Merge branch 'master' into actions-dag-f14

.gitmodules

@@ -190,3 +190,6 @@
 	path = contrib/croaring
 	url = https://github.com/RoaringBitmap/CRoaring
 	branch = v0.2.66
+[submodule "contrib/miniselect"]
+	path = contrib/miniselect
+	url = https://github.com/danlark1/miniselect

CMakeLists.txt

@@ -445,6 +445,7 @@ include (cmake/find/brotli.cmake)
 include (cmake/find/protobuf.cmake)
 include (cmake/find/grpc.cmake)
 include (cmake/find/pdqsort.cmake)
+include (cmake/find/miniselect.cmake)
 include (cmake/find/hdfs3.cmake) # uses protobuf
 include (cmake/find/poco.cmake)
 include (cmake/find/curl.cmake)

cmake/find/miniselect.cmake

@@ -0,0 +1,2 @@
+set(MINISELECT_INCLUDE_DIR ${ClickHouse_SOURCE_DIR}/contrib/miniselect/include)
+message(STATUS "Using miniselect: ${MINISELECT_INCLUDE_DIR}")

contrib/miniselect

@@ -0,0 +1 @@
+Subproject commit be0af6bd0b6eb044d1acc4f754b229972d99903a

docker/test/fasttest/run.sh

@@ -127,7 +127,7 @@ function clone_submodules
 (
 cd "$FASTTEST_SOURCE"
 
-SUBMODULES_TO_UPDATE=(contrib/boost contrib/zlib-ng contrib/libxml2 contrib/poco contrib/libunwind contrib/ryu contrib/fmtlib contrib/base64 contrib/cctz contrib/libcpuid contrib/double-conversion contrib/libcxx contrib/libcxxabi contrib/libc-headers contrib/lz4 contrib/zstd contrib/fastops contrib/rapidjson contrib/re2 contrib/sparsehash-c11 contrib/croaring)
+SUBMODULES_TO_UPDATE=(contrib/boost contrib/zlib-ng contrib/libxml2 contrib/poco contrib/libunwind contrib/ryu contrib/fmtlib contrib/base64 contrib/cctz contrib/libcpuid contrib/double-conversion contrib/libcxx contrib/libcxxabi contrib/libc-headers contrib/lz4 contrib/zstd contrib/fastops contrib/rapidjson contrib/re2 contrib/sparsehash-c11 contrib/croaring contrib/miniselect)
 
 git submodule sync
 git submodule update --init --recursive "${SUBMODULES_TO_UPDATE[@]}"

docker/test/stress/stress

@@ -17,13 +17,6 @@ def get_skip_list_cmd(path):
     return ''
 
 
-def run_perf_test(cmd, xmls_path, output_folder):
-    output_path = os.path.join(output_folder, "perf_stress_run.txt")
-    f = open(output_path, 'w')
-    p = Popen("{} --skip-tags=long --recursive --input-files {}".format(cmd, xmls_path), shell=True, stdout=f, stderr=f)
-    return p
-
-
 def get_options(i):
     options = ""
     if 0 < i:
@@ -75,8 +68,6 @@ if __name__ == "__main__":
 
     args = parser.parse_args()
     func_pipes = []
-    perf_process = None
-    perf_process = run_perf_test(args.perf_test_cmd, args.perf_test_xml_path, args.output_folder)
     func_pipes = run_func_test(args.test_cmd, args.output_folder, args.num_parallel, args.skip_func_tests, args.global_time_limit)
 
     logging.info("Will wait functests to finish")

docs/en/development/adding_test_queries.md

@@ -0,0 +1,141 @@
+# How to add test queries to ClickHouse CI
+
+ClickHouse has hundreds (or even thousands) of features. Every commit get checked by a complex set of tests containing many thousands of test cases.
+
+The core functionality is very well tested, but some corner-cases and different combinations of features can be uncovered with ClickHouse CI.
+
+Most of the bugs/regressions we see happen in that 'grey area' where test coverage is poor. 
+
+And we are very interested in covering most of the possible scenarios and feature combinations used in real life by tests. 
+
+## Why adding tests 
+
+Why/when you should add a test case into ClickHouse code:
+1) you use some complicated scenarios / feature combinations / you have some corner case which is probably not widely used
+2) you see that certain behavior gets changed between version w/o notifications in the changelog
+3) you just want to help to improve ClickHouse quality and ensure the features you use will not be broken in the future releases
+4) once the test is added/accepted, you can be sure the corner case you check will never be accidentally broken.
+5) you will be a part of great open-source community
+6) your name will be visible in the `system.contributors` table!
+7) you will make a world bit better :) 
+
+### Steps to do
+
+#### Prerequisite 
+
+I assume you run some Linux machine (you can use docker / virtual machines on other OS) and any modern browser / internet connection, and you have some basic Linux & SQL skills. 
+
+Any highly specialized knowledge is not needed (so you don't need to know C++ or know something about how ClickHouse CI works).
+
+
+#### Preparation 
+
+1) [create GitHub account](https://github.com/join) (if you haven't one yet)
+2) [setup git](https://docs.github.com/en/free-pro-team@latest/github/getting-started-with-github/set-up-git)
+```bash
+# for Ubuntu
+sudo apt-get update
+sudo apt-get install git
+
+git config --global user.name "John Doe" # fill with your name
+git config --global user.email "email@example.com" # fill with your email
+
+```
+3) [fork ClickHouse project](https://docs.github.com/en/free-pro-team@latest/github/getting-started-with-github/fork-a-repo) - just open [https://github.com/ClickHouse/ClickHouse](https://github.com/ClickHouse/ClickHouse) and press fork button in the top right corner:
+![fork repo](https://github-images.s3.amazonaws.com/help/bootcamp/Bootcamp-Fork.png)
+
+4) clone your fork to some folder on your PC, for example, `~/workspace/ClickHouse`
+```
+mkdir ~/workspace && cd ~/workspace
+git clone https://github.com/< your GitHub username>/ClickHouse
+cd ClickHouse
+git remote add upstream https://github.com/ClickHouse/ClickHouse
+```
+
+#### New branch for the test
+
+1) create a new branch from the latest clickhouse master 
+```
+cd ~/workspace/ClickHouse
+git fetch upstream
+git checkout -b name_for_a_branch_with_my_test upstream/master 
+```
+
+#### Install & run clickhouse 
+
+1) install `clickhouse-server` (follow [official docs](https://clickhouse.tech/docs/en/getting-started/install/))
+2) install test configurations (it will use Zookeeper mock implementation and adjust some settings) 
+```
+cd ~/workspace/ClickHouse/tests/config
+sudo ./install.sh
+```
+3) run clickhouse-server
+```
+sudo systemctl restart clickhouse-server
+```
+
+#### Creating the test file 
+
+
+1) find the number for your test - find the file with the biggest number in `tests/queries/0_stateless/`
+
+```sh
+$ cd ~/workspace/ClickHouse
+$ ls tests/queries/0_stateless/[0-9]*.reference | tail -n 1
+tests/queries/0_stateless/01520_client_print_query_id.reference
+```
+Currently, the last number for the test is `01520`, so my test will have the number `01521`
+
+2) create an SQL file with the next number and name of the feature you test 
+
+```sh
+touch tests/queries/0_stateless/01521_dummy_test.sql
+```
+
+3) edit SQL file with your favorite editor (see hint of creating tests below)
+```sh
+vim tests/queries/0_stateless/01521_dummy_test.sql
+```
+
+
+4) run the test, and put the result of that into the reference file:
+```
+clickhouse-client -nmT < tests/queries/0_stateless/01521_dummy_test.sql | tee tests/queries/0_stateless/01521_dummy_test.reference
+```
+
+5) ensure everything is correct, if the test output is incorrect (due to some bug for example), adjust the reference file using text editor.
+
+####  How create good test
+
+- test should be
+	- minimal - create only tables related to tested functionality, remove unrelated columns and parts of query
+	- fast - should not take longer than few seconds (better subseconds)
+	- correct - fails then feature is not working
+        - deteministic
+	- isolated / stateless 
+		- don't rely on some environment things
+		- don't rely on timing when possible 
+- try to cover corner cases (zeros / Nulls / empty sets / throwing exceptions) 
+- to test that query return errors, you can put special comment after the query: `-- { serverError 60 }` or `-- { clientError 20 }`
+- don't switch databases (unless necessary)
+- you can create several table replicas on the same node if needed
+- you can use one of the test cluster definitions when needed (see system.clusters)
+- use `number` / `numbers_mt` / `zeros` / `zeros_mt` and similar for queries / to initialize data when appliable
+- clean up the created objects after test and before the test (DROP IF EXISTS) - in case of some dirty state 
+- prefer sync mode of operations (mutations, merges, etc.)
+- use other SQL files in the `0_stateless` folder as an example
+- ensure the feature / feature combination you want to tests is not covered yet with existsing tests
+
+#### Commit / push / create PR.
+
+1) commit & push your changes 
+```sh
+cd ~/workspace/ClickHouse
+git add tests/queries/0_stateless/01521_dummy_test.sql
+git add tests/queries/0_stateless/01521_dummy_test.reference
+git commit # use some nice commit message when possible
+git push origin HEAD
+```
+2) use a link which was shown during the push, to create a PR into the main repo
+3) adjust the PR title and contents, in `Changelog category (leave one)` keep 
+`Build/Testing/Packaging Improvement`, fill the rest of the fields if you want. 

docs/en/operations/settings/settings.md

@@ -384,7 +384,7 @@ Possible values:
 
 -   `'basic'` — Use basic parser.
 
-    ClickHouse can parse only the basic `YYYY-MM-DD HH:MM:SS` format. For example, `'2019-08-20 10:18:56'`.
+    ClickHouse can parse only the basic `YYYY-MM-DD HH:MM:SS` or `YYYY-MM-DD` format. For example, `'2019-08-20 10:18:56'` or `2019-08-20`.
 
 Default value: `'basic'`.
 

docs/en/sql-reference/data-types/date.md

@@ -3,10 +3,45 @@ toc_priority: 47
 toc_title: Date
 ---
 
-# Date {#date}
+# Date {#data_type-date}
 
 A date. Stored in two bytes as the number of days since 1970-01-01 (unsigned). Allows storing values from just after the beginning of the Unix Epoch to the upper threshold defined by a constant at the compilation stage (currently, this is until the year 2106, but the final fully-supported year is 2105).
 
 The date value is stored without the time zone.
 
+## Examples {#examples}
+
+**1.** Creating a table with a `DateTime`-type column and inserting data into it:
+
+``` sql
+CREATE TABLE dt
+(
+    `timestamp` Date,
+    `event_id` UInt8
+)
+ENGINE = TinyLog;
+```
+
+``` sql
+INSERT INTO dt Values (1546300800, 1), ('2019-01-01', 2);
+```
+
+``` sql
+SELECT * FROM dt;
+```
+
+``` text
+┌──timestamp─┬─event_id─┐
+│ 2019-01-01 │        1 │
+│ 2019-01-01 │        2 │
+└────────────┴──────────┘
+```
+
+## See Also {#see-also}
+
+-   [Functions for working with dates and times](../../sql-reference/functions/date-time-functions.md)
+-   [Operators for working with dates and times](../../sql-reference/operators/index.md#operators-datetime)
+-   [`DateTime` data type](../../sql-reference/data-types/datetime.md)
+
+
 [Original article](https://clickhouse.tech/docs/en/data_types/date/) <!--hide-->

docs/en/sql-reference/table-functions/null.md

@@ -5,7 +5,7 @@ toc_title: null function
 
 # null {#null-function}
 
-Accepts an inserted data of the specified structure and immediately drops it away. The function is used for convenience writing tests and demonstrations.
+Creates a temporary table of the specified structure with the [Null](../../engines/table-engines/special/null.md) table engine. According to the `Null`-engine properties, the table data is ignored and the table itself is immediately droped right after the query execution. The function is used for the convenience of test writing and demonstrations.
 
 **Syntax** 
 
@@ -19,7 +19,7 @@ null('structure')
 
 **Returned value**
 
-A table with the specified structure, which is dropped right after the query execution.
+A temporary `Null`-engine table with the specified structure.
 
