[O2DPG,UTILS,O2-5725] Expand macros to find corrupted AO2Ds (#1913)

* [O2DPG,UTILS,O2-5725] Expand macros to find corrupted AO2Ds

- Added check for "repair" messages in the `checkCorruptedAO2Ds.C` macro to also catch files, where some problem appears and root tries to repair on the fly. This addition is made based on the recent report by Nicolas on the JIRA ticket O2-5725 that we see still some AO2D files with problems that can not be catched with the old version of `checkCorruptedAO2Ds.C`.
- Modified `findCorruptedAO2Ds.sh` to correctly read those cases and identify them as "broken" aswell.

* Update macros to find corrupted AO2Ds

* Added functionality to check for DFs which cause "repair" warning message

* Fix comment in script

Author: mhemmer-cern

UTILS/checkCorruptedAO2Ds.C

@@ -4,8 +4,21 @@
 #include <TTreeReader.h>
 #include <TGrid.h>
 #include <iostream>
+#include <TError.h>
+#include <cstring>
+
+bool gWarningDetected = false; // Global flag to track the warning
+
+void MyErrorHandler(int level, Bool_t abort, const char *location, const char *msg) {
+    if (strstr(msg, "repair") != nullptr) {
+        gWarningDetected = true;
+    }
+    DefaultErrorHandler(level, abort, location, msg); // Call ROOT’s default handler
+}
 
 int checkCorruptedAO2Ds(TString infileName = "/alice/sim/2024/LHC24h2/535545/AOD/005/AO2D.root", bool fromAlien = true) {
+    
+    SetErrorHandler(MyErrorHandler);
 
     if (fromAlien) {
         TGrid::Connect("alien://");
@@ -21,7 +34,7 @@ int checkCorruptedAO2Ds(TString infileName = "/alice/sim/2024/LHC24h2/535545/AOD
 
     // all VLA branches in the AO2Ds.root
     std::map<std::string, std::vector<std::string>> branchesToCheck = {
-        {"O2mcparticle_001", std::vector<std::string>{"fIndexArray_Mothers"}},
+        {"O2mcparticle_001", std::vector<std::string>{"fIndexArray_Mothers", "fVx", "fIndexMcCollisions"}},
         {"O2ft0", std::vector<std::string>{"fAmplitudeA", "fChannelA", "fAmplitudeC", "fChannelC"}},
         {"O2fv0a", std::vector<std::string>{"fAmplitude", "fChannel"}},
         {"O2mccalolabel_001", std::vector<std::string>{"fIndexArrayMcParticles", "fAmplitudeA"}},
@@ -42,11 +55,15 @@ int checkCorruptedAO2Ds(TString infileName = "/alice/sim/2024/LHC24h2/535545/AOD
                             std::cout << "Found corrupted file! DF: " << dirKey->GetName() << " Tree:" << pair.first.data() << " Branch:" << branchName.data() << std::endl;
                             return -1;
                         }
+                        if (gWarningDetected) {
+                            std::cout << "Found file in need of repair! DF: " << dirKey->GetName() << " Tree:" << pair.first.data() << " Branch:" << branchName.data() << std::endl;
+                            return -2;
+                        }
                     }
                 }
             }
         }
     }
 
     return 0;
-}
+}

UTILS/findCorruptedAO2Ds.sh

@@ -2,28 +2,29 @@
 
 # Simple script to find corrupted AO2Ds using the checkCorruptedAO2Ds.C macro
 
-PRODUCTION=LHC24h2
+PRODUCTION=LHC24f3c
 RUN=* # use * for all runs
-NJOBS=90
+NJOBS=20
+PRODUCTIONCYCLE=0
 
 OUTPUTFILE=corrupted_files_$PRODUCTION.txt
 if [ -e "$OUTPUTFILE" ]; then
   rm $OUTPUTFILE
 fi
 
 # find all files in alien
-if [ "$variable" == "*" ]; then
-  alien_find alien:///alice/sim/2024/${PRODUCTION} 5*/AOD/*/AO2D.root > files_to_check.txt
+if [ "$RUN" == "*" ]; then
+  alien_find alien:///alice/sim/2024/${PRODUCTION}/${PRODUCTIONCYCLE}/5*/AOD/*/AO2D.root > files_to_check.txt
 else
-  alien_find alien:///alice/sim/2024/${PRODUCTION} ${RUN}/AOD/*/AO2D.root > files_to_check.txt
+  alien_find alien:///alice/sim/2024/${PRODUCTION}/${PRODUCTIONCYCLE}/${RUN}/AOD/*/AO2D.root > files_to_check.txt
 fi
 mapfile -t FILESTOCHECK < files_to_check.txt
 
 # process AO2Ds
 process_file() {
   IFS='/' read -a num <<< "$1"
   INPUT=$1
-  echo '.x checkCorruptedAO2Ds.C("'${INPUT}'", true)' | root -l -b > log_${num[5]}_${num[7]}
+  echo '.x checkCorruptedAO2Ds.C("'${INPUT}'", true)' | root -l -b > log_${num[6]}_${num[8]}
   echo '.q'
 }
 export -f process_file
@@ -33,12 +34,15 @@ parallel -j $NJOBS process_file ::: "${FILESTOCHECK[@]}"
 # create list of corrupted files
 touch $OUTPUTFILE
 ERRORSTR="Found corrupted file!"
+REPAIRSTR="Found file in need of repair!"
 for FILE in "${FILESTOCHECK[@]}"; do
   IFS='/' read -a num <<< "$FILE"
-  if grep -q "$ERRORSTR" log_${num[5]}_${num[7]}; then
-    echo $FILE >> $OUTPUTFILE
+  if grep -q "$ERRORSTR" log_${num[6]}_${num[8]}; then
+    echo $FILE " is corrupted!" >> $OUTPUTFILE
+  elif grep -q "$REPAIRSTR" log_${num[6]}_${num[8]}; then
+    echo $FILE " is broken!" >> $OUTPUTFILE
   fi
 done
 
 rm files_to_check.txt
-rm log*
+rm log_*