Improve exception handling and remove obsolete hltinterface header includes

Added try-catch blocks around hltinterface and eformat calls where missing and improved
the existing handling. Removed include of EventId, HLTResult and DataCollector headers
where not used anymore.


Former-commit-id: 197f35fc

HLT/Event/TrigByteStreamCnvSvc/src/TrigByteStreamCnvSvc.cxx

@@ -158,24 +158,25 @@ StatusCode TrigByteStreamCnvSvc::commitOutput(const std::string& /*outputFile*/,
   }
 
   // Send output to the DataCollector
+  StatusCode result = StatusCode::SUCCESS;
   try {
     hltinterface::DataCollector::instance()->eventDone(std::move(rawEventPtr));
+    ATH_MSG_DEBUG("Serialised FullEventFragment with HLT result was returned to DataCollector successfully");
   }
   catch (const std::exception& e) {
     ATH_MSG_ERROR("Sending output to DataCollector failed, caught an unexpected std::exception " << e.what());
-    return StatusCode::FAILURE;
+    result = StatusCode::FAILURE;
   }
   catch (...) {
     ATH_MSG_ERROR("Sending output to DataCollector failed, caught an unexpected exception");
-    return StatusCode::FAILURE;
+    result = StatusCode::FAILURE;
   }
-  ATH_MSG_DEBUG("Serialised FullEventFragment with HLT result was returned to DataCollector successfully");
 
   delete m_rawEventWrite;
   m_rawEventWrite = nullptr;
 
   ATH_MSG_VERBOSE("end of " << __FUNCTION__);
-  return StatusCode::SUCCESS;
+  return result;
 }
 
 // =============================================================================
@@ -216,7 +217,17 @@ void TrigByteStreamCnvSvc::printRawEvent() {
      << std::endl;
 
   std::vector<eformat::helper::StreamTag> stream_tags;
-  eformat::helper::decode(m_rawEventWrite->nstream_tag(), m_rawEventWrite->stream_tag(), stream_tags);
+  try {
+    eformat::helper::decode(m_rawEventWrite->nstream_tag(), m_rawEventWrite->stream_tag(), stream_tags);
+  }
+  catch (const std::exception& ex) {
+    ATH_MSG_ERROR("StreamTag decoding failed, caught an unexpected std::exception " << ex.what());
+    return;
+  }
+  catch (...) {
+    ATH_MSG_ERROR("StreamTag decoding failed, caught an unexpected exception");
+    return;
+  }
   ss << "--> stream_tags         = ";
   bool first = true;
   for (const auto& st : stream_tags) {

HLT/Event/TrigByteStreamCnvSvc/src/TrigByteStreamInputSvc.cxx

@@ -88,11 +88,23 @@ const RawEvent* TrigByteStreamInputSvc::nextEvent() {
 
   eformat::write::FullEventFragment l1r;
   using DCStatus = hltinterface::DataCollector::Status;
-  auto status = DCStatus::NO_EVENT;
+  DCStatus status = DCStatus::NO_EVENT;
   do {
-    status = hltinterface::DataCollector::instance()->getNext(l1r);
-    if (status == DCStatus::NO_EVENT)
-      ATH_MSG_ERROR("Failed to read new event, DataCollector::getNext returned Status::NO_EVENT. Trying again.");
+    try {
+      status = hltinterface::DataCollector::instance()->getNext(l1r);
+      if (status == DCStatus::NO_EVENT)
+        ATH_MSG_ERROR("Failed to read new event, DataCollector::getNext returned Status::NO_EVENT. Trying again.");
+    }
+    catch (const std::exception& ex) {
+      ATH_MSG_ERROR("Failed to read new event, caught an unexpected exception: " << ex.what()
+                    << ". Throwing hltonl::Exception::EventSourceCorrupted" );
+      throw hltonl::Exception::EventSourceCorrupted();
+    }
+    catch (...) {
+      ATH_MSG_ERROR("Failed to read new event, caught an unexpected exception. "
+                    << "Throwing hltonl::Exception::EventSourceCorrupted" );
+      throw hltonl::Exception::EventSourceCorrupted();
+    }
   } while (status == DCStatus::NO_EVENT);
 
   if (status == DCStatus::STOP) {

HLT/Event/TrigByteStreamCnvSvc/src/TrigByteStreamInputSvc.h

@@ -38,20 +38,19 @@ public:
   virtual const RawEvent* currentEvent() const override;
 
 private:
-  // ------------------------- service handles ---- ----------------------------
+  // ------------------------- Service handles ---------------------------------
   ServiceHandle<IROBDataProviderSvc> m_robDataProviderSvc;
   ServiceHandle<StoreGateSvc> m_evtStore;
 
