HltEventLoopMgr.h

Go to the documentation of this file.

/*
  Copyright (C) 2002-2026 CERN for the benefit of the ATLAS collaboration
*/
 
#ifndef TRIGSERVICES_HLTEVENTLOOPMGR_H
#define TRIGSERVICES_HLTEVENTLOOPMGR_H
 
// Local includes
#include "TrigKernel/EventLoopUtils.h"
#include "TrigSORFromPtreeHelper.h"
 
// Trigger includes
#include "TrigKernel/ITrigEventLoopMgr.h"
#include "TrigOutputHandling/HLTResultMTMaker.h"
#include "TrigSteeringEvent/OnlineErrorCode.h"
#include "TrigSteerMonitor/ISchedulerMonSvc.h"
#include "TrigSteerMonitor/ITrigErrorMonTool.h"
#include "TrigT1Result/RoIBResult.h"
#include "xAODTrigger/TrigCompositeContainer.h"
 
// Athena includes
#include "AthenaBaseComps/AthService.h"
#include "AthenaKernel/EventContextClid.h"
#include "AthenaKernel/Timeout.h"
#include "AthenaMonitoringKernel/Monitored.h"
#include "xAODEventInfo/EventInfo.h"
#include "StoreGate/ReadHandleKey.h"
#include "StoreGate/WriteHandleKey.h"
 
// Gaudi includes
#include "GaudiKernel/EventIDBase.h" // number_type
#include "GaudiKernel/IEventProcessor.h"
#include "GaudiKernel/IEvtSelector.h"
#include "GaudiKernel/IConversionSvc.h"
#include "GaudiKernel/IAlgResourcePool.h"
#include "GaudiKernel/IAlgExecStateSvc.h"
#include "GaudiKernel/IHiveWhiteBoard.h"
#include "GaudiKernel/IScheduler.h"
#include "GaudiKernel/IIoComponentMgr.h"
#include "GaudiKernel/SmartIF.h"
#include "Gaudi/Interfaces/IOptionsSvc.h"
 
// TBB includes
#include "tbb/concurrent_queue.h"
#include "tbb/task_group.h"
 
// System includes
#include <atomic>
#include <chrono>
#include <condition_variable>
#include <memory>
 
// Forward declarations
class CondAttrListCollection;
class IAlgorithm;
class IIncidentSvc;
class StoreGateSvc;
class TrigCOOLUpdateHelper;
 
namespace coral {
  class AttributeList;
}
namespace HLT {
  class HLTResultMT;
}

class HltEventLoopMgr : public extends<AthService, ITrigEventLoopMgr, IEventProcessor>,
                        public Athena::TimeoutMaster
{
 
public:

  HltEventLoopMgr(const std::string& name, ISvcLocator* svcLoc);
  virtual ~HltEventLoopMgr() noexcept override;
 
  // Copy and move not allowed
  HltEventLoopMgr(const HltEventLoopMgr&) = delete;
  HltEventLoopMgr(HltEventLoopMgr&&) = delete;
  HltEventLoopMgr& operator=(const HltEventLoopMgr&) = delete;
  HltEventLoopMgr& operator=(HltEventLoopMgr&&) = delete;

  virtual StatusCode initialize() override;
  virtual StatusCode stop() override;
  virtual StatusCode finalize() override;

  virtual StatusCode prepareForStart (const boost::property_tree::ptree &) override;
  virtual StatusCode prepareForRun(const boost::property_tree::ptree& pt) override;
  virtual StatusCode hltUpdateAfterFork(const boost::property_tree::ptree& pt) override;

  virtual StatusCode executeRun(int maxevt=-1) override;

  virtual StatusCode nextEvent(int maxevt=-1) override;

  virtual StatusCode executeEvent( EventContext &&ctx ) override;

  virtual EventContext createEventContext() override;

  virtual StatusCode stopRun() override;
 
private:
  // ------------------------- Helper types ------------------------------------
  struct EventLoopStatus {
    std::atomic<bool> eventsAvailable{true};
    std::atomic<bool> loopEnded{false};
    std::condition_variable loopEndedCond;
    std::mutex loopEndedMutex;
    std::atomic<EventIDBase::number_type> maxLB{0};
    std::condition_variable coolUpdateCond;
    std::mutex coolUpdateMutex;
    bool coolUpdateOngoing{false};
    StatusCode exitCode{StatusCode::SUCCESS};
  };
 
  // ------------------------- Helper methods ----------------------------------

  void updateDFProps();
 
  // Update internally kept data from new sor
  void updateMetadataStore(const coral::AttributeList & sor_attrlist) const;

  StatusCode updateMagField(const boost::property_tree::ptree& pt) const;

