#include <AlarmHandler.h>

Public Types
enum	ALARM_LEVEL { OKAY, WARNING, ERROR, FATAL }
Public Member Functions
	AlarmHandler ()
	AlarmHandler (xdaq::Application *)
	AlarmHandler (xdaq::Application *, boost::shared_ptr< SharedResources >)
void	clearAllAlarms ()
Logger &	getLogger () const
void	localDebug (const std::string &message) const
virtual void	moveToFailedState (xcept::Exception &)
virtual void	notifySentinel (const ALARM_LEVEL, xcept::Exception &)
virtual void	raiseAlarm (const std::string name, const ALARM_LEVEL, xcept::Exception &)
virtual void	revokeAlarm (const std::string name)
virtual	~AlarmHandler ()
Private Member Functions
bool	raiseAlarm (const std::string name, const std::string level, xcept::Exception &)
Private Attributes
xdata::InfoSpace *	alarmInfoSpace_
xdaq::Application *	app_
boost::mutex	mutex_
boost::shared_ptr < SharedResources >	sharedResources_

Detailed Description

Helper class to handle sentinel alarming

Author:: mommsen

Revision:: 1.10

Date:: 2011/11/08 10:48:39

Definition at line 30 of file AlarmHandler.h.

Member Enumeration Documentation

enum stor::AlarmHandler::ALARM_LEVEL

Enumerator:

OKAY
WARNING
ERROR
FATAL

Definition at line 35 of file AlarmHandler.h.

{ OKAY, WARNING, ERROR, FATAL };

Constructor & Destructor Documentation

stor::AlarmHandler::AlarmHandler ( ) [inline]

Definition at line 38 of file AlarmHandler.h.

{};

stor::AlarmHandler::AlarmHandler ( xdaq::Application * app ) [explicit]

Definition at line 21 of file AlarmHandler.cc.

References Exception.

    :
  app_(app)
  {
    try
    {
      alarmInfoSpace_ = xdata::getInfoSpaceFactory()->get("urn:xdaq-sentinel:alarms");
    }
    catch(xdata::exception::Exception)
    {
      // sentinel is not available
      alarmInfoSpace_ = 0;
    }
  }

stor::AlarmHandler::AlarmHandler	(	xdaq::Application *	,
		boost::shared_ptr< SharedResources >
	)

virtual stor::AlarmHandler::~AlarmHandler ( ) [inline, virtual]

Definition at line 50 of file AlarmHandler.h.

{};

Member Function Documentation

void stor::AlarmHandler::clearAllAlarms ( )

Revokes all sentinel alarms

Definition at line 191 of file AlarmHandler.cc.

References alarmInfoSpace_, app_, and mutex_.

  {
    if (!alarmInfoSpace_) return;
    
    boost::mutex::scoped_lock sl( mutex_ );
    
    typedef std::map<std::string, xdata::Serializable*, std::less<std::string> > alarmList;
    alarmList alarms = alarmInfoSpace_->match(".*");
    for (alarmList::const_iterator it = alarms.begin(), itEnd = alarms.end();
         it != itEnd; ++it)
    {
      sentinel::utils::Alarm* alarm = dynamic_cast<sentinel::utils::Alarm*>(it->second);
      alarmInfoSpace_->fireItemRevoked(it->first, app_);
      delete alarm;
    }
  }

Logger& stor::AlarmHandler::getLogger ( ) const [inline]

Return the application logger

Definition at line 95 of file AlarmHandler.h.

References app_.

    { return app_->getApplicationLogger(); }

void stor::AlarmHandler::localDebug ( const std::string & message ) const

Write message to a file in /tmp (last resort when everything else fails)

Definition at line 247 of file AlarmHandler.cc.

References f, alignmentValidation::fname, dbtoconf::out, sharedResources_, and AlCaHLTBitMon_QueryRunRegistry::string.

Referenced by moveToFailedState().

  {
    std::ostringstream fname_oss;
    fname_oss << "/tmp/storage_manager_debug_" << 
      sharedResources_->configuration_->getDiskWritingParams().smInstanceString_ <<
      "_" << getpid();
    const std::string fname = fname_oss.str();
    std::ofstream f( fname.c_str(), std::ios_base::ate | std::ios_base::out | std::ios_base::app );
    if( f.is_open() )
    {
      try
      {
        f << message << std::endl;
        f.close();
      }
      catch(...)
      {}
    }
  }

void stor::AlarmHandler::moveToFailedState ( xcept::Exception & exception ) [virtual]

Add a Failed state-machine event to the command queue

Definition at line 209 of file AlarmHandler.cc.

References alignCSCRings::e, exception, Exception, Fail, FATAL, localDebug(), notifySentinel(), seconds(), sharedResources_, and AlCaHLTBitMon_QueryRunRegistry::string.

