SharedHiveEvtQueueConsumer Class Reference

#include <SharedHiveEvtQueueConsumer.h>

Inheritance diagram for SharedHiveEvtQueueConsumer:

Collaboration diagram for SharedHiveEvtQueueConsumer:

Public Member Functions
	SharedHiveEvtQueueConsumer (const std::string &type, const std::string &name, const IInterface *parent)
virtual	~SharedHiveEvtQueueConsumer () override
virtual StatusCode	initialize () override
virtual StatusCode	finalize () override
virtual int makePool	ATLAS_NOT_THREAD_SAFE (int maxevt, int nprocs, const std::string &topdir) override
virtual StatusCode exec	ATLAS_NOT_THREAD_SAFE () override
virtual StatusCode wait_once	ATLAS_NOT_THREAD_SAFE (pid_t &pid) override
virtual void	reportSubprocessStatuses () override
virtual void	subProcessLogs (std::vector< std::string > &) override
virtual std::unique_ptr< AthenaInterprocess::ScheduledWork >	bootstrap_func () override
virtual std::unique_ptr< AthenaInterprocess::ScheduledWork >	exec_func () override
virtual std::unique_ptr< AthenaInterprocess::ScheduledWork >	fin_func () override
virtual AthenaMP::AllWorkerOutputs_ptr	generateOutputReport () override
virtual void	useFdsRegistry (std::shared_ptr< AthenaInterprocess::FdsRegistry >) override
virtual void	setRandString (const std::string &randStr) override
virtual void	setMaxEvt (int maxEvt) override
virtual void	setMPRunStop (const AthenaInterprocess::IMPRunStop *runStop) override
virtual void	killChildren () override
virtual std::unique_ptr< ScheduledWork >	operator () ATLAS_NOT_THREAD_SAFE(const ScheduledWork &)=0

Protected Types
enum	ESRange_Status {   ESRANGE_SUCCESS , ESRANGE_NOTFOUND , ESRANGE_SEEKFAILED , ESRANGE_PROCFAILED ,   ESRANGE_FILENOTMADE , ESRANGE_BADINPFILE }
enum	Func_Flag { FUNC_BOOTSTRAP , FUNC_EXEC , FUNC_FIN }

Protected Member Functions
int mapAsyncFlag	ATLAS_NOT_THREAD_SAFE (Func_Flag flag, pid_t pid=0)
int	redirectLog (const std::string &rundir, bool addTimeStamp=true)
int	updateIoReg (const std::string &rundir)
std::string	fmterror (int errnum)
int	reopenFds ()
int	handleSavedPfc (const std::filesystem::path &dest_path)
void	waitForSignal ()
IEvtSelector *	evtSelector ()

Protected Attributes
int	m_nprocs {-1}
	Number of workers spawned by the master process.
int	m_maxEvt {-1}
	Maximum number of events assigned to the job.
std::string	m_subprocTopDir
	Top run directory for subprocesses.
std::string	m_subprocDirPrefix
	For ex. "worker__".
std::string	m_evtSelName
	Name of the event selector.
AthenaInterprocess::ProcessGroup *	m_processGroup {nullptr}
const AthenaInterprocess::IMPRunStop *	m_mpRunStop {nullptr}
ServiceHandle< IEventProcessor >	m_evtProcessor
ServiceHandle< IAppMgrUI >	m_appMgr
ServiceHandle< IFileMgr >	m_fileMgr
ServiceHandle< IIoComponentMgr >	m_ioMgr
SmartIF< IEvtSelector >	m_evtSelector
std::string	m_fileMgrLog
std::shared_ptr< AthenaInterprocess::FdsRegistry >	m_fdsRegistry
std::string	m_randStr
Gaudi::Property< bool >	m_isPileup {this, "IsPileup", false, "Flag for configuring PileUpEventLoopMgr"}

Private Member Functions
StatusCode	initHive ()
int decodeProcessResult	ATLAS_NOT_THREAD_SAFE (const AthenaInterprocess::ProcessResult *presult, bool doFinalize)
int	reopenFd (int fd, const std::string &name)

Private Attributes
Gaudi::Property< int >	m_nEventsBeforeFork
Gaudi::Property< bool >	m_debug
Gaudi::Property< bool >	m_useSharedWriter
int	m_rankId {-1}
ServiceHandle< IChronoStatSvc >	m_chronoStatSvc
SmartIF< IDataShare >	m_dataShare
SmartIF< IEvtSelectorSeek >	m_evtSelSeek
IEvtSelector::Context *	m_evtContext {}
AthenaInterprocess::SharedQueue *	m_sharedEventQueue {}
std::unique_ptr< AthenaInterprocess::SharedQueue >	m_sharedRankQueue
std::map< pid_t, int >	m_nProcessedEvents
std::queue< pid_t >	m_finQueue
SmartIF< IScheduler >	m_schedulerSvc

Detailed Description

Definition at line 22 of file SharedHiveEvtQueueConsumer.h.

Member Enumeration Documentation

ESRange_Status

enum AthenaMPToolBase::ESRange_Status

protectedinherited

Enumerator
ESRANGE_SUCCESS
ESRANGE_NOTFOUND
ESRANGE_SEEKFAILED
ESRANGE_PROCFAILED
ESRANGE_FILENOTMADE
ESRANGE_BADINPFILE

Definition at line 58 of file AthenaMPToolBase.h.

                      {
    ESRANGE_SUCCESS
    , ESRANGE_NOTFOUND
    , ESRANGE_SEEKFAILED
    , ESRANGE_PROCFAILED
    , ESRANGE_FILENOTMADE
    , ESRANGE_BADINPFILE
  };

Func_Flag

enum AthenaMPToolBase::Func_Flag

protectedinherited

Enumerator
FUNC_BOOTSTRAP
FUNC_EXEC
FUNC_FIN

Definition at line 67 of file AthenaMPToolBase.h.

