diff --git a/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp b/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp
index f983de91b37..cd9d6235f52 100644
--- a/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp
+++ b/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp
@@ -24,7 +24,7 @@ namespace DB
 {
 ParallelAggregatingBlockInputStream::ParallelAggregatingBlockInputStream(
     const BlockInputStreams & inputs,
-    const BlockInputStreamPtr & additional_input_at_end,
+    const BlockInputStreams & additional_inputs_at_end,
     const Aggregator::Params & params_,
     const FileProviderPtr & file_provider_,
     bool final_,
@@ -41,11 +41,10 @@ ParallelAggregatingBlockInputStream::ParallelAggregatingBlockInputStream(
     , keys_size(params.keys_size)
     , aggregates_size(params.aggregates_size)
     , handler(*this)
-    , processor(inputs, additional_input_at_end, max_threads, handler, log)
+    , processor(inputs, additional_inputs_at_end, max_threads, handler, log)
 {
     children = inputs;
-    if (additional_input_at_end)
-        children.push_back(additional_input_at_end);
+    children.insert(children.end(), additional_inputs_at_end.begin(), additional_inputs_at_end.end());
 }
 
 
diff --git a/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.h b/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.h
index 41e61786370..907622c8364 100644
--- a/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.h
+++ b/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.h
@@ -36,7 +36,7 @@ class ParallelAggregatingBlockInputStream : public IProfilingBlockInputStream
       */
     ParallelAggregatingBlockInputStream(
         const BlockInputStreams & inputs,
-        const BlockInputStreamPtr & additional_input_at_end,
+        const BlockInputStreams & additional_inputs_at_end,
         const Aggregator::Params & params_,
         const FileProviderPtr & file_provider_,
         bool final_,
diff --git a/dbms/src/DataStreams/ParallelInputsProcessor.h b/dbms/src/DataStreams/ParallelInputsProcessor.h
index 34c70a7085e..57ab37e1756 100644
--- a/dbms/src/DataStreams/ParallelInputsProcessor.h
+++ b/dbms/src/DataStreams/ParallelInputsProcessor.h
@@ -16,6 +16,7 @@
 
 #include <Common/CurrentMetrics.h>
 #include <Common/Logger.h>
+#include <Common/MPMCQueue.h>
 #include <Common/MemoryTracker.h>
 #include <Common/ThreadFactory.h>
 #include <Common/ThreadManager.h>
@@ -83,9 +84,8 @@ template <typename Handler, StreamUnionMode mode = StreamUnionMode::Basic>
 class ParallelInputsProcessor
 {
 public:
-    /** additional_input_at_end - if not nullptr,
-      *  then the blocks from this source will start to be processed only after all other sources are processed.
-      * This is done in the main thread.
+    /** additional_inputs_at_end - if not empty,
+      *  then the blocks from the sources will start to be processed only after all other sources are processed.
       *
       * Intended for implementation of FULL and RIGHT JOIN
       * - where you must first make JOIN in parallel, while noting which keys are not found,
@@ -93,19 +93,18 @@ class ParallelInputsProcessor
       */
     ParallelInputsProcessor(
         const BlockInputStreams & inputs_,
-        const BlockInputStreamPtr & additional_input_at_end_,
+        const BlockInputStreams & additional_inputs_at_end_,
         size_t max_threads_,
         Handler & handler_,
         const LoggerPtr & log_)
         : inputs(inputs_)
-        , additional_input_at_end(additional_input_at_end_)
-        , max_threads(std::min(inputs_.size(), max_threads_))
+        , additional_inputs_at_end(additional_inputs_at_end_)
+        , max_threads(std::min(std::max(inputs_.size(), additional_inputs_at_end_.size()), max_threads_))
         , handler(handler_)
+        , working_inputs(inputs_)
+        , working_additional_inputs(additional_inputs_at_end_)
         , log(log_)
-    {
-        for (size_t i = 0; i < inputs_.size(); ++i)
-            unprepared_inputs.emplace(inputs_[i], i);
-    }
+    {}
 
     ~ParallelInputsProcessor()
     {
@@ -132,36 +131,21 @@ class ParallelInputsProcessor
     /// Ask all sources to stop earlier than they run out.
     void cancel(bool kill)
     {
-        finish = true;
+        working_inputs.available_inputs.cancel();
+        working_additional_inputs.available_inputs.cancel();
 
-        for (auto & input : inputs)
-        {
-            if (IProfilingBlockInputStream * child = dynamic_cast<IProfilingBlockInputStream *>(&*input))
-            {
-                try
-                {
-                    child->cancel(kill);
-                }
-                catch (...)
-                {
-                    /** If you can not ask one or more sources to stop.
-                      * (for example, the connection is broken for distributed query processing)
-                      * - then do not care.
-                      */
-                    LOG_FMT_ERROR(log, "Exception while cancelling {}", child->getName());
-                }
-            }
-        }
+        cancelStreams(inputs, kill);
+        cancelStreams(additional_inputs_at_end, kill);
     }
 
     /// Wait until all threads are finished, before the destructor.
     void wait()
     {
-        if (joined_threads)
-            return;
         if (thread_manager)
+        {
             thread_manager->wait();
-        joined_threads = true;
+            thread_manager.reset();
+        }
     }
 
     size_t getNumActiveThreads() const
@@ -181,13 +165,78 @@ class ParallelInputsProcessor
         BlockInputStreamPtr in;
         size_t i; /// The source number (for debugging).
 
-        InputData() {}
+        InputData()
+            : i(0)
+        {}
         InputData(const BlockInputStreamPtr & in_, size_t i_)
             : in(in_)
             , i(i_)
         {}
     };
 