 **Example**
 
@@ -36,6 +36,8 @@ INSERT INTO t SELECT * FROM numbers_mt(1000000000);
 DROP TABLE IF EXISTS t;
 ```
 
-See also: format **Null**.
+See also: 
+
+-   [Null table engine](../../engines/table-engines/special/null.md)
 
 [Original article](https://clickhouse.tech/docs/en/sql-reference/table-functions/null/) <!--hide-->

docs/ru/sql-reference/table-functions/null.md

@@ -0,0 +1,43 @@
+---
+toc_priority: 53
+toc_title: null функция
+---
+
+# null {#null-function}
+
+Создает временную таблицу указанной структуры с движком [Null](../../engines/table-engines/special/null.md). В соответствии со свойствами движка, данные в таблице игнорируются, а сама таблица удаляется сразу после выполнения запроса. Функция используется для удобства написания тестов и демонстрационных примеров.
+
+**Синтаксис** 
+
+``` sql
+null('structure')
+```
+
+**Параметр** 
+
+-   `structure` — список колонок и их типов. [String](../../sql-reference/data-types/string.md).
+
+**Возвращаемое значение**
+
+Временная таблица указанной структуры с движком `Null`.
+
+**Пример**
+
+Один запрос с функцией `null`:
+
+``` sql
+INSERT INTO function null('x UInt64') SELECT * FROM numbers_mt(1000000000);
+```
+заменяет три запроса:
+
+```sql
+CREATE TABLE t (x UInt64) ENGINE = Null;
+INSERT INTO t SELECT * FROM numbers_mt(1000000000);
+DROP TABLE IF EXISTS t;
+```
+
+См. также: 
+
+-   [Движок таблиц Null](../../engines/table-engines/special/null.md)
+
+[Original article](https://clickhouse.tech/docs/en/sql-reference/table-functions/null/) <!--hide-->

programs/server/Server.cpp

@@ -568,6 +568,8 @@ int Server::main(const std::vector<std::string> & /*args*/)
             if (config->has("zookeeper"))
                 global_context->reloadZooKeeperIfChanged(config);
 
+            global_context->reloadAuxiliaryZooKeepersConfigIfChanged(config);
+
             global_context->updateStorageConfiguration(*config);
         },
         /* already_loaded = */ true);

src/AggregateFunctions/QuantileExact.h

@@ -8,6 +8,7 @@
 #include <Common/NaNUtils.h>
 #include <Common/PODArray.h>
 
+#include <miniselect/floyd_rivest_select.h>
 
 namespace DB
 {
@@ -87,7 +88,7 @@ struct QuantileExact : QuantileExactBase<Value, QuantileExact<Value>>
         {
             size_t n = level < 1 ? level * array.size() : (array.size() - 1);
 
-            std::nth_element(array.begin(), array.begin() + n, array.end()); /// NOTE You can think of the radix-select algorithm.
+            miniselect::floyd_rivest_select(array.begin(), array.begin() + n, array.end()); /// NOTE You can think of the radix-select algorithm.
             return array[n];
         }
 
@@ -107,7 +108,7 @@ struct QuantileExact : QuantileExactBase<Value, QuantileExact<Value>>
 
                 size_t n = level < 1 ? level * array.size() : (array.size() - 1);
 
-                std::nth_element(array.begin() + prev_n, array.begin() + n, array.end());
+                miniselect::floyd_rivest_select(array.begin() + prev_n, array.begin() + n, array.end());
 
                 result[indices[i]] = array[n];
                 prev_n = n;
@@ -144,7 +145,7 @@ struct QuantileExactExclusive : public QuantileExact<Value>
             else if (n < 1)
                 return static_cast<Float64>(array[0]);
 
-            std::nth_element(array.begin(), array.begin() + n - 1, array.end());
+            miniselect::floyd_rivest_select(array.begin(), array.begin() + n - 1, array.end());
             auto nth_element = std::min_element(array.begin() + n, array.end());
 
             return static_cast<Float64>(array[n - 1]) + (h - n) * static_cast<Float64>(*nth_element - array[n - 1]);
@@ -173,7 +174,7 @@ struct QuantileExactExclusive : public QuantileExact<Value>
                     result[indices[i]] = static_cast<Float64>(array[0]);
                 else
                 {
-                    std::nth_element(array.begin() + prev_n, array.begin() + n - 1, array.end());
+                    miniselect::floyd_rivest_select(array.begin() + prev_n, array.begin() + n - 1, array.end());
                     auto nth_element = std::min_element(array.begin() + n, array.end());
 
                     result[indices[i]] = static_cast<Float64>(array[n - 1]) + (h - n) * static_cast<Float64>(*nth_element - array[n - 1]);
@@ -209,7 +210,7 @@ struct QuantileExactInclusive : public QuantileExact<Value>
             else if (n < 1)
                 return static_cast<Float64>(array[0]);
 
-            std::nth_element(array.begin(), array.begin() + n - 1, array.end());
+            miniselect::floyd_rivest_select(array.begin(), array.begin() + n - 1, array.end());
             auto nth_element = std::min_element(array.begin() + n, array.end());
 
             return static_cast<Float64>(array[n - 1]) + (h - n) * static_cast<Float64>(*nth_element - array[n - 1]);
@@ -236,7 +237,7 @@ struct QuantileExactInclusive : public QuantileExact<Value>
                     result[indices[i]] = static_cast<Float64>(array[0]);
                 else
                 {
-                    std::nth_element(array.begin() + prev_n, array.begin() + n - 1, array.end());
+                    miniselect::floyd_rivest_select(array.begin() + prev_n, array.begin() + n - 1, array.end());
                     auto nth_element = std::min_element(array.begin() + n, array.end());
 
                     result[indices[i]] = static_cast<Float64>(array[n - 1]) + (h - n) * static_cast<Float64>(*nth_element - array[n - 1]);

src/AggregateFunctions/QuantileTiming.h

@@ -7,6 +7,7 @@
 #include <IO/ReadHelpers.h>
 #include <IO/WriteHelpers.h>
 
+#include <miniselect/floyd_rivest_select.h>
 
 namespace DB
 {
@@ -179,7 +180,7 @@ namespace detail
 
                 /// Sorting an array will not be considered a violation of constancy.
                 auto & array = elems;
-                std::nth_element(array.begin(), array.begin() + n, array.end());
+                miniselect::floyd_rivest_select(array.begin(), array.begin() + n, array.end());
                 quantile = array[n];
             }
 
@@ -200,7 +201,7 @@ namespace detail
                     ? level * elems.size()
                     : (elems.size() - 1);
 
-                std::nth_element(array.begin() + prev_n, array.begin() + n, array.end());
+                miniselect::floyd_rivest_select(array.begin() + prev_n, array.begin() + n, array.end());
 
                 result[level_index] = array[n];
                 prev_n = n;

src/CMakeLists.txt

@@ -321,6 +321,7 @@ target_include_directories(clickhouse_common_io PUBLIC ${CMAKE_CURRENT_BINARY_DI
 dbms_target_include_directories(PUBLIC ${CMAKE_CURRENT_BINARY_DIR}/Core/include)
 
 dbms_target_include_directories(SYSTEM BEFORE PUBLIC ${PDQSORT_INCLUDE_DIR})
+dbms_target_include_directories(SYSTEM BEFORE PUBLIC ${MINISELECT_INCLUDE_DIR})
 
 if (ZSTD_LIBRARY)
     dbms_target_link_libraries(PRIVATE ${ZSTD_LIBRARY})

src/Columns/ColumnArray.cpp

@@ -20,6 +20,7 @@
 #include <Common/WeakHash.h>
 #include <Common/HashTable/Hash.h>
 
+#include <miniselect/floyd_rivest_select.h>
 
 namespace DB
 {
@@ -782,7 +783,7 @@ void ColumnArray::getPermutationImpl(size_t limit, Permutation & res, Comparator
     auto less = [&cmp](size_t lhs, size_t rhs){ return cmp(lhs, rhs) < 0; };
 
     if (limit)
-        std::partial_sort(res.begin(), res.begin() + limit, res.end(), less);
+        miniselect::floyd_rivest_partial_sort(res.begin(), res.begin() + limit, res.end(), less);
     else
         std::sort(res.begin(), res.end(), less);
 }
@@ -835,7 +836,7 @@ void ColumnArray::updatePermutationImpl(size_t limit, Permutation & res, EqualRa
 
         /// Since then we are working inside the interval.
 
-        std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less);
+        miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less);
         auto new_first = first;
         for (auto j = first + 1; j < limit; ++j)
         {

src/Columns/ColumnDecimal.cpp

@@ -8,6 +8,7 @@
 
 #include <common/unaligned.h>
 #include <ext/scope_guard.h>
+#include <miniselect/floyd_rivest_select.h>
 
 #include <IO/WriteHelpers.h>
 
@@ -162,10 +163,10 @@ void ColumnDecimal<T>::updatePermutation(bool reverse, size_t limit, int, IColum
     {
         const auto& [first, last] = equal_ranges[i];
         if (reverse)
-            std::partial_sort(res.begin() + first, res.begin() + last, res.begin() + last,
+            std::sort(res.begin() + first, res.begin() + last,
                 [this](size_t a, size_t b) { return data[a] > data[b]; });
         else
-            std::partial_sort(res.begin() + first, res.begin() + last, res.begin() + last,
+            std::sort(res.begin() + first, res.begin() + last,
                 [this](size_t a, size_t b) { return data[a] < data[b]; });
 
         auto new_first = first;
@@ -193,10 +194,10 @@ void ColumnDecimal<T>::updatePermutation(bool reverse, size_t limit, int, IColum
         /// Since then we are working inside the interval.
 
         if (reverse)
-            std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last,
+            miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last,
                 [this](size_t a, size_t b) { return data[a] > data[b]; });
         else
-            std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last,
+            miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last,
                 [this](size_t a, size_t b) { return data[a] < data[b]; });
 
         auto new_first = first;

src/Columns/ColumnDecimal.h

@@ -7,6 +7,7 @@
 #include <Columns/IColumnImpl.h>
 #include <Columns/ColumnVectorHelper.h>
 #include <Core/Field.h>
+#include <miniselect/floyd_rivest_select.h>
 
 
 namespace DB
@@ -253,9 +254,9 @@ protected:
             sort_end = res.begin() + limit;
 
         if (reverse)
-            std::partial_sort(res.begin(), sort_end, res.end(), [this](size_t a, size_t b) { return data[a] > data[b]; });
+            miniselect::floyd_rivest_partial_sort(res.begin(), sort_end, res.end(), [this](size_t a, size_t b) { return data[a] > data[b]; });
         else
-            std::partial_sort(res.begin(), sort_end, res.end(), [this](size_t a, size_t b) { return data[a] < data[b]; });
+            miniselect::floyd_rivest_partial_sort(res.begin(), sort_end, res.end(), [this](size_t a, size_t b) { return data[a] < data[b]; });
     }
 };
 

src/Columns/ColumnFixedString.cpp

@@ -10,6 +10,7 @@
 #include <Common/HashTable/Hash.h>
 
 #include <ext/scope_guard.h>
+#include <miniselect/floyd_rivest_select.h>
 
 #include <DataStreams/ColumnGathererStream.h>
 
@@ -157,9 +158,9 @@ void ColumnFixedString::getPermutation(bool reverse, size_t limit, int /*nan_dir
     if (limit)
     {
         if (reverse)
-            std::partial_sort(res.begin(), res.begin() + limit, res.end(), less<false>(*this));
+            miniselect::floyd_rivest_partial_sort(res.begin(), res.begin() + limit, res.end(), less<false>(*this));
         else
-            std::partial_sort(res.begin(), res.begin() + limit, res.end(), less<true>(*this));
+            miniselect::floyd_rivest_partial_sort(res.begin(), res.begin() + limit, res.end(), less<true>(*this));
     }
     else
     {
@@ -217,9 +218,9 @@ void ColumnFixedString::updatePermutation(bool reverse, size_t limit, int, Permu
         /// Since then we are working inside the interval.
 
         if (reverse)
-            std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less<false>(*this));
+            miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less<false>(*this));
         else
-            std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less<true>(*this));
+            miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less<true>(*this));
 
         auto new_first = first;
         for (auto j = first + 1; j < limit; ++j)

src/Columns/ColumnLowCardinality.cpp

@@ -8,6 +8,7 @@
 #include <Common/WeakHash.h>
 
 #include <ext/scope_guard.h>
+#include <miniselect/floyd_rivest_select.h>
 
 namespace DB
 {
@@ -393,7 +394,7 @@ void ColumnLowCardinality::updatePermutationImpl(size_t limit, Permutation & res
 
         /// Since then we are working inside the interval.
 
-        std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less);
+        miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less);
         auto new_first = first;
 
         for (auto j = first + 1; j < limit; ++j)

src/Columns/ColumnString.cpp

@@ -10,6 +10,7 @@
 
 #include <common/unaligned.h>
 #include <ext/scope_guard.h>
+#include <miniselect/floyd_rivest_select.h>
 
 namespace DB
 {
@@ -313,7 +314,7 @@ void ColumnString::getPermutationImpl(size_t limit, Permutation & res, Comparato
     auto less = [&cmp](size_t lhs, size_t rhs){ return cmp(lhs, rhs) < 0; };
 
     if (limit)
-        std::partial_sort(res.begin(), res.begin() + limit, res.end(), less);
+        miniselect::floyd_rivest_partial_sort(res.begin(), res.begin() + limit, res.end(), less);
     else
         std::sort(res.begin(), res.end(), less);
 }
@@ -365,7 +366,7 @@ void ColumnString::updatePermutationImpl(size_t limit, Permutation & res, EqualR
 
         /// Since then we are working inside the interval.
 
-        std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less);
+        miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less);
 
         size_t new_first = first;
         for (size_t j = first + 1; j < limit; ++j)

src/Columns/ColumnTuple.cpp

@@ -9,7 +9,7 @@
 #include <Common/assert_cast.h>
 #include <Common/WeakHash.h>
 #include <Core/Field.h>
-
+#include <miniselect/floyd_rivest_select.h>
 
 namespace DB
 {
@@ -352,7 +352,7 @@ void ColumnTuple::getPermutationImpl(size_t limit, Permutation & res, LessOperat
 
     if (limit)
     {
-        std::partial_sort(res.begin(), res.begin() + limit, res.end(), less);
+        miniselect::floyd_rivest_partial_sort(res.begin(), res.begin() + limit, res.end(), less);
     }
     else
     {

src/Columns/ColumnVector.cpp

@@ -17,7 +17,7 @@
 #include <ext/bit_cast.h>
 #include <ext/scope_guard.h>
 #include <pdqsort.h>
-
+#include <miniselect/floyd_rivest_select.h>
 
 #ifdef __SSE2__
     #include <emmintrin.h>
@@ -156,9 +156,9 @@ void ColumnVector<T>::getPermutation(bool reverse, size_t limit, int nan_directi
             res[i] = i;
 
         if (reverse)
-            std::partial_sort(res.begin(), res.begin() + limit, res.end(), greater(*this, nan_direction_hint));
+            miniselect::floyd_rivest_partial_sort(res.begin(), res.begin() + limit, res.end(), greater(*this, nan_direction_hint));
         else
-            std::partial_sort(res.begin(), res.begin() + limit, res.end(), less(*this, nan_direction_hint));
+            miniselect::floyd_rivest_partial_sort(res.begin(), res.begin() + limit, res.end(), less(*this, nan_direction_hint));
     }
     else
     {
@@ -254,9 +254,9 @@ void ColumnVector<T>::updatePermutation(bool reverse, size_t limit, int nan_dire
         /// Since then, we are working inside the interval.
 
         if (reverse)
-            std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, greater(*this, nan_direction_hint));
+            miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, greater(*this, nan_direction_hint));
         else
-            std::partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less(*this, nan_direction_hint));
+            miniselect::floyd_rivest_partial_sort(res.begin() + first, res.begin() + limit, res.begin() + last, less(*this, nan_direction_hint));
 
         size_t new_first = first;
         for (size_t j = first + 1; j < limit; ++j)

src/Core/Settings.cpp

@@ -130,4 +130,6 @@ void Settings::checkNoSettingNamesAtTopLevel(const Poco::Util::AbstractConfigura
     }
 }
 
+IMPLEMENT_SETTINGS_TRAITS(FormatFactorySettingsTraits, FORMAT_FACTORY_SETTINGS)
+
 }

src/Core/Settings.h

@@ -399,6 +399,7 @@ class IColumn;
     \
     /** Obsolete settings that do nothing but left for compatibility reasons. Remove each one after half a year of obsolescence. */ \
     \
+    M(UInt64, max_memory_usage_for_all_queries, 0, "Obsolete. Will be removed after 2020-10-20", 0) \
     M(UInt64, multiple_joins_rewriter_version, 0, "Obsolete setting, does nothing. Will be removed after 2021-03-31", 0) \
     M(Bool, experimental_use_processors, true, "Obsolete setting, does nothing. Will be removed after 2020-11-29.", 0) \
     M(Bool, force_optimize_skip_unused_shards_no_nested, false, "Obsolete setting, does nothing. Will be removed after 2020-12-01. Use force_optimize_skip_unused_shards_nesting instead.", 0) \
@@ -514,4 +515,13 @@ struct Settings : public BaseSettings<SettingsTraits>
     static void checkNoSettingNamesAtTopLevel(const Poco::Util::AbstractConfiguration & config, const String & config_path);
 };
 
+/*
+ * User-specified file format settings for File and ULR engines.
+ */
+DECLARE_SETTINGS_TRAITS(FormatFactorySettingsTraits, FORMAT_FACTORY_SETTINGS)
+
+struct FormatFactorySettings : public BaseSettings<FormatFactorySettingsTraits>
+{
+};
+
 }

src/Databases/MySQL/MaterializeMySQLSyncThread.cpp

@@ -623,24 +623,10 @@ void MaterializeMySQLSyncThread::onEvent(Buffers & buffers, const BinlogEventPtr
     else if (receive_event->type() == MYSQL_QUERY_EVENT)
     {
         QueryEvent & query_event = static_cast<QueryEvent &>(*receive_event);
-        flushBuffersData(buffers, metadata);
-
-        try
-        {
-            Context query_context = createQueryContext(global_context);
-            String comment = "Materialize MySQL step 2: execute MySQL DDL for sync data";
-            String event_database = query_event.schema == mysql_database_name ? database_name : "";
-            tryToExecuteQuery(query_prefix + query_event.query, query_context, event_database, comment);
-        }
-        catch (Exception & exception)
-        {
-            tryLogCurrentException(log);
-
-            /// If some DDL query was not successfully parsed and executed
-            /// Then replication may fail on next binlog events anyway
-            if (exception.code() != ErrorCodes::SYNTAX_ERROR)
-                throw;
-        }
+        Position position_before_ddl;
+        position_before_ddl.update(metadata.binlog_position, metadata.binlog_file, metadata.executed_gtid_set);
+        metadata.transaction(position_before_ddl, [&]() { buffers.commit(global_context); });
+        metadata.transaction(client.getPosition(),[&](){ executeDDLAtomic(query_event); });
     }
     else if (receive_event->header.type != HEARTBEAT_EVENT)
     {
@@ -656,6 +642,26 @@ void MaterializeMySQLSyncThread::onEvent(Buffers & buffers, const BinlogEventPtr
     }
 }
 
+void MaterializeMySQLSyncThread::executeDDLAtomic(const QueryEvent & query_event)
+{
+    try
+    {
+        Context query_context = createQueryContext(global_context);
+        String comment = "Materialize MySQL step 2: execute MySQL DDL for sync data";
+        String event_database = query_event.schema == mysql_database_name ? database_name : "";
+        tryToExecuteQuery(query_prefix + query_event.query, query_context, event_database, comment);
+    }
+    catch (Exception & exception)
+    {
+        tryLogCurrentException(log);
+
+        /// If some DDL query was not successfully parsed and executed
+        /// Then replication may fail on next binlog events anyway
+        if (exception.code() != ErrorCodes::SYNTAX_ERROR)
+            throw;
+    }
+}
+
 bool MaterializeMySQLSyncThread::isMySQLSyncThread()
 {
     return getThreadName() == MYSQL_BACKGROUND_THREAD_NAME;

src/Databases/MySQL/MaterializeMySQLSyncThread.h

@@ -100,6 +100,7 @@ private:
 
     std::atomic<bool> sync_quit{false};
     std::unique_ptr<ThreadFromGlobalPool> background_thread_pool;
+    void executeDDLAtomic(const QueryEvent & query_event);
 };
 
 }

src/Formats/FormatFactory.cpp

@@ -40,100 +40,93 @@ const FormatFactory::Creators & FormatFactory::getCreators(const String & name)
     throw Exception("Unknown format " + name, ErrorCodes::UNKNOWN_FORMAT);
 }
 
+FormatSettings getFormatSettings(const Context & context)
+{
+    const auto & settings = context.getSettingsRef();
 
-static FormatSettings getInputFormatSetting(const Settings & settings, const Context & context)
+    return getFormatSettings(context, settings);
+}
+
+template <typename Settings>
+FormatSettings getFormatSettings(const Context & context,
+    const Settings & settings)
 {
     FormatSettings format_settings;
-    format_settings.csv.delimiter = settings.format_csv_delimiter;
-    format_settings.csv.allow_single_quotes = settings.format_csv_allow_single_quotes;
+
+    format_settings.avro.allow_missing_fields = settings.input_format_avro_allow_missing_fields;
+    format_settings.avro.output_codec = settings.output_format_avro_codec;
+    format_settings.avro.output_sync_interval = settings.output_format_avro_sync_interval;
+    format_settings.avro.schema_registry_url = settings.format_avro_schema_registry_url.toString();
     format_settings.csv.allow_double_quotes = settings.format_csv_allow_double_quotes;
-    format_settings.csv.unquoted_null_literal_as_null = settings.input_format_csv_unquoted_null_literal_as_null;
+    format_settings.csv.allow_single_quotes = settings.format_csv_allow_single_quotes;
+    format_settings.csv.crlf_end_of_line = settings.output_format_csv_crlf_end_of_line;
+    format_settings.csv.delimiter = settings.format_csv_delimiter;
     format_settings.csv.empty_as_default = settings.input_format_defaults_for_omitted_fields;