                 {
    FUNC_BOOTSTRAP
    , FUNC_EXEC
    , FUNC_FIN
  };

Constructor & Destructor Documentation

SharedHiveEvtQueueConsumer()

SharedHiveEvtQueueConsumer::SharedHiveEvtQueueConsumer	(	const std::string &	type,
		const std::string &	name,
		const IInterface *	parent )

Definition at line 42 of file SharedHiveEvtQueueConsumer.cxx.

  : AthenaMPToolBase(type,name,parent)
  , m_chronoStatSvc("ChronoStatSvc", name)
{
  m_subprocDirPrefix = "worker_";
}

~SharedHiveEvtQueueConsumer()

SharedHiveEvtQueueConsumer::~SharedHiveEvtQueueConsumer ( )

overridevirtual

Definition at line 53 of file SharedHiveEvtQueueConsumer.cxx.

{
}

Member Function Documentation

ATLAS_NOT_THREAD_SAFE() [1/5]

int mapAsyncFlag AthenaMPToolBase::ATLAS_NOT_THREAD_SAFE ( Func_Flag flag, pid_t pid = 0 )

protectedinherited

ATLAS_NOT_THREAD_SAFE() [2/5]

virtual StatusCode exec SharedHiveEvtQueueConsumer::ATLAS_NOT_THREAD_SAFE ( )

overridevirtual

ATLAS_NOT_THREAD_SAFE() [3/5]

int decodeProcessResult SharedHiveEvtQueueConsumer::ATLAS_NOT_THREAD_SAFE ( const AthenaInterprocess::ProcessResult * presult, bool doFinalize )

private

ATLAS_NOT_THREAD_SAFE() [4/5]

virtual int makePool SharedHiveEvtQueueConsumer::ATLAS_NOT_THREAD_SAFE ( int maxevt, int nprocs, const std::string & topdir )

overridevirtual

ATLAS_NOT_THREAD_SAFE() [5/5]

virtual StatusCode wait_once SharedHiveEvtQueueConsumer::ATLAS_NOT_THREAD_SAFE ( pid_t & pid )

overridevirtual

Reimplemented from AthenaMPToolBase.

bootstrap_func()

std::unique_ptr< AthenaInterprocess::ScheduledWork > SharedHiveEvtQueueConsumer::bootstrap_func ( )

overridevirtual

Implements AthenaMPToolBase.

Definition at line 224 of file SharedHiveEvtQueueConsumer.cxx.

{
  if (m_debug) {
    ATH_MSG_INFO("Bootstrap worker PID " << getpid() << " - waiting for SIGUSR1");
    sigset_t mask, oldmask;
    
    signal(SIGUSR1, SharedHiveEvtQueueConsumer_d::pauseForDebug);
    
    sigemptyset (&mask);
    sigaddset (&mask, SIGUSR1);
    
    sigprocmask (SIG_BLOCK, &mask, &oldmask);
    while (!SharedHiveEvtQueueConsumer_d::sig_done)
      sigsuspend (&oldmask);
    sigprocmask (SIG_UNBLOCK, &mask, NULL);
  }
 
  std::unique_ptr<AthenaInterprocess::ScheduledWork> outwork(new AthenaInterprocess::ScheduledWork);
  outwork->data = CxxUtils::xmalloc(sizeof(int));
  *(int*)(outwork->data) = 1; // Error code: for now use 0 success, 1 failure
  outwork->size = sizeof(int);
 
  // ...
  // (possible) TODO: extend outwork with some error message, which will be eventually
  // reported in the master proces
  // ...
 
  // ________________________ Get IncidentSvc and fire PostFork ________________________
  SmartIF<IIncidentSvc> p_incidentSvc(serviceLocator()->service("IncidentSvc"));
  if (!p_incidentSvc) {
    ATH_MSG_ERROR("Unable to retrieve IncidentSvc");
    return outwork;
  }
  p_incidentSvc->fireIncident(Incident(name(),"PostFork"));
 
 
  // ________________________ Get RankID ________________________
  //
  if(!m_sharedRankQueue->receive_basic<int>(m_rankId)) {
    ATH_MSG_ERROR("Unable to get rank ID!");
    return outwork;
  }
  std::ostringstream workindex;
  workindex<<m_rankId;
 
  // ________________________ Worker dir: mkdir ________________________
  std::filesystem::path worker_rundir(m_subprocTopDir);
  worker_rundir /= std::filesystem::path(m_subprocDirPrefix+workindex.str());
  // TODO: this "worker_" can be made configurable too
 
  if(mkdir(worker_rundir.string().c_str(),S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)==-1) {
    ATH_MSG_ERROR("Unable to make worker run directory: " << worker_rundir.string() << ". " << fmterror(errno));
    return outwork;
  }
 
  // ________________________ Redirect logs ________________________
  if(redirectLog(worker_rundir.string()))
    return outwork;
 
  ATH_MSG_INFO("Logs redirected in the AthenaMP event worker PID=" << getpid());
 
  // ________________________ Update Io Registry ____________________________
  if(updateIoReg(worker_rundir.string()))
    return outwork;
 
  ATH_MSG_INFO("Io registry updated in the AthenaMP event worker PID=" << getpid());
 
  // ________________________ SimParams & DigiParams & PDGTABLE.MeV ____________________________
  std::filesystem::path abs_worker_rundir = std::filesystem::absolute(worker_rundir);
  if(std::filesystem::is_regular_file("SimParams.db"))
    COPY_FILE_HACK("SimParams.db", abs_worker_rundir.string()+"/SimParams.db");
  if(std::filesystem::is_regular_file("DigitParams.db"))
    COPY_FILE_HACK("DigitParams.db", abs_worker_rundir.string()+"/DigitParams.db");
  if(std::filesystem::is_regular_file("PDGTABLE.MeV"))
    COPY_FILE_HACK("PDGTABLE.MeV", abs_worker_rundir.string()+"/PDGTABLE.MeV");
 
  // _______________________ Handle saved PFC (if any) ______________________
  if(handleSavedPfc(abs_worker_rundir))
    return outwork;
 
  // ________________________  reopen descriptors ____________________________
  if(reopenFds())
    return outwork;
 
  ATH_MSG_INFO("File descriptors re-opened in the AthenaMP event worker PID=" << getpid());
  