  StatusCode clearTemporaryStores();

  const coral::AttributeList& getSorAttrList() const;

  void printSORAttrList(const coral::AttributeList& atr) const;

  StatusCode execAtStart(const EventContext& ctx) const;

  StatusCode failedEvent(HLT::OnlineErrorCode errorCode,
                         const EventContext& eventContext);

  void resetEventTimer(const EventContext& eventContext, bool processing);

  StatusCode clearWBSlot(size_t evtSlot) const;

  void inputThreadCallback();

  void outputThreadCallback();

  void eventTimerCallback();

  StatusCode startNextEvent();

  StatusCode processFinishedEvent();
 
  // ------------------------- Handles to required services/tools --------------
  ServiceHandle<IIncidentSvc>        m_incidentSvc{this, "IncidentSvc", "IncidentSvc"};
  ServiceHandle<Gaudi::Interfaces::IOptionsSvc> m_jobOptionsSvc{this, "JobOptionsSvc", "JobOptionsSvc"};
  ServiceHandle<StoreGateSvc>        m_evtStore{this, "EventStore", "StoreGateSvc"};
  ServiceHandle<StoreGateSvc>        m_detectorStore{this, "DetectorStore", "DetectorStore"};
  ServiceHandle<StoreGateSvc>        m_inputMetaDataStore{this, "InputMetaDataStore", "StoreGateSvc/InputMetaDataStore"};
  ServiceHandle<IIoComponentMgr>     m_ioCompMgr{this, "IoComponentMgr", "IoComponentMgr"};
  ServiceHandle<IEvtSelector>        m_evtSelector{this, "EvtSel", "EvtSel"};
  ServiceHandle<IConversionSvc>      m_outputCnvSvc{this, "OutputCnvSvc", "OutputCnvSvc"};
  ServiceHandle<ISchedulerMonSvc>    m_schedulerMonSvc{this, "SchedulerMonSvc", "SchedulerMonSvc"};
  ToolHandle<TrigCOOLUpdateHelper>   m_coolHelper{this, "CoolUpdateTool", "TrigCOOLUpdateHelper"};
  ToolHandle<HLTResultMTMaker>       m_hltResultMaker{this, "ResultMaker", "HLTResultMTMaker"};
  ToolHandle<GenericMonitoringTool>  m_monTool{this, "MonTool", "", "Monitoring tool"};
  ToolHandle<ITrigErrorMonTool>      m_errorMonTool{this, "TrigErrorMonTool", "TrigErrorMonTool", "Error monitoring tool"};
 
  SmartIF<IHiveWhiteBoard> m_whiteboard;
  SmartIF<IAlgResourcePool> m_algResourcePool;
  SmartIF<IAlgExecStateSvc> m_aess;
  SmartIF<IScheduler> m_schedulerSvc;
 
  std::unique_ptr<TrigSORFromPtreeHelper> m_sorHelper;
 
  // ------------------------- Other properties --------------------------------------
  Gaudi::Property<std::string> m_schedulerName{
    this, "SchedulerSvc", "AvalancheSchedulerSvc", "Name of the scheduler"};
 
  Gaudi::Property<std::string> m_whiteboardName{
    this, "WhiteboardSvc", "EventDataSvc", "Name of the Whiteboard"};
 
  Gaudi::Property<float> m_hardTimeout{
    this, "HardTimeout", 10*60*1000/*=10min*/, "Hard event processing timeout in milliseconds"};
 
  Gaudi::Property<float> m_softTimeoutFraction{
    this, "SoftTimeoutFraction", 0.8, "Fraction of the hard timeout to be set as the soft timeout"};
 
  Gaudi::Property<unsigned int> m_timeoutThreadIntervalMs{
    this, "TimeoutThreadIntervalMs", 1000, "How often the timeout thread checks for soft timeout, in milliseconds"};
 
  Gaudi::Property<bool> m_traceOnTimeout{
    this, "TraceOnTimeout", true,
    "Print a stack trace on the first soft timeout (might take a while, holding all threads)"};
 
  Gaudi::Property<int> m_maxParallelIOTasks{
    this, "MaxParallelIOTasks", -1,
    "Maximum number of I/O tasks which can be executed in parallel. "
    "If <=0 then the number of scheduler threads is used."};
 
  Gaudi::Property<int> m_maxIOWakeUpIntervalMs{
    this, "MaxIOWakeUpIntervalMs", 5000,
    "Maximum time input or output handling thread will sleep unless notified. Negative value (default) means no limit, "
    "i.e. threads will only wake up on notifications. Zero means threads will never wait for notifications. "
    "Positive value means the number of milliseconds after which a thread will wake up if it's not notified earlier."};
 