     format_settings.csv.input_format_enum_as_number = settings.input_format_csv_enum_as_number;
-    format_settings.null_as_default = settings.input_format_null_as_default;
-    format_settings.values.interpret_expressions = settings.input_format_values_interpret_expressions;
-    format_settings.values.deduce_templates_of_expressions = settings.input_format_values_deduce_templates_of_expressions;
-    format_settings.values.accurate_types_of_literals = settings.input_format_values_accurate_types_of_literals;
-    format_settings.with_names_use_header = settings.input_format_with_names_use_header;
-    format_settings.skip_unknown_fields = settings.input_format_skip_unknown_fields;
-    format_settings.import_nested_json = settings.input_format_import_nested_json;
+    format_settings.csv.unquoted_null_literal_as_null = settings.input_format_csv_unquoted_null_literal_as_null;
+    format_settings.custom.escaping_rule = settings.format_custom_escaping_rule;
+    format_settings.custom.field_delimiter = settings.format_custom_field_delimiter;
+    format_settings.custom.result_after_delimiter = settings.format_custom_result_after_delimiter;
+    format_settings.custom.result_after_delimiter = settings.format_custom_result_after_delimiter;
+    format_settings.custom.result_before_delimiter = settings.format_custom_result_before_delimiter;
+    format_settings.custom.row_after_delimiter = settings.format_custom_row_after_delimiter;
+    format_settings.custom.row_before_delimiter = settings.format_custom_row_before_delimiter;
+    format_settings.custom.row_between_delimiter = settings.format_custom_row_between_delimiter;
     format_settings.date_time_input_format = settings.date_time_input_format;
+    format_settings.date_time_output_format = settings.date_time_output_format;
+    format_settings.enable_streaming = settings.output_format_enable_streaming;
+    format_settings.import_nested_json = settings.input_format_import_nested_json;
     format_settings.input_allow_errors_num = settings.input_format_allow_errors_num;
     format_settings.input_allow_errors_ratio = settings.input_format_allow_errors_ratio;
-    format_settings.template_settings.resultset_format = settings.format_template_resultset;
-    format_settings.template_settings.row_format = settings.format_template_row;
-    format_settings.template_settings.row_between_delimiter = settings.format_template_rows_between_delimiter;
-    format_settings.tsv.empty_as_default = settings.input_format_tsv_empty_as_default;
-    format_settings.tsv.input_format_enum_as_number = settings.input_format_tsv_enum_as_number;
+    format_settings.json.escape_forward_slashes = settings.output_format_json_escape_forward_slashes;
+    format_settings.json.quote_64bit_integers = settings.output_format_json_quote_64bit_integers;
+    format_settings.json.quote_denormals = settings.output_format_json_quote_denormals;
+    format_settings.null_as_default = settings.input_format_null_as_default;
+    format_settings.parquet.row_group_size = settings.output_format_parquet_row_group_size;
+    format_settings.pretty.charset = settings.output_format_pretty_grid_charset.toString() == "ASCII" ? FormatSettings::Pretty::Charset::ASCII : FormatSettings::Pretty::Charset::UTF8;
+    format_settings.pretty.color = settings.output_format_pretty_color;
+    format_settings.pretty.max_column_pad_width = settings.output_format_pretty_max_column_pad_width;
+    format_settings.pretty.max_rows = settings.output_format_pretty_max_rows;
+    format_settings.pretty.max_value_width = settings.output_format_pretty_max_value_width;
+    format_settings.pretty.output_format_pretty_row_numbers = settings.output_format_pretty_row_numbers;
+    format_settings.regexp.escaping_rule = settings.format_regexp_escaping_rule;
+    format_settings.regexp.regexp = settings.format_regexp;
+    format_settings.regexp.skip_unmatched = settings.format_regexp_skip_unmatched;
     format_settings.schema.format_schema = settings.format_schema;
     format_settings.schema.format_schema_path = context.getFormatSchemaPath();
     format_settings.schema.is_server = context.hasGlobalContext() && (context.getGlobalContext().getApplicationType() == Context::ApplicationType::SERVER);
-    format_settings.custom.result_before_delimiter = settings.format_custom_result_before_delimiter;
-    format_settings.custom.result_after_delimiter = settings.format_custom_result_after_delimiter;
-    format_settings.custom.escaping_rule = settings.format_custom_escaping_rule;
-    format_settings.custom.field_delimiter = settings.format_custom_field_delimiter;
-    format_settings.custom.row_before_delimiter = settings.format_custom_row_before_delimiter;
-    format_settings.custom.row_after_delimiter = settings.format_custom_row_after_delimiter;
-    format_settings.custom.row_between_delimiter = settings.format_custom_row_between_delimiter;
-    format_settings.regexp.regexp = settings.format_regexp;
-    format_settings.regexp.escaping_rule = settings.format_regexp_escaping_rule;
-    format_settings.regexp.skip_unmatched = settings.format_regexp_skip_unmatched;
+    format_settings.skip_unknown_fields = settings.input_format_skip_unknown_fields;
+    format_settings.template_settings.resultset_format = settings.format_template_resultset;
+    format_settings.template_settings.row_between_delimiter = settings.format_template_rows_between_delimiter;
+    format_settings.template_settings.row_format = settings.format_template_row;
+    format_settings.tsv.crlf_end_of_line = settings.output_format_tsv_crlf_end_of_line;
+    format_settings.tsv.empty_as_default = settings.input_format_tsv_empty_as_default;
+    format_settings.tsv.input_format_enum_as_number = settings.input_format_tsv_enum_as_number;
+    format_settings.tsv.null_representation = settings.output_format_tsv_null_representation;
+    format_settings.values.accurate_types_of_literals = settings.input_format_values_accurate_types_of_literals;
+    format_settings.values.deduce_templates_of_expressions = settings.input_format_values_deduce_templates_of_expressions;
+    format_settings.values.interpret_expressions = settings.input_format_values_interpret_expressions;
+    format_settings.with_names_use_header = settings.input_format_with_names_use_header;
+    format_settings.write_statistics = settings.output_format_write_statistics;
 
     /// Validate avro_schema_registry_url with RemoteHostFilter when non-empty and in Server context
-    if (context.hasGlobalContext() && (context.getGlobalContext().getApplicationType() == Context::ApplicationType::SERVER))
+    if (format_settings.schema.is_server)
     {
         const Poco::URI & avro_schema_registry_url = settings.format_avro_schema_registry_url;
         if (!avro_schema_registry_url.empty())
             context.getRemoteHostFilter().checkURL(avro_schema_registry_url);
     }
-    format_settings.avro.schema_registry_url = settings.format_avro_schema_registry_url.toString();
-    format_settings.avro.allow_missing_fields = settings.input_format_avro_allow_missing_fields;
 
     return format_settings;
 }
 
-static FormatSettings getOutputFormatSetting(const Settings & settings, const Context & context)
-{
-    FormatSettings format_settings;
-    format_settings.enable_streaming = settings.output_format_enable_streaming;
-    format_settings.json.quote_64bit_integers = settings.output_format_json_quote_64bit_integers;
-    format_settings.json.quote_denormals = settings.output_format_json_quote_denormals;
-    format_settings.json.escape_forward_slashes = settings.output_format_json_escape_forward_slashes;
-    format_settings.csv.delimiter = settings.format_csv_delimiter;
-    format_settings.csv.allow_single_quotes = settings.format_csv_allow_single_quotes;
-    format_settings.csv.allow_double_quotes = settings.format_csv_allow_double_quotes;
-    format_settings.csv.crlf_end_of_line = settings.output_format_csv_crlf_end_of_line;
-    format_settings.pretty.max_rows = settings.output_format_pretty_max_rows;
-    format_settings.pretty.max_column_pad_width = settings.output_format_pretty_max_column_pad_width;
-    format_settings.pretty.max_value_width = settings.output_format_pretty_max_value_width;
-    format_settings.pretty.color = settings.output_format_pretty_color;
-    format_settings.pretty.charset = settings.output_format_pretty_grid_charset.toString() == "ASCII" ?
-                                     FormatSettings::Pretty::Charset::ASCII :
-                                     FormatSettings::Pretty::Charset::UTF8;
-    format_settings.pretty.output_format_pretty_row_numbers = settings.output_format_pretty_row_numbers;
-    format_settings.template_settings.resultset_format = settings.format_template_resultset;
-    format_settings.template_settings.row_format = settings.format_template_row;
-    format_settings.template_settings.row_between_delimiter = settings.format_template_rows_between_delimiter;
-    format_settings.tsv.crlf_end_of_line = settings.output_format_tsv_crlf_end_of_line;
-    format_settings.tsv.null_representation = settings.output_format_tsv_null_representation;
-    format_settings.write_statistics = settings.output_format_write_statistics;
-    format_settings.parquet.row_group_size = settings.output_format_parquet_row_group_size;
-    format_settings.schema.format_schema = settings.format_schema;
-    format_settings.schema.format_schema_path = context.getFormatSchemaPath();
-    format_settings.schema.is_server = context.hasGlobalContext() && (context.getGlobalContext().getApplicationType() == Context::ApplicationType::SERVER);
-    format_settings.custom.result_before_delimiter = settings.format_custom_result_before_delimiter;
-    format_settings.custom.result_after_delimiter = settings.format_custom_result_after_delimiter;
-    format_settings.custom.escaping_rule = settings.format_custom_escaping_rule;
-    format_settings.custom.field_delimiter = settings.format_custom_field_delimiter;
-    format_settings.custom.row_before_delimiter = settings.format_custom_row_before_delimiter;
-    format_settings.custom.row_after_delimiter = settings.format_custom_row_after_delimiter;
-    format_settings.custom.row_between_delimiter = settings.format_custom_row_between_delimiter;
-    format_settings.avro.output_codec = settings.output_format_avro_codec;
-    format_settings.avro.output_sync_interval = settings.output_format_avro_sync_interval;
-    format_settings.date_time_output_format = settings.date_time_output_format;
+template
+FormatSettings getFormatSettings<FormatFactorySettings>(const Context & context,
+    const FormatFactorySettings & settings);
 
-    return format_settings;
-}
+template
+FormatSettings getFormatSettings<Settings>(const Context & context,
+    const Settings & settings);
 
 
 BlockInputStreamPtr FormatFactory::getInput(
@@ -142,21 +135,22 @@ BlockInputStreamPtr FormatFactory::getInput(
     const Block & sample,
     const Context & context,
     UInt64 max_block_size,
-    ReadCallback callback) const
+    const std::optional<FormatSettings> & _format_settings) const
 {
     if (name == "Native")
         return std::make_shared<NativeBlockInputStream>(buf, sample, 0);
 
+    auto format_settings = _format_settings
+        ? *_format_settings : getFormatSettings(context);
+
     if (!getCreators(name).input_processor_creator)
     {
         const auto & input_getter = getCreators(name).input_creator;
         if (!input_getter)
             throw Exception("Format " + name + " is not suitable for input", ErrorCodes::FORMAT_IS_NOT_SUITABLE_FOR_INPUT);
 
-        const Settings & settings = context.getSettingsRef();
-        FormatSettings format_settings = getInputFormatSetting(settings, context);
 
-        return input_getter(buf, sample, max_block_size, callback ? callback : ReadCallback(), format_settings);
+        return input_getter(buf, sample, max_block_size, {}, format_settings);
     }
 
     const Settings & settings = context.getSettingsRef();
@@ -182,17 +176,16 @@ BlockInputStreamPtr FormatFactory::getInput(
         if (!input_getter)
             throw Exception("Format " + name + " is not suitable for input", ErrorCodes::FORMAT_IS_NOT_SUITABLE_FOR_INPUT);
 
-        FormatSettings format_settings = getInputFormatSetting(settings, context);
-
         RowInputFormatParams row_input_format_params;
         row_input_format_params.max_block_size = max_block_size;
         row_input_format_params.allow_errors_num = format_settings.input_allow_errors_num;
         row_input_format_params.allow_errors_ratio = format_settings.input_allow_errors_ratio;
-        row_input_format_params.callback = std::move(callback);
         row_input_format_params.max_execution_time = settings.max_execution_time;
         row_input_format_params.timeout_overflow_mode = settings.timeout_overflow_mode;
 
-        auto input_creator_params = ParallelParsingBlockInputStream::InputCreatorParams{sample, row_input_format_params, format_settings};
+        auto input_creator_params =
+            ParallelParsingBlockInputStream::InputCreatorParams{sample,
+                row_input_format_params, format_settings};
         ParallelParsingBlockInputStream::Params params{buf, input_getter,
             input_creator_params, file_segmentation_engine,
             static_cast<int>(settings.max_threads),
@@ -200,32 +193,37 @@ BlockInputStreamPtr FormatFactory::getInput(
         return std::make_shared<ParallelParsingBlockInputStream>(params);
     }
 
-    auto format = getInputFormat(name, buf, sample, context, max_block_size, std::move(callback));
+    auto format = getInputFormat(name, buf, sample, context, max_block_size,
+        format_settings);
     return std::make_shared<InputStreamFromInputFormat>(std::move(format));
 }
 
 
-BlockOutputStreamPtr FormatFactory::getOutput(
-    const String & name, WriteBuffer & buf, const Block & sample, const Context & context, WriteCallback callback, const bool ignore_no_row_delimiter) const
+BlockOutputStreamPtr FormatFactory::getOutput(const String & name,
+    WriteBuffer & buf, const Block & sample, const Context & context,
+    WriteCallback callback, const std::optional<FormatSettings> & _format_settings) const
 {
+    auto format_settings = _format_settings
+        ? *_format_settings : getFormatSettings(context);
+
     if (!getCreators(name).output_processor_creator)
     {
         const auto & output_getter = getCreators(name).output_creator;
         if (!output_getter)
             throw Exception("Format " + name + " is not suitable for output", ErrorCodes::FORMAT_IS_NOT_SUITABLE_FOR_OUTPUT);
 
-        const Settings & settings = context.getSettingsRef();
-        FormatSettings format_settings = getOutputFormatSetting(settings, context);
-
         /**  Materialization is needed, because formats can use the functions `IDataType`,
           *  which only work with full columns.
           */
         return std::make_shared<MaterializingBlockOutputStream>(
-                output_getter(buf, sample, std::move(callback), format_settings), sample);
+            output_getter(buf, sample, std::move(callback), format_settings),
+            sample);
     }
 
-    auto format = getOutputFormat(name, buf, sample, context, std::move(callback), ignore_no_row_delimiter);
-    return std::make_shared<MaterializingBlockOutputStream>(std::make_shared<OutputStreamToOutputFormat>(format), sample);
+    auto format = getOutputFormat(name, buf, sample, context, std::move(callback),
+        format_settings);
+    return std::make_shared<MaterializingBlockOutputStream>(
+        std::make_shared<OutputStreamToOutputFormat>(format), sample);
 }
 
 
@@ -235,25 +233,27 @@ InputFormatPtr FormatFactory::getInputFormat(
     const Block & sample,
     const Context & context,
     UInt64 max_block_size,
-    ReadCallback callback) const
+    const std::optional<FormatSettings> & _format_settings) const
 {
     const auto & input_getter = getCreators(name).input_processor_creator;
     if (!input_getter)
         throw Exception("Format " + name + " is not suitable for input", ErrorCodes::FORMAT_IS_NOT_SUITABLE_FOR_INPUT);
 
     const Settings & settings = context.getSettingsRef();
-    FormatSettings format_settings = getInputFormatSetting(settings, context);
+
+    auto format_settings = _format_settings
+        ? *_format_settings : getFormatSettings(context);
 
     RowInputFormatParams params;
     params.max_block_size = max_block_size;
     params.allow_errors_num = format_settings.input_allow_errors_num;
     params.allow_errors_ratio = format_settings.input_allow_errors_ratio;
-    params.callback = std::move(callback);
     params.max_execution_time = settings.max_execution_time;
     params.timeout_overflow_mode = settings.timeout_overflow_mode;
 
     auto format = input_getter(buf, sample, params, format_settings);
 
+
     /// It's a kludge. Because I cannot remove context from values format.
     if (auto * values = typeid_cast<ValuesBlockInputFormat *>(format.get()))
         values->setContext(context);
@@ -263,19 +263,20 @@ InputFormatPtr FormatFactory::getInputFormat(
 
 
 OutputFormatPtr FormatFactory::getOutputFormat(
-    const String & name, WriteBuffer & buf, const Block & sample, const Context & context, WriteCallback callback, const bool ignore_no_row_delimiter) const
+    const String & name, WriteBuffer & buf, const Block & sample,
+    const Context & context, WriteCallback callback,
+    const std::optional<FormatSettings> & _format_settings) const
 {
     const auto & output_getter = getCreators(name).output_processor_creator;
     if (!output_getter)
         throw Exception("Format " + name + " is not suitable for output", ErrorCodes::FORMAT_IS_NOT_SUITABLE_FOR_OUTPUT);
 
-    const Settings & settings = context.getSettingsRef();
-    FormatSettings format_settings = getOutputFormatSetting(settings, context);
-
     RowOutputFormatParams params;
-    params.ignore_no_row_delimiter = ignore_no_row_delimiter;
     params.callback = std::move(callback);
 
+    auto format_settings = _format_settings
+        ? *_format_settings : getFormatSettings(context);
+
     /** TODO: Materialization is needed, because formats can use the functions `IDataType`,
       *  which only work with full columns.
       */

src/Formats/FormatFactory.h

@@ -3,6 +3,7 @@
 #include <common/types.h>
 #include <Columns/IColumn.h>
 #include <DataStreams/IBlockStream_fwd.h>
+#include <Formats/FormatSettings.h>
 #include <IO/BufferWithOwnMemory.h>
 
 #include <functional>
@@ -16,6 +17,8 @@ namespace DB
 class Block;
 class Context;
 struct FormatSettings;
+struct Settings;
+struct FormatFactorySettings;
 
 class ReadBuffer;
 class WriteBuffer;
@@ -32,6 +35,11 @@ struct RowOutputFormatParams;
 using InputFormatPtr = std::shared_ptr<IInputFormat>;
 using OutputFormatPtr = std::shared_ptr<IOutputFormat>;
 
+FormatSettings getFormatSettings(const Context & context);
+
+template <typename T>
+FormatSettings getFormatSettings(const Context & context,
+    const T & settings);
 
 /** Allows to create an IBlockInputStream or IBlockOutputStream by the name of the format.
   * Note: format and compression are independent things.
@@ -104,10 +112,11 @@ public:
         const Block & sample,
         const Context & context,
         UInt64 max_block_size,
-        ReadCallback callback = {}) const;
+        const std::optional<FormatSettings> & format_settings = std::nullopt) const;
 
     BlockOutputStreamPtr getOutput(const String & name, WriteBuffer & buf,
-        const Block & sample, const Context & context, WriteCallback callback = {}, const bool ignore_no_row_delimiter = false) const;
+        const Block & sample, const Context & context, WriteCallback callback = {},
+        const std::optional<FormatSettings> & format_settings = std::nullopt) const;
 
     InputFormatPtr getInputFormat(
         const String & name,
@@ -115,10 +124,12 @@ public:
         const Block & sample,
         const Context & context,
         UInt64 max_block_size,
-        ReadCallback callback = {}) const;
+        const std::optional<FormatSettings> & format_settings = std::nullopt) const;
 
     OutputFormatPtr getOutputFormat(
-        const String & name, WriteBuffer & buf, const Block & sample, const Context & context, WriteCallback callback = {}, const bool ignore_no_row_delimiter = false) const;
+        const String & name, WriteBuffer & buf, const Block & sample,
+        const Context & context, WriteCallback callback = {},
+        const std::optional<FormatSettings> & format_settings = std::nullopt) const;
 
     /// Register format by its name.
     void registerInputFormat(const String & name, InputCreator input_creator);

src/Formats/FormatSettings.h

@@ -6,10 +6,16 @@
 namespace DB
 {
 
-/** Various tweaks for input/output formats.
-  * Text serialization/deserialization of data types also depend on some of these settings.
-  * NOTE Parameters for unrelated formats and unrelated data types
-  *  are collected in this struct - it prevents modularity, but they are difficult to separate.
+/**
+  * Various tweaks for input/output formats. Text serialization/deserialization
+  * of data types also depend on some of these settings. It is different from
+  * FormatFactorySettings in that it has all necessary user-provided settings
+  * combined with information from context etc, that we can use directly during