  {
    std::string errorMsg = "Failed to process FAIL exception: "
      + xcept::stdformat_exception_history(exception) + " due to ";

    try
    {
      notifySentinel(AlarmHandler::FATAL, exception);
      sharedResources_->statisticsReporter_->getStateMachineMonitorCollection().setStatusMessage( 
        xcept::stdformat_exception_history(exception)
      );
      EventPtr_t stMachEvent( new Fail() );
      // wait maximum 5 seconds until enqueuing succeeds
      if ( ! sharedResources_->commandQueue_->enqTimedWait( stMachEvent, boost::posix_time::seconds(5) ) )
      {
        XCEPT_DECLARE_NESTED( stor::exception::StateTransition,
          sentinelException, "Failed to enqueue FAIL event", exception );
        notifySentinel(AlarmHandler::FATAL, sentinelException);
      }
    }
    catch(xcept::Exception &e)
    {
      errorMsg += xcept::stdformat_exception_history(e);
      localDebug( errorMsg );
    }
    catch(std::exception &e)
    {
      errorMsg += e.what();
      localDebug( errorMsg );
    }
    catch( ... )
    {
      errorMsg += "an unknown exception.";
      localDebug( errorMsg );
    }
  }

void stor::AlarmHandler::notifySentinel	(	const ALARM_LEVEL	level,
		xcept::Exception &	exception
	)		`[virtual]`

Notifies the sentinel

Definition at line 98 of file AlarmHandler.cc.

References dqm::qstatus::ERROR, and dqm::qstatus::WARNING.

Referenced by moveToFailedState().

  {
    
    switch( level )
    {
      case OKAY:
        LOG4CPLUS_INFO(app_->getApplicationLogger(),
          xcept::stdformat_exception_history(exception));
        break;
        
      case WARNING:
        LOG4CPLUS_WARN(app_->getApplicationLogger(),
          xcept::stdformat_exception_history(exception));
        app_->notifyQualified("warning", exception);
        break;
        
      case ERROR:
        LOG4CPLUS_ERROR(app_->getApplicationLogger(),
          xcept::stdformat_exception_history(exception));
        app_->notifyQualified("error", exception);
        break;
        
      case FATAL:
        LOG4CPLUS_FATAL(app_->getApplicationLogger(),
          xcept::stdformat_exception_history(exception));
        app_->notifyQualified("fatal", exception);
        break;
        
      default:
        LOG4CPLUS_WARN(app_->getApplicationLogger(),
          "Unknown alarm level received for exception: " <<
          xcept::stdformat_exception_history(exception));
    }
  }

void stor::AlarmHandler::raiseAlarm	(	const std::string	name,
		const ALARM_LEVEL	level,
		xcept::Exception &	exception
	)		`[virtual]`

Raises a sentinel alarm

Definition at line 59 of file AlarmHandler.cc.

References dqm::qstatus::ERROR, and dqm::qstatus::WARNING.

  {
    
    switch( level )
    {
      case OKAY:
        revokeAlarm(name);
        break;
        
      case WARNING:
        if ( raiseAlarm(name, "warning", exception) )
          LOG4CPLUS_WARN(app_->getApplicationLogger(),
            "Raising warning alarm " << name << ": " << exception.message());
        break;
        
      case ERROR:
        if ( raiseAlarm(name, "error", exception) )
          LOG4CPLUS_ERROR(app_->getApplicationLogger(),
            "Raising error alarm " << name << ": " << exception.message());
        break;
        
      case FATAL:
        if ( raiseAlarm(name, "fatal", exception) )
          LOG4CPLUS_FATAL(app_->getApplicationLogger(),
            "Raising fatal alarm " << name << ": " << exception.message());
        break;
        
      default:
        LOG4CPLUS_WARN(app_->getApplicationLogger(),
          "Unknown alarm level received for " << name << ": " << exception.message());
    }
  }

bool stor::AlarmHandler::raiseAlarm	(	const std::string	name,
		const std::string	level,
		xcept::Exception &	exception
	)		`[private]`

Definition at line 138 of file AlarmHandler.cc.

References Exception.

  {
    
    if (!alarmInfoSpace_) return false;
    
    boost::mutex::scoped_lock sl( mutex_ );
    
    sentinel::utils::Alarm *alarm =
      new sentinel::utils::Alarm(level, exception, app_);
    try
    {
      alarmInfoSpace_->fireItemAvailable(name, alarm);
    }
    catch(xdata::exception::Exception)
    {
      // Alarm is already set or sentinel not available
      return false;
    }
    return true;
  }

void stor::AlarmHandler::revokeAlarm ( const std::string name ) [virtual]

Revokes a sentinel alarm

Definition at line 165 of file AlarmHandler.cc.

References Exception, and mergeVDriftHistosByStation::name.

  {
    if (!alarmInfoSpace_) return;
    
    boost::mutex::scoped_lock sl( mutex_ );
    
    sentinel::utils::Alarm *alarm;
    try
    {
      alarm = dynamic_cast<sentinel::utils::Alarm*>( alarmInfoSpace_->find( name ) );
    }
    catch(xdata::exception::Exception)
    {
      // Alarm has not been set or sentinel not available
      return;
    }
    
    LOG4CPLUS_INFO(app_->getApplicationLogger(), "Revoking alarm " << name);
    
    alarmInfoSpace_->fireItemRevoked(name, app_);
    delete alarm;
  }