  Gaudi::Property<int> m_maxFrameworkErrors{
    this, "MaxFrameworkErrors", 10,
    "Tolerable number of recovered framework errors before exiting (<0 means all are tolerated)"};
 
  Gaudi::Property<std::string> m_fwkErrorDebugStreamName{
    this, "FwkErrorDebugStreamName", "HLTMissingData",
    "Debug stream name for events with HLT framework errors"};
 
  Gaudi::Property<std::string> m_algErrorDebugStreamName{
    this, "AlgErrorDebugStreamName", "HltError",
    "Debug stream name for events with HLT algorithm errors"};
 
  Gaudi::Property<std::string> m_timeoutDebugStreamName{
    this, "TimeoutDebugStreamName", "HltTimeout",
    "Debug stream name for events with HLT timeout"};
 
  Gaudi::Property<std::string> m_truncationDebugStreamName{
    this, "TruncationDebugStreamName", "TruncatedHLTResult",
    "Debug stream name for events with HLT result truncation"};
 
  Gaudi::Property<std::string> m_sorPath{
    this, "SORPath", "/TDAQ/RunCtrl/SOR_Params", "Path to StartOfRun parameters in detector store"};
 
  Gaudi::Property<std::vector<std::string>> m_execAtStart{
    this, "execAtStart", {}, "List of algorithms/sequences to execute during prepareForRun"};
 
  Gaudi::Property<bool> m_setMagFieldFromPtree{
    this, "setMagFieldFromPtree", true, "Read magnet currents from ptree"};
 
  Gaudi::Property<unsigned int> m_forceRunNumber{
    this, "forceRunNumber", 0, "Override run number"};
 
  Gaudi::Property<unsigned int> m_forceLumiblock{
    this, "forceLumiblock", 0, "Override lumiblock number"};
 
  Gaudi::Property<unsigned long long> m_forceSOR_ns{
    this, "forceStartOfRunTime", 0, "Override SOR time (epoch in nano-seconds)"};
 
  Gaudi::Property<bool> m_rewriteLVL1{
    this, "RewriteLVL1", false,
    "Encode L1 results to ByteStream and write to the output. Possible only with athenaHLT, not online."};
 
  Gaudi::Property<bool> m_monitorScheduler{
    this, "MonitorScheduler", false, "Enable SchedulerMonSvc to collect scheduler status data in online histograms"};
 
  SG::WriteHandleKey<EventContext> m_eventContextWHKey{
    this, "EventContextWHKey", "EventContext", "StoreGate key for recording EventContext"};
 
  SG::ReadHandleKey<xAOD::EventInfo> m_eventInfoRHKey{
    this, "EventInfoRHKey", "EventInfo", "StoreGate key for reading xAOD::EventInfo"};
 
  SG::ReadHandleKey<xAOD::TrigCompositeContainer> m_l1TriggerResultRHKey{
    this, "L1TriggerResultRHKey", "", "StoreGate key for reading L1TriggerResult for RewriteLVL1"};
 
  SG::ReadHandleKey<ROIB::RoIBResult> m_roibResultRHKey{
    this, "RoIBResultRHKey", "", "StoreGate key for reading RoIBResult for RewriteLVL1 with legacy (Run-2) L1 simulation"};
 
  SG::ReadHandleKey<HLT::HLTResultMT> m_hltResultRHKey;    
 
  // ------------------------- Other private members ---------------------------

  EventContext m_currentRunCtx{0,0};
  std::atomic<size_t> m_localEventNumber{0};
  IEvtSelector::Context* m_evtSelContext{nullptr};
  std::vector<std::chrono::steady_clock::time_point> m_eventTimerStartPoint;
  std::vector<std::chrono::steady_clock::time_point> m_freeSlotStartPoint;
  std::vector<bool> m_isSlotProcessing; // be aware of vector<bool> specialisation
  std::atomic<size_t> m_freeSlots{0};
  std::unique_ptr<HLT::LoopThread> m_inputThread;
  std::unique_ptr<HLT::LoopThread> m_outputThread;
  std::unique_ptr<HLT::LoopThread> m_timeoutThread;
  std::chrono::milliseconds m_softTimeoutValue{0};
  tbb::task_group m_parallelIOTaskGroup;
  tbb::concurrent_bounded_queue<bool> m_parallelIOQueue;
  tbb::concurrent_bounded_queue<EventContext*> m_finishedEventsQueue;
  EventLoopStatus m_loopStatus{};
  bool m_timeoutTraceGenerated{false};
  std::atomic<int> m_nFrameworkErrors{0};
  std::string m_applicationName;
  int m_workerID{0};
  int m_workerPID{0};
 
};
 
#endif // TRIGSERVICES_HLTEVENTLOOPMGR_H