fix(CEJ, segfault): MCOL-6198 - segfault during crossengine join

* fix(CEJ, segfault): MCOL-6198 - segfault during crossengine join

The patch moves joiners' initialization to a place after all possible
allocations of smallSideRGs vector so pointer to it's data does not
change anymore. This makes crash to cease.

An appropriate test is added to bugfixes suite.

* Change to test

* Another dangling pointer

* A change to test

* A change to test

mysql-test/columnstore/bugfixes/MCOL-6198-segfault-crossengine-join.result

@@ -0,0 +1,17 @@
+DROP DATABASE IF EXISTS MCOL_6198;
+CREATE DATABASE MCOL_6198;
+SELECT 
+v.c36 AS hst,
+m.p6 AS g,
+COALESCE(g0.eg_id,'MARIA') AS g01,
+COALESCE(g1.eg_id,'MARIA') AS g02,
+SUM(v.c758 * m.p42 / 100 + v.c759 * m.p42 / 100 + v.c760 * m.p42 / 100) AS sval
+FROM 
+c AS v 
+JOIN p m on (v.c4 = m.a4)
+LEFT OUTER JOIN group_g01 AS g0 ON g0.key_id=m.p6 
+LEFT OUTER JOIN group_g02 AS g1 ON g1.key_id=m.p6 
+WHERE 
+1=1 
+GROUP BY c36,p6,g01,g02;
+DROP DATABASE MCOL_6198;

primitives/primproc/batchprimitiveprocessor.cpp

@@ -72,6 +72,24 @@ using namespace logging;
 using namespace utils;
 using namespace joblist;
 
+#define idblog(x)                                                                       \
+  do                                                                                       \
+  {                                                                                        \
+    {                                                                                      \
+      std::ostringstream os;                                                               \
+                                                                                           \
+      os << __FILE__ << "@" << __LINE__ << ": \'" << x << "\'"; \
+      std::cerr << os.str() << std::endl;                                                  \
+      logging::MessageLog logger((logging::LoggingID()));                                  \
+      logging::Message message;                                                            \
+      logging::Message::Args args;                                                         \
+                                                                                           \
+      args.add(os.str());                                                                  \
+      message.format(args);                                                                \
+      logger.logErrorMessage(message);                                                     \
+    }                                                                                      \
+  } while (0)
+
 namespace primitiveprocessor
 {
 #ifdef PRIMPROC_STOPWATCH
@@ -386,15 +404,6 @@ void BatchPrimitiveProcessor::initBPP(ByteStream& bs)
             smallSideRGRecvd = true;
           }
 
-          for (uint j = 0; j < processorThreads; ++j)
-          {
-            auto tlHasher = TupleJoiner::TypelessDataHasher(&outputRG, &tlLargeSideKeyColumns[i],
-                                                            mSmallSideKeyColumnsPtr, mSmallSideRGPtr);
-            auto tlComparator = TupleJoiner::TypelessDataComparator(&outputRG, &tlLargeSideKeyColumns[i],
-                                                                    mSmallSideKeyColumnsPtr, mSmallSideRGPtr);
-            tlJoiners[i][j].reset(new TLJoiner(10, tlHasher, tlComparator,
-                                               utils::STLPoolAllocator<TLJoiner::value_type>(resourceManager)));
-          }
         }
       }
 
@@ -408,6 +417,7 @@ void BatchPrimitiveProcessor::initBPP(ByteStream& bs)
       {
         deserializeVector(bs, smallSideRGs);
         idbassert(smallSideRGs.size() == joinerCount);
+        mSmallSideRGPtr = mSmallSideRGPtr ? &smallSideRGs[0] : nullptr;
         smallSideRowLengths.reset(new uint32_t[joinerCount]);
         smallSideRowData.reset(new RGData[joinerCount]);
         smallNullRowData.reset(new RGData[joinerCount]);
@@ -438,6 +448,24 @@ void BatchPrimitiveProcessor::initBPP(ByteStream& bs)
         bs >> largeSideRG;
         bs >> joinedRG;
       }
+
+      for (i = 0; i < joinerCount; i++)
+      {
+        if (!typelessJoin[i])
+        {
+          continue;
+        }
+        for (uint j = 0; j < processorThreads; ++j)
+        {
+          auto tlHasher = TupleJoiner::TypelessDataHasher(&outputRG, &tlLargeSideKeyColumns[i],
+                                                          mSmallSideKeyColumnsPtr, mSmallSideRGPtr);
+          auto tlComparator = TupleJoiner::TypelessDataComparator(&outputRG, &tlLargeSideKeyColumns[i],
+                                                                  mSmallSideKeyColumnsPtr, mSmallSideRGPtr);
+          tlJoiners[i][j].reset(new TLJoiner(10, tlHasher, tlComparator,
+                                             utils::STLPoolAllocator<TLJoiner::value_type>(resourceManager)));
+        }
+      }
+
     }
 
     pthread_mutex_unlock(&objLock);
@@ -2412,8 +2440,7 @@ SBPP BatchPrimitiveProcessor::duplicate()
     bpp->hasJoinFEFilters = hasJoinFEFilters;
     bpp->hasSmallOuterJoin = hasSmallOuterJoin;
     bpp->mJOINHasSkewedKeyColumn = mJOINHasSkewedKeyColumn;
-    bpp->mSmallSideRGPtr = mSmallSideRGPtr;
+    bpp->mSmallSideKeyColumnsPtr = &(*bpp->tlSmallSideKeyColumns);
-    bpp->mSmallSideKeyColumnsPtr = mSmallSideKeyColumnsPtr;
     if (!getTupleJoinRowGroupData && mJOINHasSkewedKeyColumn)
     {
       idbassert(!smallSideRGs.empty());
@@ -2440,6 +2467,7 @@ SBPP BatchPrimitiveProcessor::duplicate()
       bpp->smallNullPointers = smallNullPointers;
       bpp->joinedRG = joinedRG;
     }
+    bpp->mSmallSideRGPtr = &bpp->smallSideRGs[0];
 
 #ifdef __FreeBSD__
     pthread_mutex_unlock(&bpp->objLock);