+    struct WorkingInputs
+    {
+        explicit WorkingInputs(const BlockInputStreams & inputs_)
+            : available_inputs(inputs_.size())
+            , active_inputs(inputs_.size())
+            , unprepared_inputs(inputs_.size())
+        {
+            for (size_t i = 0; i < inputs_.size(); ++i)
+                unprepared_inputs.emplace(inputs_[i], i);
+        }
+        /** A set of available sources that are not currently processed by any thread.
+          * Each thread takes one source from this set, takes a block out of the source (at this moment the source does the calculations)
+          *  and (if the source is not run out), puts it back into the set of available sources.
+          *
+          * The question arises what is better to use:
+          * - the queue (just processed source will be processed the next time later than the rest)
+          * - stack (just processed source will be processed as soon as possible).
+          *
+          * The stack is better than the queue when you need to do work on reading one source more consequentially,
+          *  and theoretically, this allows you to achieve more consequent/consistent reads from the disk.
+          *
+          * But when using the stack, there is a problem with distributed query processing:
+          *  data is read only from a part of the servers, and on the other servers
+          * a timeout occurs during send, and the request processing ends with an exception.
+          *
+          * Therefore, a queue is used. This can be improved in the future.
+          */
+        using AvailableInputs = MPMCQueue<InputData>;
+        AvailableInputs available_inputs;
+
+        /// How many active input streams.
+        std::atomic<size_t> active_inputs;
+
+        /** For parallel preparing (readPrefix) child streams.
+          * First, streams are located here.
+          * After a stream was prepared, it is moved to "available_inputs" for reading.
+          */
+        using UnpreparedInputs = MPMCQueue<InputData>;
+        UnpreparedInputs unprepared_inputs;
+    };
+
+    void cancelStreams(const BlockInputStreams & streams, bool kill)
+    {
+        for (const auto & input : streams)
+        {
+            if (auto * p_child = dynamic_cast<IProfilingBlockInputStream *>(&*input))
+            {
+                try
+                {
+                    p_child->cancel(kill);
+                }
+                catch (...)
+                {
+                    /** If you can not ask one or more sources to stop.
+                      * (for example, the connection is broken for distributed query processing)
+                      * - then do not care.
+                      */
+                    LOG_FMT_ERROR(log, "Exception while cancelling {}", p_child->getName());
+                }
+            }
+        }
+    }
+
     void publishPayload(BlockInputStreamPtr & stream, Block & block, size_t thread_num)
     {
         if constexpr (mode == StreamUnionMode::Basic)
@@ -201,32 +250,24 @@ class ParallelInputsProcessor
 
     void thread(size_t thread_num)
     {
-        std::exception_ptr exception;
+        work(thread_num, working_inputs);
+        work(thread_num, working_additional_inputs);
 
-        try
-        {
-            while (!finish)
-            {
-                InputData unprepared_input;
-                {
-                    std::lock_guard lock(unprepared_inputs_mutex);
-
-                    if (unprepared_inputs.empty())
-                        break;
-
-                    unprepared_input = unprepared_inputs.front();
-                    unprepared_inputs.pop();
-                }
+        handler.onFinishThread(thread_num);
 
-                unprepared_input.in->readPrefix();
+        if (0 == --active_threads)
+        {
+            handler.onFinish();
+        }
+    }
 
-                {
-                    std::lock_guard lock(available_inputs_mutex);
-                    available_inputs.push(unprepared_input);
-                }
-            }
+    void work(size_t thread_num, WorkingInputs & work)
+    {
+        std::exception_ptr exception;
 
-            loop(thread_num);
+        try
+        {
+            loop(thread_num, work);
         }
         catch (...)
         {
@@ -237,134 +278,63 @@ class ParallelInputsProcessor
         {
             handler.onException(exception, thread_num);
         }
-
-        handler.onFinishThread(thread_num);
-
-        /// The last thread on the output indicates that there is no more data.
-        if (0 == --active_threads)
-        {
-            /// And then it processes an additional source, if there is one.
-            if (additional_input_at_end)
-            {
-                try
-                {
-                    additional_input_at_end->readPrefix();
-                    while (Block block = additional_input_at_end->read())
-                        publishPayload(additional_input_at_end, block, thread_num);
-                }
-                catch (...)
-                {
-                    exception = std::current_exception();
-                }
-
-                if (exception)
-                {
-                    handler.onException(exception, thread_num);
-                }
-            }
-
-            handler.onFinish(); /// TODO If in `onFinish` or `onFinishThread` there is an exception, then std::terminate is called.
-        }
     }
 
-    void loop(size_t thread_num)
+    /// This function may be called in different threads.
+    /// If no exception occurs, we can ensure that the work is all done when the function
+    /// returns in any thread.
+    void loop(size_t thread_num, WorkingInputs & work)
     {
-        while (!finish) /// You may need to stop work earlier than all sources run out.
+        if (work.active_inputs == 0)
         {
-            InputData input;
+            return;
+        }
 
-            /// Select the next source.
-            {
-                std::lock_guard lock(available_inputs_mutex);
+        InputData input;
 
-                /// If there are no free sources, then this thread is no longer needed. (But other threads can work with their sources.)
-                if (available_inputs.empty())
-                    break;
-
-                input = available_inputs.front();
+        while (work.unprepared_inputs.tryPop(input))
+        {
+            input.in->readPrefix();
 
-                /// We remove the source from the queue of available sources.
-                available_inputs.pop();
-            }
+            work.available_inputs.push(input);
+        }
 
+        // The condition is false when all input streams are exhausted or
+        // an exception occurred then the queue was cancelled.
+        while (work.available_inputs.pop(input))
+        {
             /// The main work.
             Block block = input.in->read();
 
+            if (block)
             {
-                if (finish)
-                    break;
-
-                /// If this source is not run out yet, then put the resulting block in the ready queue.
+                work.available_inputs.push(input);
+                publishPayload(input.in, block, thread_num);
+            }
+            else
+            {
+                if (0 == --work.active_inputs)
                 {
-                    std::lock_guard lock(available_inputs_mutex);
-
-                    if (block)
-                    {
-                        available_inputs.push(input);
-                    }
-                    else
-                    {
-                        if (available_inputs.empty())
-                            break;
-                    }
-                }
-
-                if (finish)
+                    work.available_inputs.finish();
                     break;
-
-                if (block)
-                    publishPayload(input.in, block, thread_num);
+                }
             }
         }
     }
 
-    BlockInputStreams inputs;
-    BlockInputStreamPtr additional_input_at_end;
+    const BlockInputStreams inputs;
+    const BlockInputStreams additional_inputs_at_end;
     unsigned max_threads;
 
     Handler & handler;
 
     std::shared_ptr<ThreadManager> thread_manager;
 
