QueryPlan for StorageBuffer and StorageMaterializedView read.

src/Processors/QueryPlan/AddingMissedStep.cpp

@@ -0,0 +1,45 @@
+#include <Processors/QueryPlan/AddingMissedStep.h>
+#include <Processors/QueryPipeline.h>
+#include <Processors/Transforms/AddingMissedTransform.h>
+#include <IO/Operators.h>
+
+namespace DB
+{
+
+static ITransformingStep::Traits getTraits()
+{
+    return ITransformingStep::Traits
+    {
+        {
+            .preserves_distinct_columns = true,
+            .returns_single_stream = false,
+            .preserves_number_of_streams = true,
+            .preserves_sorting = true,
+        },
+        {
+            .preserves_number_of_rows = true,
+        }
+    };
+}
+
+AddingMissedStep::AddingMissedStep(
+    const DataStream & input_stream_,
+    Block result_header_,
+    const ColumnDefaults & column_defaults_,
+    const Context & context_)
+    : ITransformingStep(input_stream_, result_header_, getTraits())
+    , column_defaults(column_defaults_)
+    , context(context_)
+{
+    updateDistinctColumns(output_stream->header, output_stream->distinct_columns);
+}
+
+void AddingMissedStep::transformPipeline(QueryPipeline & pipeline)
+{
+    pipeline.addSimpleTransform([&](const Block & header)
+    {
+        return std::make_shared<AddingMissedTransform>(header, output_stream->header, column_defaults, context);
+    });
+}
+
+}

src/Processors/QueryPlan/AddingMissedStep.h

@@ -0,0 +1,28 @@
+#pragma once
+#include <Processors/QueryPlan/ITransformingStep.h>
+
+namespace DB
+{
+
+struct ColumnDefault;
+using ColumnDefaults = std::unordered_map<std::string, ColumnDefault>;
+
+/// Convert one block structure to another. See ConvertingTransform.
+class AddingMissedStep : public ITransformingStep
+{
+public:
+    AddingMissedStep(const DataStream & input_stream_,
+                     Block result_header_,
+                     const ColumnDefaults & column_defaults_,
+                     const Context & context_);
+
+    String getName() const override { return "AddingMissed"; }
+
+    void transformPipeline(QueryPipeline & pipeline) override;
+
+private:
+    const ColumnDefaults column_defaults;
+    const Context & context;
+};
+
+}

src/Processors/QueryPlan/SettingQuotaAndLimitsStep.cpp

@@ -59,7 +59,10 @@ void SettingQuotaAndLimitsStep::transformPipeline(QueryPipeline & pipeline)
     if (quota)
         pipeline.setQuota(quota);
 
+    if (context)
         pipeline.addInterpreterContext(std::move(context));
+
+    if (storage)
         pipeline.addStorageHolder(std::move(storage));
 }
 

src/Processors/ya.make

@@ -89,6 +89,7 @@ SRCS(
     printPipeline.cpp
     QueryPipeline.cpp
     QueryPlan/AddingDelayedSourceStep.cpp
+    QueryPlan/AddingMissedStep.cpp
     QueryPlan/AggregatingStep.cpp
     QueryPlan/ArrayJoinStep.cpp
     QueryPlan/ConvertingStep.cpp

src/Storages/StorageBuffer.cpp

@@ -4,7 +4,7 @@
 #include <Interpreters/InterpreterAlterQuery.h>
 #include <Interpreters/castColumn.h>
 #include <Interpreters/evaluateConstantExpression.h>
-#include <Processors/Transforms/AddingMissedTransform.h>
+#include <Processors/QueryPlan/AddingMissedStep.h>
 #include <DataStreams/IBlockInputStream.h>
 #include <Storages/StorageBuffer.h>
 #include <Storages/StorageFactory.h>
@@ -22,10 +22,13 @@
 #include <common/logger_useful.h>
 #include <common/getThreadId.h>
 #include <ext/range.h>
-#include <Processors/Transforms/ConvertingTransform.h>
+#include <Processors/QueryPlan/ConvertingStep.h>
 #include <Processors/Transforms/FilterTransform.h>
 #include <Processors/Transforms/ExpressionTransform.h>
 #include <Processors/Sources/SourceFromInputStream.h>
+#include <Processors/QueryPlan/SettingQuotaAndLimitsStep.h>
+#include <Processors/QueryPlan/ReadFromPreparedSource.h>
+#include <Processors/QueryPlan/UnionStep.h>
 
 
 namespace ProfileEvents
@@ -147,6 +150,21 @@ QueryProcessingStage::Enum StorageBuffer::getQueryProcessingStage(const Context
 