-  // ------------------------- private data members ----------------------------
+  // ------------------------- Private data members ----------------------------
   struct EventCache {
     ~EventCache();
-    RawEvent* rawEvent = 0; //!< current event
-    // unsigned int eventStatus = 0; //!< status of the current event
+    RawEvent* rawEvent {nullptr}; //!< Current event
   };
 
-  SG::SlotSpecificObj<EventCache> m_eventsCache;
+  SG::SlotSpecificObj<EventCache> m_eventsCache; //!< Cache of RawEvent pointer per event slot
 
-  // ------------------------- private helper methods --------------------------
+  // ------------------------- Private helper methods --------------------------
   /// Clean up parts of previous event and re-init them
   void releaseEvent(EventCache* cache);
 };

HLT/Event/TrigByteStreamCnvSvc/src/TrigEventSelectorByteStream.cxx

@@ -14,9 +14,6 @@
 #include "EventInfo/EventInfo.h"
 #include "StoreGate/StoreGateSvc.h"
 
-// TDAQ includes
-#include "hltinterface/DataCollector.h"
-
 // =============================================================================
 // Standard constructor
 // =============================================================================
@@ -78,12 +75,16 @@ StatusCode TrigEventSelectorByteStream::next(IEvtSelector::Context& /*c*/) const
     ATH_MSG_INFO(e.what());
     throw; // rethrow NoMoreEvents
   }
+  catch (const hltonl::Exception::EventSourceCorrupted& e) {
+    ATH_MSG_ERROR(e.what());
+    throw; // rethrow EventSourceCorrupted
+  }
   catch (const std::exception& e) {
     ATH_MSG_ERROR("Failed to get next event from the event source, std::exception caught: " << e.what());
     return StatusCode::FAILURE;
   }
   catch (...) {
-    ATH_MSG_ERROR("Failed to get next event from the event source, unhandled exception caught");
+    ATH_MSG_ERROR("Failed to get next event from the event source, unknown exception caught");
     return StatusCode::FAILURE;
   }
 

HLT/Trigger/TrigControl/TrigKernel/CMakeLists.txt

@@ -11,7 +11,7 @@ atlas_depends_on_subdirs( PUBLIC
 
 # External dependencies:
 find_package( Boost COMPONENTS filesystem thread system )
-find_package( tdaq-common COMPONENTS eformat hltinterface )
+find_package( tdaq-common COMPONENTS eformat )
 
 # Component(s) in the package:
 atlas_add_library( TrigKernel

HLT/Trigger/TrigControl/TrigKernel/TrigKernel/HltExceptions.h

@@ -11,7 +11,7 @@ namespace hltonl {
   namespace Exception {
     /**
      * @class NoMoreEvents
-     * @brief Can be thrown if all events are already read from the input and another one is requested
+     * @brief Thrown if all events are already read from the input and another one is requested
      */
     class NoMoreEvents : public std::exception {
       public:
@@ -19,6 +19,16 @@ namespace hltonl {
         virtual ~NoMoreEvents() {}
         const char* what() const noexcept override {return "No more events to be processed";}
     };
+    /**
+     * @class EventSourceCorrupted
+     * @brief Thrown if event source keeps throwing exceptions when new event is requested
+     */
+    class EventSourceCorrupted : public std::exception {
+      public:
+        EventSourceCorrupted() {}
+        virtual ~EventSourceCorrupted() {}
+        const char* what() const noexcept override {return "Event source corrupted and cannot provide events";}
+    };
   }
 }
 

HLT/Trigger/TrigControl/TrigKernel/TrigKernel/ITrigEventLoopMgr.h

@@ -8,12 +8,6 @@
 #include "GaudiKernel/IInterface.h"
 #include <boost/property_tree/ptree.hpp>
 
-namespace hltinterface
-{
-  struct EventId;
-  struct HLTResult;
-}
-
 /**@class ITrigEventLoopMgr
  * @brief EventLoopMgr interface implemented by the HLT event loop manager
  *

HLT/Trigger/TrigControl/TrigPSC/TrigPSC/Psc.h

@@ -15,10 +15,6 @@
 
 // Package includes
 #include "hltinterface/HLTInterface.h"
-#include "hltinterface/EventId.h"
-
-// to be removed together with the Psc::process method
-#include "hltinterface/HLTResult.h"
 
 // Gaudi Includes
 #include "GaudiKernel/StatusCode.h"

HLT/Trigger/TrigControl/TrigServices/src/HltEventLoopMgr.cxx

@@ -36,7 +36,6 @@
 
 // TDAQ includes
 #include "eformat/StreamTag.h"
-#include "hltinterface/DataCollector.h"
 #include "owl/time.h"
 
