNextLoop\n";
  }

  bool EventProcessor::shouldWeCloseOutput() const {
    FDEBUG(1) << "\tshouldWeCloseOutput\n";
    return hasSubProcess() ? subProcess_->shouldWeCloseOutput() : schedule_->shouldWeCloseOutput();
  }

  void EventProcessor::doErrorStuff() {
    FDEBUG(1) << "\tdoErrorStuff\n";
    LogError("StateMachine")
      << "The EventProcessor state machine encountered an unexpected event\n"
      << "and went to the error state\n"
      << "Will attempt to terminate processing normally\n"
      << "(IF using the looper the next loop will be attempted)\n"
      << "This likely indicates a bug in an input module or corrupted input or both\n";
    stateMachineWasInErrorState_ = true;
  }

  void EventProcessor::beginRun(statemachine::Run const& run) {
    RunPrincipal& runPrincipal = principalCache_.runPrincipal(run.processHistoryID(), run.runNumber());
    input_->doBeginRun(runPrincipal);
    IOVSyncValue ts(EventID(runPrincipal.run(), 0, 0),
                    runPrincipal.beginTime());
    if(forceESCacheClearOnNewRun_){
      espController_->forceCacheClear();
    }
    espController_->eventSetupForInstance(ts);
    EventSetup const& es = esp_->eventSetup();
    if(looper_ && looperBeginJobRun_== false) {
      looper_->copyInfo(ScheduleInfo(schedule_.get()));
      looper_->beginOfJob(es);
      looperBeginJobRun_ = true;
      looper_->doStartingNewLoop();
    }
    {
      typedef OccurrenceTraits<RunPrincipal, BranchActionBegin> Traits;
      ScheduleSignalSentry<Traits> sentry(actReg_.get(), &runPrincipal, &es);
      schedule_->processOneOccurrence<Traits>(runPrincipal, es);
      if(hasSubProcess()) {
        subProcess_->doBeginRun(runPrincipal, ts);
      }
      sentry.allowThrow();
    }
    FDEBUG(1) << "\tbeginRun " << run.runNumber() << "\n";
    if(looper_) {
      looper_->doBeginRun(runPrincipal, es);
    }
  }

  void EventProcessor::endRun(statemachine::Run const& run, bool cleaningUpAfterException) {
    RunPrincipal& runPrincipal = principalCache_.runPrincipal(run.processHistoryID(), run.runNumber());
    input_->doEndRun(runPrincipal, cleaningUpAfterException);
    IOVSyncValue ts(EventID(runPrincipal.run(), LuminosityBlockID::maxLuminosityBlockNumber(), EventID::maxEventNumber()),
                    runPrincipal.endTime());
    espController_->eventSetupForInstance(ts);
    EventSetup const& es = esp_->eventSetup();
    {
      typedef OccurrenceTraits<RunPrincipal, BranchActionEnd> Traits;
      ScheduleSignalSentry<Traits> sentry(actReg_.get(), &runPrincipal, &es);
      schedule_->processOneOccurrence<Traits>(runPrincipal, es, cleaningUpAfterException);
      if(hasSubProcess()) {
        subProcess_->doEndRun(runPrincipal, ts, cleaningUpAfterException);
      }
      sentry.allowThrow();
    }
    FDEBUG(1) << "\tendRun " << run.runNumber() << "\n";
    if(looper_) {
      looper_->doEndRun(runPrincipal, es);
    }
  }

  void EventProcessor::beginLumi(ProcessHistoryID const& phid, RunNumber_t run, LuminosityBlockNumber_t lumi) {
    LuminosityBlockPrincipal& lumiPrincipal = principalCache_.lumiPrincipal(phid, run, lumi);
    input_->doBeginLumi(lumiPrincipal);

    Service<RandomNumberGenerator> rng;
    if(rng.isAvailable()) {
      LuminosityBlock lb(lumiPrincipal, ModuleDescription());
      rng->preBeginLumi(lb);
    }

