disable by default

Author: liujiayi771

velox/core/QueryConfig.h

@@ -743,7 +743,7 @@ class QueryConfig {
   }
 
   int32_t abandonBuildNoDupHashMinPct() const {
-    return get<int32_t>(kAbandonBuildNoDupHashMinPct, 70);
+    return get<int32_t>(kAbandonBuildNoDupHashMinPct, 0);
   }
 
   int32_t maxElementsSizeInRepeatAndSequence() const {

velox/docs/configs.rst

@@ -46,11 +46,13 @@ Generic Configuration
    * - abandon_build_no_dup_hash_min_rows
      - integer
      - 100,000
-     - Number of input rows to receive before starting to check whether to abandon building a HashTable without duplicates in HashBuild for left semi/anti join.
+     - Number of input rows to receive before starting to check whether to abandon building a HashTable without
+       duplicates in HashBuild for left semi/anti join.
    * - abandon_build_no_dup_hash_min_pct
      - integer
-     - 70
-     - Abandons building a HashTable without duplicates in HashBuild for left semi/anti join if the percentage of distinct keys in the HashTable exceeds this threshold.
+     - 0
+     - Abandons building a HashTable without duplicates in HashBuild for left semi/anti join if the percentage of
+       distinct keys in the HashTable exceeds this threshold. Zero means 'disable this optimization'.
    * - session_timezone
      - string
      -

velox/exec/HashBuild.cpp

@@ -183,6 +183,12 @@ void HashBuild::setupTable() {
   }
   lookup_ = std::make_unique<HashLookup>(table_->hashers(), pool());
   analyzeKeys_ = table_->hashMode() != BaseHashTable::HashMode::kHash;
+  if (abandonBuildNoDupHashMinPct_ == 0) {
+    // Building a HashTable without duplicates is disabled if
+    // abandonBuildNoDupHashMinPct_ is 0.
+    abandonBuildNoDupHash_ = true;
+    table_->joinTableMayHaveDuplicates();
+  }
 }
 
 void HashBuild::setupSpiller(SpillPartition* spillPartition) {