Merge pull request #32751 from ClickHouse/fix-32668

Apply some more optimizations to NO_QUERY ast.

src/Interpreters/SelectQueryOptions.h

@@ -41,6 +41,9 @@ struct SelectQueryOptions
     /// It is needed because lazy normal projections require special planning in FetchColumns stage, such as adding WHERE transform.
     /// It is also used to avoid adding aggregating step when aggregate projection is chosen.
     bool is_projection_query = false;
+    /// This flag is needed for projection description.
+    /// Otherwise, keys for GROUP BY may be removed as constants.
+    bool ignore_ast_optimizations = false;
     bool ignore_alias = false;
     bool is_internal = false;
     bool is_subquery = false; // non-subquery can also have subquery_depth > 0, e.g. insert select
@@ -120,6 +123,12 @@ struct SelectQueryOptions
         return *this;
     }
 
+    SelectQueryOptions & ignoreASTOptimizationsAlias(bool value = true)
+    {
+        ignore_ast_optimizations = value;
+        return *this;
+    }
+
     SelectQueryOptions & setInternal(bool value = false)
     {
         is_internal = value;

src/Interpreters/TreeRewriter.cpp

@@ -1123,7 +1123,7 @@ TreeRewriterResultPtr TreeRewriter::analyzeSelect(
     TreeOptimizer::optimizeIf(query, result.aliases, settings.optimize_if_chain_to_multiif);
 
     /// Only apply AST optimization for initial queries.
-    if (getContext()->getClientInfo().query_kind == ClientInfo::QueryKind::INITIAL_QUERY)
+    if (getContext()->getClientInfo().query_kind != ClientInfo::QueryKind::SECONDARY_QUERY && !select_options.ignore_ast_optimizations)
         TreeOptimizer::apply(query, result, tables_with_columns, getContext());
 
     /// array_join_alias_to_name, array_join_result_to_source.

src/Storages/ProjectionsDescription.cpp

@@ -201,7 +201,7 @@ ProjectionDescription ProjectionDescription::getMinMaxCountProjection(
     select_expression_list->children.push_back(makeASTFunction("count"));
     select_query->setExpression(ASTProjectionSelectQuery::Expression::SELECT, std::move(select_expression_list));
 
-    if (partition_columns)
+    if (partition_columns && !partition_columns->children.empty())
         select_query->setExpression(ASTProjectionSelectQuery::Expression::GROUP_BY, partition_columns->clone());
 
     result.definition_ast = select_query;
@@ -211,7 +211,9 @@ ProjectionDescription ProjectionDescription::getMinMaxCountProjection(
     auto external_storage_holder = std::make_shared<TemporaryTableHolder>(query_context, columns, ConstraintsDescription{});
     StoragePtr storage = external_storage_holder->getTable();
     InterpreterSelectQuery select(
-        result.query_ast, query_context, storage, {}, SelectQueryOptions{QueryProcessingStage::WithMergeableState}.modify().ignoreAlias());
+        result.query_ast, query_context, storage, {},
+        /// Here we ignore ast optimizations because otherwise aggregation keys may be removed from result header as constants.
+        SelectQueryOptions{QueryProcessingStage::WithMergeableState}.modify().ignoreAlias().ignoreASTOptimizationsAlias());
     result.required_columns = select.getRequiredColumns();
     result.sample_block = select.getSampleBlock();
 

tests/integration/test_storage_rabbitmq/test.py

@@ -284,6 +284,12 @@ def test_rabbitmq_materialized_view(rabbitmq_cluster):
             ORDER BY key;
         CREATE MATERIALIZED VIEW test.consumer TO test.view AS
             SELECT * FROM test.rabbitmq;
+
+        CREATE TABLE test.view2 (key UInt64, value UInt64)
+            ENGINE = MergeTree()
+            ORDER BY key;
+        CREATE MATERIALIZED VIEW test.consumer2 TO test.view2 AS
+            SELECT * FROM test.rabbitmq group by (key, value);
     ''')
 
     credentials = pika.PlainCredentials('root', 'clickhouse')
@@ -297,14 +303,26 @@ def test_rabbitmq_materialized_view(rabbitmq_cluster):
     for message in messages:
         channel.basic_publish(exchange='mv', routing_key='', body=message)
 
-    while True:
+    time_limit_sec = 60
+    deadline = time.monotonic() + time_limit_sec
+
+    while time.monotonic() < deadline:
         result = instance.query('SELECT * FROM test.view ORDER BY key')
         if (rabbitmq_check_result(result)):
             break
 
-    connection.close()
     rabbitmq_check_result(result, True)
 
+    deadline = time.monotonic() + time_limit_sec
+
+    while time.monotonic() < deadline:
+        result = instance.query('SELECT * FROM test.view2 ORDER BY key')
+        if (rabbitmq_check_result(result)):
+            break
+
+    rabbitmq_check_result(result, True)
+    connection.close()
+
 
 def test_rabbitmq_materialized_view_with_subquery(rabbitmq_cluster):
     instance.query('''