-    /** A set of available sources that are not currently processed by any thread.
-      * Each thread takes one source from this set, takes a block out of the source (at this moment the source does the calculations)
-      *  and (if the source is not run out), puts it back into the set of available sources.
-      *
-      * The question arises what is better to use:
-      * - the queue (just processed source will be processed the next time later than the rest)
-      * - stack (just processed source will be processed as soon as possible).
-      *
-      * The stack is better than the queue when you need to do work on reading one source more consequentially,
-      *  and theoretically, this allows you to achieve more consequent/consistent reads from the disk.
-      *
-      * But when using the stack, there is a problem with distributed query processing:
-      *  data is read only from a part of the servers, and on the other servers
-      * a timeout occurs during send, and the request processing ends with an exception.
-      *
-      * Therefore, a queue is used. This can be improved in the future.
-      */
-    using AvailableInputs = std::queue<InputData>;
-    AvailableInputs available_inputs;
-
-    /** For parallel preparing (readPrefix) child streams.
-      * First, streams are located here.
-      * After a stream was prepared, it is moved to "available_inputs" for reading.
-      */
-    using UnpreparedInputs = std::queue<InputData>;
-    UnpreparedInputs unprepared_inputs;
-
-    /// For operations with available_inputs.
-    std::mutex available_inputs_mutex;
-
-    /// For operations with unprepared_inputs.
-    std::mutex unprepared_inputs_mutex;
+    WorkingInputs working_inputs;
+    WorkingInputs working_additional_inputs;
 
     /// How many sources ran out.
     std::atomic<size_t> active_threads{0};
-    /// Finish the threads work (before the sources run out).
-    std::atomic<bool> finish{false};
-    /// Wait for the completion of all threads.
-    std::atomic<bool> joined_threads{false};
 
     const LoggerPtr log;
 };
diff --git a/dbms/src/DataStreams/UnionBlockInputStream.h b/dbms/src/DataStreams/UnionBlockInputStream.h
index a782c3dd087..ffcc8d77c10 100644
--- a/dbms/src/DataStreams/UnionBlockInputStream.h
+++ b/dbms/src/DataStreams/UnionBlockInputStream.h
@@ -94,20 +94,19 @@ class UnionBlockInputStream final : public IProfilingBlockInputStream
 public:
     UnionBlockInputStream(
         BlockInputStreams inputs,
-        BlockInputStreamPtr additional_input_at_end,
+        BlockInputStreams additional_inputs_at_end,
         size_t max_threads,
         const String & req_id,
         ExceptionCallback exception_callback_ = ExceptionCallback())
-        : output_queue(std::min(inputs.size(), max_threads) * 5) // reduce contention
+        : output_queue(std::min(std::max(inputs.size(), additional_inputs_at_end.size()), max_threads) * 5) // reduce contention
         , log(Logger::get(NAME, req_id))
         , handler(*this)
-        , processor(inputs, additional_input_at_end, max_threads, handler, log)
+        , processor(inputs, additional_inputs_at_end, max_threads, handler, log)
         , exception_callback(exception_callback_)
     {
         // TODO: assert capacity of output_queue is not less than processor.getMaxThreads()
         children = inputs;
-        if (additional_input_at_end)
-            children.push_back(additional_input_at_end);
+        children.insert(children.end(), additional_inputs_at_end.begin(), additional_inputs_at_end.end());
 
         size_t num_children = children.size();
         if (num_children > 1)
diff --git a/dbms/src/DataStreams/tests/union_stream2.cpp b/dbms/src/DataStreams/tests/union_stream2.cpp
index f939cda4e14..fb3f7238414 100644
--- a/dbms/src/DataStreams/tests/union_stream2.cpp
+++ b/dbms/src/DataStreams/tests/union_stream2.cpp
@@ -51,7 +51,7 @@ try
     for (size_t i = 0, size = streams.size(); i < size; ++i)
         streams[i] = std::make_shared<AsynchronousBlockInputStream>(streams[i]);
 
-    BlockInputStreamPtr stream = std::make_shared<UnionBlockInputStream<>>(streams, nullptr, settings.max_threads, /*req_id=*/"");
+    BlockInputStreamPtr stream = std::make_shared<UnionBlockInputStream<>>(streams, BlockInputStreams{}, settings.max_threads, /*req_id=*/"");
     stream = std::make_shared<LimitBlockInputStream>(stream, 10, 0, "");
 
     WriteBufferFromFileDescriptor wb(STDERR_FILENO);
diff --git a/dbms/src/Debug/astToExecutor.cpp b/dbms/src/Debug/astToExecutor.cpp
index edadc7a9940..6cbba6efe99 100644
--- a/dbms/src/Debug/astToExecutor.cpp
+++ b/dbms/src/Debug/astToExecutor.cpp
@@ -1554,7 +1554,7 @@ ExecutorPtr compileAggregation(ExecutorPtr input, size_t & executor_index, ASTPt
                 ci.tp = TiDB::TypeLongLong;
                 ci.flag = TiDB::ColumnFlagUnsigned | TiDB::ColumnFlagNotNull;
             }
-            else if (func->name == "max" || func->name == "min" || func->name == "first_row")
+            else if (func->name == "max" || func->name == "min" || func->name == "first_row" || func->name == "sum")
             {
                 ci = children_ci[0];
                 ci.flag &= ~TiDB::ColumnFlagNotNull;
diff --git a/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp b/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp
index 30d033c870f..4fb98add2a9 100644
--- a/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp
+++ b/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp
@@ -379,34 +379,39 @@ void DAGQueryBlockInterpreter::executeAggregation(
         is_final_agg);
 
     /// If there are several sources, then we perform parallel aggregation
-    if (pipeline.streams.size() > 1)
+    if (pipeline.streams.size() > 1 || pipeline.streams_with_non_joined_data.size() > 1)
     {
         const Settings & settings = context.getSettingsRef();
-        BlockInputStreamPtr stream_with_non_joined_data = combinedNonJoinedDataStream(pipeline, max_streams, log);
-        pipeline.firstStream() = std::make_shared<ParallelAggregatingBlockInputStream>(
+        BlockInputStreamPtr stream = std::make_shared<ParallelAggregatingBlockInputStream>(
             pipeline.streams,
-            stream_with_non_joined_data,
+            pipeline.streams_with_non_joined_data,
             params,
             context.getFileProvider(),
             true,
             max_streams,
             settings.aggregation_memory_efficient_merge_threads ? static_cast<size_t>(settings.aggregation_memory_efficient_merge_threads) : static_cast<size_t>(settings.max_threads),
             log->identifier());
+
         pipeline.streams.resize(1);
+        pipeline.streams_with_non_joined_data.clear();
+        pipeline.firstStream() = std::move(stream);
+
         // should record for agg before restore concurrency. See #3804.
         recordProfileStreams(pipeline, query_block.aggregation_name);
         restorePipelineConcurrency(pipeline);
     }
     else
     {
-        BlockInputStreamPtr stream_with_non_joined_data = combinedNonJoinedDataStream(pipeline, max_streams, log);
         BlockInputStreams inputs;
         if (!pipeline.streams.empty())
             inputs.push_back(pipeline.firstStream());
-        else
-            pipeline.streams.resize(1);
-        if (stream_with_non_joined_data)
-            inputs.push_back(stream_with_non_joined_data);
+
+        if (!pipeline.streams_with_non_joined_data.empty())
+            inputs.push_back(pipeline.streams_with_non_joined_data.at(0));
+
+        pipeline.streams.resize(1);
+        pipeline.streams_with_non_joined_data.clear();
+
         pipeline.firstStream() = std::make_shared<AggregatingBlockInputStream>(
             std::make_shared<ConcatBlockInputStream>(inputs, log->identifier()),
             params,
diff --git a/dbms/src/Flash/Coprocessor/InterpreterUtils.cpp b/dbms/src/Flash/Coprocessor/InterpreterUtils.cpp
index 9de5b83626f..c747823b69d 100644
--- a/dbms/src/Flash/Coprocessor/InterpreterUtils.cpp
+++ b/dbms/src/Flash/Coprocessor/InterpreterUtils.cpp
@@ -42,32 +42,6 @@ void restoreConcurrency(
     }
 }
 