  // ________________________ Make Shared Writer Client ________________________
 
  if(m_useSharedWriter && m_dataShare) {
    SmartIF<IProperty> propertyServer(m_dataShare);
    if (!propertyServer || propertyServer->setProperty("MakeStreamingToolClient", m_rankId + 1).isFailure()) {
      ATH_MSG_ERROR("Could not change AthenaPoolSharedIOCnvSvc MakeClient Property");
      return outwork;
    } else {
      ATH_MSG_DEBUG("Successfully made the conversion service a share client");
    }
  }
 
  // ________________________ I/O reinit ________________________
  ATH_CHECK( m_ioMgr->io_reinitialize(), outwork );
 
  // ________________________ Event selector restart ________________________
  SmartIF<IService> evtSelSvc(m_evtSelector);
  ATH_CHECK( evtSelSvc.isValid(), outwork );
  ATH_CHECK( evtSelSvc->start(), outwork );
 
  // ________________________ Worker dir: chdir ________________________
  if(chdir(worker_rundir.string().c_str())==-1) {
    ATH_MSG_ERROR("Failed to chdir to " << worker_rundir.string());
    return outwork;
  }
 
  // ___________________ Fire UpdateAfterFork incident _________________
  p_incidentSvc->fireIncident(AthenaInterprocess::UpdateAfterFork(m_rankId,getpid(),name()));
 
  // Declare success and return
  *(int*)(outwork->data) = 0;
  return outwork;
}

evtSelector()

IEvtSelector * AthenaMPToolBase::evtSelector ( )

inlineprotectedinherited

Definition at line 83 of file AthenaMPToolBase.h.

{ return m_evtSelector; }

exec_func()

std::unique_ptr< AthenaInterprocess::ScheduledWork > SharedHiveEvtQueueConsumer::exec_func ( )

overridevirtual

Implements AthenaMPToolBase.

Definition at line 346 of file SharedHiveEvtQueueConsumer.cxx.

{
  ATH_MSG_INFO("Exec function in the AthenaMP worker PID=" << getpid());
 
  bool all_ok(true);
 
  if (!initHive().isSuccess()) {
    ATH_MSG_FATAL("unable to initialize Hive");
    all_ok = false;
  }
 
  // Get the value of SkipEvent
  int skipEvents(0);
  SmartIF<IProperty> propertyServer(m_evtSelector);
  if(propertyServer==0) {
    ATH_MSG_ERROR("Unable to cast event selector to IProperty");
    all_ok = false;
  }
  else {
    std::string propertyName("SkipEvents");
    IntegerProperty skipEventsProp(std::move(propertyName),skipEvents);
    if(propertyServer->getProperty(&skipEventsProp).isFailure()) {
      ATH_MSG_INFO("Event Selector does not have SkipEvents property");
    }
    else {
      skipEvents = skipEventsProp.value();
    }
  }
 
  IHybridProcessorHelper* hybridHelper = dynamic_cast<IHybridProcessorHelper*>(m_evtProcessor.get());
  if(!hybridHelper) {
    ATH_MSG_FATAL("Failed to acquire IHybridProcessorHelper interface");
    all_ok = false;
    return std::unique_ptr<AthenaInterprocess::ScheduledWork>();
  }
  // Reset the application return code.
  hybridHelper->resetAppReturnCode();
 
  int finishedEvts =0;
  int createdEvts =0;
  long intmask = pow(0x100,sizeof(int))-1; // Mask for decoding event number from the value posted to the queue
  long evtnumAndChunk(0);
//  unsigned evtCounter(0);
  int evtnum(0), chunkSize(1);
 
  ATH_MSG_INFO("Starting loop on events");
 
  StatusCode sc(StatusCode::SUCCESS);
 
  while(!m_sharedEventQueue->try_receive_basic<long>(evtnumAndChunk)) {
    ATH_MSG_DEBUG("Event queue is empty");
    usleep(1000);
  }
  bool loop_ended = (evtnumAndChunk<0);
  if(!loop_ended) {
    ATH_MSG_DEBUG("Received value from the queue 0x" << std::hex << evtnumAndChunk << std::dec);
    chunkSize = evtnumAndChunk >> (sizeof(int)*8);
    evtnum = evtnumAndChunk & intmask;
    ATH_MSG_INFO("Received from the queue: event num=" << evtnum << " chunk size=" << chunkSize);
    hybridHelper->setCurrentEventNum(++evtnum);
  }
 
  bool no_more_events = false;
 
  while(!loop_ended) {
    ATH_MSG_DEBUG(" -> createdEvts: " << createdEvts);
 
    if(!hybridHelper->terminateLoop()         // No scheduled loop termination
       && !no_more_events                     // We are not yet done getting events
       && m_schedulerSvc->freeSlots()>0) {    // There are still free slots in the scheduler
      ATH_MSG_DEBUG("createdEvts: " << createdEvts << ", freeslots: " << m_schedulerSvc->freeSlots());
 
      auto ctx = m_evtProcessor->createEventContext();
      if(!ctx.valid()) {
        sc = StatusCode::FAILURE;
      }
      else {
        sc = m_evtProcessor->executeEvent(std::move(ctx));
      }
 
      if (sc.isFailure()) {
        ATH_MSG_ERROR("Terminating event processing loop due to errors");
        loop_ended = true;
      }
      else {
    ++createdEvts;
    if(--chunkSize==0) {
      // Fetch next chunk
      while(!m_sharedEventQueue->try_receive_basic<long>(evtnumAndChunk)) {
        ATH_MSG_DEBUG("Event queue is empty");
        usleep(1000);
      }
      if(evtnumAndChunk<0) {
        no_more_events = true;
        evtnumAndChunk *= -1;
        ATH_MSG_DEBUG("No more events are expected. The total number of events for this job = " << evtnumAndChunk);
      }
      else {
        ATH_MSG_DEBUG("Received value from the queue 0x" << std::hex << evtnumAndChunk << std::dec);
        chunkSize = evtnumAndChunk >> (sizeof(int)*8);
        evtnum = evtnumAndChunk & intmask;
        ATH_MSG_INFO("Received from the queue: event num=" << evtnum << " chunk size=" << chunkSize);
      }
    }
    // Advance to the next event
    if(!no_more_events) {
      hybridHelper->setCurrentEventNum(++evtnum);
    }
      }
    }
    else {
      // all the events were created but not all finished or the slots were
      // all busy: the scheduler should finish its job
      ATH_MSG_DEBUG("Draining the scheduler");
 