+  * serialization. In contrast, FormatFactorySettings' job is to reflect the
+  * changes made to user-visible format settings, such as when tweaking the
+  * the format for File engine.
+  * NOTE Parameters for unrelated formats and unrelated data types are collected
+  * in this struct - it prevents modularity, but they are difficult to separate.
   */
 struct FormatSettings
 {
@@ -17,76 +23,6 @@ struct FormatSettings
     /// Option means that each chunk of data need to be formatted independently. Also each chunk will be flushed at the end of processing.
     bool enable_streaming = false;
 
-    struct JSON
-    {
-        bool quote_64bit_integers = true;
-        bool quote_denormals = true;
-        bool escape_forward_slashes = true;
-    };
-
-    JSON json;
-
-    struct CSV
-    {
-        char delimiter = ',';
-        bool allow_single_quotes = true;
-        bool allow_double_quotes = true;
-        bool unquoted_null_literal_as_null = false;
-        bool empty_as_default = false;
-        bool crlf_end_of_line = false;
-        bool input_format_enum_as_number = false;
-    };
-
-    CSV csv;
-
-    struct Pretty
-    {
-        UInt64 max_rows = 10000;
-        UInt64 max_column_pad_width = 250;
-        UInt64 max_value_width = 10000;
-        bool color = true;
-
-        bool output_format_pretty_row_numbers = false;
-
-        enum class Charset
-        {
-            UTF8,
-            ASCII,
-        };
-
-        Charset charset = Charset::UTF8;
-    };
-
-    Pretty pretty;
-
-    struct Values
-    {
-        bool interpret_expressions = true;
-        bool deduce_templates_of_expressions = true;
-        bool accurate_types_of_literals = true;
-    };
-
-    Values values;
-
-    struct Template
-    {
-        String resultset_format;
-        String row_format;
-        String row_between_delimiter;
-    };
-
-    Template template_settings;
-
-    struct TSV
-    {
-        bool empty_as_default = false;
-        bool crlf_end_of_line = false;
-        String null_representation = "\\N";
-        bool input_format_enum_as_number = false;
-    };
-
-    TSV tsv;
-
     bool skip_unknown_fields = false;
     bool with_names_use_header = false;
     bool write_statistics = true;
@@ -113,24 +49,29 @@ struct FormatSettings
     UInt64 input_allow_errors_num = 0;
     Float32 input_allow_errors_ratio = 0;
 
-    struct Arrow
+    struct
     {
         UInt64 row_group_size = 1000000;
     } arrow;
 
-    struct Parquet
+    struct
     {
-        UInt64 row_group_size = 1000000;
-    } parquet;
+        String schema_registry_url;
+        String output_codec;
+        UInt64 output_sync_interval = 16 * 1024;
+        bool allow_missing_fields = false;
+    } avro;
 
-    struct Schema
+    struct CSV
     {
-        std::string format_schema;
-        std::string format_schema_path;
-        bool is_server = false;
-    };
-
-    Schema schema;
+        char delimiter = ',';
+        bool allow_single_quotes = true;
+        bool allow_double_quotes = true;
+        bool unquoted_null_literal_as_null = false;
+        bool empty_as_default = false;
+        bool crlf_end_of_line = false;
+        bool input_format_enum_as_number = false;
+    } csv;
 
     struct Custom
     {
@@ -141,29 +82,87 @@ struct FormatSettings
         std::string row_between_delimiter;
         std::string field_delimiter;
         std::string escaping_rule;
-    };
+    } custom;
 
-    Custom custom;
-
-    struct Avro
+    struct
     {
-        String schema_registry_url;
-        String output_codec;
-        UInt64 output_sync_interval = 16 * 1024;
-        bool allow_missing_fields = false;
-    };
+        bool quote_64bit_integers = true;
+        bool quote_denormals = true;
+        bool escape_forward_slashes = true;
+        bool serialize_as_strings = false;
+    } json;
 
-    Avro avro;
+    struct
+    {
+        UInt64 row_group_size = 1000000;
+    } parquet;
 
-    struct Regexp
+    struct Pretty
+    {
+        UInt64 max_rows = 10000;
+        UInt64 max_column_pad_width = 250;
+        UInt64 max_value_width = 10000;
+        bool color = true;
+
+        bool output_format_pretty_row_numbers = false;
+
+        enum class Charset
+        {
+            UTF8,
+            ASCII,
+        };
+
+        Charset charset = Charset::UTF8;
+    } pretty;
+
+    struct
+    {
+        bool write_row_delimiters = true;
+        /**
+         * Some buffers (kafka / rabbit) split the rows internally using callback,
+         * and always send one row per message, so we can push there formats
+         * without framing / delimiters (like ProtobufSingle). In other cases,
+         * we have to enforce exporting at most one row in the format output,
+         * because Protobuf without delimiters is not generally useful.
+         */
+        bool allow_many_rows_no_delimiters = false;
+    } protobuf;
+
+    struct
     {
         std::string regexp;
         std::string escaping_rule;
         bool skip_unmatched = false;
-    };
+    } regexp;
 
-    Regexp regexp;
+    struct
+    {
+        std::string format_schema;
+        std::string format_schema_path;
+        bool is_server = false;
+    } schema;
 
+    struct
+    {
+        String resultset_format;
+        String row_format;
+        String row_between_delimiter;
+    } template_settings;
+
+    struct
+    {
+        bool empty_as_default = false;
+        bool crlf_end_of_line = false;
+        String null_representation = "\\N";
+        bool input_format_enum_as_number = false;
+    } tsv;
+
+    struct
+    {
+        bool interpret_expressions = true;
+        bool deduce_templates_of_expressions = true;
+        bool accurate_types_of_literals = true;
+    } values;
 };
 
 }

src/Formats/tests/tab_separated_streams.cpp

@@ -38,8 +38,8 @@ try
 
     FormatSettings format_settings;
 
-    RowInputFormatParams in_params{DEFAULT_INSERT_BLOCK_SIZE, 0, 0, []{}};
-    RowOutputFormatParams out_params{[](const Columns & /* columns */, size_t /* row */){},false};
+    RowInputFormatParams in_params{DEFAULT_INSERT_BLOCK_SIZE, 0, 0};
+    RowOutputFormatParams out_params{[](const Columns & /* columns */, size_t /* row */){}};
 
     InputFormatPtr input_format = std::make_shared<TabSeparatedRowInputFormat>(sample, in_buf, in_params, false, false, format_settings);
     BlockInputStreamPtr block_input = std::make_shared<InputStreamFromInputFormat>(std::move(input_format));

src/IO/ReadHelpers.cpp

@@ -817,7 +817,11 @@ ReturnType readDateTimeTextFallback(time_t & datetime, ReadBuffer & buf, const D
 {
     static constexpr bool throw_exception = std::is_same_v<ReturnType, void>;
 
+    /// YYYY-MM-DD hh:mm:ss
     static constexpr auto date_time_broken_down_length = 19;
+    /// YYYY-MM-DD
+    static constexpr auto date_broken_down_length = 10;
+    /// unix timestamp max length
     static constexpr auto unix_timestamp_max_length = 10;
 
     char s[date_time_broken_down_length];
@@ -831,12 +835,15 @@ ReturnType readDateTimeTextFallback(time_t & datetime, ReadBuffer & buf, const D
         ++buf.position();
     }
 
-    /// 2015-01-01 01:02:03
+    /// 2015-01-01 01:02:03 or 2015-01-01
     if (s_pos == s + 4 && !buf.eof() && (*buf.position() < '0' || *buf.position() > '9'))
     {
-        const size_t remaining_size = date_time_broken_down_length - (s_pos - s);
-        size_t size = buf.read(s_pos, remaining_size);
-        if (remaining_size != size)
+        const auto already_read_length = s_pos - s;
+        const size_t remaining_date_time_size = date_time_broken_down_length - already_read_length;
+        const size_t remaining_date_size = date_broken_down_length - already_read_length;
+
+        size_t size = buf.read(s_pos, remaining_date_time_size);
+        if (size != remaining_date_time_size && size != remaining_date_size)
         {
             s_pos[size] = 0;
 
@@ -850,9 +857,16 @@ ReturnType readDateTimeTextFallback(time_t & datetime, ReadBuffer & buf, const D
         UInt8 month = (s[5] - '0') * 10 + (s[6] - '0');
         UInt8 day = (s[8] - '0') * 10 + (s[9] - '0');
 
-        UInt8 hour = (s[11] - '0') * 10 + (s[12] - '0');
-        UInt8 minute = (s[14] - '0') * 10 + (s[15] - '0');
-        UInt8 second = (s[17] - '0') * 10 + (s[18] - '0');
+        UInt8 hour = 0;
+        UInt8 minute = 0;
+        UInt8 second = 0;
+
+        if (size == remaining_date_time_size)
+        {
+            hour = (s[11] - '0') * 10 + (s[12] - '0');
+            minute = (s[14] - '0') * 10 + (s[15] - '0');
+            second = (s[17] - '0') * 10 + (s[18] - '0');
+        }
 
         if (unlikely(year == 0))
             datetime = 0;

src/IO/ReadHelpers.h

@@ -700,7 +700,7 @@ UInt128 stringToUUID(const String & str);
 template <typename ReturnType = void>
 ReturnType readDateTimeTextFallback(time_t & datetime, ReadBuffer & buf, const DateLUTImpl & date_lut);
 
-/** In YYYY-MM-DD hh:mm:ss format, according to specified time zone.
+/** In YYYY-MM-DD hh:mm:ss or YYYY-MM-DD format, according to specified time zone.
   * As an exception, also supported parsing of unix timestamp in form of decimal number.
   */
 template <typename ReturnType = void>
@@ -709,12 +709,17 @@ inline ReturnType readDateTimeTextImpl(time_t & datetime, ReadBuffer & buf, cons
     /** Read 10 characters, that could represent unix timestamp.
       * Only unix timestamp of 5-10 characters is supported.
       * Then look at 5th character. If it is a number - treat whole as unix timestamp.
-      * If it is not a number - then parse datetime in YYYY-MM-DD hh:mm:ss format.
+      * If it is not a number - then parse datetime in YYYY-MM-DD hh:mm:ss or YYYY-MM-DD format.
       */
 
     /// Optimistic path, when whole value is in buffer.
     const char * s = buf.position();
-    if (s + 19 <= buf.buffer().end())
+
+    /// YYYY-MM-DD hh:mm:ss
+    static constexpr auto DateTimeStringInputSize = 19;
+    bool optimistic_path_for_date_time_input = s + DateTimeStringInputSize <= buf.buffer().end();
+
+    if (optimistic_path_for_date_time_input)
     {
         if (s[4] < '0' || s[4] > '9')
         {
@@ -731,7 +736,7 @@ inline ReturnType readDateTimeTextImpl(time_t & datetime, ReadBuffer & buf, cons
             else
                 datetime = date_lut.makeDateTime(year, month, day, hour, minute, second);
 
-            buf.position() += 19;
+            buf.position() += DateTimeStringInputSize;
             return ReturnType(true);
         }
         else

src/Interpreters/ClusterProxy/IStreamFactory.h

@@ -16,6 +16,9 @@ struct SelectQueryInfo;
 class Pipe;
 using Pipes = std::vector<Pipe>;
 
+class QueryPlan;
+using QueryPlanPtr = std::unique_ptr<QueryPlan>;
+
 namespace ClusterProxy
 {
 
@@ -31,7 +34,9 @@ public:
             const String & query, const ASTPtr & query_ast,
             const Context & context, const ThrottlerPtr & throttler,
             const SelectQueryInfo & query_info,
-            Pipes & res) = 0;
+            std::vector<QueryPlanPtr> & res,
+            Pipes & remote_pipes,
+            Pipes & delayed_pipes) = 0;
 };
 
 }

src/Interpreters/ClusterProxy/SelectStreamFactory.cpp

@@ -14,6 +14,8 @@
 #include <Processors/Sources/RemoteSource.h>
 #include <Processors/Sources/DelayedSource.h>
 #include <Processors/QueryPlan/QueryPlan.h>
+#include <Processors/QueryPlan/ConvertingStep.h>
+
 
 namespace ProfileEvents
 {
@@ -69,37 +71,27 @@ SelectStreamFactory::SelectStreamFactory(
 namespace
 {
 
-auto createLocalPipe(
-    const ASTPtr & query_ast, const Block & header, const Context & context, QueryProcessingStage::Enum processed_stage)
+std::unique_ptr<QueryPlan> createLocalPlan(
+    const ASTPtr & query_ast,
+    const Block & header,
+    const Context & context,
+    QueryProcessingStage::Enum processed_stage)
 {
     checkStackSize();
 
-    InterpreterSelectQuery interpreter(query_ast, context, SelectQueryOptions(processed_stage));
     auto query_plan = std::make_unique<QueryPlan>();
 
+    InterpreterSelectQuery interpreter(query_ast, context, SelectQueryOptions(processed_stage));
     interpreter.buildQueryPlan(*query_plan);
-    auto pipeline = std::move(*query_plan->buildQueryPipeline());
 
-    /// Avoid going it out-of-scope for EXPLAIN
-    pipeline.addQueryPlan(std::move(query_plan));
+    /// Convert header structure to expected.
+    /// Also we ignore constants from result and replace it with constants from header.
+    /// It is needed for functions like `now64()` or `randConstant()` because their values may be different.
+    auto converting = std::make_unique<ConvertingStep>(query_plan->getCurrentDataStream(), header, true);
+    converting->setStepDescription("Convert block structure for query from local replica");
+    query_plan->addStep(std::move(converting));
 
-    pipeline.addSimpleTransform([&](const Block & source_header)
-    {
-        return std::make_shared<ConvertingTransform>(
-                source_header, header, ConvertingTransform::MatchColumnsMode::Name, true);
-    });
-
-    /** Materialization is needed, since from remote servers the constants come materialized.
-      * If you do not do this, different types (Const and non-Const) columns will be produced in different threads,
-      * And this is not allowed, since all code is based on the assumption that in the block stream all types are the same.
-      */
-
-    /* Now we don't need to materialize constants, because RemoteBlockInputStream will ignore constant and take it from header.
-     * So, streams from different threads will always have the same header.
-     */
-
-    pipeline.setMaxThreads(1);
-    return QueryPipeline::getPipe(std::move(pipeline));
+    return query_plan;
 }
 
 String formattedAST(const ASTPtr & ast)
@@ -119,7 +111,9 @@ void SelectStreamFactory::createForShard(
     const String &, const ASTPtr & query_ast,
     const Context & context, const ThrottlerPtr & throttler,
     const SelectQueryInfo &,
-    Pipes & pipes)
+    std::vector<QueryPlanPtr> & plans,
+    Pipes & remote_pipes,
+    Pipes & delayed_pipes)
 {
     bool add_agg_info = processed_stage == QueryProcessingStage::WithMergeableState;
     bool add_totals = false;
@@ -136,7 +130,7 @@ void SelectStreamFactory::createForShard(
 
     auto emplace_local_stream = [&]()
     {
-        pipes.emplace_back(createLocalPipe(modified_query_ast, header, context, processed_stage));
+        plans.emplace_back(createLocalPlan(modified_query_ast, header, context, processed_stage));
     };
 
     String modified_query = formattedAST(modified_query_ast);
@@ -149,7 +143,7 @@ void SelectStreamFactory::createForShard(
         if (!table_func_ptr)
             remote_query_executor->setMainTable(main_table);
 
-        pipes.emplace_back(createRemoteSourcePipe(remote_query_executor, add_agg_info, add_totals, add_extremes));
+        remote_pipes.emplace_back(createRemoteSourcePipe(remote_query_executor, add_agg_info, add_totals, add_extremes));
     };
 
     const auto & settings = context.getSettingsRef();
@@ -275,7 +269,10 @@ void SelectStreamFactory::createForShard(
             }
 
             if (try_results.empty() || local_delay < max_remote_delay)
-                return createLocalPipe(modified_query_ast, header, context, stage);
+            {
+                auto plan = createLocalPlan(modified_query_ast, header, context, stage);
+                return QueryPipeline::getPipe(std::move(*plan->buildQueryPipeline()));
+            }
             else
             {
                 std::vector<IConnectionPool::Entry> connections;
@@ -290,7 +287,7 @@ void SelectStreamFactory::createForShard(
             }
         };
 
-        pipes.emplace_back(createDelayedPipe(header, lazily_create_stream, add_totals, add_extremes));
+        delayed_pipes.emplace_back(createDelayedPipe(header, lazily_create_stream, add_totals, add_extremes));
     }
     else
         emplace_remote_stream();

src/Interpreters/ClusterProxy/SelectStreamFactory.h

@@ -39,7 +39,9 @@ public:
         const String & query, const ASTPtr & query_ast,
         const Context & context, const ThrottlerPtr & throttler,
         const SelectQueryInfo & query_info,
-        Pipes & pipes) override;
+        std::vector<QueryPlanPtr> & plans,
+        Pipes & remote_pipes,
+        Pipes & delayed_pipes) override;
 
 private:
     const Block header;

src/Interpreters/ClusterProxy/executeQuery.cpp

@@ -7,6 +7,9 @@
 #include <Interpreters/ProcessList.h>
 #include <Parsers/queryToString.h>
 #include <Processors/Pipe.h>
+#include <Processors/QueryPlan/QueryPlan.h>
+#include <Processors/QueryPlan/ReadFromPreparedSource.h>
+#include <Processors/QueryPlan/UnionStep.h>
 #include <Storages/SelectQueryInfo.h>
 
 
@@ -81,15 +84,19 @@ Context updateSettingsForCluster(const Cluster & cluster, const Context & contex
     return new_context;
 }
 
-Pipe executeQuery(
+void executeQuery(
+    QueryPlan & query_plan,
     IStreamFactory & stream_factory, Poco::Logger * log,
     const ASTPtr & query_ast, const Context & context, const SelectQueryInfo & query_info)
 {
     assert(log);
 
-    Pipes res;
     const Settings & settings = context.getSettingsRef();
 
+    std::vector<QueryPlanPtr> plans;
+    Pipes remote_pipes;
+    Pipes delayed_pipes;
+
     const std::string query = queryToString(query_ast);
 
     Context new_context = updateSettingsForCluster(*query_info.cluster, context, settings, log);
@@ -112,9 +119,43 @@ Pipe executeQuery(
         throttler = user_level_throttler;
 
     for (const auto & shard_info : query_info.cluster->getShardsInfo())
-        stream_factory.createForShard(shard_info, query, query_ast, new_context, throttler, query_info, res);
+        stream_factory.createForShard(shard_info, query, query_ast, new_context, throttler, query_info, plans, remote_pipes, delayed_pipes);
 
-    return Pipe::unitePipes(std::move(res));
+    if (!remote_pipes.empty())
+    {
+        auto plan = std::make_unique<QueryPlan>();
+        auto read_from_remote = std::make_unique<ReadFromPreparedSource>(Pipe::unitePipes(std::move(remote_pipes)));
+        read_from_remote->setStepDescription("Read from remote replica");
+        plan->addStep(std::move(read_from_remote));
+        plans.emplace_back(std::move(plan));
+    }
+
+    if (!delayed_pipes.empty())
+    {
+        auto plan = std::make_unique<QueryPlan>();
+        auto read_from_remote = std::make_unique<ReadFromPreparedSource>(Pipe::unitePipes(std::move(delayed_pipes)));
+        read_from_remote->setStepDescription("Read from delayed local replica");
+        plan->addStep(std::move(read_from_remote));
+        plans.emplace_back(std::move(plan));
+    }
+
+    if (plans.empty())
+        return;
+
+    if (plans.size() == 1)
+    {
+        query_plan = std::move(*plans.front());
+        return;
+    }
+
+    DataStreams input_streams;
+    input_streams.reserve(plans.size());
+    for (auto & plan : plans)
+        input_streams.emplace_back(plan->getCurrentDataStream());
+
+    auto header = input_streams.front().header;
+    auto union_step = std::make_unique<UnionStep>(std::move(input_streams), header);
+    query_plan.unitePlans(std::move(union_step), std::move(plans));
 }
 
 }

src/Interpreters/ClusterProxy/executeQuery.h

@@ -11,6 +11,7 @@ class Cluster;
 struct SelectQueryInfo;
 
 class Pipe;
+class QueryPlan;
 
 namespace ClusterProxy
 {
@@ -31,8 +32,10 @@ Context updateSettingsForCluster(const Cluster & cluster, const Context & contex
 /// Execute a distributed query, creating a vector of BlockInputStreams, from which the result can be read.
 /// `stream_factory` object encapsulates the logic of creating streams for a different type of query
 /// (currently SELECT, DESCRIBE).
-Pipe executeQuery(
-    IStreamFactory & stream_factory, Poco::Logger * log, const ASTPtr & query_ast, const Context & context, const SelectQueryInfo & query_info);
+void executeQuery(
+    QueryPlan & query_plan,
+    IStreamFactory & stream_factory, Poco::Logger * log,
+    const ASTPtr & query_ast, const Context & context, const SelectQueryInfo & query_info);
 
 }
 

src/Interpreters/Context.cpp

@@ -302,9 +302,11 @@ struct ContextShared
     mutable std::mutex zookeeper_mutex;
 
     mutable zkutil::ZooKeeperPtr zookeeper;                 /// Client for ZooKeeper.
+    ConfigurationPtr zookeeper_config;                      /// Stores zookeeper configs
 
     mutable std::mutex auxiliary_zookeepers_mutex;
     mutable std::map<String, zkutil::ZooKeeperPtr> auxiliary_zookeepers;    /// Map for auxiliary ZooKeeper clients.
+    ConfigurationPtr auxiliary_zookeepers_config;           /// Stores auxiliary zookeepers configs
 
     String interserver_io_host;                             /// The host name by which this server is available for other servers.
     UInt16 interserver_io_port = 0;                         /// and port.
@@ -364,8 +366,7 @@ struct ContextShared
     /// Initialized on demand (on distributed storages initialization) since Settings should be initialized
     std::unique_ptr<Clusters> clusters;
     ConfigurationPtr clusters_config;                        /// Stores updated configs
-    ConfigurationPtr zookeeper_config;                        /// Stores zookeeper configs
-    mutable std::mutex clusters_mutex;                        /// Guards clusters and clusters_config
+    mutable std::mutex clusters_mutex;                       /// Guards clusters and clusters_config
 
 #if USE_EMBEDDED_COMPILER