-BlockInputStreamPtr combinedNonJoinedDataStream(
-    DAGPipeline & pipeline,
-    size_t max_threads,
-    const LoggerPtr & log,
-    bool ignore_block)
-{
-    BlockInputStreamPtr ret = nullptr;
-    if (pipeline.streams_with_non_joined_data.size() == 1)
-        ret = pipeline.streams_with_non_joined_data.at(0);
-    else if (pipeline.streams_with_non_joined_data.size() > 1)
-    {
-        if (ignore_block)
-        {
-            ret = std::make_shared<UnionWithoutBlock>(pipeline.streams_with_non_joined_data, nullptr, max_threads, log->identifier());
-            ret->setExtraInfo("combine non joined(ignore block)");
-        }
-        else
-        {
-            ret = std::make_shared<UnionWithBlock>(pipeline.streams_with_non_joined_data, nullptr, max_threads, log->identifier());
-            ret->setExtraInfo("combine non joined");
-        }
-    }
-    pipeline.streams_with_non_joined_data.clear();
-    return ret;
-}
-
 void executeUnion(
     DAGPipeline & pipeline,
     size_t max_streams,
@@ -75,21 +49,33 @@ void executeUnion(
     bool ignore_block,
     const String & extra_info)
 {
-    if (pipeline.streams.size() == 1 && pipeline.streams_with_non_joined_data.empty())
-        return;
-    auto non_joined_data_stream = combinedNonJoinedDataStream(pipeline, max_streams, log, ignore_block);
-    if (!pipeline.streams.empty())
+    switch (pipeline.streams.size() + pipeline.streams_with_non_joined_data.size())
     {
+    case 0:
+        break;
+    case 1:
+    {
+        if (pipeline.streams.size() == 1)
+            break;
+        // streams_with_non_joined_data's size is 1.
+        pipeline.streams.push_back(pipeline.streams_with_non_joined_data.at(0));
+        pipeline.streams_with_non_joined_data.clear();
+        break;
+    }
+    default:
+    {
+        BlockInputStreamPtr stream;
         if (ignore_block)
-            pipeline.firstStream() = std::make_shared<UnionWithoutBlock>(pipeline.streams, non_joined_data_stream, max_streams, log->identifier());
+            stream = std::make_shared<UnionWithoutBlock>(pipeline.streams, pipeline.streams_with_non_joined_data, max_streams, log->identifier());
         else
-            pipeline.firstStream() = std::make_shared<UnionWithBlock>(pipeline.streams, non_joined_data_stream, max_streams, log->identifier());
-        pipeline.firstStream()->setExtraInfo(extra_info);
+            stream = std::make_shared<UnionWithBlock>(pipeline.streams, pipeline.streams_with_non_joined_data, max_streams, log->identifier());
+        stream->setExtraInfo(extra_info);
+
         pipeline.streams.resize(1);
+        pipeline.streams_with_non_joined_data.clear();
+        pipeline.firstStream() = std::move(stream);
+        break;
     }
-    else if (non_joined_data_stream != nullptr)
-    {
-        pipeline.streams.push_back(non_joined_data_stream);
     }
 }
 
diff --git a/dbms/src/Flash/tests/bench_exchange.cpp b/dbms/src/Flash/tests/bench_exchange.cpp
index fbb53bfd4a4..cbbdf060580 100644
--- a/dbms/src/Flash/tests/bench_exchange.cpp
+++ b/dbms/src/Flash/tests/bench_exchange.cpp
@@ -215,7 +215,7 @@ std::vector<BlockInputStreamPtr> ReceiverHelper::buildExchangeReceiverStream()
 BlockInputStreamPtr ReceiverHelper::buildUnionStream()
 {
     auto streams = buildExchangeReceiverStream();
-    return std::make_shared<UnionBlockInputStream<>>(streams, nullptr, concurrency, /*req_id=*/"");
+    return std::make_shared<UnionBlockInputStream<>>(streams, BlockInputStreams{}, concurrency, /*req_id=*/"");
 }
 
 void ReceiverHelper::finish()
@@ -290,7 +290,7 @@ BlockInputStreamPtr SenderHelper::buildUnionStream(
         send_streams.push_back(std::make_shared<ExchangeSenderBlockInputStream>(stream, std::move(response_writer), /*req_id=*/""));
     }
 
-    return std::make_shared<UnionBlockInputStream<>>(send_streams, nullptr, concurrency, /*req_id=*/"");
+    return std::make_shared<UnionBlockInputStream<>>(send_streams, BlockInputStreams{}, concurrency, /*req_id=*/"");
 }
 
 BlockInputStreamPtr SenderHelper::buildUnionStream(size_t total_rows, const std::vector<Block> & blocks)
@@ -312,7 +312,7 @@ BlockInputStreamPtr SenderHelper::buildUnionStream(size_t total_rows, const std:
         send_streams.push_back(std::make_shared<ExchangeSenderBlockInputStream>(stream, std::move(response_writer), /*req_id=*/""));
     }
 