    // NOTE: Using 0 as the event number for the begin of a lumi block is a bad idea
    // lumi blocks know their start and end times why not also start and end events?
    IOVSyncValue ts(EventID(lumiPrincipal.run(), lumiPrincipal.luminosityBlock(), 0), lumiPrincipal.beginTime());
    espController_->eventSetupForInstance(ts);
    EventSetup const& es = esp_->eventSetup();
    {
      typedef OccurrenceTraits<LuminosityBlockPrincipal, BranchActionBegin> Traits;
      ScheduleSignalSentry<Traits> sentry(actReg_.get(), &lumiPrincipal, &es);
      schedule_->processOneOccurrence<Traits>(lumiPrincipal, es);
      if(hasSubProcess()) {
        subProcess_->doBeginLuminosityBlock(lumiPrincipal, ts);
      }
      sentry.allowThrow();
    }
    FDEBUG(1) << "\tbeginLumi " << run << "/" << lumi << "\n";
    if(looper_) {
      looper_->doBeginLuminosityBlock(lumiPrincipal, es);
    }
  }

  void EventProcessor::endLumi(ProcessHistoryID const& phid, RunNumber_t run, LuminosityBlockNumber_t lumi, bool cleaningUpAfterException) {
    LuminosityBlockPrincipal& lumiPrincipal = principalCache_.lumiPrincipal(phid, run, lumi);
    input_->doEndLumi(lumiPrincipal, cleaningUpAfterException);
    //NOTE: Using the max event number for the end of a lumi block is a bad idea
    // lumi blocks know their start and end times why not also start and end events?
    IOVSyncValue ts(EventID(lumiPrincipal.run(), lumiPrincipal.luminosityBlock(), EventID::maxEventNumber()),
                    lumiPrincipal.endTime());
    espController_->eventSetupForInstance(ts);
    EventSetup const& es = esp_->eventSetup();
    {
      typedef OccurrenceTraits<LuminosityBlockPrincipal, BranchActionEnd> Traits;
      ScheduleSignalSentry<Traits> sentry(actReg_.get(), &lumiPrincipal, &es);
      schedule_->processOneOccurrence<Traits>(lumiPrincipal, es, cleaningUpAfterException);
      if(hasSubProcess()) {
        subProcess_->doEndLuminosityBlock(lumiPrincipal, ts, cleaningUpAfterException);
      }
      sentry.allowThrow();
    }
    FDEBUG(1) << "\tendLumi " << run << "/" << lumi << "\n";
    if(looper_) {
      looper_->doEndLuminosityBlock(lumiPrincipal, es);
    }
  }

  statemachine::Run EventProcessor::readAndCacheRun() {
    if (principalCache_.hasRunPrincipal()) {
      throw edm::Exception(edm::errors::LogicError)
        << "EventProcessor::readAndCacheRun\n"
        << "Illegal attempt to insert run into cache\n"
        << "Contact a Framework Developer\n";
    }
    principalCache_.insert(input_->readAndCacheRun(*historyAppender_));
    return statemachine::Run(input_->reducedProcessHistoryID(), input_->run());
  }

  statemachine::Run EventProcessor::readAndMergeRun() {
    principalCache_.merge(input_->runAuxiliary(), preg_);
    input_->readAndMergeRun(principalCache_.runPrincipalPtr());
    return statemachine::Run(input_->reducedProcessHistoryID(), input_->run());
  }

  int EventProcessor::readAndCacheLumi() {
    if (principalCache_.hasLumiPrincipal()) {
      throw edm::Exception(edm::errors::LogicError)
        << "EventProcessor::readAndCacheRun\n"
        << "Illegal attempt to insert lumi into cache\n"
        << "Contact a Framework Developer\n";
    }
    if (!principalCache_.hasRunPrincipal()) {
      throw edm::Exception(edm::errors::LogicError)
        << "EventProcessor::readAndCacheRun\n"
        << "Illegal attempt to insert lumi into cache\n"
        << "Run is invalid\n"
        << "Contact a Framework Developer\n";
    }
    principalCache_.insert(input_->readAndCacheLumi(*historyAppender_));
    principalCache_.lumiPrincipalPtr()->setRunPrincipal(principalCache_.runPrincipalPtr());
    return input_->luminosityBlock();
  }

  int EventProcessor::readAndMergeLumi() {
    principalCache_.merge(input_->luminosityBlockAuxiliary(), preg_);
    input_->readAndMergeLumi(principalCache_.lumiPrincipalPtr());
    return input_->luminosityBlock();
  }

  void EventProcessor::writeRun(statemachine::Run const& run) {
    schedule_->writeRun(principalCache_.runPrincipal(run.processHistoryID(), run.runNumber()));
    if(hasSubProcess()) subProcess_->writeRun(run.processHistoryID(), run.runNumber());
    FDEBUG(1) << "\twriteRun " << run.runNumber() << "\n";
  }

  void EventProcessor::deleteRunFromCache(statemachine::Run const& run) {
    principalCache_.deleteRun(run.processHistoryID(), run.runNumber());
    if(hasSubProcess()) subProcess_->deleteRunFromCache(run.processHistoryID(), run.runNumber());
    FDEBUG(1) << "\tdeleteRunFromCache " << run.runNumber() << "\n";
  }