     std::shared_ptr<CompiledExpressionCache> compiled_expression_cache;
@@ -1498,10 +1499,16 @@ zkutil::ZooKeeperPtr Context::getAuxiliaryZooKeeper(const String & name) const
     auto zookeeper = shared->auxiliary_zookeepers.find(name);
     if (zookeeper == shared->auxiliary_zookeepers.end())
     {
-        if (!getConfigRef().has("auxiliary_zookeepers." + name))
-            throw Exception(ErrorCodes::BAD_ARGUMENTS, "Unknown auxiliary ZooKeeper name '{}'. If it's required it can be added to the section <auxiliary_zookeepers> in config.xml", name);
+        const auto & config = shared->auxiliary_zookeepers_config ? *shared->auxiliary_zookeepers_config : getConfigRef();
+        if (!config.has("auxiliary_zookeepers." + name))
+            throw Exception(
+                ErrorCodes::BAD_ARGUMENTS,
+                "Unknown auxiliary ZooKeeper name '{}'. If it's required it can be added to the section <auxiliary_zookeepers> in "
+                "config.xml",
+                name);
 
-        zookeeper->second = std::make_shared<zkutil::ZooKeeper>(getConfigRef(), "auxiliary_zookeepers." + name);
+        zookeeper
+            = shared->auxiliary_zookeepers.emplace(name, std::make_shared<zkutil::ZooKeeper>(config, "auxiliary_zookeepers." + name)).first;
     }
     else if (zookeeper->second->expired())
         zookeeper->second = zookeeper->second->startNewSession();
@@ -1515,17 +1522,38 @@ void Context::resetZooKeeper() const
     shared->zookeeper.reset();
 }
 
+static void reloadZooKeeperIfChangedImpl(const ConfigurationPtr & config, const std::string & config_name, zkutil::ZooKeeperPtr & zk)
+{
+    if (!zk || zk->configChanged(*config, config_name))
+        zk = std::make_shared<zkutil::ZooKeeper>(*config, config_name);
+}
+
 void Context::reloadZooKeeperIfChanged(const ConfigurationPtr & config) const
 {
     std::lock_guard lock(shared->zookeeper_mutex);
     shared->zookeeper_config = config;
+    reloadZooKeeperIfChangedImpl(config, "zookeeper", shared->zookeeper);
+}
 
-    if (!shared->zookeeper || shared->zookeeper->configChanged(*config, "zookeeper"))
+void Context::reloadAuxiliaryZooKeepersConfigIfChanged(const ConfigurationPtr & config)
+{
+    std::lock_guard lock(shared->auxiliary_zookeepers_mutex);
+
+    shared->auxiliary_zookeepers_config = config;
+
+    for (auto it = shared->auxiliary_zookeepers.begin(); it != shared->auxiliary_zookeepers.end();)
     {
-        shared->zookeeper = std::make_shared<zkutil::ZooKeeper>(*config, "zookeeper");
+        if (!config->has("auxiliary_zookeepers." + it->first))
+            it = shared->auxiliary_zookeepers.erase(it);
+        else
+        {
+            reloadZooKeeperIfChangedImpl(config, "auxiliary_zookeepers." + it->first, it->second);
+            ++it;
+        }
     }
 }
 
+
 bool Context::hasZooKeeper() const
 {
     return getConfigRef().has("zookeeper");

src/Interpreters/Context.h

@@ -487,6 +487,9 @@ public:
     std::shared_ptr<zkutil::ZooKeeper> getZooKeeper() const;
     /// Same as above but return a zookeeper connection from auxiliary_zookeepers configuration entry.
     std::shared_ptr<zkutil::ZooKeeper> getAuxiliaryZooKeeper(const String & name) const;
+
+    /// Set auxiliary zookeepers configuration at server starting or configuration reloading.
+    void reloadAuxiliaryZooKeepersConfigIfChanged(const ConfigurationPtr & config);
     /// Has ready or expired ZooKeeper
     bool hasZooKeeper() const;
     /// Reset current zookeeper session. Do not create a new one.

src/Interpreters/InterpreterAlterQuery.cpp

@@ -87,7 +87,7 @@ BlockIO InterpreterAlterQuery::execute()
     if (!partition_commands.empty())
     {
         table->checkAlterPartitionIsPossible(partition_commands, metadata_snapshot, context.getSettingsRef());
-        auto partition_commands_pipe = table->alterPartition(query_ptr, metadata_snapshot, partition_commands, context);
+        auto partition_commands_pipe = table->alterPartition(metadata_snapshot, partition_commands, context);
         if (!partition_commands_pipe.empty())
             res.pipeline.init(std::move(partition_commands_pipe));
     }

src/Interpreters/InterpreterSelectQuery.cpp

@@ -33,11 +33,13 @@
 
 #include <Processors/Pipe.h>
 #include <Processors/Sources/SourceFromInputStream.h>
+#include <Processors/Sources/NullSource.h>
 #include <Processors/Transforms/ExpressionTransform.h>
 #include <Processors/Transforms/JoiningTransform.h>
 #include <Processors/Transforms/AggregatingTransform.h>
+#include <Processors/Transforms/FilterTransform.h>
 #include <Processors/QueryPlan/ArrayJoinStep.h>
-#include <Processors/QueryPlan/ReadFromStorageStep.h>
+#include <Processors/QueryPlan/SettingQuotaAndLimitsStep.h>
 #include <Processors/QueryPlan/ExpressionStep.h>
 #include <Processors/QueryPlan/FilterStep.h>
 #include <Processors/QueryPlan/ReadNothingStep.h>
@@ -1108,6 +1110,48 @@ static StreamLocalLimits getLimitsForStorage(const Settings & settings, const Se
     return limits;
 }
 
+void InterpreterSelectQuery::addEmptySourceToQueryPlan(QueryPlan & query_plan, const Block & source_header, const SelectQueryInfo & query_info)
+{
+    Pipe pipe(std::make_shared<NullSource>(source_header));
+
+    if (query_info.prewhere_info)
+    {
+        if (query_info.prewhere_info->alias_actions)
+        {
+            pipe.addSimpleTransform([&](const Block & header)
+            {
+                return std::make_shared<ExpressionTransform>(header, query_info.prewhere_info->alias_actions);
+            });
+        }
+
+        pipe.addSimpleTransform([&](const Block & header)
+        {
+            return std::make_shared<FilterTransform>(
+                header,
+                query_info.prewhere_info->prewhere_actions,
+                query_info.prewhere_info->prewhere_column_name,
+                query_info.prewhere_info->remove_prewhere_column);
+        });
+
+        // To remove additional columns
+        // In some cases, we did not read any marks so that the pipeline.streams is empty
+        // Thus, some columns in prewhere are not removed as expected
+        // This leads to mismatched header in distributed table
+        if (query_info.prewhere_info->remove_columns_actions)
+        {
+            pipe.addSimpleTransform([&](const Block & header)
+            {
+                return std::make_shared<ExpressionTransform>(
+                        header, query_info.prewhere_info->remove_columns_actions);
+            });
+        }
+    }
+
+    auto read_from_pipe = std::make_unique<ReadFromPreparedSource>(std::move(pipe));
+    read_from_pipe->setStepDescription("Read from NullSource");
+    query_plan.addStep(std::move(read_from_pipe));
+}
+
 void InterpreterSelectQuery::executeFetchColumns(
     QueryProcessingStage::Enum processing_stage, QueryPlan & query_plan,
     const PrewhereDAGInfoPtr & prewhere_info, const NameSet & columns_to_remove_after_prewhere)
@@ -1347,7 +1391,7 @@ void InterpreterSelectQuery::executeFetchColumns(
             ErrorCodes::TOO_MANY_COLUMNS);
 
     /// General limit for the number of threads.
-    query_plan.setMaxThreads(settings.max_threads);
+    size_t max_threads_execute_query = settings.max_threads;
 
     /** With distributed query processing, almost no computations are done in the threads,
      *  but wait and receive data from remote servers.
@@ -1360,8 +1404,7 @@ void InterpreterSelectQuery::executeFetchColumns(
     if (storage && storage->isRemote())
     {
         is_remote = true;
-        max_streams = settings.max_distributed_connections;
-        query_plan.setMaxThreads(max_streams);
+        max_threads_execute_query = max_streams = settings.max_distributed_connections;
     }
 
     UInt64 max_block_size = settings.max_block_size;
@@ -1386,8 +1429,7 @@ void InterpreterSelectQuery::executeFetchColumns(
         && limit_length + limit_offset < max_block_size)
     {
         max_block_size = std::max(UInt64(1), limit_length + limit_offset);
-        max_streams = 1;
-        query_plan.setMaxThreads(max_streams);
+        max_threads_execute_query = max_streams = 1;
     }
 
     if (!max_block_size)
@@ -1481,12 +1523,36 @@ void InterpreterSelectQuery::executeFetchColumns(
         if (!options.ignore_quota && (options.to_stage == QueryProcessingStage::Complete))
             quota = context->getQuota();
 
-        storage->read(query_plan, table_lock, metadata_snapshot, limits, leaf_limits, std::move(quota),
-                      required_columns, query_info, context, processing_stage, max_block_size, max_streams);
+        storage->read(query_plan, required_columns, metadata_snapshot,
+                      query_info, *context, processing_stage, max_block_size, max_streams);
+
+        /// Create step which reads from empty source if storage has no data.
+        if (!query_plan.isInitialized())
+        {
+            auto header = metadata_snapshot->getSampleBlockForColumns(
+                    required_columns, storage->getVirtuals(), storage->getStorageID());
+            addEmptySourceToQueryPlan(query_plan, header, query_info);
+        }
+
+        /// Extend lifetime of context, table lock, storage. Set limits and quota.
+        auto adding_limits_and_quota = std::make_unique<SettingQuotaAndLimitsStep>(
+                query_plan.getCurrentDataStream(),
+                storage,
+                std::move(table_lock),
+                limits,
+                leaf_limits,
+                std::move(quota),
+                context);
+        adding_limits_and_quota->setStepDescription("Set limits and quota after reading from storage");
+        query_plan.addStep(std::move(adding_limits_and_quota));
     }
     else
         throw Exception("Logical error in InterpreterSelectQuery: nowhere to read", ErrorCodes::LOGICAL_ERROR);
 
+    /// Specify the number of threads only if it wasn't specified in storage.
+    if (!query_plan.getMaxThreads())
+        query_plan.setMaxThreads(max_threads_execute_query);
+
     /// Aliases in table declaration.
     if (processing_stage == QueryProcessingStage::FetchColumns && alias_actions)
     {

src/Interpreters/InterpreterSelectQuery.h

@@ -94,6 +94,8 @@ public:
 
     const SelectQueryInfo & getQueryInfo() const { return query_info; }
 
+    static void addEmptySourceToQueryPlan(QueryPlan & query_plan, const Block & source_header, const SelectQueryInfo & query_info);
+
 private:
     InterpreterSelectQuery(
         const ASTPtr & query_ptr_,

src/Interpreters/MutationsInterpreter.cpp

@@ -5,6 +5,7 @@
 #include <Interpreters/MutationsInterpreter.h>
 #include <Interpreters/TreeRewriter.h>
 #include <Storages/MergeTree/MergeTreeData.h>
+#include <Storages/MergeTree/StorageFromMergeTreeDataPart.h>
 #include <Processors/Transforms/FilterTransform.h>
 #include <Processors/Transforms/ExpressionTransform.h>
 #include <Processors/Transforms/CreatingSetsTransform.h>
@@ -32,6 +33,7 @@ namespace DB
 
 namespace ErrorCodes
 {
+    extern const int NOT_IMPLEMENTED;
     extern const int BAD_ARGUMENTS;
     extern const int LOGICAL_ERROR;
     extern const int UNKNOWN_MUTATION_COMMAND;
@@ -92,6 +94,7 @@ std::optional<String> findFirstNonDeterministicFunctionName(const MutationComman
             if (finder_data.nondeterministic_function_name)
                 return finder_data.nondeterministic_function_name;
 
+            /// Currently UPDATE and DELETE both always have predicates so we can use fallthrough
             [[fallthrough]];
         }
 
@@ -110,7 +113,7 @@ std::optional<String> findFirstNonDeterministicFunctionName(const MutationComman
     return {};
 }
 
-ASTPtr prepareQueryAffectedAST(const std::vector<MutationCommand> & commands)
+ASTPtr prepareQueryAffectedAST(const std::vector<MutationCommand> & commands, const StoragePtr & storage, const Context & context)
 {
     /// Execute `SELECT count() FROM storage WHERE predicate1 OR predicate2 OR ...` query.
     /// The result can differ from the number of affected rows (e.g. if there is an UPDATE command that
@@ -125,20 +128,23 @@ ASTPtr prepareQueryAffectedAST(const std::vector<MutationCommand> & commands)
     count_func->arguments = std::make_shared<ASTExpressionList>();
     select->select()->children.push_back(count_func);
 
-    if (commands.size() == 1)
-        select->setExpression(ASTSelectQuery::Expression::WHERE, commands[0].predicate->clone());
-    else
+    ASTs conditions;
+    for (const MutationCommand & command : commands)
     {
-        auto coalesced_predicates = std::make_shared<ASTFunction>();
-        coalesced_predicates->name = "or";
-        coalesced_predicates->arguments = std::make_shared<ASTExpressionList>();
-        coalesced_predicates->children.push_back(coalesced_predicates->arguments);
-
-        for (const MutationCommand & command : commands)
-            coalesced_predicates->arguments->children.push_back(command.predicate->clone());
+        if (ASTPtr condition = getPartitionAndPredicateExpressionForMutationCommand(command, storage, context))
+            conditions.push_back(std::move(condition));
+    }
 
+    if (conditions.size() > 1)
+    {
+        auto coalesced_predicates = makeASTFunction("or");
+        coalesced_predicates->arguments->children = std::move(conditions);
         select->setExpression(ASTSelectQuery::Expression::WHERE, std::move(coalesced_predicates));
     }
+    else if (conditions.size() == 1)
+    {
+        select->setExpression(ASTSelectQuery::Expression::WHERE, std::move(conditions.front()));
+    }
 
     return select;
 }
@@ -167,8 +173,9 @@ ColumnDependencies getAllColumnDependencies(const StorageMetadataPtr & metadata_
 
 }
 
+
 bool isStorageTouchedByMutations(
-    StoragePtr storage,
+    const StoragePtr & storage,
     const StorageMetadataPtr & metadata_snapshot,
     const std::vector<MutationCommand> & commands,
     Context context_copy)
@@ -176,16 +183,33 @@ bool isStorageTouchedByMutations(
     if (commands.empty())
         return false;
 
+    bool all_commands_can_be_skipped = true;
+    auto storage_from_merge_tree_data_part = std::dynamic_pointer_cast<StorageFromMergeTreeDataPart>(storage);
     for (const MutationCommand & command : commands)
     {
         if (!command.predicate) /// The command touches all rows.
             return true;
+
+        if (command.partition && !storage_from_merge_tree_data_part)
+            throw Exception("ALTER UPDATE/DELETE ... IN PARTITION is not supported for non-MergeTree tables", ErrorCodes::NOT_IMPLEMENTED);
+
+        if (command.partition && storage_from_merge_tree_data_part)
+        {
+            const String partition_id = storage_from_merge_tree_data_part->getPartitionIDFromQuery(command.partition, context_copy);
+            if (partition_id == storage_from_merge_tree_data_part->getPartitionId())
+                all_commands_can_be_skipped = false;
+        }
+        else
+            all_commands_can_be_skipped = false;
     }
 
+    if (all_commands_can_be_skipped)
+        return false;
+
     context_copy.setSetting("max_streams_to_max_threads_ratio", 1);
     context_copy.setSetting("max_threads", 1);
 
-    ASTPtr select_query = prepareQueryAffectedAST(commands);
+    ASTPtr select_query = prepareQueryAffectedAST(commands, storage, context_copy);
 
     /// Interpreter must be alive, when we use result of execute() method.
     /// For some reason it may copy context and and give it into ExpressionBlockInputStream
@@ -202,9 +226,42 @@ bool isStorageTouchedByMutations(
 
     auto count = (*block.getByName("count()").column)[0].get<UInt64>();
     return count != 0;
-
 }
 
+
+ASTPtr getPartitionAndPredicateExpressionForMutationCommand(
+    const MutationCommand & command,
+    const StoragePtr & storage,
+    const Context & context
+)
+{
+    ASTPtr partition_predicate_as_ast_func;
+    if (command.partition)
+    {
+        String partition_id;
+
+        auto storage_merge_tree = std::dynamic_pointer_cast<MergeTreeData>(storage);
+        auto storage_from_merge_tree_data_part = std::dynamic_pointer_cast<StorageFromMergeTreeDataPart>(storage);
+        if (storage_merge_tree)
+            partition_id = storage_merge_tree->getPartitionIDFromQuery(command.partition, context);
+        else if (storage_from_merge_tree_data_part)
+            partition_id = storage_from_merge_tree_data_part->getPartitionIDFromQuery(command.partition, context);
+        else
+            throw Exception("ALTER UPDATE/DELETE ... IN PARTITION is not supported for non-MergeTree tables", ErrorCodes::NOT_IMPLEMENTED);
+
+        partition_predicate_as_ast_func = makeASTFunction("equals",
+                    std::make_shared<ASTIdentifier>("_partition_id"),
+                    std::make_shared<ASTLiteral>(partition_id)
+        );
+    }
+
+    if (command.predicate && command.partition)
+        return makeASTFunction("and", command.predicate->clone(), std::move(partition_predicate_as_ast_func));
+    else
+        return command.predicate ? command.predicate->clone() : partition_predicate_as_ast_func;
+}
+
+
 MutationsInterpreter::MutationsInterpreter(
     StoragePtr storage_,
     const StorageMetadataPtr & metadata_snapshot_,
@@ -349,7 +406,7 @@ ASTPtr MutationsInterpreter::prepare(bool dry_run)
             if (stages.empty() || !stages.back().column_to_updated.empty())
                 stages.emplace_back(context);
 
-            auto negated_predicate = makeASTFunction("isZeroOrNull", command.predicate->clone());
+            auto negated_predicate = makeASTFunction("isZeroOrNull", getPartitionAndPredicateExpressionForMutationCommand(command));
             stages.back().filters.push_back(negated_predicate);
         }
         else if (command.type == MutationCommand::UPDATE)
@@ -387,7 +444,7 @@ ASTPtr MutationsInterpreter::prepare(bool dry_run)
                 const auto & update_expr = kv.second;
                 auto updated_column = makeASTFunction("CAST",
                     makeASTFunction("if",
-                        command.predicate->clone(),
+                        getPartitionAndPredicateExpressionForMutationCommand(command),
                         makeASTFunction("CAST",
                             update_expr->clone(),
                             type_literal),
@@ -592,7 +649,7 @@ ASTPtr MutationsInterpreter::prepareInterpreterSelectQuery(std::vector<Stage> &
         for (const String & column : stage.output_columns)
             all_asts->children.push_back(std::make_shared<ASTIdentifier>(column));
 
-        auto syntax_result = TreeRewriter(context).analyze(all_asts, all_columns);
+        auto syntax_result = TreeRewriter(context).analyze(all_asts, all_columns, storage, metadata_snapshot);
         if (context.hasQueryContext())
             for (const auto & it : syntax_result->getScalars())
                 context.getQueryContext().addScalar(it.first, it.second);
@@ -763,10 +820,10 @@ const Block & MutationsInterpreter::getUpdatedHeader() const
 size_t MutationsInterpreter::evaluateCommandsSize()
 {
     for (const MutationCommand & command : commands)
-        if (unlikely(!command.predicate)) /// The command touches all rows.
+        if (unlikely(!command.predicate && !command.partition)) /// The command touches all rows.
             return mutation_ast->size();
 
-    return std::max(prepareQueryAffectedAST(commands)->size(), mutation_ast->size());
+    return std::max(prepareQueryAffectedAST(commands, storage, context)->size(), mutation_ast->size());
 }
 
 std::optional<SortDescription> MutationsInterpreter::getStorageSortDescriptionIfPossible(const Block & header) const
@@ -787,6 +844,11 @@ std::optional<SortDescription> MutationsInterpreter::getStorageSortDescriptionIf
     return sort_description;
 }
 
+ASTPtr MutationsInterpreter::getPartitionAndPredicateExpressionForMutationCommand(const MutationCommand & command) const
+{
+    return DB::getPartitionAndPredicateExpressionForMutationCommand(command, storage, context);
+}
+
 bool MutationsInterpreter::Stage::isAffectingAllColumns(const Names & storage_columns) const
 {
     /// is subset

src/Interpreters/MutationsInterpreter.h

@@ -20,7 +20,17 @@ using QueryPipelinePtr = std::unique_ptr<QueryPipeline>;
 
 /// Return false if the data isn't going to be changed by mutations.
 bool isStorageTouchedByMutations(
-    StoragePtr storage, const StorageMetadataPtr & metadata_snapshot, const std::vector<MutationCommand> & commands, Context context_copy);
+    const StoragePtr & storage,
+    const StorageMetadataPtr & metadata_snapshot,
+    const std::vector<MutationCommand> & commands,
+    Context context_copy
+);
+
+ASTPtr getPartitionAndPredicateExpressionForMutationCommand(
+    const MutationCommand & command,
+    const StoragePtr & storage,
+    const Context & context
+);
 
 /// Create an input stream that will read data from storage and apply mutation commands (UPDATEs, DELETEs, MATERIALIZEs)
 /// to this data.
@@ -59,6 +69,8 @@ private:
 
     std::optional<SortDescription> getStorageSortDescriptionIfPossible(const Block & header) const;
 
+    ASTPtr getPartitionAndPredicateExpressionForMutationCommand(const MutationCommand & command) const;
+
     StoragePtr storage;
     StorageMetadataPtr metadata_snapshot;
     MutationCommands commands;

src/Interpreters/SystemLog.cpp

@@ -25,6 +25,7 @@ namespace
 {
 
 constexpr size_t DEFAULT_SYSTEM_LOG_FLUSH_INTERVAL_MILLISECONDS = 7500;
+constexpr size_t DEFAULT_METRIC_LOG_COLLECT_INTERVAL_MILLISECONDS = 1000;
 