 
 Pipe StorageBuffer::read(
+    const Names & column_names,
+    const StorageMetadataPtr & metadata_snapshot,
+    const SelectQueryInfo & query_info,
+    const Context & context,
+    QueryProcessingStage::Enum processed_stage,
+    const size_t max_block_size,
+    const unsigned num_streams)
+{
+    QueryPlan plan;
+    read(plan, column_names, metadata_snapshot, query_info, context, processed_stage, max_block_size, num_streams);
+    return QueryPipeline::getPipe(std::move(*plan.buildQueryPipeline()));
+}
+
+void StorageBuffer::read(
+    QueryPlan & query_plan,
     const Names & column_names,
     const StorageMetadataPtr & metadata_snapshot,
     const SelectQueryInfo & query_info,
@@ -155,8 +173,6 @@ Pipe StorageBuffer::read(
     size_t max_block_size,
     unsigned num_streams)
 {
-    Pipe pipe_from_dst;
-
     if (destination_id)
     {
         auto destination = DatabaseCatalog::instance().getTable(destination_id, context);
@@ -182,8 +198,8 @@ Pipe StorageBuffer::read(
                 query_info.input_order_info = query_info.order_optimizer->getInputOrder(destination, destination_metadata_snapshot);
 
             /// The destination table has the same structure of the requested columns and we can simply read blocks from there.
-            pipe_from_dst = destination->read(
-                column_names, destination_metadata_snapshot, query_info,
+            destination->read(
+                query_plan, column_names, destination_metadata_snapshot, query_info,
                 context, processed_stage, max_block_size, num_streams);
         }
         else
@@ -217,25 +233,45 @@ Pipe StorageBuffer::read(
             }
             else
             {
-                pipe_from_dst = destination->read(
-                    columns_intersection, destination_metadata_snapshot, query_info,
+                destination->read(
+                        query_plan, columns_intersection, destination_metadata_snapshot, query_info,
                         context, processed_stage, max_block_size, num_streams);
 
-                pipe_from_dst.addSimpleTransform([&](const Block & stream_header)
-                {
-                    return std::make_shared<AddingMissedTransform>(stream_header, header_after_adding_defaults,
+                auto adding_missed = std::make_unique<AddingMissedStep>(
+                        query_plan.getCurrentDataStream(),
+                        header_after_adding_defaults,
                         metadata_snapshot->getColumns().getDefaults(), context);
-                });
 
-                pipe_from_dst.addSimpleTransform([&](const Block & stream_header)
+                adding_missed->setStepDescription("Add columns missing in destination table");
+                query_plan.addStep(std::move(adding_missed));
+
+                auto converting = std::make_unique<ConvertingStep>(
+                        query_plan.getCurrentDataStream(),
+                        header);
+
+                converting->setStepDescription("Convert destination table columns to Buffer table structure");
+                query_plan.addStep(std::move(converting));
+            }
+        }
+
+        if (query_plan.isInitialized())
         {
-                    return std::make_shared<ConvertingTransform>(
-                        stream_header, header, ConvertingTransform::MatchColumnsMode::Name);
-                });
-            }
-        }
+            StreamLocalLimits limits;
+            SizeLimits leaf_limits;
 
-        pipe_from_dst.addTableLock(destination_lock);
+            /// Add table lock for destination table.
+            auto adding_limits_and_quota = std::make_unique<SettingQuotaAndLimitsStep>(
+                    query_plan.getCurrentDataStream(),
+                    destination,
+                    std::move(destination_lock),
+                    limits,
+                    leaf_limits,
+                    nullptr,
+                    nullptr);
+
+            adding_limits_and_quota->setStepDescription("Lock destination table for Buffer");
+            query_plan.addStep(std::move(adding_limits_and_quota));
+        }
     }
 