      // Pull out of the scheduler the finished events
      int ir = hybridHelper->drainScheduler(finishedEvts,true);
      if(ir < 0) {
        // some sort of error draining scheduler;
        loop_ended = true;
        sc = StatusCode::FAILURE;
      }
      else if(ir == 0) {
    // no more events in scheduler
        if(no_more_events) {
          // We are done
          loop_ended = true;
          sc = StatusCode::SUCCESS;
        }
      }
      else {
        // keep going!
      }
    }
  } // end main loop on finished events
 
  if(all_ok) {
    if(m_evtProcessor->executeRun(0).isFailure()) {
      ATH_MSG_ERROR("Could not finalize the Run");
      all_ok=false;
    } else {
      if(m_evtSelSeek->seek(*m_evtContext, evtnumAndChunk+skipEvents).isFailure()) { 
    ATH_MSG_DEBUG("Seek past maxevt to " << evtnumAndChunk+skipEvents << " returned failure. As expected...");
      }
    }
  }
 
  std::unique_ptr<AthenaInterprocess::ScheduledWork> outwork(new AthenaInterprocess::ScheduledWork);
 
  // Return value: "ERRCODE|Func_Flag|NEvt"
  int outsize = 2*sizeof(int)+sizeof(AthenaMPToolBase::Func_Flag);
  void* outdata = CxxUtils::xmalloc(outsize);
  *(int*)(outdata) = (all_ok?0:1); // Error code: for now use 0 success, 1 failure
  AthenaMPToolBase::Func_Flag func = AthenaMPToolBase::FUNC_EXEC;
  memcpy((char*)outdata+sizeof(int),&func,sizeof(func));
  memcpy((char*)outdata+sizeof(int)+sizeof(func),&createdEvts,sizeof(int));
 
  outwork->data = outdata;
  outwork->size = outsize;
  // ...
  // (possible) TODO: extend outwork with some error message, which will be eventually
  // reported in the master proces
  // ...
  return outwork;
}

fin_func()

std::unique_ptr< AthenaInterprocess::ScheduledWork > SharedHiveEvtQueueConsumer::fin_func ( )

overridevirtual

Implements AthenaMPToolBase.

Definition at line 515 of file SharedHiveEvtQueueConsumer.cxx.

{
  ATH_MSG_INFO("Fin function in the AthenaMP worker PID=" << getpid());
 
  bool all_ok(true);
 
  if(m_appMgr->stop().isFailure()) {
    ATH_MSG_ERROR("Unable to stop AppMgr"); 
    all_ok=false;
  } else { 
    if(m_appMgr->finalize().isFailure()) {
      std::cerr << "Unable to finalize AppMgr" << std::endl;
      all_ok=false;
    }
  }
 
  std::unique_ptr<AthenaInterprocess::ScheduledWork> outwork(new AthenaInterprocess::ScheduledWork);
 
  // Return value: "ERRCODE|Func_Flag|NEvt"  (Here NEvt=-1)
  int outsize = 2*sizeof(int)+sizeof(AthenaMPToolBase::Func_Flag);
  void* outdata = CxxUtils::xmalloc(outsize);
  *(int*)(outdata) = (all_ok?0:1); // Error code: for now use 0 success, 1 failure
  AthenaMPToolBase::Func_Flag func = AthenaMPToolBase::FUNC_FIN;
  memcpy((char*)outdata+sizeof(int),&func,sizeof(func));
  int nEvt = -1;
  memcpy((char*)outdata+sizeof(int)+sizeof(func),&nEvt,sizeof(int));
 
  outwork->data = outdata;
  outwork->size = outsize;
 
  return outwork;
}

finalize()

StatusCode SharedHiveEvtQueueConsumer::finalize ( )

overridevirtual

Reimplemented from AthenaMPToolBase.

Definition at line 85 of file SharedHiveEvtQueueConsumer.cxx.

{
  if (m_evtContext) {
    ATH_CHECK( evtSelector()->releaseContext (m_evtContext) );
    m_evtContext = nullptr;
  }
 
  return StatusCode::SUCCESS;
}

fmterror()

std::string AthenaMPToolBase::fmterror ( int errnum )

protectedinherited

Definition at line 333 of file AthenaMPToolBase.cxx.

{
  char buf[256];
  strerror_r(errnum, buf, sizeof(buf));
  return std::string(buf);
}

generateOutputReport()

AthenaMP::AllWorkerOutputs_ptr AthenaMPToolBase::generateOutputReport ( )

overridevirtualinherited

Reimplemented in EvtRangeProcessor, EvtRangeScatterer, and SharedWriterTool.

Definition at line 119 of file AthenaMPToolBase.cxx.

{
  AthenaMP::AllWorkerOutputs_ptr jobOutputs(new AthenaMP::AllWorkerOutputs());
 
  if(m_fileMgrLog.empty()) {
    ATH_MSG_WARNING(name() << " cannot make output report because FileMgr has not been configured to write log file!");
  }
  else {
    // Collect output files made by the workers
    std::string line;
    for(int i=0;i<m_nprocs;++i) {
      // Get the name of FileMgr log
      std::ostringstream workindex;
      workindex<<i;
      std::filesystem::path logFilePath(m_subprocTopDir);
      logFilePath /= std::filesystem::path(m_subprocDirPrefix+workindex.str());
      std::filesystem::path logFile(logFilePath);
      logFile /= std::filesystem::path(m_fileMgrLog);
      if(!(std::filesystem::exists(logFile)&&std::filesystem::is_regular_file(logFile))) {
        ATH_MSG_WARNING(logFile.string() << " either does not exist or is not a regular file. Skipping");
        continue;
      }
 
