Merge pull request ClickHouse#97990 from ClickHouse/revert-97498-improve-join-side-flipping-rules

Revert "Support swapping sides of ANTI, SEMI and FULL joins"

Author: tavplubix

src/Processors/QueryPlan/Optimizations/joinOrder.cpp

@@ -550,9 +550,6 @@ std::optional<JoinKind> JoinOrderOptimizer::isValidJoinOrder(const BitSet & left
         return right_join_type;
     if (right_join_type == JoinKind::Inner)
         return left_join_type;
-    /// Allow FULL join as it's restricted to table swapping and no reordering
-    if (left_join_type == JoinKind::Full && right_join_type == JoinKind::Full)
-        return JoinKind::Full;
 
     /// Conflict, join is not possible:
     /// FROM t1 LEFT JOIN t2 LEFT JOIN t3

src/Processors/QueryPlan/Optimizations/optimizeJoin.cpp

@@ -811,16 +811,6 @@ static const ActionsDAG::Node * trackInputColumn(const ActionsDAG::Node * node)
     return node;
 }
 
-constexpr bool isSwapOnlyJoinKind(JoinKind kind)
-{
-    return kind == JoinKind::Full;
-}
-
-constexpr bool isSwapOnlyJoinStrictness(JoinStrictness strictness)
-{
-    return strictness == JoinStrictness::Any || strictness == JoinStrictness::Semi || strictness == JoinStrictness::Anti;
-}
-
 QueryPlan::Node chooseJoinOrder(QueryGraphBuilder query_graph_builder, QueryPlan::Nodes & nodes, JoinStrictness join_strictness)
 {
     QueryGraph query_graph;
@@ -961,17 +951,17 @@ QueryPlan::Node chooseJoinOrder(QueryGraphBuilder query_graph_builder, QueryPlan
             bool flip_join = has_prepared_storage_at_left || (!has_prepared_storage_at_right && swap_on_sizes);
 
             /// fixme: USING clause handled specially in join algorithm, so swap breaks it
-            /// At the time of writing, we're not able to swap inputs for ANY or SEMI partial merge join, because it only supports inner or left joins, but not right
-            /// ANTI partial merge join is not supported for any join kind
+            /// fixme: Swapping for SEMI and ANTI joins should be alright, need to try to enable it and test
+            /// At the time of writing, we're not able to swap inputs for ANY partial merge join, because it only supports ANY inner or left joins, but not right.
             const bool partial_merge_join_can_be_selected = std::ranges::any_of(
                 join_settings.join_algorithms,
                 [](JoinAlgorithm alg)
                 { return alg == JoinAlgorithm::PARTIAL_MERGE || alg == JoinAlgorithm::PREFER_PARTIAL_MERGE || alg == JoinAlgorithm::AUTO; });
             const bool should_worry_about_partial_merge_join = partial_merge_join_can_be_selected
                 && (!MergeJoin::isSupported(join_operator.kind, join_operator.strictness)
                     || !MergeJoin::isSupported(reverseJoinKind(join_operator.kind), join_operator.strictness));
-            const bool suitable_swap_only_join = isSwapOnlyJoinStrictness(join_operator.strictness) && !should_worry_about_partial_merge_join;
-            if (join_operator.strictness != JoinStrictness::All && !suitable_swap_only_join)
+            const bool suitable_any_join = join_operator.strictness == JoinStrictness::Any && !should_worry_about_partial_merge_join;
+            if (join_operator.strictness != JoinStrictness::All && !suitable_any_join)
                 flip_join = false;
 
             if (flip_join)
@@ -1182,9 +1172,10 @@ void optimizeJoinLogicalImpl(JoinStepLogical * join_step, QueryPlan::Node & node
     auto kind = join_operator.kind;
     auto locality = join_operator.locality;
     if (!optimization_settings.query_plan_optimize_join_order_limit
-        || (strictness != JoinStrictness::All && !isSwapOnlyJoinStrictness(strictness))
+        || (strictness != JoinStrictness::All && strictness != JoinStrictness::Any)
         || locality != JoinLocality::Unspecified
         || kind == JoinKind::Paste
+        || kind == JoinKind::Full
         || !join_operator.residual_filter.empty()
     )
     {
@@ -1196,8 +1187,8 @@ void optimizeJoinLogicalImpl(JoinStepLogical * join_step, QueryPlan::Node & node
     query_graph_builder.context->dummy_stats = join_step->getDummyStats();
 
     int query_graph_size_limit = safe_cast<int>(optimization_settings.query_plan_optimize_join_order_limit);
-    if ((isSwapOnlyJoinStrictness(strictness) || isSwapOnlyJoinKind(kind)) && query_graph_size_limit > 2)
-        /// Do not reorder joins, only allow swap
+    if (strictness == JoinStrictness::Any)
+        /// Do not reorder ANY joins, only allow swap
         query_graph_size_limit = 2;
 
     buildQueryGraph(query_graph_builder, node, nodes, query_graph_size_limit);

tests/queries/0_stateless/03623_convert_any_join_to_semi_or_anti_2.reference

@@ -67,8 +67,8 @@ Positions: 0
                FUNCTION greater(__table3.age :: 0, 1_UInt8 :: 1) -> greater(__table3.age, 1_UInt8) UInt8 : 2
       Positions: 2
         JoinLogical
-        Join: u1[100000] ⋉ u2[1000]
-        ResultRows: 100000
+        Join: u1 ⋉ u2
+        ResultRows: unknown
         Type: LEFT
         Strictness: SEMI
         Locality: UNSPECIFIED