Fix race condition in HitMerger

shahor02 · sawenzel · commit f206cf5b216e · 2020-09-28T09:16:50.000+02:00
lock regions where mEventToTTreeMap and mEventToTMemFileMap are modified
diff --git a/run/O2HitMerger.h b/run/O2HitMerger.h
@@ -61,6 +61,7 @@
 #include <map>
 #include <vector>
 #include <csignal>
+#include <mutex>
 
 #ifdef ENABLE_UPGRADES
 #include <ITS3Simulation/Detector.h>
@@ -192,7 +193,7 @@ class O2HitMerger : public FairMQDevice
   void fillBranch(int eventID, std::string const& name, T* ptr)
   {
     // fetch tree into which to fill
-
+    const std::lock_guard<std::mutex> lock(mMapsMtx);
     auto iter = mEventToTTreeMap.find(eventID);
     if (iter == mEventToTTreeMap.end()) {
       {
@@ -312,11 +313,11 @@ class O2HitMerger : public FairMQDevice
   void cleanEvent(int eventID)
   {
     // remove tree for that eventID
+    const std::lock_guard<std::mutex> lock(mMapsMtx);
     delete mEventToTTreeMap[eventID];
-    mEventToTTreeMap.erase(eventID);
-    // remove memfile
     delete mEventToTMemFileMap[eventID];
-    mEventToTMemFileMap.erase(eventID);
+    mEventToTTreeMap.erase(eventID);
+    mEventToTMemFileMap.erase(eventID); // remove memfile
   }
 
   template <typename T>
@@ -648,7 +649,7 @@ class O2HitMerger : public FairMQDevice
   std::unordered_map<int, TTree*> mEventToTTreeMap;       //! in memory trees to collect / presort incoming data per event
   std::unordered_map<int, TMemFile*> mEventToTMemFileMap; //! files associated to the TTrees
   std::thread mMergerIOThread;                            //! a thread used to do hit merging and IO flushing asynchronously
-
+  std::mutex mMapsMtx;                                    //!
   int mEntries = 0;         //! counts the number of entries in the branches
   int mEventChecksum = 0;   //! checksum for events
   int mNExpectedEvents = 0; //! number of events that we expect to receive