     Pipe pipe_from_buffers;
@@ -248,27 +284,23 @@ Pipe StorageBuffer::read(
         pipe_from_buffers = Pipe::unitePipes(std::move(pipes_from_buffers));
     }
 
-    /// Convert pipes from table to structure from buffer.
-    if (!pipe_from_buffers.empty() && !pipe_from_dst.empty()
-        && !blocksHaveEqualStructure(pipe_from_buffers.getHeader(), pipe_from_dst.getHeader()))
-    {
-        pipe_from_dst.addSimpleTransform([&](const Block & header)
-        {
-            return std::make_shared<ConvertingTransform>(
-                   header,
-                   pipe_from_buffers.getHeader(),
-                   ConvertingTransform::MatchColumnsMode::Name);
-        });
-    }
+    if (pipe_from_buffers.empty())
+        return;
+
+    QueryPlan buffers_plan;
 
     /** If the sources from the table were processed before some non-initial stage of query execution,
       * then sources from the buffers must also be wrapped in the processing pipeline before the same stage.
       */
     if (processed_stage > QueryProcessingStage::FetchColumns)
-        pipe_from_buffers = QueryPipeline::getPipe(
-                InterpreterSelectQuery(query_info.query, context, std::move(pipe_from_buffers),
-                                               SelectQueryOptions(processed_stage)).execute().pipeline);
-
+    {
+        auto interpreter = InterpreterSelectQuery(
+                query_info.query, context, std::move(pipe_from_buffers),
+                SelectQueryOptions(processed_stage));
+        interpreter.buildQueryPlan(buffers_plan);
+    }
+    else
+    {
         if (query_info.prewhere_info)
         {
             pipe_from_buffers.addSimpleTransform([&](const Block & header)
@@ -287,10 +319,38 @@ Pipe StorageBuffer::read(
             }
         }
 
-    Pipes pipes;
-    pipes.emplace_back(std::move(pipe_from_dst));
-    pipes.emplace_back(std::move(pipe_from_buffers));
-    return Pipe::unitePipes(std::move(pipes));
+        auto read_from_buffers = std::make_unique<ReadFromPreparedSource>(std::move(pipe_from_buffers));
+        read_from_buffers->setStepDescription("Read from buffers of Buffer table");
+        buffers_plan.addStep(std::move(read_from_buffers));
+    }
+
+    if (!query_plan.isInitialized())
+    {
+        query_plan = std::move(buffers_plan);
+        return;
+    }
+
+    auto result_header = buffers_plan.getCurrentDataStream().header;
+
+    /// Convert structure from table to structure from buffer.
+    if (!blocksHaveEqualStructure(query_plan.getCurrentDataStream().header, result_header))
+    {
+        auto converting = std::make_unique<ConvertingStep>(query_plan.getCurrentDataStream(), result_header);
+        query_plan.addStep(std::move(converting));
+    }
+
+    DataStreams input_streams;
+    input_streams.emplace_back(query_plan.getCurrentDataStream());
+    input_streams.emplace_back(buffers_plan.getCurrentDataStream());
+
+    std::vector<std::unique_ptr<QueryPlan>> plans;
+    plans.emplace_back(std::make_unique<QueryPlan>(std::move(query_plan)));
+    plans.emplace_back(std::make_unique<QueryPlan>(std::move(buffers_plan)));
+    query_plan = QueryPlan();
+
+    auto union_step = std::make_unique<UnionStep>(std::move(input_streams), result_header);
+    union_step->setStepDescription("Unite sources from Buffer table");
+    query_plan.unitePlans(std::move(union_step), std::move(plans));
 }
 
 

src/Storages/StorageBuffer.h

@@ -65,6 +65,16 @@ public:
         size_t max_block_size,
         unsigned num_streams) override;
 