 /// Creates a system log with MergeTree engine using parameters from config
 template <typename TSystemLog>
@@ -125,7 +126,8 @@ SystemLogs::SystemLogs(Context & global_context, const Poco::Util::AbstractConfi
 
     if (metric_log)
     {
-        size_t collect_interval_milliseconds = config.getUInt64("metric_log.collect_interval_milliseconds");
+        size_t collect_interval_milliseconds = config.getUInt64("metric_log.collect_interval_milliseconds",
+                                                                DEFAULT_METRIC_LOG_COLLECT_INTERVAL_MILLISECONDS);
         metric_log->startCollectMetric(collect_interval_milliseconds);
     }
 

src/Interpreters/TreeRewriter.cpp

@@ -63,7 +63,7 @@ struct CustomizeFunctionsData
 
     const String & customized_func_name;
 
-    void visit(ASTFunction & func, ASTPtr &)
+    void visit(ASTFunction & func, ASTPtr &) const
     {
         if (Poco::toLower(func.name) == func_name)
         {
@@ -97,7 +97,7 @@ struct CustomizeFunctionsSuffixData
 
     const String & customized_func_suffix;
 
-    void visit(ASTFunction & func, ASTPtr &)
+    void visit(ASTFunction & func, ASTPtr &) const
     {
         if (endsWith(Poco::toLower(func.name), func_suffix))
         {
@@ -118,7 +118,7 @@ struct CustomizeAggregateFunctionsSuffixData
 
     const String & customized_func_suffix;
 
-    void visit(ASTFunction & func, ASTPtr &)
+    void visit(ASTFunction & func, ASTPtr &) const
     {
         const auto & instance = AggregateFunctionFactory::instance();
         if (instance.isAggregateFunctionName(func.name) && !endsWith(func.name, customized_func_suffix))

src/Parsers/ASTAlterQuery.cpp

@@ -90,7 +90,7 @@ void ASTAlterCommand::formatImpl(
         column->formatImpl(settings, state, frame);
         if (partition)
         {
-            settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str<< " IN PARTITION " << (settings.hilite ? hilite_none : "");
+            settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str << " IN PARTITION " << (settings.hilite ? hilite_none : "");
             partition->formatImpl(settings, state, frame);
         }
     }
@@ -150,7 +150,7 @@ void ASTAlterCommand::formatImpl(
         index->formatImpl(settings, state, frame);
         if (partition)
         {
-            settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str<< " IN PARTITION " << (settings.hilite ? hilite_none : "");
+            settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str << " IN PARTITION " << (settings.hilite ? hilite_none : "");
             partition->formatImpl(settings, state, frame);
         }
     }
@@ -161,7 +161,7 @@ void ASTAlterCommand::formatImpl(
         index->formatImpl(settings, state, frame);
         if (partition)
         {
-            settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str<< " IN PARTITION " << (settings.hilite ? hilite_none : "");
+            settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str << " IN PARTITION " << (settings.hilite ? hilite_none : "");
             partition->formatImpl(settings, state, frame);
         }
     }
@@ -178,7 +178,8 @@ void ASTAlterCommand::formatImpl(
     }
     else if (type == ASTAlterCommand::DROP_PARTITION)
     {
-        settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str << (detach ? "DETACH" : "DROP") << " PARTITION "
+        settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str
+                      << (detach ? "DETACH" : "DROP") << (part ? " PART " : " PARTITION ")
                       << (settings.hilite ? hilite_none : "");
         partition->formatImpl(settings, state, frame);
     }
@@ -271,7 +272,15 @@ void ASTAlterCommand::formatImpl(
     }
     else if (type == ASTAlterCommand::DELETE)
     {
-        settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str << "DELETE WHERE " << (settings.hilite ? hilite_none : "");
+        settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str << "DELETE" << (settings.hilite ? hilite_none : "");
+
+        if (partition)
+        {
+            settings.ostr << (settings.hilite ? hilite_keyword : "") << " IN PARTITION " << (settings.hilite ? hilite_none : "");
+            partition->formatImpl(settings, state, frame);
+        }
+
+        settings.ostr << (settings.hilite ? hilite_keyword : "") << " WHERE " << (settings.hilite ? hilite_none : "");
         predicate->formatImpl(settings, state, frame);
     }
     else if (type == ASTAlterCommand::UPDATE)
@@ -279,6 +288,12 @@ void ASTAlterCommand::formatImpl(
         settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str << "UPDATE " << (settings.hilite ? hilite_none : "");
         update_assignments->formatImpl(settings, state, frame);
 
+        if (partition)
+        {
+            settings.ostr << (settings.hilite ? hilite_keyword : "") << " IN PARTITION " << (settings.hilite ? hilite_none : "");
+            partition->formatImpl(settings, state, frame);
+        }
+
         settings.ostr << (settings.hilite ? hilite_keyword : "") << " WHERE " << (settings.hilite ? hilite_none : "");
         predicate->formatImpl(settings, state, frame);
     }
@@ -297,7 +312,7 @@ void ASTAlterCommand::formatImpl(
                       << (settings.hilite ? hilite_none : "");
         if (partition)
         {
-            settings.ostr << (settings.hilite ? hilite_keyword : "") << indent_str<< " IN PARTITION " << (settings.hilite ? hilite_none : "");
+            settings.ostr << (settings.hilite ? hilite_keyword : "") << " IN PARTITION " << (settings.hilite ? hilite_none : "");
             partition->formatImpl(settings, state, frame);
         }
     }

src/Parsers/ASTAlterQuery.h

@@ -103,7 +103,7 @@ public:
     */
     ASTPtr constraint;
 
-    /** Used in DROP PARTITION and ATTACH PARTITION FROM queries.
+    /** Used in DROP PARTITION, ATTACH PARTITION FROM, UPDATE, DELETE queries.
      *  The value or ID of the partition is stored here.
      */
     ASTPtr partition;

src/Parsers/ExpressionListParsers.cpp

@@ -55,6 +55,12 @@ const char * ParserComparisonExpression::operators[] =
     nullptr
 };
 
+const char * ParserComparisonExpression::overlapping_operators_to_skip[] =
+{
+    "IN PARTITION",
+    nullptr
+};
+
 const char * ParserLogicalNotExpression::operators[] =
 {
     "NOT", "not",
@@ -137,6 +143,14 @@ bool ParserLeftAssociativeBinaryOperatorList::parseImpl(Pos & pos, ASTPtr & node
             /// try to find any of the valid operators
 
             const char ** it;
+            Expected stub;
+            for (it = overlapping_operators_to_skip; *it; ++it)
+                if (ParserKeyword{*it}.checkWithoutMoving(pos, stub))
+                    break;
+
+            if (*it)
+                break;
+
             for (it = operators; *it; it += 2)
                 if (parseOperator(pos, *it, expected))
                     break;

src/Parsers/ExpressionListParsers.h

@@ -82,6 +82,7 @@ class ParserLeftAssociativeBinaryOperatorList : public IParserBase
 {
 private:
     Operators_t operators;
+    Operators_t overlapping_operators_to_skip = { (const char *[]){ nullptr } };
     ParserPtr first_elem_parser;
     ParserPtr remaining_elem_parser;
 
@@ -93,6 +94,11 @@ public:
     {
     }
 
+    ParserLeftAssociativeBinaryOperatorList(Operators_t operators_, Operators_t overlapping_operators_to_skip_, ParserPtr && first_elem_parser_)
+        : operators(operators_), overlapping_operators_to_skip(overlapping_operators_to_skip_), first_elem_parser(std::move(first_elem_parser_))
+    {
+    }
+
     ParserLeftAssociativeBinaryOperatorList(Operators_t operators_, ParserPtr && first_elem_parser_,
         ParserPtr && remaining_elem_parser_)
         : operators(operators_), first_elem_parser(std::move(first_elem_parser_)),
@@ -284,7 +290,8 @@ class ParserComparisonExpression : public IParserBase
 {
 private:
     static const char * operators[];
-    ParserLeftAssociativeBinaryOperatorList operator_parser {operators, std::make_unique<ParserBetweenExpression>()};
+    static const char * overlapping_operators_to_skip[];
+    ParserLeftAssociativeBinaryOperatorList operator_parser {operators, overlapping_operators_to_skip, std::make_unique<ParserBetweenExpression>()};
 
 protected:
     const char * getName() const  override{ return "comparison expression"; }

src/Parsers/ParserAlterQuery.cpp

@@ -52,13 +52,15 @@ bool ParserAlterCommand::parseImpl(Pos & pos, ASTPtr & node, Expected & expected
     ParserKeyword s_modify("MODIFY");
 
     ParserKeyword s_attach_partition("ATTACH PARTITION");
+    ParserKeyword s_attach_part("ATTACH PART");
     ParserKeyword s_detach_partition("DETACH PARTITION");
+    ParserKeyword s_detach_part("DETACH PART");
     ParserKeyword s_drop_partition("DROP PARTITION");
+    ParserKeyword s_drop_part("DROP PART");
     ParserKeyword s_move_partition("MOVE PARTITION");
+    ParserKeyword s_move_part("MOVE PART");
     ParserKeyword s_drop_detached_partition("DROP DETACHED PARTITION");
     ParserKeyword s_drop_detached_part("DROP DETACHED PART");
-    ParserKeyword s_attach_part("ATTACH PART");
-    ParserKeyword s_move_part("MOVE PART");
     ParserKeyword s_fetch_partition("FETCH PARTITION");
     ParserKeyword s_replace_partition("REPLACE PARTITION");
     ParserKeyword s_freeze("FREEZE");
@@ -77,7 +79,7 @@ bool ParserAlterCommand::parseImpl(Pos & pos, ASTPtr & node, Expected & expected
     ParserKeyword s_to_volume("TO VOLUME");
     ParserKeyword s_to_table("TO TABLE");
 
-    ParserKeyword s_delete_where("DELETE WHERE");
+    ParserKeyword s_delete("DELETE");
     ParserKeyword s_update("UPDATE");
     ParserKeyword s_where("WHERE");
     ParserKeyword s_to("TO");
@@ -161,6 +163,14 @@ bool ParserAlterCommand::parseImpl(Pos & pos, ASTPtr & node, Expected & expected
 
             command->type = ASTAlterCommand::DROP_PARTITION;
         }
+        else if (s_drop_part.ignore(pos, expected))
+        {
+            if (!parser_string_literal.parse(pos, command->partition, expected))
+                return false;
+
+            command->type = ASTAlterCommand::DROP_PARTITION;
+            command->part = true;
+        }
         else if (s_drop_detached_partition.ignore(pos, expected))
         {
             if (!parser_partition.parse(pos, command->partition, expected))
@@ -354,6 +364,15 @@ bool ParserAlterCommand::parseImpl(Pos & pos, ASTPtr & node, Expected & expected
             command->type = ASTAlterCommand::DROP_PARTITION;
             command->detach = true;
         }
+        else if (s_detach_part.ignore(pos, expected))
+        {
+            if (!parser_string_literal.parse(pos, command->partition, expected))
+                return false;
+
+            command->type = ASTAlterCommand::DROP_PARTITION;
+            command->part = true;
+            command->detach = true;
+        }
         else if (s_attach_partition.ignore(pos, expected))
         {
             if (!parser_partition.parse(pos, command->partition, expected))
@@ -487,8 +506,17 @@ bool ParserAlterCommand::parseImpl(Pos & pos, ASTPtr & node, Expected & expected
 
             command->type = ASTAlterCommand::MODIFY_SAMPLE_BY;
         }
-        else if (s_delete_where.ignore(pos, expected))
+        else if (s_delete.ignore(pos, expected))
         {
+            if (s_in_partition.ignore(pos, expected))
+            {
+                if (!parser_partition.parse(pos, command->partition, expected))
+                    return false;
+            }
+
+            if (!s_where.ignore(pos, expected))
+                return false;
+
             if (!parser_exp_elem.parse(pos, command->predicate, expected))
                 return false;
 
@@ -499,6 +527,12 @@ bool ParserAlterCommand::parseImpl(Pos & pos, ASTPtr & node, Expected & expected
             if (!parser_assignment_list.parse(pos, command->update_assignments, expected))
                 return false;
 
+            if (s_in_partition.ignore(pos, expected))
+            {
+                if (!parser_partition.parse(pos, command->partition, expected))
+                    return false;
+            }
+
             if (!s_where.ignore(pos, expected))
                 return false;
 

src/Parsers/ParserAlterQuery.h

@@ -10,7 +10,7 @@ namespace DB
   * ALTER TABLE [db.]name [ON CLUSTER cluster]
   *     [ADD COLUMN [IF NOT EXISTS] col_name type [AFTER col_after],]
   *     [DROP COLUMN [IF EXISTS] col_to_drop, ...]
-  *     [CLEAR COLUMN [IF EXISTS] col_to_clear [IN PARTITION partition],]
+  *     [CLEAR COLUMN [IF EXISTS] col_to_clear[ IN PARTITION partition],]
   *     [MODIFY COLUMN [IF EXISTS] col_to_modify type, ...]
   *     [RENAME COLUMN [IF EXISTS] col_name TO col_name]
   *     [MODIFY PRIMARY KEY (a, b, c...)]
@@ -19,8 +19,12 @@ namespace DB
   *     [DROP|DETACH|ATTACH PARTITION|PART partition, ...]
   *     [FETCH PARTITION partition FROM ...]
   *     [FREEZE [PARTITION] [WITH NAME name]]
-  *     [DELETE WHERE ...]
-  *     [UPDATE col_name = expr, ... WHERE ...]
+  *     [DELETE[ IN PARTITION partition] WHERE ...]
+  *     [UPDATE col_name = expr, ...[ IN PARTITION partition] WHERE ...]
+  *     [ADD INDEX [IF NOT EXISTS] index_name [AFTER index_name]]
+  *     [DROP INDEX [IF EXISTS] index_name]
+  *     [CLEAR INDEX [IF EXISTS] index_name IN PARTITION partition]
+  *     [MATERIALIZE INDEX [IF EXISTS] index_name [IN PARTITION partition]]
   * ALTER LIVE VIEW [db.name]
   *     [REFRESH]
   */

src/Processors/Formats/IRowInputFormat.cpp

@@ -65,8 +65,6 @@ Chunk IRowInputFormat::generate()
                 info.read_columns.clear();
                 if (!readRow(columns, info))
                     break;
-                if (params.callback)
-                    params.callback();
 
                 for (size_t column_idx = 0; column_idx < info.read_columns.size(); ++column_idx)
                 {

src/Processors/Formats/IRowInputFormat.h

@@ -27,9 +27,6 @@ struct RowInputFormatParams
     UInt64 allow_errors_num;
     Float64 allow_errors_ratio;
 
-    using ReadCallback = std::function<void()>;
-    ReadCallback callback;
-
     Poco::Timespan max_execution_time = 0;
     OverflowMode timeout_overflow_mode = OverflowMode::THROW;
 };

src/Processors/Formats/IRowOutputFormat.h

@@ -15,14 +15,6 @@ struct RowOutputFormatParams
 
     // Callback used to indicate that another row is written.
     WriteCallback callback;
-
-    /**
-     * some buffers (kafka / rabbit) split the rows internally using callback
-     * so we can push there formats without framing / delimiters
-     * (like ProtobufSingle). In other cases you can't write more than single row
-     * in unframed format.
-     */
-    bool ignore_no_row_delimiter = false;
 };
 
 class WriteBuffer;

src/Processors/Formats/Impl/ProtobufRowOutputFormat.cpp

@@ -23,18 +23,22 @@ ProtobufRowOutputFormat::ProtobufRowOutputFormat(
     const Block & header,
     const RowOutputFormatParams & params_,
     const FormatSchemaInfo & format_schema,
-    const bool use_length_delimiters_)
+    const FormatSettings & settings)
     : IRowOutputFormat(header, out_, params_)
     , data_types(header.getDataTypes())
-    , writer(out, ProtobufSchemas::instance().getMessageTypeForFormatSchema(format_schema), header.getNames(), use_length_delimiters_)
-    , throw_on_multiple_rows_undelimited(!use_length_delimiters_ && !params_.ignore_no_row_delimiter)
+    , writer(out,
+        ProtobufSchemas::instance().getMessageTypeForFormatSchema(format_schema),
+        header.getNames(), settings.protobuf.write_row_delimiters)
+    , allow_only_one_row(
+        !settings.protobuf.write_row_delimiters
+            && !settings.protobuf.allow_many_rows_no_delimiters)
 {
     value_indices.resize(header.columns());
 }
 
 void ProtobufRowOutputFormat::write(const Columns & columns, size_t row_num)
 {
-    if (throw_on_multiple_rows_undelimited && !first_row)
+    if (allow_only_one_row && !first_row)
     {
         throw Exception("The ProtobufSingle format can't be used to write multiple rows because this format doesn't have any row delimiter.", ErrorCodes::NO_ROW_DELIMITER);
     }
@@ -51,19 +55,23 @@ void ProtobufRowOutputFormat::write(const Columns & columns, size_t row_num)
 
 void registerOutputFormatProcessorProtobuf(FormatFactory & factory)
 {
-    for (bool use_length_delimiters : {false, true})
+    for (bool write_row_delimiters : {false, true})
     {
         factory.registerOutputFormatProcessor(
-            use_length_delimiters ? "Protobuf" : "ProtobufSingle",
-            [use_length_delimiters](WriteBuffer & buf,
+            write_row_delimiters ? "Protobuf" : "ProtobufSingle",
+            [write_row_delimiters](WriteBuffer & buf,
                const Block & header,
                const RowOutputFormatParams & params,
-               const FormatSettings & settings)
+               const FormatSettings & _settings)
             {
-                return std::make_shared<ProtobufRowOutputFormat>(buf, header, params,
-                    FormatSchemaInfo(settings.schema.format_schema, "Protobuf", true,
-                                     settings.schema.is_server, settings.schema.format_schema_path),
-                                     use_length_delimiters);
+                FormatSettings settings = _settings;
+                settings.protobuf.write_row_delimiters = write_row_delimiters;
+                return std::make_shared<ProtobufRowOutputFormat>(
+                    buf, header, params,
+                    FormatSchemaInfo(settings.schema.format_schema, "Protobuf",
+                        true, settings.schema.is_server,
+                        settings.schema.format_schema_path),
+                    settings);
             });
     }
 }

src/Processors/Formats/Impl/ProtobufRowOutputFormat.h

@@ -41,7 +41,7 @@ public:
         const Block & header,
         const RowOutputFormatParams & params_,
         const FormatSchemaInfo & format_schema,
-        const bool use_length_delimiters_);
+        const FormatSettings & settings);
 
     String getName() const override { return "ProtobufRowOutputFormat"; }
 
@@ -53,7 +53,7 @@ private:
     DataTypes data_types;
     ProtobufWriter writer;
     std::vector<size_t> value_indices;
-    const bool throw_on_multiple_rows_undelimited;
+    const bool allow_only_one_row;
 };
 
 }

src/Processors/Formats/Impl/ValuesBlockInputFormat.cpp

@@ -54,8 +54,6 @@ Chunk ValuesBlockInputFormat::generate()
             if (buf.eof() || *buf.position() == ';')
                 break;
             readRow(columns, rows_in_block);
-            if (params.callback)
-                params.callback();
         }
         catch (Exception & e)
         {

src/Processors/QueryPipeline.cpp

@@ -95,6 +95,12 @@ void QueryPipeline::addTransform(ProcessorPtr transform)
     pipe.addTransform(std::move(transform));
 }
 
+void QueryPipeline::transform(const Transformer & transformer)
+{
+    checkInitializedAndNotCompleted();
+    pipe.transform(transformer);
+}
+
 void QueryPipeline::setSinks(const Pipe::ProcessorGetterWithStreamKind & getter)
 {
     checkInitializedAndNotCompleted();

src/Processors/QueryPipeline.h

@@ -53,6 +53,11 @@ public:
     void addSimpleTransform(const Pipe::ProcessorGetterWithStreamKind & getter);
     /// Add transform with getNumStreams() input ports.
     void addTransform(ProcessorPtr transform);
+
+    using Transformer = std::function<Processors(OutputPortRawPtrs ports)>;
+    /// Transform pipeline in general way.
+    void transform(const Transformer & transformer);
+
     /// Add TotalsHavingTransform. Resize pipeline to single input. Adds totals port.
     void addTotalsHavingTransform(ProcessorPtr transform);
     /// Add transform which calculates extremes. This transform adds extremes port and doesn't change inputs number.
@@ -105,6 +110,9 @@ public:
     void addInterpreterContext(std::shared_ptr<Context> context) { pipe.addInterpreterContext(std::move(context)); }
     void addStorageHolder(StoragePtr storage) { pipe.addStorageHolder(std::move(storage)); }
     void addQueryPlan(std::unique_ptr<QueryPlan> plan) { pipe.addQueryPlan(std::move(plan)); }
+    void setLimits(const StreamLocalLimits & limits) { pipe.setLimits(limits); }
+    void setLeafLimits(const SizeLimits & limits) { pipe.setLeafLimits(limits); }
+    void setQuota(const std::shared_ptr<const EnabledQuota> & quota) { pipe.setQuota(quota); }
 
     /// For compatibility with IBlockInputStream.
     void setProgressCallback(const ProgressCallback & callback);

src/Processors/QueryPlan/AddingConstColumnStep.cpp

@@ -0,0 +1,41 @@
+#include <Processors/QueryPlan/AddingConstColumnStep.h>
+#include <Processors/QueryPipeline.h>
+#include <Processors/Transforms/AddingConstColumnTransform.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,
+        }
+    };
+}
+
+AddingConstColumnStep::AddingConstColumnStep(const DataStream & input_stream_, ColumnWithTypeAndName column_)
+    : ITransformingStep(input_stream_,
+                        AddingConstColumnTransform::transformHeader(input_stream_.header, column_),
+                        getTraits())
+    , column(std::move(column_))
+{
+}
+
+void AddingConstColumnStep::transformPipeline(QueryPipeline & pipeline)
+{
+    pipeline.addSimpleTransform([&](const Block & header)
+    {
+        return std::make_shared<AddingConstColumnTransform>(header, column);
+    });
+}
+
+}

src/Processors/QueryPlan/AddingConstColumnStep.h

@@ -0,0 +1,22 @@