      ATH_MSG_DEBUG("FileMgr log file (" << i << ") " << logFile);
 
      std::ifstream inpStream(logFile.string().c_str());
      std::set<std::string> reportedFiles; // Don't report the same file twice
      while(!inpStream.eof()) {
    std::getline(inpStream,line);
        if(line.find("WRITE")!=std::string::npos) {
          // Parse the entry
          size_t startpos(0);
      std::vector<std::string> entries;
          while(startpos<line.size()) {
            while(line[startpos]==' ')
              startpos++;
 
            size_t endpos = line.find(' ',startpos);
            if(endpos==std::string::npos) endpos = line.size();
            entries.push_back(line.substr(startpos,endpos-startpos));
            startpos=endpos+1;
          }
 
          // enties[0] is filename
      std::filesystem::path filenamePath(entries[0]);
      std::filesystem::path basename = filenamePath.filename();
          if(reportedFiles.find(basename.string())==reportedFiles.end())
            reportedFiles.insert(basename.string());
          else
            continue;
      std::filesystem::path absolutename = basename.is_absolute() ? basename : std::filesystem::absolute(std::filesystem::path(logFilePath)/=basename);
      AthenaMP::AllWorkerOutputsIterator it1 = jobOutputs->find(basename.string());
          if(it1==jobOutputs->end()) {
            (*jobOutputs)[basename.string()] = AthenaMP::SingleWorkerOutputs();
            (*jobOutputs)[basename.string()].reserve(m_nprocs);
          }
 
      AthenaMP::WorkerOutput newOutput;
          newOutput.filename = absolutename.string();
          newOutput.technology = entries[1];
          newOutput.description = entries[2];
          newOutput.access_mode = entries[3];
          newOutput.shared = (line.find("SHARED")!=std::string::npos);
 
          (*jobOutputs)[basename.string()].emplace_back(std::move(newOutput));
        }
      }
    }
  }
  return jobOutputs;
}

handleSavedPfc()

int AthenaMPToolBase::handleSavedPfc ( const std::filesystem::path & dest_path )

protectedinherited

Definition at line 396 of file AthenaMPToolBase.cxx.

{
  if(std::filesystem::is_regular_file("PoolFileCatalog.xml.AthenaMP-saved"))
    COPY_FILE_HACK("PoolFileCatalog.xml.AthenaMP-saved",dest_path.string()+"/PoolFileCatalog.xml");
  return 0;
}

initHive()

StatusCode SharedHiveEvtQueueConsumer::initHive ( )

private

Definition at line 614 of file SharedHiveEvtQueueConsumer.cxx.

                                     {
 
    if (m_evtProcessor.release().isFailure()) {
      ATH_MSG_INFO("could not release old EventProcessor ");
    }
 
    ISvcManager* pISM(dynamic_cast<ISvcManager*>(serviceLocator().get()));
    if (pISM == 0) {
      ATH_MSG_ERROR("initHive: Could not get SvcManager");
    } else {
      if (pISM->removeService(m_evtProcessor.name()).isFailure()) {
    ATH_MSG_ERROR("initHive: could not remove " << m_evtProcessor.name() 
              << " from SvcManager");
      }
    }      
 
    m_evtProcessor = ServiceHandle<IEventProcessor>("AthenaMtesEventLoopMgr",name());
 
    if (m_evtProcessor.retrieve().isFailure()) {
      ATH_MSG_ERROR("could not setup " << m_evtProcessor.typeAndName());
      return StatusCode::FAILURE;
    }
 
    m_schedulerSvc = serviceLocator()->service("AvalancheSchedulerSvc");
 
  // m_whiteboard = serviceLocator()->service(m_whiteboardName);
  // if( !m_whiteboard.isValid() )  {
  //   ATH_MSG_FATAL( "Error retrieving " << m_whiteboardName 
  //           << " interface IHiveWhiteBoard." );
  //   return StatusCode::FAILURE;
  // }
  
  // m_schedulerSvc = serviceLocator()->service(m_schedulerName);
  // if ( !m_schedulerSvc.isValid()){
  //   ATH_MSG_FATAL( "Error retrieving SchedulerSvc interface ISchedulerSvc." );
  //   return StatusCode::FAILURE;    
  // }
  // // Setup algorithm resource pool
  // m_algResourcePool = serviceLocator()->service("AlgResourcePool");
  // if( !m_algResourcePool.isValid() ) {
  //   ATH_MSG_FATAL ("Error retrieving AlgResourcePool");
  //   return StatusCode::FAILURE;
  // }
 
  // sc = m_eventStore.retrieve();
  // if( !sc.isSuccess() )  {
  //   ATH_MSG_FATAL("Error retrieving pointer to StoreGateSvc");
  //   return sc;
  // }
 
 
  return StatusCode::SUCCESS;
 
}

initialize()

StatusCode SharedHiveEvtQueueConsumer::initialize ( )

overridevirtual

Reimplemented from AthenaMPToolBase.

Definition at line 59 of file SharedHiveEvtQueueConsumer.cxx.

{
  ATH_MSG_DEBUG("In initialize");
 
  ATH_CHECK( AthenaMPToolBase::initialize() );
 
  m_evtSelSeek = serviceLocator()->service(m_evtSelName);
  ATH_CHECK( m_evtSelSeek.isValid() );
  ATH_CHECK( evtSelector()->createContext (m_evtContext) );
 
  ATH_CHECK(m_chronoStatSvc.retrieve());
 
  if(m_useSharedWriter) {
    m_dataShare = SmartIF<IDataShare>(serviceLocator()->service("AthenaPoolSharedIOCnvSvc"));
    if(!m_dataShare) {
      ATH_MSG_ERROR("Error retrieving AthenaPoolSharedIOCnvSvc");
      return StatusCode::FAILURE;
    }
  }
 
  return StatusCode::SUCCESS;
}

killChildren()

void AthenaMPToolBase::killChildren ( )

overridevirtualinherited

Definition at line 201 of file AthenaMPToolBase.cxx.