+    void read(
+        QueryPlan & query_plan,
+        const Names & column_names,
+        const StorageMetadataPtr & metadata_snapshot,
+        const SelectQueryInfo & query_info,
+        const Context & context,
+        QueryProcessingStage::Enum processed_stage,
+        size_t max_block_size,
+        unsigned num_streams) override;
+
     bool supportsParallelInsert() const override { return true; }
 
     BlockOutputStreamPtr write(const ASTPtr & query, const StorageMetadataPtr & /*metadata_snapshot*/, const Context & context) override;

src/Storages/StorageMaterializedView.cpp

@@ -107,6 +107,21 @@ QueryProcessingStage::Enum StorageMaterializedView::getQueryProcessingStage(cons
 }
 
 Pipe StorageMaterializedView::read(
+    const Names & column_names,
+    const StorageMetadataPtr & metadata_snapshot,
+    const SelectQueryInfo & query_info,
+    const Context & context,
+    QueryProcessingStage::Enum processed_stage,
+    const size_t max_block_size,
+    const unsigned num_streams)
+{
+    QueryPlan plan;
+    read(plan, column_names, metadata_snapshot, query_info, context, processed_stage, max_block_size, num_streams);
+    return QueryPipeline::getPipe(std::move(*plan.buildQueryPipeline()));
+}
+
+void StorageMaterializedView::read(
+    QueryPlan & query_plan,
     const Names & column_names,
     const StorageMetadataPtr & /*metadata_snapshot*/,
     const SelectQueryInfo & query_info,
@@ -122,10 +137,7 @@ Pipe StorageMaterializedView::read(
     if (query_info.order_optimizer)
         query_info.input_order_info = query_info.order_optimizer->getInputOrder(storage, metadata_snapshot);
 
-    Pipe pipe = storage->read(column_names, metadata_snapshot, query_info, context, processed_stage, max_block_size, num_streams);
-    pipe.addTableLock(lock);
-
-    return pipe;
+    storage->read(query_plan, column_names, metadata_snapshot, query_info, context, processed_stage, max_block_size, num_streams);
 }
 
 BlockOutputStreamPtr StorageMaterializedView::write(const ASTPtr & query, const StorageMetadataPtr & /*metadata_snapshot*/, const Context & context)

src/Storages/StorageMaterializedView.h

@@ -80,6 +80,16 @@ public:
         size_t max_block_size,
         unsigned num_streams) override;
 
+    void read(
+        QueryPlan & query_plan,
+        const Names & column_names,
+        const StorageMetadataPtr & metadata_snapshot,
+        const SelectQueryInfo & query_info,
+        const Context & context,
+        QueryProcessingStage::Enum processed_stage,
+        size_t max_block_size,
+        unsigned num_streams) override;
+
     Strings getDataPaths() const override;
 
 private:

tests/queries/0_stateless/00184_shard_distributed_group_by_no_merge.sql

@@ -10,9 +10,9 @@ SET max_threads=1;
 SET optimize_move_functions_out_of_any=0;
 
 SELECT 'LIMIT';
-SELECT any(_shard_num) shard_num, count(), uniq(dummy) FROM remote('127.0.0.{2,3}', system.one) LIMIT 1 SETTINGS distributed_group_by_no_merge=2;
+SELECT * FROM (SELECT any(_shard_num) shard_num, count(), uniq(dummy) FROM remote('127.0.0.{2,3}', system.one) LIMIT 1 ) ORDER BY shard_num SETTINGS distributed_group_by_no_merge=2;
 SELECT 'OFFSET';
-SELECT any(_shard_num) shard_num, count(), uniq(dummy) FROM remote('127.0.0.{2,3}', system.one) LIMIT 1, 1 SETTINGS distributed_group_by_no_merge=2;
+SELECT * FROM (SELECT any(_shard_num) shard_num, count(), uniq(dummy) FROM remote('127.0.0.{2,3}', system.one) LIMIT 1, 1) ORDER BY shard_num  SETTINGS distributed_group_by_no_merge=2;
 
 SELECT 'ALIAS';
 SELECT dummy AS d FROM remote('127.0.0.{2,3}', system.one) ORDER BY d SETTINGS distributed_group_by_no_merge=2;