From 3c29f27dd2a1cc8d4649860a9b5021f8784e467b Mon Sep 17 00:00:00 2001
From: Igor Nikonov <igor@clickhouse.com>
Date: Fri, 30 Aug 2024 20:04:43 +0000
Subject: [PATCH] Fix 03228_pr_subquery_view_order_by

---
 .../QueryPlan/Optimizations/liftUpUnion.cpp   |  1 +
 .../Optimizations/optimizeReadInOrder.cpp     | 23 +++++++++++++++++++
 .../03228_pr_subquery_view_order_by.sql       |  2 +-
 3 files changed, 25 insertions(+), 1 deletion(-)

diff --git a/src/Processors/QueryPlan/Optimizations/liftUpUnion.cpp b/src/Processors/QueryPlan/Optimizations/liftUpUnion.cpp
index c48551732c9..43cf166002e 100644
--- a/src/Processors/QueryPlan/Optimizations/liftUpUnion.cpp
+++ b/src/Processors/QueryPlan/Optimizations/liftUpUnion.cpp
@@ -50,6 +50,7 @@ size_t tryLiftUpUnion(QueryPlan::Node * parent_node, QueryPlan::Nodes & nodes)
             expr_node.step = std::make_unique<ExpressionStep>(
                 expr_node.children.front()->step->getOutputStream(),
                 expression->getExpression().clone());
+            expr_node.step->setStepDescription(expression->getStepDescription());
         }
 
         ///       - Expression - Something
diff --git a/src/Processors/QueryPlan/Optimizations/optimizeReadInOrder.cpp b/src/Processors/QueryPlan/Optimizations/optimizeReadInOrder.cpp
index 5df7d7b4e82..2eac1896066 100644
--- a/src/Processors/QueryPlan/Optimizations/optimizeReadInOrder.cpp
+++ b/src/Processors/QueryPlan/Optimizations/optimizeReadInOrder.cpp
@@ -18,6 +18,7 @@
 #include <Processors/QueryPlan/Optimizations/Optimizations.h>
 #include <Processors/QueryPlan/Optimizations/actionsDAGUtils.h>
 #include <Processors/QueryPlan/ReadFromMergeTree.h>
+#include <Processors/QueryPlan/ReadFromRemote.h>
 #include <Processors/QueryPlan/SortingStep.h>
 #include <Processors/QueryPlan/TotalsHavingStep.h>
 #include <Processors/QueryPlan/UnionStep.h>
@@ -899,6 +900,18 @@ AggregationInputOrder buildInputOrderInfo(AggregatingStep & aggregating, QueryPl
     return {};
 }
 
+static bool readingFromParallelReplicas(const QueryPlan::Node * node)
+{
+    IQueryPlanStep * step = node->step.get();
+    while (!node->children.empty())
+    {
+        step = node->children.front()->step.get();
+        node = node->children.front();
+    }
+
+    return typeid_cast<const ReadFromParallelRemoteReplicasStep *>(step);
+}
+
 void optimizeReadInOrder(QueryPlan::Node & node, QueryPlan::Nodes & nodes)
 {
     if (node.children.size() != 1)
@@ -924,6 +937,16 @@ void optimizeReadInOrder(QueryPlan::Node & node, QueryPlan::Nodes & nodes)
         std::vector<InputOrderInfoPtr> infos;
         infos.reserve(node.children.size());
 
+        for (const auto * child : union_node->children)
+        {
+            /// in case of parallel replicas
+            /// avoid applying read-in-order optimization for local replica
+            /// since it will lead to different parallel replicas modes
+            /// between local and remote nodes
+            if (readingFromParallelReplicas(child))
+                return;
+        }
+
         for (auto * child : union_node->children)
         {
             infos.push_back(buildInputOrderInfo(*sorting, *child, steps_to_update));
diff --git a/tests/queries/0_stateless/03228_pr_subquery_view_order_by.sql b/tests/queries/0_stateless/03228_pr_subquery_view_order_by.sql
index b85392e0521..804a97f737f 100644
--- a/tests/queries/0_stateless/03228_pr_subquery_view_order_by.sql
+++ b/tests/queries/0_stateless/03228_pr_subquery_view_order_by.sql
@@ -12,7 +12,7 @@ FROM
 )
 ORDER BY number DESC
 LIMIT 20
-SETTINGS cluster_for_parallel_replicas = 'parallel_replicas', allow_experimental_parallel_reading_from_replicas = 1, max_parallel_replicas = 3, parallel_replicas_for_non_replicated_merge_tree = 1, parallel_replicas_local_plan = 1, query_plan_lift_up_union = 0;
+SETTINGS cluster_for_parallel_replicas = 'parallel_replicas', allow_experimental_parallel_reading_from_replicas = 1, max_parallel_replicas = 3, parallel_replicas_for_non_replicated_merge_tree = 1, parallel_replicas_local_plan = 1;
 
 DROP TABLE view1;
 DROP TABLE table1;