{
  for(const AthenaInterprocess::Process& child : m_processGroup->getChildren()) {
    kill(child.getProcessID(),SIGKILL);
  }
}

operator()

virtual std::unique_ptr< ScheduledWork > AthenaInterprocess::IMessageDecoder::operator ( ) const &

pure virtualinherited

redirectLog()

int AthenaMPToolBase::redirectLog ( const std::string & rundir, bool addTimeStamp = true )

protectedinherited

Definition at line 269 of file AthenaMPToolBase.cxx.

{
  // Redirect both stdout and stderr to the same file AthenaMP.log
  int dup2result1(0), dup2result2(0);
 
  int newout = open(std::string(rundir+"/AthenaMP.log").c_str(),O_CREAT | O_RDWR, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH);
  if(newout==-1) {
    ATH_MSG_ERROR("Unable to open log file in the run directory. " << fmterror(errno));
    return -1;
  }
  dup2result1 = dup2(newout, STDOUT_FILENO);
  dup2result2 = dup2(newout, STDERR_FILENO);
  TEMP_FAILURE_RETRY(close(newout));
  if(dup2result1==-1) {
    ATH_MSG_ERROR("Unable to redirect standard output. " << fmterror(errno));
    return -1;
  }
  if(dup2result2==-1) {
    ATH_MSG_ERROR("Unable to redirect standard error. " << fmterror(errno));
    return -1;
  }
 
  if(addTimeStamp) {
    SmartIF<IProperty> propertyServer(msgSvc());
    if(propertyServer==0) {
      ATH_MSG_ERROR("Unable to cast message svc to IProperty");
      return -1;
    }
 
    std::string propertyName("Format");
    std::string oldFormat("");
    StringProperty formatProp(propertyName,oldFormat);
    StatusCode sc = propertyServer->getProperty(&formatProp);
    if(sc.isFailure()) {
      ATH_MSG_WARNING("Message Service does not have Format property");
    }
    else {
      oldFormat = formatProp.value();
      if(oldFormat.find("%t")==std::string::npos) {
        // Add time stamps
        std::string newFormat("%t " + oldFormat);
        StringProperty newFormatProp(std::move(propertyName),newFormat);
        ATH_CHECK(propertyServer->setProperty(newFormatProp), -1);
      }
      else {
        ATH_MSG_DEBUG("MsgSvc format already contains timestamps. Nothing to be done");
      }
    }
  }
 
  return 0;
}

reopenFd()

int AthenaMPToolBase::reopenFd ( int fd, const std::string & name )

privateinherited

Definition at line 419 of file AthenaMPToolBase.cxx.

{
  ATH_MSG_DEBUG("Attempting to reopen descriptor for " << name);
  int old_openflags = fcntl(fd,F_GETFL,0);
  switch(old_openflags & O_ACCMODE) {
  case O_RDONLY: {
    ATH_MSG_DEBUG("The File Access Mode is RDONLY");
    break;
  }
  case O_WRONLY: {
    ATH_MSG_DEBUG("The File Access Mode is WRONLY");
    break;
  }
  case O_RDWR: {
    ATH_MSG_DEBUG("The File Access Mode is RDWR");
    break;
  }
  }
   
  int old_descflags = fcntl(fd,F_GETFD,0);
  off_t oldpos = lseek(fd,0,SEEK_CUR);
  if(oldpos==-1) {
    if(errno==ESPIPE) {
      ATH_MSG_WARNING("Dealing with PIPE. Skipping ... (FIXME!)");
    }
    else {
      ATH_MSG_ERROR("When re-opening file descriptors lseek failed on " << name << ". " << fmterror(errno));
      return -1;
    }
  }
  else {
    Io::Fd newfd = open(name.c_str(),old_openflags);
    if(newfd==-1) {
      ATH_MSG_ERROR("When re-opening file descriptors unable to open " << name << " for reading. " << fmterror(errno));
      return -1;
    }
    if(lseek(newfd,oldpos,SEEK_SET)==-1){
      ATH_MSG_ERROR("When re-opening file descriptors lseek failed on the newly opened " << name << ". " << fmterror(errno));
      TEMP_FAILURE_RETRY(close(newfd));
      return -1;
    }
    TEMP_FAILURE_RETRY(close(fd));
    if(dup2(newfd,fd)==-1) {
      ATH_MSG_ERROR("When re-opening file descriptors unable to duplicate descriptor for " << name << ". " << fmterror(errno));
      TEMP_FAILURE_RETRY(close(newfd));
      return -1;
    }
    if(fcntl(fd,F_SETFD,old_descflags)==-1) {
      ATH_MSG_ERROR("When re-opening file descriptors unable to set descriptor flags for " << name << ". " << fmterror(errno));
      TEMP_FAILURE_RETRY(close(newfd));
      return -1;
    }
    TEMP_FAILURE_RETRY(close(newfd));
  }
  return 0;
}

reopenFds()

int AthenaMPToolBase::reopenFds ( )

protectedinherited

Definition at line 340 of file AthenaMPToolBase.cxx.

{
  // Reopen file descriptors.
  // First go over all open files, which have been registered with the FileMgr
  // Then also check the FdsRegistry, in case it contains some files not registered with the FileMgr
  std::set<int> fdLog;
  