-    return std::make_shared<UnionBlockInputStream<>>(send_streams, nullptr, concurrency, /*req_id=*/"");
+    return std::make_shared<UnionBlockInputStream<>>(send_streams, BlockInputStreams{}, concurrency, /*req_id=*/"");
 }
 
 void SenderHelper::finish()
diff --git a/dbms/src/Flash/tests/bench_window.cpp b/dbms/src/Flash/tests/bench_window.cpp
index dfdb358c46c..356f544a836 100644
--- a/dbms/src/Flash/tests/bench_window.cpp
+++ b/dbms/src/Flash/tests/bench_window.cpp
@@ -71,7 +71,7 @@ class WindowFunctionBench : public ExchangeBench
         pipeline.transform([&](auto & stream) {
             stream = std::make_shared<SquashingBlockInputStream>(stream, 8192, 0, "mock_executor_id_squashing");
         });
-        receiver_stream = std::make_shared<UnionBlockInputStream<>>(pipeline.streams, nullptr, concurrency, /*req_id=*/"");
+        receiver_stream = std::make_shared<UnionBlockInputStream<>>(pipeline.streams, BlockInputStreams{}, concurrency, /*req_id=*/"");
     }
 
     tipb::Window window;
diff --git a/dbms/src/Flash/tests/gtest_qb_interpreter.cpp b/dbms/src/Flash/tests/gtest_qb_interpreter.cpp
index 5529fc358db..1228eaa6201 100644
--- a/dbms/src/Flash/tests/gtest_qb_interpreter.cpp
+++ b/dbms/src/Flash/tests/gtest_qb_interpreter.cpp
@@ -33,8 +33,8 @@ class QBInterpreterExecuteTest : public DB::tests::ExecutorTest
         context.addMockTable({"test_db", "r_table"}, {{"r_a", TiDB::TP::TypeLong}, {"r_b", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}});
         context.addMockTable({"test_db", "l_table"}, {{"l_a", TiDB::TP::TypeLong}, {"l_b", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}});
         context.addExchangeRelationSchema("sender_1", {{"s1", TiDB::TP::TypeString}, {"s2", TiDB::TP::TypeString}, {"s3", TiDB::TP::TypeString}});
-        context.addExchangeRelationSchema("sender_l", {{"l_a", TiDB::TP::TypeString}, {"l_b", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}});
-        context.addExchangeRelationSchema("sender_r", {{"r_a", TiDB::TP::TypeString}, {"r_b", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}});
+        context.addExchangeRelationSchema("sender_l", {{"l_a", TiDB::TP::TypeLong}, {"l_b", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}});
+        context.addExchangeRelationSchema("sender_r", {{"r_a", TiDB::TP::TypeLong}, {"r_b", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}});
     }
 };
 
@@ -200,47 +200,6 @@ Union: <for test>
         ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
     }
 
-    // Join Source.
-    DAGRequestBuilder table1 = context.scan("test_db", "r_table");
-    DAGRequestBuilder table2 = context.scan("test_db", "l_table");
-    DAGRequestBuilder table3 = context.scan("test_db", "r_table");
-    DAGRequestBuilder table4 = context.scan("test_db", "l_table");
-
-    request = table1.join(
-                        table2.join(
-                            table3.join(table4,
-                                        {col("join_c")},
-                                        ASTTableJoin::Kind::Left),
-                            {col("join_c")},
-                            ASTTableJoin::Kind::Left),
-                        {col("join_c")},
-                        ASTTableJoin::Kind::Left)
-                  .build(context);
-    {
-        String expected = R"(
-CreatingSets
- Union: <for join>
-  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = table_scan_3>, join_kind = Left
-   Expression: <append join key and join filters for build side>
-    Expression: <final projection>
-     MockTableScan
- Union x 2: <for join>
-  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = Join_4>, join_kind = Left
-   Expression: <append join key and join filters for build side>
-    Expression: <final projection>
-     Expression: <remove useless column after join>
-      HashJoinProbe: <join probe, join_executor_id = Join_4>
-       Expression: <final projection>
-        MockTableScan
- Union: <for test>
-  Expression x 10: <final projection>
-   Expression: <remove useless column after join>
-    HashJoinProbe: <join probe, join_executor_id = Join_6>
-     Expression: <final projection>
-      MockTableScan)";
-        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
-    }
-
     request = context.receive("sender_1")
                   .project({"s1", "s2", "s3"})
                   .project({"s1", "s2"})
@@ -280,90 +239,6 @@ Union: <for test>
          MockExchangeReceiver)";
         ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
     }
-
-    // only join + ExchangeReceiver
-    DAGRequestBuilder receiver1 = context.receive("sender_l");
-    DAGRequestBuilder receiver2 = context.receive("sender_r");
-    DAGRequestBuilder receiver3 = context.receive("sender_l");
-    DAGRequestBuilder receiver4 = context.receive("sender_r");
-
-    request = receiver1.join(
-                           receiver2.join(
-                               receiver3.join(receiver4,
-                                              {col("join_c")},
-                                              ASTTableJoin::Kind::Left),
-                               {col("join_c")},
-                               ASTTableJoin::Kind::Left),
-                           {col("join_c")},
-                           ASTTableJoin::Kind::Left)
-                  .build(context);
-    {
-        String expected = R"(
-CreatingSets
- Union: <for join>
-  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = exchange_receiver_3>, join_kind = Left
-   Expression: <append join key and join filters for build side>
-    Expression: <final projection>
-     MockExchangeReceiver
- Union x 2: <for join>
-  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = Join_4>, join_kind = Left
-   Expression: <append join key and join filters for build side>
-    Expression: <final projection>
-     Expression: <remove useless column after join>
-      HashJoinProbe: <join probe, join_executor_id = Join_4>
-       Expression: <final projection>
-        MockExchangeReceiver
- Union: <for test>
-  Expression x 10: <final projection>
-   Expression: <remove useless column after join>
-    HashJoinProbe: <join probe, join_executor_id = Join_6>
-     Expression: <final projection>
-      MockExchangeReceiver)";
-        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
-    }
-
-    // join + receiver + sender
-    // TODO: Find a way to write the request easier.
-    DAGRequestBuilder receiver5 = context.receive("sender_l");
-    DAGRequestBuilder receiver6 = context.receive("sender_r");
-    DAGRequestBuilder receiver7 = context.receive("sender_l");
-    DAGRequestBuilder receiver8 = context.receive("sender_r");
-    request = receiver5.join(
-                           receiver6.join(
-                               receiver7.join(receiver8,
-                                              {col("join_c")},
-                                              ASTTableJoin::Kind::Left),
-                               {col("join_c")},
-                               ASTTableJoin::Kind::Left),
-                           {col("join_c")},
-                           ASTTableJoin::Kind::Left)
-                  .exchangeSender(tipb::PassThrough)
-                  .build(context);
-    {
-        String expected = R"(
-CreatingSets
- Union: <for join>
-  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = exchange_receiver_3>, join_kind = Left
-   Expression: <append join key and join filters for build side>
-    Expression: <final projection>
-     MockExchangeReceiver
- Union x 2: <for join>
-  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = Join_4>, join_kind = Left
-   Expression: <append join key and join filters for build side>
-    Expression: <final projection>
-     Expression: <remove useless column after join>
-      HashJoinProbe: <join probe, join_executor_id = Join_4>
-       Expression: <final projection>
-        MockExchangeReceiver
- Union: <for test>
-  MockExchangeSender x 10
-   Expression: <final projection>
-    Expression: <remove useless column after join>
-     HashJoinProbe: <join probe, join_executor_id = Join_6>
-      Expression: <final projection>
-       MockExchangeReceiver)";
-        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
-    }
 }
 CATCH
 
