evf::rb_statemachine::Stopping Class Reference

#include <RBStateMachine.h>

Inheritance diagram for evf::rb_statemachine::Stopping:

Public Types
typedef boost::mpl::list< SD >	reactions
typedef bsc::transition < StopDone, Stopped >	SD

Public Member Functions
virtual bool	discardDataEvent (MemRef_t *bufRef) const
virtual bool	discardDqmEvent (MemRef_t *bufRef) const
virtual void	do_stateAction () const
virtual void	do_stateNotify ()
virtual int	stateID () const
	Stopping (my_context)
virtual	~Stopping ()
Public Member Functions inherited from evf::rb_statemachine::BaseState
	BaseState ()
void	moveToFailedState (xcept::Exception &exception) const
std::string	stateName () const
virtual	~BaseState ()=0

Private Member Functions
bool	destructionIsDone () const
virtual void	do_entryActionWork ()
virtual void	do_exitActionWork ()
virtual void	do_moveToFailedState (xcept::Exception &exception) const
virtual std::string	do_stateName () const
void	emergencyStop () const

Additional Inherited Members
Protected Member Functions inherited from evf::rb_statemachine::BaseState
void	fail ()
void	safeEntryAction ()
void	safeExitAction ()

Detailed Description

Stopping state

Definition at line 598 of file RBStateMachine.h.

Member Typedef Documentation

typedef boost::mpl::list<SD> evf::rb_statemachine::Stopping::reactions

Definition at line 603 of file RBStateMachine.h.

typedef bsc::transition<StopDone, Stopped> evf::rb_statemachine::Stopping::SD

Definition at line 602 of file RBStateMachine.h.

Constructor & Destructor Documentation

Stopping::Stopping

(

my_context

c

)

Definition at line 132 of file Stopping.cc.

References evf::rb_statemachine::BaseState::safeEntryAction().

                               :
    my_base(c) {
    safeEntryAction();
}

Stopping::~Stopping ( )

virtual

Definition at line 137 of file Stopping.cc.

References evf::rb_statemachine::BaseState::safeExitAction().

                    {
    safeExitAction();
}

Member Function Documentation

bool evf::rb_statemachine::Stopping::destructionIsDone ( ) const

private

bool Stopping::discardDataEvent ( MemRef_t *  bufRef ) const

virtual

Base callback on process buffer received via I2O_SM_DATA_DISCARD message

Reimplemented from evf::rb_statemachine::BaseState.

Definition at line 108 of file Stopping.cc.

References alignCSCRings::e, and evf::rb_statemachine::BaseState::moveToFailedState().

                                                      {
    SharedResourcesPtr_t res = outermost_context().getSharedResources();
    bool returnValue = false;
    try {
        returnValue = res->resourceStructure_->discardDataEvent(bufRef);
    } catch (evf::Exception& e) {
        moveToFailedState(e);
    }
    return returnValue;
}

bool Stopping::discardDqmEvent ( MemRef_t *  bufRef ) const

virtual

Base callback on process buffer received via I2O_SM_DQM_DISCARD message

Reimplemented from evf::rb_statemachine::BaseState.

Definition at line 118 of file Stopping.cc.

References alignCSCRings::e, and evf::rb_statemachine::BaseState::moveToFailedState().

                                                     {
    SharedResourcesPtr_t res = outermost_context().getSharedResources();
    bool returnValue = false;
    try {
        returnValue = res->resourceStructure_->discardDqmEvent(bufRef);
        //returnValue = res->resourceStructure_->discardDqmEventWhileHalting(bufRef);
    } catch (evf::Exception& e) {
        moveToFailedState(e);
    }
    return returnValue;
}

void Stopping::do_entryActionWork ( )

privatevirtual

Implements evf::rb_statemachine::BaseState.

Definition at line 23 of file Stopping.cc.

                                  {
}

void Stopping::do_exitActionWork ( )

privatevirtual

Implements evf::rb_statemachine::BaseState.

Definition at line 203 of file Stopping.cc.

                                 {
}

void Stopping::do_moveToFailedState ( xcept::Exception &  exception ) const

privatevirtual

Implements evf::rb_statemachine::BaseState.

Definition at line 210 of file Stopping.cc.

References edm::hlt::Fail, and evf::rb_statemachine::BaseState::fail().

                                                                   {
    SharedResourcesPtr_t res = outermost_context().getSharedResources();
    res->reasonForFailed_ = exception.what();
    LOG4CPLUS_FATAL(res->log_,
            "Moving to FAILED state! Reason: " << exception.what());
    EventPtr fail(new Fail());
    res->commands_.enqEvent(fail);
}

void Stopping::do_stateAction ( ) const

virtual

State-dependent actions

EMERGENCY STOP IS TRIGGERED

Reimplemented from evf::rb_statemachine::BaseState.

Definition at line 36 of file Stopping.cc.

References prof2calltree::count, gather_cfg::cout, alignCSCRings::e, emergencyStop(), edm::hlt::Exception, evf::rb_statemachine::BaseState::moveToFailedState(), and cmsPerfSuiteHarvest::now.