  // Query the FileMgr contents
  std::vector<const Io::FileAttr*> filemgrFiles;
  std::vector<const Io::FileAttr*>::const_iterator itFile;
  unsigned filenum = m_fileMgr->getFiles(filemgrFiles); // Get attributes for open files only. We don't care about closed ones at this point
  if(filenum!=filemgrFiles.size())
    ATH_MSG_WARNING("getFiles returned " << filenum << " while vector size is " << filemgrFiles.size());
  
  for(itFile=filemgrFiles.begin();itFile!=filemgrFiles.end();++itFile) {
    ATH_MSG_DEBUG("* " << **itFile);
    const std::string& filename = (**itFile).name();
    Io::Fd fd = (**itFile).fd();
 
    if(fd==-1) {
      // It is legal to have fd=-1 for remote inputs
      // On the other hand, these inputs should not remain open after fork. The issue being tracked at ATEAM-434.
      // So, this hopefully is a temporary patch
      ATH_MSG_WARNING("FD=-1 detected on an open file retrieved from FileMgr. Skip FD reopening. File name: " << filename);
      continue;
    }
    
    if(reopenFd(fd,filename))
      return -1;
    
    fdLog.insert(fd);
  }
  
  // Check the FdsRegistry
  for(const AthenaInterprocess::FdsRegistryEntry& regEntry : *m_fdsRegistry) {
    if(fdLog.find(regEntry.fd)!=fdLog.end()) {
      ATH_MSG_DEBUG("The file from FdsRegistry " << regEntry.name << " was registered with FileMgr. Skip reopening");
    }
    else {
      ATH_MSG_WARNING("The file " << regEntry.name << " has not been registered with the FileMgr!");
      
      if(regEntry.fd==-1) {
      // Same protection as the one above
    ATH_MSG_WARNING("FD=-1 detected on an open file retrieved from FD Registry. Skip FD reopening. File name: " << regEntry.name);
    continue;
      }
 
      if(reopenFd(regEntry.fd,regEntry.name))
        return -1;
      
      fdLog.insert(regEntry.fd);
    }
  }
  return 0;
}

reportSubprocessStatuses()

void SharedHiveEvtQueueConsumer::reportSubprocessStatuses ( )

overridevirtual

Reimplemented from AthenaMPToolBase.

Definition at line 189 of file SharedHiveEvtQueueConsumer.cxx.

{
  ATH_MSG_INFO("Statuses of event processors");
  const std::vector<AthenaInterprocess::ProcessStatus>& statuses = m_processGroup->getStatuses();
  for(size_t i=0; i<statuses.size(); ++i) {
    // Get the number of events processed by this worker
    std::map<pid_t,int>::const_iterator it = m_nProcessedEvents.find(statuses[i].pid);
    msg(MSG::INFO) << "*** Process PID=" << statuses[i].pid 
           << ". Status " << ((statuses[i].exitcode)?"FAILURE":"SUCCESS") 
           << ". Number of events processed: ";
    if(it==m_nProcessedEvents.end())
      msg(MSG::INFO) << "N/A" << endmsg;
    else
      msg(MSG::INFO) << it->second << endmsg;
  }
}

setMaxEvt()

virtual void AthenaMPToolBase::setMaxEvt ( int maxEvt )

inlineoverridevirtualinherited

Definition at line 44 of file AthenaMPToolBase.h.

{m_maxEvt=maxEvt;}

setMPRunStop()

virtual void AthenaMPToolBase::setMPRunStop ( const AthenaInterprocess::IMPRunStop * runStop )

inlineoverridevirtualinherited

Definition at line 45 of file AthenaMPToolBase.h.

{m_mpRunStop=runStop;}

setRandString()

void AthenaMPToolBase::setRandString ( const std::string & randStr )

overridevirtualinherited

Definition at line 196 of file AthenaMPToolBase.cxx.

{
  m_randStr = randStr;
}

subProcessLogs()

void SharedHiveEvtQueueConsumer::subProcessLogs ( std::vector< std::string > & filenames )

overridevirtual

Definition at line 209 of file SharedHiveEvtQueueConsumer.cxx.

{
  filenames.clear();
  for(int i=0; i<m_nprocs; ++i) {
    std::ostringstream workerIndex;
    workerIndex << i;
    std::filesystem::path worker_rundir(m_subprocTopDir);
    worker_rundir /= std::filesystem::path(m_subprocDirPrefix+workerIndex.str());
    filenames.push_back(worker_rundir.string()+std::string("/AthenaMP.log"));
  }
}

updateIoReg()

int AthenaMPToolBase::updateIoReg ( const std::string & rundir )

protectedinherited

Definition at line 322 of file AthenaMPToolBase.cxx.

{
  ATH_CHECK(m_ioMgr.retrieve(), -1);
 
  // update the IoRegistry for the new workdir - make sure we use absolute path
  std::filesystem::path abs_rundir = std::filesystem::absolute(rundir);
  ATH_CHECK(m_ioMgr->io_update_all(abs_rundir.string()), -1);
 
  return 0;
}

useFdsRegistry()

void AthenaMPToolBase::useFdsRegistry ( std::shared_ptr< AthenaInterprocess::FdsRegistry > registry )

overridevirtualinherited

Definition at line 191 of file AthenaMPToolBase.cxx.

{
  m_fdsRegistry = std::move(registry);
}

waitForSignal()

void AthenaMPToolBase::waitForSignal ( )

protectedinherited

Definition at line 403 of file AthenaMPToolBase.cxx.

{
  ATH_MSG_INFO("Bootstrap worker PID " << getpid() << " - waiting for SIGUSR1");
  sigset_t mask, oldmask;
  
  signal(SIGUSR1, AthenaMPToolBase_d::pauseForDebug);
  
  sigemptyset (&mask);
  sigaddset (&mask, SIGUSR1);
  
  sigprocmask (SIG_BLOCK, &mask, &oldmask);
  while (!AthenaMPToolBase_d::sig_done)
    sigsuspend (&oldmask);
  sigprocmask (SIG_UNBLOCK, &mask, NULL);
}

Member Data Documentation

m_appMgr

ServiceHandle<IAppMgrUI> AthenaMPToolBase::m_appMgr

protectedinherited

Definition at line 95 of file AthenaMPToolBase.h.

m_chronoStatSvc

ServiceHandle<IChronoStatSvc> SharedHiveEvtQueueConsumer::m_chronoStatSvc

private

Definition at line 72 of file SharedHiveEvtQueueConsumer.h.

m_dataShare

SmartIF<IDataShare> SharedHiveEvtQueueConsumer::m_dataShare

private

Definition at line 73 of file SharedHiveEvtQueueConsumer.h.

m_debug

Gaudi::Property<bool> SharedHiveEvtQueueConsumer::m_debug

private

Initial value:

{
      this, "Debug", false,
      "Perform extra debugging if true. The default is false."}

Definition at line 61 of file SharedHiveEvtQueueConsumer.h.