@@ -447,5 +322,250 @@ Union: <for test>
 }
 CATCH
 
+TEST_F(InterpreterExecuteTest, Join)
+try
+{
+    // TODO: Find a way to write the request easier.
+    {
+        // Join Source.
+        DAGRequestBuilder table1 = context.scan("test_db", "r_table");
+        DAGRequestBuilder table2 = context.scan("test_db", "l_table");
+        DAGRequestBuilder table3 = context.scan("test_db", "r_table");
+        DAGRequestBuilder table4 = context.scan("test_db", "l_table");
+
+        auto request = table1.join(
+                                 table2.join(
+                                     table3.join(table4,
+                                                 {col("join_c")},
+                                                 ASTTableJoin::Kind::Left),
+                                     {col("join_c")},
+                                     ASTTableJoin::Kind::Left),
+                                 {col("join_c")},
+                                 ASTTableJoin::Kind::Left)
+                           .build(context);
+
+        String expected = R"(
+CreatingSets
+ Union: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = table_scan_3>, join_kind = Left
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     MockTableScan
+ Union x 2: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = Join_4>, join_kind = Left
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     Expression: <remove useless column after join>
+      HashJoinProbe: <join probe, join_executor_id = Join_4>
+       Expression: <final projection>
+        MockTableScan
+ Union: <for test>
+  Expression x 10: <final projection>
+   Expression: <remove useless column after join>
+    HashJoinProbe: <join probe, join_executor_id = Join_6>
+     Expression: <final projection>
+      MockTableScan)";
+        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
+    }
+
+    {
+        // only join + ExchangeReceiver
+        DAGRequestBuilder receiver1 = context.receive("sender_l");
+        DAGRequestBuilder receiver2 = context.receive("sender_r");
+        DAGRequestBuilder receiver3 = context.receive("sender_l");
+        DAGRequestBuilder receiver4 = context.receive("sender_r");
+
+        auto request = receiver1.join(
+                                    receiver2.join(
+                                        receiver3.join(receiver4,
+                                                       {col("join_c")},
+                                                       ASTTableJoin::Kind::Left),
+                                        {col("join_c")},
+                                        ASTTableJoin::Kind::Left),
+                                    {col("join_c")},
+                                    ASTTableJoin::Kind::Left)
+                           .build(context);
+
+        String expected = R"(
+CreatingSets
+ Union: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = exchange_receiver_3>, join_kind = Left
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     MockExchangeReceiver
+ Union x 2: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = Join_4>, join_kind = Left
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     Expression: <remove useless column after join>
+      HashJoinProbe: <join probe, join_executor_id = Join_4>
+       Expression: <final projection>
+        MockExchangeReceiver
+ Union: <for test>
+  Expression x 10: <final projection>
+   Expression: <remove useless column after join>
+    HashJoinProbe: <join probe, join_executor_id = Join_6>
+     Expression: <final projection>
+      MockExchangeReceiver)";
+        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
+    }
+
+    {
+        // join + receiver + sender
+        DAGRequestBuilder receiver1 = context.receive("sender_l");
+        DAGRequestBuilder receiver2 = context.receive("sender_r");
+        DAGRequestBuilder receiver3 = context.receive("sender_l");
+        DAGRequestBuilder receiver4 = context.receive("sender_r");
+
+        auto request = receiver1.join(
+                                    receiver2.join(
+                                        receiver3.join(receiver4,
+                                                       {col("join_c")},
+                                                       ASTTableJoin::Kind::Left),
+                                        {col("join_c")},
+                                        ASTTableJoin::Kind::Left),
+                                    {col("join_c")},
+                                    ASTTableJoin::Kind::Left)
+                           .exchangeSender(tipb::PassThrough)
+                           .build(context);
+
+        String expected = R"(
+CreatingSets
+ Union: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = exchange_receiver_3>, join_kind = Left
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     MockExchangeReceiver
+ Union x 2: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = Join_4>, join_kind = Left
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     Expression: <remove useless column after join>
+      HashJoinProbe: <join probe, join_executor_id = Join_4>
+       Expression: <final projection>
+        MockExchangeReceiver
+ Union: <for test>
+  MockExchangeSender x 10
+   Expression: <final projection>
+    Expression: <remove useless column after join>
+     HashJoinProbe: <join probe, join_executor_id = Join_6>
+      Expression: <final projection>
+       MockExchangeReceiver)";
+        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
+    }
+}
+CATCH
+
+TEST_F(InterpreterExecuteTest, JoinThenAgg)
+try
+{
+    {
+        // Left Join.
+        DAGRequestBuilder table1 = context.scan("test_db", "r_table");
+        DAGRequestBuilder table2 = context.scan("test_db", "l_table");
+
+        auto request = table1.join(
+                                 table2,
+                                 {col("join_c")},
+                                 ASTTableJoin::Kind::Left)
+                           .aggregation({Max(col("r_a"))}, {col("join_c")})
+                           .build(context);
+        String expected = R"(
+CreatingSets
+ Union: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = table_scan_1>, join_kind = Left
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     MockTableScan
+ Union: <for test>
+  Expression x 10: <final projection>
+   SharedQuery: <restore concurrency>
+    ParallelAggregating, max_threads: 10, final: true
+     Expression x 10: <before aggregation>
+      Expression: <remove useless column after join>
+       HashJoinProbe: <join probe, join_executor_id = Join_2>
+        Expression: <final projection>
+         MockTableScan)";
+        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
+    }
+
+    {
+        // Right Join
+        DAGRequestBuilder table1 = context.scan("test_db", "r_table");
+        DAGRequestBuilder table2 = context.scan("test_db", "l_table");
+
+        auto request = table1.join(
+                                 table2,
+                                 {col("join_c")},
+                                 ASTTableJoin::Kind::Right)
+                           .aggregation({Max(col("r_a"))}, {col("join_c")})
+                           .build(context);
+        String expected = R"(
+CreatingSets
+ Union: <for join>
+  HashJoinBuildBlockInputStream x 10: <join build, build_side_root_executor_id = table_scan_1>, join_kind = Right
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     MockTableScan
+ Union: <for test>
+  Expression x 10: <final projection>
+   SharedQuery: <restore concurrency>
+    ParallelAggregating, max_threads: 10, final: true
+     Expression x 10: <before aggregation>
+      Expression: <remove useless column after join>
+       HashJoinProbe: <join probe, join_executor_id = Join_2>
+        Expression: <append join key and join filters for probe side>
+         Expression: <final projection>
+          MockTableScan
+     Expression x 10: <before aggregation>
+      Expression: <remove useless column after join>
+       NonJoined: <add stream with non_joined_data if full_or_right_join>)";
+        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 10);
+    }
+
+    {
+        // Right join + receiver + sender
+        DAGRequestBuilder receiver1 = context.receive("sender_l");
+        DAGRequestBuilder receiver2 = context.receive("sender_r");
+
+        auto request = receiver1.join(
+                                    receiver2,
+                                    {col("join_c")},
+                                    ASTTableJoin::Kind::Right)
+                           .aggregation({Sum(col("r_a"))}, {col("join_c")})
+                           .exchangeSender(tipb::PassThrough)
+                           .limit(10)
+                           .build(context);
+        String expected = R"(
+CreatingSets
+ Union: <for join>
+  HashJoinBuildBlockInputStream x 20: <join build, build_side_root_executor_id = exchange_receiver_1>, join_kind = Right
+   Expression: <append join key and join filters for build side>
+    Expression: <final projection>
+     MockExchangeReceiver
+ Union: <for test>
+  MockExchangeSender x 20
+   SharedQuery: <restore concurrency>
+    Limit, limit = 10
+     Union: <for partial limit>
+      Limit x 20, limit = 10
+       Expression: <final projection>
+        Expression: <before order and select>
+         SharedQuery: <restore concurrency>
+          ParallelAggregating, max_threads: 20, final: true
+           Expression x 20: <before aggregation>
+            Expression: <remove useless column after join>
+             HashJoinProbe: <join probe, join_executor_id = Join_2>
+              Expression: <append join key and join filters for probe side>
+               Expression: <final projection>
+                MockExchangeReceiver
+           Expression x 20: <before aggregation>
+            Expression: <remove useless column after join>
+             NonJoined: <add stream with non_joined_data if full_or_right_join>)";
+        ASSERT_BLOCKINPUTSTREAM_EQAUL(expected, request, 20);
+    }
+}
+CATCH
+
 } // namespace tests
 } // namespace DB
diff --git a/dbms/src/Interpreters/InterpreterSelectQuery.cpp b/dbms/src/Interpreters/InterpreterSelectQuery.cpp
index fe8f04427a0..3514f915626 100644
--- a/dbms/src/Interpreters/InterpreterSelectQuery.cpp
+++ b/dbms/src/Interpreters/InterpreterSelectQuery.cpp
@@ -512,13 +512,13 @@ void InterpreterSelectQuery::executeImpl(Pipeline & pipeline, const BlockInputSt
             {
                 const auto & join = static_cast<const ASTTableJoin &>(*query.join()->table_join);
                 if (join.kind == ASTTableJoin::Kind::Full || join.kind == ASTTableJoin::Kind::Right)
-                    pipeline.stream_with_non_joined_data = expressions.before_join->createStreamWithNonJoinedDataIfFullOrRightJoin(
+                    pipeline.streams_with_non_joined_data.push_back(expressions.before_join->createStreamWithNonJoinedDataIfFullOrRightJoin(
                         pipeline.firstStream()->getHeader(),
                         0,
                         1,
-                        settings.max_block_size);
+                        settings.max_block_size));
 