                                    {
    SharedResourcesPtr_t res = outermost_context().getSharedResources();

    try {
        LOG4CPLUS_INFO(res->log_, "Start stopping :) ...");
        res->resourceStructure_->setStopFlag(true);
        res->resourceStructure_->shutDownClients();
        timeval now;
        timeval then;
        gettimeofday(&then, 0);
        while (!res->resourceStructure_->isReadyToShutDown()) {
            ::usleep(res->resourceStructureTimeout_.value_ * 10);
            gettimeofday(&now, 0);
            if ((unsigned int) (now.tv_sec - then.tv_sec)
                    > res->resourceStructureTimeout_.value_ / 10000) {
                cout << "times: " << now.tv_sec << " " << then.tv_sec << " "
                        << res->resourceStructureTimeout_.value_ / 10000
                        << endl;
                LOG4CPLUS_WARN(res->log_,
                        "Some Process did not detach - going to Emergency stop! resource status:" 
                        << res->resourceStructure_->printStatus() );

                //try to acquire RS lock
                int count=5;
                do {
                  if (!res->tryLockRSAccess()) break;
                  usleep(100000);
                } while (--count);
                if (!count) XCEPT_RAISE(evf::Exception,"Can not acquire RS lock for the emergency stop!");

                emergencyStop();

                res->unlockRSAccess();

                break;
            }
        }

        if (res->resourceStructure_->isReadyToShutDown()) {
            // lock access to I2O discards (data & dqm)
            res->lockRSAccess();

            // if emergency stop was not triggered
            if (res->allowI2ODiscards_) {
                // any I2O discards after this point are ignored
                res->allowI2ODiscards_ = false;
                // UPDATED: release resources
                res->resourceStructure_->releaseResources();
                // UPDATED: forget pending allocates to BU
                res->resourceStructure_->resetPendingAllocates();
                // UPDATE: reset the underlying IPC method
                res->resourceStructure_->resetIPC();
            }

            res->unlockRSAccess();

            LOG4CPLUS_INFO(res->log_, "Finished stopping!");
            EventPtr stopDone(new StopDone());
            res->commands_.enqEvent(stopDone);
        }
    } catch (xcept::Exception &e) {
        moveToFailedState(e);
    }
}

string Stopping::do_stateName ( ) const

privatevirtual

Implements evf::rb_statemachine::BaseState.

Definition at line 206 of file Stopping.cc.

                                    {
    return std::string("Stopping");
}

void Stopping::do_stateNotify ( )

virtual

State entry notifications

Implements evf::rb_statemachine::BaseState.

Definition at line 26 of file Stopping.cc.

References evf::rb_statemachine::BaseState::stateName().

                              {
    SharedResourcesPtr_t res = outermost_context().getSharedResources();
    LOG4CPLUS_INFO(res->log_, "--> ResourceBroker: NEW STATE: " << stateName());
    outermost_context().setExternallyVisibleState(stateName());
    outermost_context().setInternalStateName(stateName());
    // RCMS notification no longer required here
    // this is done in FUResourceBroker in SOAP reply
    //outermost_context().rcmsStateChangeNotify();
}

void Stopping::emergencyStop ( ) const

private

Definition at line 141 of file Stopping.cc.

References evf::IPCMethod::clientPrcIds(), gather_cfg::cout, evf::IPCMethod::handleCrashedEP(), i, evf::IPCMethod::isReadyToShutDown(), evf::IPCMethod::lastResort(), evf::utils::pid, evf::IPCMethod::shutdownStatus(), and stor::utils::sleep().

Referenced by do_stateAction().

                                   {
    SharedResourcesPtr_t res = outermost_context().getSharedResources();
    IPCMethod* resourceStructure = res->resourceStructure_;

    LOG4CPLUS_WARN(res->log_, "in Emergency stop - handle non-clean stops");

    // UPDATE: while in emergency stop I2O discards from SM are not allowed
    // they are re-allowed after a new enable
    res->allowI2ODiscards_ = false;
    {
                #ifdef linux
        auto lk = resourceStructure->lockCrashHandlerTimed(10);
                #else
                bool lk=true;
                #endif
        if (lk) { 
            vector < pid_t > client_prc_ids = resourceStructure->clientPrcIds();
            for (UInt_t i = 0; i < client_prc_ids.size(); i++) {
                pid_t pid = client_prc_ids[i];
                cout << "B: killing process " << i << " pid= " << pid << endl;
                if (pid != 0) {
                    //assume processes are dead by now
                    if (!resourceStructure->handleCrashedEP(res->runNumber_, pid))
                        res->nbTimeoutsWithoutEvent_++;
                    else
                        res->nbTimeoutsWithEvent_++;
                }
            }
        }
        else {
          XCEPT_RAISE(evf::Exception, 
            "Timed out accessing the EP Crash Handler in emergency stop. SM discards not arriving?");
        }
    }
    LOG4CPLUS_WARN(res->log_, "in Emergency stop - running lastResort");
    resourceStructure->lastResort();
    ::sleep(1);
    if (!resourceStructure->isReadyToShutDown()) {
        UInt_t shutdownStatus = resourceStructure->shutdownStatus();
        std::ostringstream ostr;
        ostr << "EmergencyStop: failed to shut down ResourceTable. Debug info mask:" << std::hex <<  shutdownStatus;
        res->reasonForFailed_ = ostr.str();
        XCEPT_RAISE(evf::Exception, res->reasonForFailed_);
    }

    res->printWorkLoopStatus();
    res->lock();

    LOG4CPLUS_WARN(res->log_, "Deleting the resource structure!");
    delete res->resourceStructure_;
    res->resourceStructure_ = 0;

    cout << "cycle through resourcetable config " << endl;
    res->configureResources(outermost_context().getApp());
    res->unlock();
    if (res->shmInconsistent_)
        XCEPT_RAISE(evf::Exception, "Inconsistent shm state");
    cout << "done with emergency stop" << endl;
}

virtual int evf::rb_statemachine::Stopping::stateID ( ) const

inlinevirtual

Return the current state ID

Implements evf::rb_statemachine::BaseState.

Definition at line 614 of file RBStateMachine.h.

References evf::rb_statemachine::STOPPING.

                                {
        return rb_statemachine::STOPPING;
    }