                             {
      this, "Debug", false,
      "Perform extra debugging if true. The default is false."};

m_evtContext

IEvtSelector::Context* SharedHiveEvtQueueConsumer::m_evtContext {}

private

Definition at line 75 of file SharedHiveEvtQueueConsumer.h.

{};

m_evtProcessor

ServiceHandle<IEventProcessor> AthenaMPToolBase::m_evtProcessor

protectedinherited

Definition at line 94 of file AthenaMPToolBase.h.

m_evtSelector

SmartIF<IEvtSelector> AthenaMPToolBase::m_evtSelector

protectedinherited

Definition at line 98 of file AthenaMPToolBase.h.

m_evtSelName

std::string AthenaMPToolBase::m_evtSelName

protectedinherited

Name of the event selector.

Definition at line 89 of file AthenaMPToolBase.h.

m_evtSelSeek

SmartIF<IEvtSelectorSeek> SharedHiveEvtQueueConsumer::m_evtSelSeek

private

Definition at line 74 of file SharedHiveEvtQueueConsumer.h.

m_fdsRegistry

std::shared_ptr<AthenaInterprocess::FdsRegistry> AthenaMPToolBase::m_fdsRegistry

protectedinherited

Definition at line 100 of file AthenaMPToolBase.h.

m_fileMgr

ServiceHandle<IFileMgr> AthenaMPToolBase::m_fileMgr

protectedinherited

Definition at line 96 of file AthenaMPToolBase.h.

m_fileMgrLog

std::string AthenaMPToolBase::m_fileMgrLog

protectedinherited

Definition at line 99 of file AthenaMPToolBase.h.

m_finQueue

std::queue<pid_t> SharedHiveEvtQueueConsumer::m_finQueue

private

Definition at line 81 of file SharedHiveEvtQueueConsumer.h.

m_ioMgr

ServiceHandle<IIoComponentMgr> AthenaMPToolBase::m_ioMgr

protectedinherited

Definition at line 97 of file AthenaMPToolBase.h.

m_isPileup

Gaudi::Property<bool> AthenaMPToolBase::m_isPileup {this, "IsPileup", false, "Flag for configuring PileUpEventLoopMgr"}

protectedinherited

Definition at line 103 of file AthenaMPToolBase.h.

{this, "IsPileup", false, "Flag for configuring PileUpEventLoopMgr"};

m_maxEvt

int AthenaMPToolBase::m_maxEvt {-1}

protectedinherited

Maximum number of events assigned to the job.

Definition at line 86 of file AthenaMPToolBase.h.

{-1};       

m_mpRunStop

const AthenaInterprocess::IMPRunStop* AthenaMPToolBase::m_mpRunStop {nullptr}

protectedinherited

Definition at line 92 of file AthenaMPToolBase.h.

{nullptr};

m_nEventsBeforeFork

Gaudi::Property<int> SharedHiveEvtQueueConsumer::m_nEventsBeforeFork

private

Initial value:

{
      this, "EventsBeforeFork", 0,
      "The number of events before forking the workers. The default is 0."}

Definition at line 57 of file SharedHiveEvtQueueConsumer.h.

                                         {
      this, "EventsBeforeFork", 0,
      "The number of events before forking the workers. The default is 0."};

m_nProcessedEvents

std::map<pid_t,int> SharedHiveEvtQueueConsumer::m_nProcessedEvents

private

Definition at line 80 of file SharedHiveEvtQueueConsumer.h.

m_nprocs

int AthenaMPToolBase::m_nprocs {-1}

protectedinherited

Number of workers spawned by the master process.

Definition at line 85 of file AthenaMPToolBase.h.

{-1};       

m_processGroup

AthenaInterprocess::ProcessGroup* AthenaMPToolBase::m_processGroup {nullptr}

protectedinherited

Definition at line 91 of file AthenaMPToolBase.h.

{nullptr};

m_randStr

std::string AthenaMPToolBase::m_randStr

protectedinherited

Definition at line 101 of file AthenaMPToolBase.h.

m_rankId

int SharedHiveEvtQueueConsumer::m_rankId {-1}

private

Definition at line 70 of file SharedHiveEvtQueueConsumer.h.

{-1};          // Each worker has its own unique RankID from the range (0,...,m_nprocs-1)

m_schedulerSvc

SmartIF<IScheduler> SharedHiveEvtQueueConsumer::m_schedulerSvc

private

Definition at line 83 of file SharedHiveEvtQueueConsumer.h.

m_sharedEventQueue

AthenaInterprocess::SharedQueue* SharedHiveEvtQueueConsumer::m_sharedEventQueue {}

private

Definition at line 77 of file SharedHiveEvtQueueConsumer.h.

{};          

m_sharedRankQueue

std::unique_ptr<AthenaInterprocess::SharedQueue> SharedHiveEvtQueueConsumer::m_sharedRankQueue

private

Definition at line 78 of file SharedHiveEvtQueueConsumer.h.

m_subprocDirPrefix

std::string AthenaMPToolBase::m_subprocDirPrefix

protectedinherited

For ex. "worker__".

Definition at line 88 of file AthenaMPToolBase.h.

m_subprocTopDir

std::string AthenaMPToolBase::m_subprocTopDir

protectedinherited

Top run directory for subprocesses.

Definition at line 87 of file AthenaMPToolBase.h.

m_useSharedWriter

Gaudi::Property<bool> SharedHiveEvtQueueConsumer::m_useSharedWriter

private

Initial value:

{
      this, "UseSharedWriter", false,
      "Use SharedWriter to merge worker outputs on-the-fly if true. The default is false."}

Definition at line 65 of file SharedHiveEvtQueueConsumer.h.

                                       {
      this, "UseSharedWriter", false,
      "Use SharedWriter to merge worker outputs on-the-fly if true. The default is false."};

---

The documentation for this class was generated from the following files:

• SharedHiveEvtQueueConsumer.h
• SharedHiveEvtQueueConsumer.cxx