-                for (auto & stream : pipeline.streams) /// Applies to all sources except stream_with_non_joined_data.
+                for (auto & stream : pipeline.streams) /// Applies to all sources except streams_with_non_joined_data.
                     stream = std::make_shared<ExpressionBlockInputStream>(stream, expressions.before_join, /*req_id=*/"");
             }
 
@@ -603,7 +603,7 @@ void InterpreterSelectQuery::executeImpl(Pipeline & pipeline, const BlockInputSt
             if (need_second_distinct_pass
                 || query.limit_length
                 || query.limit_by_expression_list
-                || pipeline.stream_with_non_joined_data)
+                || !pipeline.streams_with_non_joined_data.empty())
             {
                 need_merge_streams = true;
             }
@@ -987,11 +987,11 @@ void InterpreterSelectQuery::executeAggregation(Pipeline & pipeline, const Expre
     Aggregator::Params params(header, keys, aggregates, overflow_row, settings.max_rows_to_group_by, settings.group_by_overflow_mode, allow_to_use_two_level_group_by ? settings.group_by_two_level_threshold : SettingUInt64(0), allow_to_use_two_level_group_by ? settings.group_by_two_level_threshold_bytes : SettingUInt64(0), settings.max_bytes_before_external_group_by, settings.empty_result_for_aggregation_by_empty_set, context.getTemporaryPath());
 
     /// If there are several sources, then we perform parallel aggregation
-    if (pipeline.streams.size() > 1)
+    if (pipeline.streams.size() > 1 || pipeline.streams_with_non_joined_data.size() > 1)
     {
-        pipeline.firstStream() = std::make_shared<ParallelAggregatingBlockInputStream>(
+        auto stream = std::make_shared<ParallelAggregatingBlockInputStream>(
             pipeline.streams,
-            pipeline.stream_with_non_joined_data,
+            pipeline.streams_with_non_joined_data,
             params,
             file_provider,
             final,
@@ -1001,19 +1001,21 @@ void InterpreterSelectQuery::executeAggregation(Pipeline & pipeline, const Expre
                 : static_cast<size_t>(settings.max_threads),
             /*req_id=*/"");
 