 // System includes
@@ -696,7 +695,12 @@ StatusCode HltEventLoopMgr::nextEvent(int /*maxevt*/)
         }
         continue;
       }
+      catch (const std::exception& e) {
+        ATH_MSG_ERROR("Failed to get next event from the event source, std::exception caught: " << e.what());
+        sc = StatusCode::FAILURE;
+      }
       catch (...) {
+        ATH_MSG_ERROR("Failed to get next event from the event source, unknown exception caught");
         sc = StatusCode::FAILURE;
       }
       HLT_EVTLOOP_CHECK(sc, "Failed to get the next event",
@@ -1055,14 +1059,14 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << "Failure occurred with PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode)
       << " meaning there was a framework error before requesting a new event. No output will be produced and all slots"
       << " of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
   else if (errorCode==hltonl::PSCErrorCode::AFTER_RESULT_SENT) {
     ATH_REPORT_MESSAGE(MSG::ERROR)
       << "Failure occurred with PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode)
       << " meaning there was a framework error after HLT result was already sent out."
       << " All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
   else if (errorCode==hltonl::PSCErrorCode::CANNOT_ACCESS_SLOT) {
     ATH_REPORT_MESSAGE(MSG::ERROR)
@@ -1070,14 +1074,26 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << hltonl::PrintPscErrorCode(errorCode)
       << ". All slots of this HltEventLoopMgr instance will be drained before proceeding, then either the loop will"
       << " exit with a failure code or the failed event will reach a hard timeout.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
   else if (!eventContext.valid()) {
     ATH_REPORT_MESSAGE(MSG::ERROR)
       << "Failure occurred with an invalid EventContext. Likely there was a framework error before requesting a new"
       << " event or after sending the result of a finished event. PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode)
       << ". All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
+  }
+
+  //----------------------------------------------------------------------------
+  // In case of event source failure, drain the scheduler and break the loop
+  //----------------------------------------------------------------------------
+  if (errorCode==hltonl::PSCErrorCode::CANNOT_RETRIEVE_EVENT) {
+    ATH_REPORT_MESSAGE(MSG::ERROR)
+      << "Failure occurred with PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode)
+      << " meaning a new event could not be correctly read. No output will be produced for this event. All slots of"
+      << " this HltEventLoopMgr instance will be drained and the loop will exit.";
+    ATH_CHECK(drainAllSlots());
+    return StatusCode::FAILURE;
   }
 
   //----------------------------------------------------------------------------
@@ -1106,7 +1122,7 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << " event processing started or not. Current local event number is " << eventContext.evt()
       << ", slot " << eventContext.slot() << ". " << eventInfoString
       << " All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
 
   //----------------------------------------------------------------------------
@@ -1133,7 +1149,7 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << " PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode) << ", local event number " << eventContext.evt()
       << ", slot " << eventContext.slot() << ". " << eventInfoString
       << " All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
 
   //----------------------------------------------------------------------------
@@ -1153,7 +1169,7 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << " recorded for this event. PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode) << ", local event number "
       << eventContext.evt() << ", slot " << eventContext.slot() << ". " << eventInfoString
       << " All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
 
   DataObject* hltResultDO = m_evtStore->accessData(hltResultWH.clid(),hltResultWH.key());
@@ -1163,7 +1179,7 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << " can be recorded for this event. PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode)
       << ", local event number " << eventContext.evt() << ", slot " << eventContext.slot() << ". " << eventInfoString
       << " All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
 
   IOpaqueAddress* addr = nullptr;
@@ -1173,7 +1189,7 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << " can be recorded for this event. PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode)
       << ", local event number " << eventContext.evt() << ", slot " << eventContext.slot() << ". " << eventInfoString
       << " All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
 
   if (m_outputCnvSvc->commitOutput("",true).isFailure()) {
@@ -1182,7 +1198,7 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
       << " recorded for this event. PSCErrorCode=" << hltonl::PrintPscErrorCode(errorCode) << ", local event number "
       << eventContext.evt() << ", slot " << eventContext.slot() << ". " << eventInfoString
       << " All slots of this HltEventLoopMgr instance will be drained before proceeding.";
-      return drainAllAndProceed();
+    return drainAllAndProceed();
   }
 
   //----------------------------------------------------------------------------
@@ -1206,8 +1222,8 @@ StatusCode HltEventLoopMgr::failedEvent(hltonl::PSCErrorCode errorCode, const Ev
         << m_maxFrameworkErrors.value() << ", was exceeded. Current local event number is " << eventContextCopy.evt()
         << ", slot " << eventContextCopy.slot() << ". " << eventInfoString
         << " All slots of this HltEventLoopMgr instance will be drained and the loop will exit.";
-        ATH_CHECK(drainAllSlots());
-        return StatusCode::FAILURE;
+      ATH_CHECK(drainAllSlots());
+      return StatusCode::FAILURE;
     }
   }