From d1de121476c00a88a27cb83999d2e5be087e263d Mon Sep 17 00:00:00 2001
From: drrtuy <drrtuy@gmail.com>
Date: Tue, 25 Mar 2025 17:50:35 +0000
Subject: [PATCH] fix(): periodically return from UM JOIN loop that can produce
 a lot of RGData-s and thus overcommit for RAM

---
 dbcon/joblist/tuplehashjoin.cpp | 5 ++++-
 dbcon/joblist/tuplehashjoin.h   | 2 ++
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/dbcon/joblist/tuplehashjoin.cpp b/dbcon/joblist/tuplehashjoin.cpp
index cefddc8c7..9e7b11a30 100644
--- a/dbcon/joblist/tuplehashjoin.cpp
+++ b/dbcon/joblist/tuplehashjoin.cpp
@@ -1828,7 +1828,10 @@ void TupleHashJoinStep::generateJoinResultSet(const vector<vector<Row::Pointer>
   }
   else
   {
+    // NB In case of OUTER JOIN this loop can produce a lot of RGDatas,
+    // so it is a must to periodically flush from this loop.   
     l_outputRG.getRow(l_outputRG.getRowCount(), &joinedRow);
+    auto flushThreshold = outputDL->maxElements();
 
     for (i = 0; i < joinerOutput[depth].size(); i++, joinedRow.nextRow(), l_outputRG.incRowCount())
     {
@@ -1840,7 +1843,7 @@ void TupleHashJoinStep::generateJoinResultSet(const vector<vector<Row::Pointer>
         uint64_t baseRid = l_outputRG.getBaseRid();
         outputData.push_back(rgData);
         // Count the memory
-        if (UNLIKELY(!getMemory(l_outputRG.getSizeWithStrings())))
+        if (UNLIKELY(outputData.size() > flushThreshold || !getMemory(l_outputRG.getSizeWithStrings())))
         {
           // MCOL-5512
           if (fe2)
diff --git a/dbcon/joblist/tuplehashjoin.h b/dbcon/joblist/tuplehashjoin.h
index f65391e25..e588b8a39 100644
--- a/dbcon/joblist/tuplehashjoin.h
+++ b/dbcon/joblist/tuplehashjoin.h
@@ -529,6 +529,8 @@ class TupleHashJoinStep : public JobStep, public TupleDeliveryStep
     }
     void operator()()
     {
+      std::string name = "HJSJoinRun" + std::to_string(index);
+      utils::setThreadName(name.c_str());
       HJ->joinRunnerFcn(index);
     }
     TupleHashJoinStep* HJ;