-        pipeline.stream_with_non_joined_data = nullptr;
         pipeline.streams.resize(1);
+        pipeline.streams_with_non_joined_data.clear();
+        pipeline.firstStream() = std::move(stream);
     }
     else
     {
         BlockInputStreams inputs;
         if (!pipeline.streams.empty())
             inputs.push_back(pipeline.firstStream());
-        else
-            pipeline.streams.resize(1);
 
-        if (pipeline.stream_with_non_joined_data)
-            inputs.push_back(pipeline.stream_with_non_joined_data);
+        if (!pipeline.streams_with_non_joined_data.empty())
+            inputs.push_back(pipeline.streams_with_non_joined_data.at(0));
+
+        pipeline.streams.resize(1);
+        pipeline.streams_with_non_joined_data.clear();
 
         pipeline.firstStream() = std::make_shared<AggregatingBlockInputStream>(
             std::make_shared<ConcatBlockInputStream>(inputs, /*req_id=*/""),
@@ -1021,8 +1023,6 @@ void InterpreterSelectQuery::executeAggregation(Pipeline & pipeline, const Expre
             file_provider,
             final,
             /*req_id=*/"");
-
-        pipeline.stream_with_non_joined_data = nullptr;
     }
 }
 
@@ -1244,21 +1244,33 @@ void InterpreterSelectQuery::executeDistinct(Pipeline & pipeline, bool before_or
 
 void InterpreterSelectQuery::executeUnion(Pipeline & pipeline)
 {
-    /// If there are still several streams, then we combine them into one
-    if (pipeline.hasMoreThanOneStream())
+    switch (pipeline.streams.size() + pipeline.streams_with_non_joined_data.size())
     {
-        pipeline.firstStream() = std::make_shared<UnionBlockInputStream<>>(
+    case 0:
+        break;
+    case 1:
+    {
+        if (pipeline.streams.size() == 1)
+            break;
+        // streams_with_non_joined_data's size is 1.
+        pipeline.streams.push_back(pipeline.streams_with_non_joined_data.at(0));
+        pipeline.streams_with_non_joined_data.clear();
+        break;
+    }
+    default:
+    {
+        BlockInputStreamPtr stream = std::make_shared<UnionBlockInputStream<>>(
             pipeline.streams,
-            pipeline.stream_with_non_joined_data,
+            pipeline.streams_with_non_joined_data,
             max_streams,
             /*req_id=*/"");
-        pipeline.stream_with_non_joined_data = nullptr;
+        ;
+
         pipeline.streams.resize(1);
+        pipeline.streams_with_non_joined_data.clear();
+        pipeline.firstStream() = std::move(stream);
+        break;
     }
-    else if (pipeline.stream_with_non_joined_data)
-    {
-        pipeline.streams.push_back(pipeline.stream_with_non_joined_data);
-        pipeline.stream_with_non_joined_data = nullptr;
     }
 }
 
diff --git a/dbms/src/Interpreters/InterpreterSelectQuery.h b/dbms/src/Interpreters/InterpreterSelectQuery.h
index 474ace7ee84..d1bcec2a3dd 100644
--- a/dbms/src/Interpreters/InterpreterSelectQuery.h
+++ b/dbms/src/Interpreters/InterpreterSelectQuery.h
@@ -95,7 +95,7 @@ class InterpreterSelectQuery : public IInterpreter
           * It has a special meaning, since reading from it should be done after reading from the main streams.
           * It is appended to the main streams in UnionBlockInputStream or ParallelAggregatingBlockInputStream.
           */
-        BlockInputStreamPtr stream_with_non_joined_data;
+        BlockInputStreams streams_with_non_joined_data;
 
         BlockInputStreamPtr & firstStream() { return streams.at(0); }
 
@@ -105,13 +105,13 @@ class InterpreterSelectQuery : public IInterpreter
             for (auto & stream : streams)
                 transform(stream);
 
-            if (stream_with_non_joined_data)
-                transform(stream_with_non_joined_data);
+            for (auto & stream : streams_with_non_joined_data)
+                transform(stream);
         }
 
         bool hasMoreThanOneStream() const
         {
-            return streams.size() + (stream_with_non_joined_data ? 1 : 0) > 1;
+            return streams.size() + streams_with_non_joined_data.size() > 1;
         }
     };
 
diff --git a/dbms/src/Interpreters/InterpreterSelectWithUnionQuery.cpp b/dbms/src/Interpreters/InterpreterSelectWithUnionQuery.cpp
index 5e73b1e5f3e..076c290cc9d 100644
--- a/dbms/src/Interpreters/InterpreterSelectWithUnionQuery.cpp
+++ b/dbms/src/Interpreters/InterpreterSelectWithUnionQuery.cpp
@@ -224,7 +224,7 @@ BlockIO InterpreterSelectWithUnionQuery::execute()
     }
     else
     {
-        result_stream = std::make_shared<UnionBlockInputStream<>>(nested_streams, nullptr, settings.max_threads, /*req_id=*/"");
+        result_stream = std::make_shared<UnionBlockInputStream<>>(nested_streams, BlockInputStreams{}, settings.max_threads, /*req_id=*/"");
         nested_streams.clear();
     }
 
diff --git a/dbms/src/TestUtils/mockExecutor.h b/dbms/src/TestUtils/mockExecutor.h
index 1f78a3278af..1f7bd2322fa 100644
--- a/dbms/src/TestUtils/mockExecutor.h
+++ b/dbms/src/TestUtils/mockExecutor.h
@@ -83,7 +83,7 @@ class DAGRequestBuilder
 
     DAGRequestBuilder & exchangeSender(tipb::ExchangeType exchange_type);
 
-    // Currentlt only support inner join, left join and right join.
+    // Currently only support inner join, left join and right join.
     // TODO support more types of join.
     DAGRequestBuilder & join(const DAGRequestBuilder & right, MockAstVec exprs);
     DAGRequestBuilder & join(const DAGRequestBuilder & right, MockAstVec exprs, ASTTableJoin::Kind kind);
@@ -175,6 +175,7 @@ MockWindowFrame buildDefaultRowsFrame();
 #define Or(expr1, expr2) makeASTFunction("or", (expr1), (expr2))
 #define NOT(expr) makeASTFunction("not", (expr))
 #define Max(expr) makeASTFunction("max", (expr))
+#define Sum(expr) makeASTFunction("sum", (expr))
 /// Window functions
 #define RowNumber() makeASTFunction("RowNumber")
 #define Rank() makeASTFunction("Rank")