+#pragma once
+#include <Processors/QueryPlan/ITransformingStep.h>
+
+namespace DB
+{
+
+/// Adds a materialized const column with a specified value.
+class AddingConstColumnStep : public ITransformingStep
+{
+public:
+    AddingConstColumnStep(const DataStream & input_stream_, ColumnWithTypeAndName column_);
+
+    String getName() const override { return "AddingConstColumn"; }
+
+    void transformPipeline(QueryPipeline & pipeline) override;
+
+private:
+    ColumnWithTypeAndName column;
+};
+
+}
+

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_,
+    ColumnsDescription columns_,
+    const Context & context_)
+    : ITransformingStep(input_stream_, result_header_, getTraits())
+    , columns(std::move(columns_))
+    , 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, columns, context);
+    });
+}
+
+}

src/Processors/QueryPlan/AddingMissedStep.h

@@ -0,0 +1,26 @@
+#pragma once
+#include <Processors/QueryPlan/ITransformingStep.h>
+#include <Storages/ColumnsDescription.h>
+
+namespace DB
+{
+
+/// Convert one block structure to another. See ConvertingTransform.
+class AddingMissedStep : public ITransformingStep
+{
+public:
+    AddingMissedStep(const DataStream & input_stream_,
+                     Block result_header_,
+                     ColumnsDescription columns_,
+                     const Context & context_);
+
+    String getName() const override { return "AddingMissed"; }
+
+    void transformPipeline(QueryPipeline & pipeline) override;
+
+private:
+    ColumnsDescription columns;
+    const Context & context;
+};
+
+}

src/Processors/QueryPlan/ISourceStep.cpp

@@ -14,7 +14,8 @@ QueryPipelinePtr ISourceStep::updatePipeline(QueryPipelines)
     auto pipeline = std::make_unique<QueryPipeline>();
     QueryPipelineProcessorsCollector collector(*pipeline, this);
     initializePipeline(*pipeline);
-    processors = collector.detachProcessors();
+    auto added_processors = collector.detachProcessors();
+    processors.insert(processors.end(), added_processors.begin(), added_processors.end());
     return pipeline;
 }
 

src/Processors/QueryPlan/ISourceStep.h

@@ -16,7 +16,7 @@ public:
 
     void describePipeline(FormatSettings & settings) const override;
 
-private:
+protected:
     /// We collect processors got after pipeline transformation.
     Processors processors;
 };

src/Processors/QueryPlan/MaterializingStep.cpp

@@ -0,0 +1,39 @@
+#include <Processors/QueryPlan/MaterializingStep.h>
+#include <Processors/QueryPipeline.h>
+#include <Processors/Transforms/MaterializingTransform.h>
+
+#include <DataStreams/materializeBlock.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,
+        }
+    };
+}
+
+MaterializingStep::MaterializingStep(const DataStream & input_stream_)
+    : ITransformingStep(input_stream_, materializeBlock(input_stream_.header), getTraits())
+{
+}
+
+void MaterializingStep::transformPipeline(QueryPipeline & pipeline)
+{
+    pipeline.addSimpleTransform([&](const Block & header)
+    {
+        return std::make_shared<MaterializingTransform>(header);
+    });
+}
+
+}

src/Processors/QueryPlan/MaterializingStep.h

@@ -0,0 +1,18 @@
+#pragma once
+#include <Processors/QueryPlan/ITransformingStep.h>
+
+namespace DB
+{
+
+/// Materialize constants. See MaterializingTransform.
+class MaterializingStep : public ITransformingStep
+{
+public:
+    explicit MaterializingStep(const DataStream & input_stream_);
+
+    String getName() const override { return "Materializing"; }
+
+    void transformPipeline(QueryPipeline & pipeline) override;
+};
+
+}

src/Processors/QueryPlan/MergingFinal.cpp

@@ -0,0 +1,164 @@
+#include <Processors/QueryPlan/MergingFinal.h>
+#include <Processors/QueryPipeline.h>
+#include <Processors/Merges/AggregatingSortedTransform.h>
+#include <Processors/Merges/CollapsingSortedTransform.h>
+#include <Processors/Merges/MergingSortedTransform.h>
+#include <Processors/Merges/ReplacingSortedTransform.h>
+#include <Processors/Merges/SummingSortedTransform.h>
+#include <Processors/Merges/VersionedCollapsingTransform.h>
+#include <Processors/Transforms/AddingSelectorTransform.h>
+#include <Processors/Transforms/CopyTransform.h>
+#include <IO/Operators.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int LOGICAL_ERROR;
+}
+
+static ITransformingStep::Traits getTraits()
+{
+    return ITransformingStep::Traits
+    {
+        {
+            .preserves_distinct_columns = true,
+            .returns_single_stream = false,
+            .preserves_number_of_streams = false,
+            .preserves_sorting = false,
+        },
+        {
+            .preserves_number_of_rows = true,
+        }
+    };
+}
+
+MergingFinal::MergingFinal(
+    const DataStream & input_stream,
+    size_t num_output_streams_,
+    SortDescription sort_description_,
+    MergeTreeData::MergingParams params_,
+    Names partition_key_columns_,
+    size_t max_block_size_)
+    : ITransformingStep(input_stream, input_stream.header, getTraits())
+    , num_output_streams(num_output_streams_)
+    , sort_description(std::move(sort_description_))
+    , merging_params(std::move(params_))
+    , partition_key_columns(std::move(partition_key_columns_))
+    , max_block_size(max_block_size_)
+{
+    /// TODO: check input_stream is partially sorted (each port) by the same description.
+//    output_stream->sort_description = sort_description;
+//    output_stream->sort_mode = DataStream::SortMode::Stream;
+}
+
+void MergingFinal::transformPipeline(QueryPipeline & pipeline)
+{
+    const auto & header = pipeline.getHeader();
+    size_t num_outputs = pipeline.getNumStreams();
+
+    auto get_merging_processor = [&]() -> MergingTransformPtr
+    {
+        switch (merging_params.mode)
+        {
+            case MergeTreeData::MergingParams::Ordinary:
+            {
+                return std::make_shared<MergingSortedTransform>(header, num_outputs,
+                           sort_description, max_block_size);
+            }
+
+            case MergeTreeData::MergingParams::Collapsing:
+                return std::make_shared<CollapsingSortedTransform>(header, num_outputs,
+                           sort_description, merging_params.sign_column, true, max_block_size);
+
+            case MergeTreeData::MergingParams::Summing:
+                return std::make_shared<SummingSortedTransform>(header, num_outputs,
+                           sort_description, merging_params.columns_to_sum, partition_key_columns, max_block_size);
+
+            case MergeTreeData::MergingParams::Aggregating:
+                return std::make_shared<AggregatingSortedTransform>(header, num_outputs,
+                           sort_description, max_block_size);
+
+            case MergeTreeData::MergingParams::Replacing:
+                return std::make_shared<ReplacingSortedTransform>(header, num_outputs,
+                           sort_description, merging_params.version_column, max_block_size);
+
+            case MergeTreeData::MergingParams::VersionedCollapsing:
+                return std::make_shared<VersionedCollapsingTransform>(header, num_outputs,
+                           sort_description, merging_params.sign_column, max_block_size);
+
+            case MergeTreeData::MergingParams::Graphite:
+                throw Exception("GraphiteMergeTree doesn't support FINAL", ErrorCodes::LOGICAL_ERROR);
+        }
+
+        __builtin_unreachable();
+    };
+
+    if (num_output_streams <= 1 || sort_description.empty())
+    {
+        pipeline.addTransform(get_merging_processor());
+        return;
+    }
+
+    ColumnNumbers key_columns;
+    key_columns.reserve(sort_description.size());
+
+    for (auto & desc : sort_description)
+    {
+        if (!desc.column_name.empty())
+            key_columns.push_back(header.getPositionByName(desc.column_name));
+        else
+            key_columns.emplace_back(desc.column_number);
+    }
+
+    pipeline.addSimpleTransform([&](const Block & stream_header)
+    {
+        return std::make_shared<AddingSelectorTransform>(stream_header, num_output_streams, key_columns);
+    });
+
+    pipeline.transform([&](OutputPortRawPtrs ports)
+    {
+        Processors transforms;
+        std::vector<OutputPorts::iterator> output_ports;
+        transforms.reserve(ports.size() + num_output_streams);
+        output_ports.reserve(ports.size());
+
+        for (auto & port : ports)
+        {
+            auto copier = std::make_shared<CopyTransform>(header, num_output_streams);
+            connect(*port, copier->getInputPort());
+            output_ports.emplace_back(copier->getOutputs().begin());
+            transforms.emplace_back(std::move(copier));
+        }
+
+        for (size_t i = 0; i < num_output_streams; ++i)
+        {
+            auto merge = get_merging_processor();
+            merge->setSelectorPosition(i);
+            auto input = merge->getInputs().begin();
+
+            /// Connect i-th merge with i-th input port of every copier.
+            for (size_t j = 0; j < ports.size(); ++j)
+            {
+                connect(*output_ports[j], *input);
+                ++output_ports[j];
+                ++input;
+            }
+
+            transforms.emplace_back(std::move(merge));
+        }
+
+        return transforms;
+    });
+}
+
+void MergingFinal::describeActions(FormatSettings & settings) const
+{
+    String prefix(settings.offset, ' ');
+    settings.out << prefix << "Sort description: ";
+    dumpSortDescription(sort_description, input_streams.front().header, settings.out);
+    settings.out << '\n';
+}
+
+}

src/Processors/QueryPlan/MergingFinal.h

@@ -0,0 +1,35 @@
+#pragma once
+#include <Processors/QueryPlan/ITransformingStep.h>
+#include <Core/SortDescription.h>
+#include <Storages/MergeTree/MergeTreeData.h>
+
+namespace DB
+{
+
+/// Merge streams of data into single sorted stream.
+class MergingFinal : public ITransformingStep
+{
+public:
+    explicit MergingFinal(
+            const DataStream & input_stream,
+            size_t num_output_streams_,
+            SortDescription sort_description_,
+            MergeTreeData::MergingParams params_,
+            Names partition_key_columns_,
+            size_t max_block_size_);
+
+    String getName() const override { return "MergingFinal"; }
+
+    void transformPipeline(QueryPipeline & pipeline) override;
+
+    void describeActions(FormatSettings & settings) const override;
+
+private:
+    size_t num_output_streams;
+    SortDescription sort_description;
+    MergeTreeData::MergingParams merging_params;
+    Names partition_key_columns;
+    size_t max_block_size;
+};
+
+}

src/Processors/QueryPlan/QueryPlan.cpp

@@ -185,6 +185,17 @@ QueryPipelinePtr QueryPlan::buildQueryPipeline()
     return last_pipeline;
 }
 
+Pipe QueryPlan::convertToPipe()
+{
+    if (!isInitialized())
+        return {};
+
+    if (isCompleted())
+        throw Exception("Cannot convert completed QueryPlan to Pipe", ErrorCodes::LOGICAL_ERROR);
+
+    return QueryPipeline::getPipe(std::move(*buildQueryPipeline()));
+}
+
 void QueryPlan::addInterpreterContext(std::shared_ptr<Context> context)
 {
     interpreter_context.emplace_back(std::move(context));

src/Processors/QueryPlan/QueryPlan.h

@@ -17,6 +17,11 @@ using QueryPipelinePtr = std::unique_ptr<QueryPipeline>;
 class Context;
 class WriteBuffer;
 
+class QueryPlan;
+using QueryPlanPtr = std::unique_ptr<QueryPlan>;
+
+class Pipe;
+
 /// A tree of query steps.
 /// The goal of QueryPlan is to build QueryPipeline.
 /// QueryPlan let delay pipeline creation which is helpful for pipeline-level optimisations.
@@ -28,7 +33,7 @@ public:
     QueryPlan(QueryPlan &&);
     QueryPlan & operator=(QueryPlan &&);
 
-    void unitePlans(QueryPlanStepPtr step, std::vector<std::unique_ptr<QueryPlan>> plans);
+    void unitePlans(QueryPlanStepPtr step, std::vector<QueryPlanPtr> plans);
     void addStep(QueryPlanStepPtr step);
 
     bool isInitialized() const { return root != nullptr; } /// Tree is not empty
@@ -39,6 +44,9 @@ public:
 
     QueryPipelinePtr buildQueryPipeline();
 
+    /// If initialized, build pipeline and convert to pipe. Otherwise, return empty pipe.
+    Pipe convertToPipe();
+
     struct ExplainPlanOptions
     {
         /// Add output header to step.
@@ -61,6 +69,7 @@ public:
     /// Set upper limit for the recommend number of threads. Will be applied to the newly-created pipelines.
     /// TODO: make it in a better way.
     void setMaxThreads(size_t max_threads_) { max_threads = max_threads_; }
+    size_t getMaxThreads() const { return max_threads; }
 
     void addInterpreterContext(std::shared_ptr<Context> context);
 

src/Processors/QueryPlan/ReadFromPreparedSource.cpp

@@ -5,7 +5,7 @@ namespace DB
 {
 
 ReadFromPreparedSource::ReadFromPreparedSource(Pipe pipe_, std::shared_ptr<Context> context_)
-    : ISourceStep(DataStream{.header = pipe_.getHeader(), .has_single_port = true})
+    : ISourceStep(DataStream{.header = pipe_.getHeader()})
     , pipe(std::move(pipe_))
     , context(std::move(context_))
 {
@@ -13,7 +13,11 @@ ReadFromPreparedSource::ReadFromPreparedSource(Pipe pipe_, std::shared_ptr<Conte
 
 void ReadFromPreparedSource::initializePipeline(QueryPipeline & pipeline)
 {
+    for (const auto & processor : pipe.getProcessors())
+        processors.emplace_back(processor);
+
     pipeline.init(std::move(pipe));
+
     if (context)
         pipeline.addInterpreterContext(std::move(context));
 }

src/Processors/QueryPlan/ReadFromPreparedSource.h

@@ -11,7 +11,7 @@ class ReadFromPreparedSource : public ISourceStep
 public:
     explicit ReadFromPreparedSource(Pipe pipe_, std::shared_ptr<Context> context_ = nullptr);
 
-    String getName() const override { return "ReadNothing"; }
+    String getName() const override { return "ReadFromPreparedSource"; }
 
     void initializePipeline(QueryPipeline & pipeline) override;
 
@@ -20,4 +20,16 @@ private:
     std::shared_ptr<Context> context;
 };
 
+class ReadFromStorageStep : public ReadFromPreparedSource
+{
+public:
+    ReadFromStorageStep(Pipe pipe_, String storage_name)
+        : ReadFromPreparedSource(std::move(pipe_))
+    {
+        setStepDescription(storage_name);
+    }
+
+    String getName() const override { return "ReadFromStorage"; }
+};
+
 }

src/Processors/QueryPlan/ReadFromStorageStep.cpp

@@ -1,114 +0,0 @@
-#include <Processors/QueryPlan/ReadFromStorageStep.h>
-
-#include <Interpreters/Context.h>
-#include <Processors/Sources/NullSource.h>
-#include <Processors/Transforms/ExpressionTransform.h>
-#include <Processors/Transforms/FilterTransform.h>
-#include <Processors/Pipe.h>
-#include <Processors/QueryPipeline.h>
-#include <Storages/IStorage.h>
-
-namespace DB
-{
-
-ReadFromStorageStep::ReadFromStorageStep(
-    TableLockHolder table_lock,
-    StorageMetadataPtr metadata_snapshot,
-    StreamLocalLimits & limits,
-    SizeLimits & leaf_limits,
-    std::shared_ptr<const EnabledQuota> quota,
-    StoragePtr storage,
-    const Names & required_columns,
-    SelectQueryInfo & query_info,
-    std::shared_ptr<Context> context,
-    QueryProcessingStage::Enum processing_stage,
-    size_t max_block_size,
-    size_t max_streams)
-{
-    /// Note: we read from storage in constructor of step because we don't know real header before reading.
-    /// It will be fixed when storage return QueryPlanStep itself.
-
-    Pipe pipe = storage->read(required_columns, metadata_snapshot, query_info, *context, processing_stage, max_block_size, max_streams);
-
-    if (pipe.empty())
-    {
-        pipe = Pipe(std::make_shared<NullSource>(metadata_snapshot->getSampleBlockForColumns(required_columns, storage->getVirtuals(), storage->getStorageID())));
-
-        if (query_info.prewhere_info)
-        {
-            if (query_info.prewhere_info->alias_actions)
-            {
-                pipe.addSimpleTransform([&](const Block & header)
-                {
-                    return std::make_shared<ExpressionTransform>(header, query_info.prewhere_info->alias_actions);
-                });
-            }
-
-            pipe.addSimpleTransform([&](const Block & header)
-            {
-                return std::make_shared<FilterTransform>(
-                    header,
-                    query_info.prewhere_info->prewhere_actions,
-                    query_info.prewhere_info->prewhere_column_name,
-                    query_info.prewhere_info->remove_prewhere_column);
-            });
-
-            // To remove additional columns
-            // In some cases, we did not read any marks so that the pipeline.streams is empty
-            // Thus, some columns in prewhere are not removed as expected
-            // This leads to mismatched header in distributed table
-            if (query_info.prewhere_info->remove_columns_actions)
-            {
-                pipe.addSimpleTransform([&](const Block & header)
-                {
-                    return std::make_shared<ExpressionTransform>(
-                            header, query_info.prewhere_info->remove_columns_actions);
-                });
-            }
-        }
-    }
-
-    pipeline = std::make_unique<QueryPipeline>();
-    QueryPipelineProcessorsCollector collector(*pipeline, this);
-
-    pipe.setLimits(limits);
-
-    /**
-      * Leaf size limits should be applied only for local processing of distributed queries.
-      * Such limits allow to control the read stage on leaf nodes and exclude the merging stage.
-      * Consider the case when distributed query needs to read from multiple shards. Then leaf
-      * limits will be applied on the shards only (including the root node) but will be ignored
-      * on the results merging stage.
-      */
-    if (!storage->isRemote())
-        pipe.setLeafLimits(leaf_limits);
-
-    if (quota)
-        pipe.setQuota(quota);
-
-    pipeline->init(std::move(pipe));
-
-    /// Add resources to pipeline. The order is important.
-    /// Add in reverse order of destruction. Pipeline will be destroyed at the end in case of exception.
-    pipeline->addInterpreterContext(std::move(context));
-    pipeline->addStorageHolder(std::move(storage));
-    pipeline->addTableLock(std::move(table_lock));
-
-    processors = collector.detachProcessors();
-
-    output_stream = DataStream{.header = pipeline->getHeader(), .has_single_port = pipeline->getNumStreams() == 1};
-}
-
-ReadFromStorageStep::~ReadFromStorageStep() = default;
-
-QueryPipelinePtr ReadFromStorageStep::updatePipeline(QueryPipelines)
-{
-    return std::move(pipeline);
-}
-
-void ReadFromStorageStep::describePipeline(FormatSettings & settings) const
-{
-    IQueryPlanStep::describePipeline(processors, settings);
-}
-
-}

src/Processors/QueryPlan/ReadFromStorageStep.h

@@ -1,53 +0,0 @@
-#pragma once
-#include <Processors/QueryPlan/IQueryPlanStep.h>
-#include <Core/QueryProcessingStage.h>
-#include <Storages/TableLockHolder.h>
-#include <DataStreams/StreamLocalLimits.h>
-
-namespace DB
-{
-
-class IStorage;
-using StoragePtr = std::shared_ptr<IStorage>;
-
-struct StorageInMemoryMetadata;
-using StorageMetadataPtr = std::shared_ptr<const StorageInMemoryMetadata>;
-
-struct SelectQueryInfo;
-
-struct PrewhereInfo;
-
-class EnabledQuota;
-
-/// Reads from storage.
-class ReadFromStorageStep : public IQueryPlanStep
-{
-public:
-    ReadFromStorageStep(
-        TableLockHolder table_lock,
-        StorageMetadataPtr metadata_snapshot,
-        StreamLocalLimits & limits,
-        SizeLimits & leaf_limits,
-        std::shared_ptr<const EnabledQuota> quota,
-        StoragePtr storage,
-        const Names & required_columns,
-        SelectQueryInfo & query_info,
-        std::shared_ptr<Context> context,
-        QueryProcessingStage::Enum processing_stage,
-        size_t max_block_size,
-        size_t max_streams);
-
-    ~ReadFromStorageStep() override;
-
-    String getName() const override { return "ReadFromStorage"; }
-
-    QueryPipelinePtr updatePipeline(QueryPipelines) override;
-
-    void describePipeline(FormatSettings & settings) const override;
-
-private:
-    QueryPipelinePtr pipeline;
-    Processors processors;
-};
-
-}

src/Processors/QueryPlan/ReverseRowsStep.cpp

@@ -0,0 +1,37 @@
+#include <Processors/QueryPlan/ReverseRowsStep.h>
+#include <Processors/QueryPipeline.h>
+#include <Processors/Transforms/ReverseTransform.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 = false,
+        },
+        {
+            .preserves_number_of_rows = true,
+        }
+    };
+}
+
+ReverseRowsStep::ReverseRowsStep(const DataStream & input_stream_)
+    : ITransformingStep(input_stream_, input_stream_.header, getTraits())
+{
+}
+
+void ReverseRowsStep::transformPipeline(QueryPipeline & pipeline)
+{
+    pipeline.addSimpleTransform([&](const Block & header)
+    {
+        return std::make_shared<ReverseTransform>(header);
+    });
+}
+
+}

src/Processors/QueryPlan/ReverseRowsStep.h

@@ -0,0 +1,18 @@
+#pragma once
+#include <Processors/QueryPlan/ITransformingStep.h>
+
+namespace DB
+{
+
+/// Reverse rows in chunk.
+class ReverseRowsStep : public ITransformingStep
+{
+public:
+    ReverseRowsStep(const DataStream & input_stream_);
+
+    String getName() const override { return "ReverseRows"; }
+
+    void transformPipeline(QueryPipeline & pipeline) override;
+};
+
+}

src/Processors/QueryPlan/SettingQuotaAndLimitsStep.cpp

@@ -0,0 +1,71 @@
+#include <Processors/QueryPlan/SettingQuotaAndLimitsStep.h>
+#include <Processors/QueryPipeline.h>
+#include <Storages/IStorage.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,