  void EventProcessor::writeLumi(ProcessHistoryID const& phid, RunNumber_t run, LuminosityBlockNumber_t lumi) {
    schedule_->writeLumi(principalCache_.lumiPrincipal(phid, run, lumi));
    if(hasSubProcess()) subProcess_->writeLumi(phid, run, lumi);
    FDEBUG(1) << "\twriteLumi " << run << "/" << lumi << "\n";
  }

  void EventProcessor::deleteLumiFromCache(ProcessHistoryID const& phid, RunNumber_t run, LuminosityBlockNumber_t lumi) {
    principalCache_.deleteLumi(phid, run, lumi);
    if(hasSubProcess()) subProcess_->deleteLumiFromCache(phid, run, lumi);
    FDEBUG(1) << "\tdeleteLumiFromCache " << run << "/" << lumi << "\n";
  }

  void EventProcessor::readAndProcessEvent() {
    EventPrincipal *pep = input_->readEvent(principalCache_.eventPrincipal());
    FDEBUG(1) << "\treadEvent\n";
    assert(pep != 0);
    pep->setLuminosityBlockPrincipal(principalCache_.lumiPrincipalPtr());
    assert(pep->luminosityBlockPrincipalPtrValid());
    assert(principalCache_.lumiPrincipalPtr()->run() == pep->run());
    assert(principalCache_.lumiPrincipalPtr()->luminosityBlock() == pep->luminosityBlock());

    IOVSyncValue ts(pep->id(), pep->time());
    espController_->eventSetupForInstance(ts);
    EventSetup const& es = esp_->eventSetup();
    {
      typedef OccurrenceTraits<EventPrincipal, BranchActionBegin> Traits;
      ScheduleSignalSentry<Traits> sentry(actReg_.get(), pep, &es);
      schedule_->processOneOccurrence<Traits>(*pep, es);
      if(hasSubProcess()) {
        subProcess_->doEvent(*pep, ts);
      }
      sentry.allowThrow();
    }

    if(looper_) {
      bool randomAccess = input_->randomAccess();
      ProcessingController::ForwardState forwardState = input_->forwardState();
      ProcessingController::ReverseState reverseState = input_->reverseState();
      ProcessingController pc(forwardState, reverseState, randomAccess);

      EDLooperBase::Status status = EDLooperBase::kContinue;
      do {
        status = looper_->doDuringLoop(*pep, esp_->eventSetup(), pc);

        bool succeeded = true;
        if(randomAccess) {
          if(pc.requestedTransition() == ProcessingController::kToPreviousEvent) {
            input_->skipEvents(-2);
          }
          else if(pc.requestedTransition() == ProcessingController::kToSpecifiedEvent) {
            succeeded = input_->goToEvent(pc.specifiedEventTransition());
          }
        }
        pc.setLastOperationSucceeded(succeeded);
      } while(!pc.lastOperationSucceeded());
      if(status != EDLooperBase::kContinue) shouldWeStop_ = true;

    }

    FDEBUG(1) << "\tprocessEvent\n";
    pep->clearEventPrincipal();
  }

  bool EventProcessor::shouldWeStop() const {
    FDEBUG(1) << "\tshouldWeStop\n";
    if(shouldWeStop_) return true;
    return (schedule_->terminate() || (hasSubProcess() && subProcess_->terminate()));
  }

  void EventProcessor::setExceptionMessageFiles(std::string& message) {
    exceptionMessageFiles_ = message;
  }

  void EventProcessor::setExceptionMessageRuns(std::string& message) {
    exceptionMessageRuns_ = message;
  }

  void EventProcessor::setExceptionMessageLumis(std::string& message) {
    exceptionMessageLumis_ = message;
  }

  bool EventProcessor::alreadyHandlingException() const {
    return alreadyHandlingException_;
  }

  void EventProcessor::terminateMachine(std::auto_ptr<statemachine::Machine>& iMachine) {
    if(iMachine.get() != 0) {
      if(!iMachine->terminated()) {
        forceLooperToEnd_ = true;
        iMachine->process_event(statemachine::Stop());
        forceLooperToEnd_ = false;
      }
      else {
        FDEBUG(1) << "EventProcess::terminateMachine  The state machine was already terminated \n";
      }
      if(iMachine->terminated()) {
        FDEBUG(1) << "The state machine reports it has been terminated (3)\n";
      }
      iMachine.reset();
    }
  }
}