+        }
+    };
+}
+
+SettingQuotaAndLimitsStep::SettingQuotaAndLimitsStep(
+    const DataStream & input_stream_,
+    StoragePtr storage_,
+    TableLockHolder table_lock_,
+    StreamLocalLimits & limits_,
+    SizeLimits & leaf_limits_,
+    std::shared_ptr<const EnabledQuota> quota_,
+    std::shared_ptr<Context> context_)
+    : ITransformingStep(input_stream_, input_stream_.header, getTraits())
+    , context(std::move(context_))
+    , storage(std::move(storage_))
+    , table_lock(std::move(table_lock_))
+    , limits(limits_)
+    , leaf_limits(leaf_limits_)
+    , quota(std::move(quota_))
+{
+}
+
+void SettingQuotaAndLimitsStep::transformPipeline(QueryPipeline & pipeline)
+{
+    /// Table lock is stored inside pipeline here.
+    pipeline.setLimits(limits);
+
+    /**
+      * Leaf size limits should be applied only for local processing of distributed queries.
+      * Such limits allow to control the read stage on leaf nodes and exclude the merging stage.
+      * Consider the case when distributed query needs to read from multiple shards. Then leaf
+      * limits will be applied on the shards only (including the root node) but will be ignored
+      * on the results merging stage.
+      */
+    if (!storage->isRemote())
+        pipeline.setLeafLimits(leaf_limits);
+
+    if (quota)
+        pipeline.setQuota(quota);
+
+    /// Order of resources below is important.
+    if (context)
+        pipeline.addInterpreterContext(std::move(context));
+
+    if (storage)
+        pipeline.addStorageHolder(std::move(storage));
+
+    if (table_lock)
+        pipeline.addTableLock(std::move(table_lock));
+}
+
+}

src/Processors/QueryPlan/SettingQuotaAndLimitsStep.h

@@ -0,0 +1,44 @@
+#pragma once
+#include <Processors/QueryPlan/ITransformingStep.h>
+#include <Storages/TableLockHolder.h>
+#include <DataStreams/StreamLocalLimits.h>
+
+namespace DB
+{
+
+class IStorage;
+using StoragePtr = std::shared_ptr<IStorage>;
+
+struct StorageInMemoryMetadata;
+using StorageMetadataPtr = std::shared_ptr<const StorageInMemoryMetadata>;
+
+class EnabledQuota;
+
+/// Add limits, quota, table_lock and other stuff to pipeline.
+/// Doesn't change DataStream.
+class SettingQuotaAndLimitsStep : public ITransformingStep
+{
+public:
+    SettingQuotaAndLimitsStep(
+        const DataStream & input_stream_,
+        StoragePtr storage_,
+        TableLockHolder table_lock_,
+        StreamLocalLimits & limits_,
+        SizeLimits & leaf_limits_,
+        std::shared_ptr<const EnabledQuota> quota_,
+        std::shared_ptr<Context> context_);
+
+    String getName() const override { return "SettingQuotaAndLimits"; }
+
+    void transformPipeline(QueryPipeline & pipeline) override;
+
+private:
+    std::shared_ptr<Context> context;
+    StoragePtr storage;
+    TableLockHolder table_lock;
+    StreamLocalLimits limits;
+    SizeLimits leaf_limits;
+    std::shared_ptr<const EnabledQuota> quota;
+};
+
+}

src/Processors/QueryPlan/UnionStep.cpp

@@ -30,7 +30,7 @@ QueryPipelinePtr UnionStep::updatePipeline(QueryPipelines pipelines)
         return pipeline;
     }
 
-    *pipeline = QueryPipeline::unitePipelines(std::move(pipelines), output_stream->header ,max_threads);
+    *pipeline = QueryPipeline::unitePipelines(std::move(pipelines), output_stream->header, max_threads);
 
     processors = collector.detachProcessors();
     return pipeline;

src/Processors/QueryPlan/UnionStep.h

@@ -9,7 +9,7 @@ class UnionStep : public IQueryPlanStep
 {
 public:
     /// max_threads is used to limit the number of threads for result pipeline.
-    UnionStep(DataStreams input_streams_, Block result_header, size_t max_threads_);
+    UnionStep(DataStreams input_streams_, Block result_header, size_t max_threads_ = 0);
 
     String getName() const override { return "Union"; }
 

src/Processors/Transforms/AddingConstColumnTransform.h

@@ -4,33 +4,40 @@
 namespace DB
 {
 
+namespace ErrorCodes
+{
+    extern const int LOGICAL_ERROR;
+}
+
 /// Adds a materialized const column to the chunk with a specified value.
-template <typename T>
 class AddingConstColumnTransform : public ISimpleTransform
 {
 public:
-    AddingConstColumnTransform(const Block & header, DataTypePtr data_type_, T value_, const String & column_name_)
-        : ISimpleTransform(header, addColumn(header, data_type_, column_name_), false)
-        , data_type(std::move(data_type_)), value(value_) {}
+    AddingConstColumnTransform(const Block & header, ColumnWithTypeAndName column_)
+        : ISimpleTransform(header, transformHeader(header, column_), false)
+        , column(std::move(column_))
+    {
+        if (!column.column || !isColumnConst(*column.column) || !column.column->empty())
+            throw Exception("AddingConstColumnTransform expected empty const column", ErrorCodes::LOGICAL_ERROR);
+    }
 
     String getName() const override { return "AddingConstColumnTransform"; }
 
+    static Block transformHeader(Block header, ColumnWithTypeAndName & column_)
+    {
+        header.insert(column_);
+        return header;
+    }
+
 protected:
     void transform(Chunk & chunk) override
     {
         auto num_rows = chunk.getNumRows();
-        chunk.addColumn(data_type->createColumnConst(num_rows, value)->convertToFullColumnIfConst());
+        chunk.addColumn(column.column->cloneResized(num_rows)->convertToFullColumnIfConst());
     }
 
 private:
-    static Block addColumn(Block header, const DataTypePtr & data_type, const String & column_name)
-    {
-        header.insert({data_type->createColumn(), data_type, column_name});
-        return header;
-    }
-
-    DataTypePtr data_type;
-    T value;
+    ColumnWithTypeAndName column;
 };
 
 }

src/Processors/ya.make

@@ -88,7 +88,9 @@ SRCS(
     Pipe.cpp
     Port.cpp
     QueryPipeline.cpp
+    QueryPlan/AddingConstColumnStep.cpp
     QueryPlan/AddingDelayedSourceStep.cpp
+    QueryPlan/AddingMissedStep.cpp
     QueryPlan/AggregatingStep.cpp
     QueryPlan/ArrayJoinStep.cpp
     QueryPlan/ConvertingStep.cpp
@@ -105,16 +107,19 @@ SRCS(
     QueryPlan/ITransformingStep.cpp
     QueryPlan/LimitByStep.cpp
     QueryPlan/LimitStep.cpp
+    QueryPlan/MaterializingStep.cpp
     QueryPlan/MergeSortingStep.cpp
     QueryPlan/MergingAggregatedStep.cpp
+    QueryPlan/MergingFinal.cpp
     QueryPlan/MergingSortedStep.cpp
     QueryPlan/OffsetStep.cpp
     QueryPlan/PartialSortingStep.cpp
     QueryPlan/QueryPlan.cpp
     QueryPlan/ReadFromPreparedSource.cpp
-    QueryPlan/ReadFromStorageStep.cpp
     QueryPlan/ReadNothingStep.cpp
+    QueryPlan/ReverseRowsStep.cpp
     QueryPlan/RollupStep.cpp
+    QueryPlan/SettingQuotaAndLimitsStep.cpp
     QueryPlan/TotalsHavingStep.cpp
     QueryPlan/UnionStep.cpp
     ResizeProcessor.cpp

src/Storages/IStorage.cpp

@@ -7,11 +7,12 @@
 #include <Parsers/ASTCreateQuery.h>
 #include <Parsers/ASTSetQuery.h>
 #include <Processors/Pipe.h>
-#include <Processors/QueryPlan/ReadFromStorageStep.h>
+#include <Processors/QueryPlan/ReadFromPreparedSource.h>
 #include <Interpreters/Context.h>
 #include <Common/StringUtils/StringUtils.h>
 #include <Common/quoteString.h>
 #include <Interpreters/ExpressionActions.h>
+#include <Interpreters/InterpreterSelectQuery.h>
 
 
 namespace DB
@@ -94,28 +95,28 @@ Pipe IStorage::read(
 
 void IStorage::read(
         QueryPlan & query_plan,
-        TableLockHolder table_lock,
-        StorageMetadataPtr metadata_snapshot,
-        StreamLocalLimits & limits,
-        SizeLimits & leaf_limits,
-        std::shared_ptr<const EnabledQuota> quota,
         const Names & column_names,
+        const StorageMetadataPtr & metadata_snapshot,
         SelectQueryInfo & query_info,
-        std::shared_ptr<Context> context,
+        const Context & context,
         QueryProcessingStage::Enum processed_stage,
         size_t max_block_size,
         unsigned num_streams)
 {
-    auto read_step = std::make_unique<ReadFromStorageStep>(
-            std::move(table_lock), std::move(metadata_snapshot), limits, leaf_limits, std::move(quota), shared_from_this(),
-            column_names, query_info, std::move(context), processed_stage, max_block_size, num_streams);
-
-    read_step->setStepDescription("Read from " + getName());
-    query_plan.addStep(std::move(read_step));
+    auto pipe = read(column_names, metadata_snapshot, query_info, context, processed_stage, max_block_size, num_streams);
+    if (pipe.empty())
+    {
+        auto header = metadata_snapshot->getSampleBlockForColumns(column_names, getVirtuals(), getStorageID());
+        InterpreterSelectQuery::addEmptySourceToQueryPlan(query_plan, header, query_info);
+    }
+    else
+    {
+        auto read_step = std::make_unique<ReadFromStorageStep>(std::move(pipe), getName());
+        query_plan.addStep(std::move(read_step));
+    }
 }
 
 Pipe IStorage::alterPartition(
-    const ASTPtr & /* query */,
     const StorageMetadataPtr & /* metadata_snapshot */,
     const PartitionCommands & /* commands */,
     const Context & /* context */)

src/Storages/IStorage.h

@@ -48,6 +48,7 @@ using Processors = std::vector<ProcessorPtr>;
 
 class Pipe;
 class QueryPlan;
+using QueryPlanPtr = std::unique_ptr<QueryPlan>;
 
 class StoragePolicy;
 using StoragePolicyPtr = std::shared_ptr<const StoragePolicy>;
@@ -285,17 +286,13 @@ public:
     /// Default implementation creates ReadFromStorageStep and uses usual read.
     virtual void read(
         QueryPlan & query_plan,
-        TableLockHolder table_lock,
-        StorageMetadataPtr metadata_snapshot,
-        StreamLocalLimits & limits,
-        SizeLimits & leaf_limits,
-        std::shared_ptr<const EnabledQuota> quota,
-        const Names & column_names,
-        SelectQueryInfo & query_info,
-        std::shared_ptr<Context> context,
-        QueryProcessingStage::Enum processed_stage,
-        size_t max_block_size,
-        unsigned num_streams);
+        const Names & /*column_names*/,
+        const StorageMetadataPtr & /*metadata_snapshot*/,
+        SelectQueryInfo & /*query_info*/,
+        const Context & /*context*/,
+        QueryProcessingStage::Enum /*processed_stage*/,
+        size_t /*max_block_size*/,
+        unsigned /*num_streams*/);
 
     /** Writes the data to a table.
       * Receives a description of the query, which can contain information about the data write method.
@@ -367,7 +364,6 @@ public:
       * Should handle locks for each command on its own.
       */
     virtual Pipe alterPartition(
-        const ASTPtr & /* query */,
         const StorageMetadataPtr & /* metadata_snapshot */,
         const PartitionCommands & /* commands */,
         const Context & /* context */);

src/Storages/Kafka/KafkaBlockOutputStream.cpp

@@ -32,13 +32,16 @@ void KafkaBlockOutputStream::writePrefix()
     if (!buffer)
         throw Exception("Failed to create Kafka producer!", ErrorCodes::CANNOT_CREATE_IO_BUFFER);
 
-    child = FormatFactory::instance().getOutput(
-            storage.getFormatName(), *buffer, getHeader(), *context, [this](const Columns & columns, size_t row)
-            {
-                buffer->countRow(columns, row);
-            },
-            /* ignore_no_row_delimiter = */ true
-            );
+    auto format_settings = getFormatSettings(*context);
+    format_settings.protobuf.allow_many_rows_no_delimiters = true;
+
+    child = FormatFactory::instance().getOutput(storage.getFormatName(), *buffer,
+        getHeader(), *context,
+        [this](const Columns & columns, size_t row)
+        {
+            buffer->countRow(columns, row);
+        },
+        format_settings);
 }
 
 void KafkaBlockOutputStream::write(const Block & block)

src/Storages/MergeTree/BoolMask.h

@@ -9,17 +9,17 @@ struct BoolMask
     BoolMask() {}
     BoolMask(bool can_be_true_, bool can_be_false_) : can_be_true(can_be_true_), can_be_false(can_be_false_) {}
 
-    BoolMask operator &(const BoolMask & m)
+    BoolMask operator &(const BoolMask & m) const
     {
-        return BoolMask(can_be_true && m.can_be_true, can_be_false || m.can_be_false);
+        return {can_be_true && m.can_be_true, can_be_false || m.can_be_false};
     }
-    BoolMask operator |(const BoolMask & m)
+    BoolMask operator |(const BoolMask & m) const
     {
-        return BoolMask(can_be_true || m.can_be_true, can_be_false && m.can_be_false);
+        return {can_be_true || m.can_be_true, can_be_false && m.can_be_false};
     }
-    BoolMask operator !()
+    BoolMask operator !() const
     {
-        return BoolMask(can_be_false, can_be_true);
+        return {can_be_false, can_be_true};
     }
 
     /// If mask is (true, true), then it can no longer change under operation |.

src/Storages/MergeTree/EphemeralLockInZooKeeper.cpp

@@ -1,6 +1,7 @@
 #include <Storages/MergeTree/EphemeralLockInZooKeeper.h>
 #include <Common/ZooKeeper/KeeperException.h>
 #include <common/logger_useful.h>
+#include <common/types.h>
 
 
 namespace DB
@@ -71,13 +72,13 @@ EphemeralLockInZooKeeper::~EphemeralLockInZooKeeper()
 EphemeralLocksInAllPartitions::EphemeralLocksInAllPartitions(
     const String & block_numbers_path, const String & path_prefix, const String & temp_path,
     zkutil::ZooKeeper & zookeeper_)
-    : zookeeper(zookeeper_)
+    : zookeeper(&zookeeper_)
 {
     std::vector<String> holders;
     while (true)
     {
         Coordination::Stat partitions_stat;
-        Strings partitions = zookeeper.getChildren(block_numbers_path, &partitions_stat);
+        Strings partitions = zookeeper->getChildren(block_numbers_path, &partitions_stat);
 
         if (holders.size() < partitions.size())
         {
@@ -85,7 +86,7 @@ EphemeralLocksInAllPartitions::EphemeralLocksInAllPartitions(
             for (size_t i = 0; i < partitions.size() - holders.size(); ++i)
             {
                 String path = temp_path + "/abandonable_lock-";
-                holder_futures.push_back(zookeeper.asyncCreate(path, {}, zkutil::CreateMode::EphemeralSequential));
+                holder_futures.push_back(zookeeper->asyncCreate(path, {}, zkutil::CreateMode::EphemeralSequential));
             }
             for (auto & future : holder_futures)
             {
@@ -104,7 +105,7 @@ EphemeralLocksInAllPartitions::EphemeralLocksInAllPartitions(
         lock_ops.push_back(zkutil::makeCheckRequest(block_numbers_path, partitions_stat.version));
 
         Coordination::Responses lock_responses;
-        Coordination::Error rc = zookeeper.tryMulti(lock_ops, lock_responses);
+        Coordination::Error rc = zookeeper->tryMulti(lock_ops, lock_responses);
         if (rc == Coordination::Error::ZBADVERSION)
         {
             LOG_TRACE(&Poco::Logger::get("EphemeralLocksInAllPartitions"), "Someone has inserted a block in a new partition while we were creating locks. Retry.");
@@ -131,13 +132,16 @@ EphemeralLocksInAllPartitions::EphemeralLocksInAllPartitions(
 
 void EphemeralLocksInAllPartitions::unlock()
 {
+    if (!zookeeper)
+        return;
+
     std::vector<zkutil::ZooKeeper::FutureMulti> futures;
     for (const auto & lock : locks)
     {
         Coordination::Requests unlock_ops;
         unlock_ops.emplace_back(zkutil::makeRemoveRequest(lock.path, -1));
         unlock_ops.emplace_back(zkutil::makeRemoveRequest(lock.holder_path, -1));
-        futures.push_back(zookeeper.asyncMulti(unlock_ops));
+        futures.push_back(zookeeper->asyncMulti(unlock_ops));
     }
 
     for (auto & future : futures)

src/Storages/MergeTree/EphemeralLockInZooKeeper.h

@@ -1,9 +1,14 @@
 #pragma once
 
+#include "ReplicatedMergeTreeMutationEntry.h"
+
 #include <Common/ZooKeeper/ZooKeeper.h>
 #include <Common/Exception.h>
 #include <IO/ReadHelpers.h>
 
+#include <map>
+#include <optional>
+
 
 namespace DB
 {
@@ -87,13 +92,30 @@ private:
 
 
 /// Acquires block number locks in all partitions.
-class EphemeralLocksInAllPartitions : private boost::noncopyable
+class EphemeralLocksInAllPartitions : public boost::noncopyable
 {
 public:
     EphemeralLocksInAllPartitions(
         const String & block_numbers_path, const String & path_prefix, const String & temp_path,
         zkutil::ZooKeeper & zookeeper_);
 
+    EphemeralLocksInAllPartitions() = default;
+
+    EphemeralLocksInAllPartitions(EphemeralLocksInAllPartitions && rhs) noexcept
+        : zookeeper(rhs.zookeeper)
+        , locks(std::move(rhs.locks))
+    {
+        rhs.zookeeper = nullptr;
+    }
+
+    EphemeralLocksInAllPartitions & operator=(EphemeralLocksInAllPartitions && rhs) noexcept
+    {
+        zookeeper = rhs.zookeeper;
+        rhs.zookeeper = nullptr;
+        locks = std::move(rhs.locks);
+        return *this;
+    }
+
     struct LockInfo
     {
         String path;
@@ -110,8 +132,51 @@ public:
     ~EphemeralLocksInAllPartitions();
 
 private:
-    zkutil::ZooKeeper & zookeeper;
+    zkutil::ZooKeeper * zookeeper = nullptr;
     std::vector<LockInfo> locks;
 };
 
+
+/// This class allows scoped manipulations with block numbers locked in certain partitions
+/// See StorageReplicatedMergeTree::allocateBlockNumbersInAffectedPartitions and alter()/mutate() methods
+class PartitionBlockNumbersHolder
+{
+public:
+    PartitionBlockNumbersHolder(const PartitionBlockNumbersHolder &) = delete;
+    PartitionBlockNumbersHolder & operator=(const PartitionBlockNumbersHolder &) = delete;
+
+    using BlockNumbersType = ReplicatedMergeTreeMutationEntry::BlockNumbersType;
+
+    PartitionBlockNumbersHolder() = default;
+    PartitionBlockNumbersHolder(
+        BlockNumbersType block_numbers_, std::optional<EphemeralLocksInAllPartitions> locked_block_numbers_holder)
+        : block_numbers(std::move(block_numbers_))
+        , multiple_partitions_holder(std::move(locked_block_numbers_holder))
+    {
+    }
+    PartitionBlockNumbersHolder(
+        BlockNumbersType block_numbers_, std::optional<EphemeralLockInZooKeeper> locked_block_numbers_holder)
+        : block_numbers(std::move(block_numbers_))
+        , single_partition_holder(std::move(locked_block_numbers_holder))
+    {
+    }
+
+    PartitionBlockNumbersHolder & operator=(PartitionBlockNumbersHolder &&) = default;
+
+    const BlockNumbersType & getBlockNumbers() const { return block_numbers; }
+
+    void reset()
+    {
+        multiple_partitions_holder.reset();
+        single_partition_holder.reset();
+        block_numbers.clear();
+    }
+
+private:
+    BlockNumbersType block_numbers;
+
+    std::optional<EphemeralLocksInAllPartitions> multiple_partitions_holder;
+    std::optional<EphemeralLockInZooKeeper> single_partition_holder;
+};
+
 }

src/Storages/MergeTree/MergeTreeData.cpp

@@ -2643,6 +2643,17 @@ void MergeTreeData::checkPartitionCanBeDropped(const ASTPtr & partition)
     global_context.checkPartitionCanBeDropped(table_id.database_name, table_id.table_name, partition_size);
 }
 
+void MergeTreeData::checkPartCanBeDropped(const ASTPtr & part_ast)
+{
+    String part_name = part_ast->as<ASTLiteral &>().value.safeGet<String>();
+    auto part = getPartIfExists(part_name, {MergeTreeDataPartState::Committed});
+    if (!part)
+        throw Exception(ErrorCodes::NO_SUCH_DATA_PART, "No part {} in commited state", part_name);
+
+    auto table_id = getStorageID();
+    global_context.checkPartitionCanBeDropped(table_id.database_name, table_id.table_name, part->getBytesOnDisk());
+}
+
 void MergeTreeData::movePartitionToDisk(const ASTPtr & partition, const String & name, bool moving_part, const Context & context)
 {
     String partition_id;

src/Storages/MergeTree/MergeTreeData.h

@@ -561,6 +561,8 @@ public:
 
     void checkPartitionCanBeDropped(const ASTPtr & partition) override;
 
+    void checkPartCanBeDropped(const ASTPtr & part);
+
     size_t getColumnCompressedSize(const std::string & name) const
     {
         auto lock = lockParts();

src/Storages/MergeTree/MergeTreeDataSelectExecutor.cpp

@@ -23,24 +23,20 @@
 #include <Parsers/parseIdentifierOrStringLiteral.h>
 #include <Interpreters/ExpressionAnalyzer.h>
 #include <Interpreters/Context.h>
+#include <Processors/ConcatProcessor.h>
+#include <Processors/QueryPlan/QueryPlan.h>
+#include <Processors/QueryPlan/FilterStep.h>
+#include <Processors/QueryPlan/ExpressionStep.h>
+#include <Processors/QueryPlan/ReadFromPreparedSource.h>
+#include <Processors/QueryPlan/AddingConstColumnStep.h>
+#include <Processors/QueryPlan/ReverseRowsStep.h>
+#include <Processors/QueryPlan/MergingSortedStep.h>
+#include <Processors/QueryPlan/UnionStep.h>
+#include <Processors/QueryPlan/MergingFinal.h>
 
 #include <DataTypes/DataTypeDate.h>
 #include <DataTypes/DataTypeEnum.h>
 #include <DataTypes/DataTypesNumber.h>
-#include <Processors/ConcatProcessor.h>
-#include <Processors/Merges/AggregatingSortedTransform.h>
-#include <Processors/Merges/CollapsingSortedTransform.h>
-#include <Processors/Merges/MergingSortedTransform.h>
-#include <Processors/Merges/ReplacingSortedTransform.h>
-#include <Processors/Merges/SummingSortedTransform.h>
-#include <Processors/Merges/VersionedCollapsingTransform.h>
-#include <Processors/Sources/SourceFromInputStream.h>
-#include <Processors/Transforms/AddingConstColumnTransform.h>
-#include <Processors/Transforms/AddingSelectorTransform.h>
-#include <Processors/Transforms/CopyTransform.h>
-#include <Processors/Transforms/ExpressionTransform.h>
-#include <Processors/Transforms/FilterTransform.h>
-#include <Processors/Transforms/ReverseTransform.h>
 #include <Storages/VirtualColumnUtils.h>
 
 namespace ProfileEvents
@@ -83,17 +79,6 @@ static Block getBlockWithPartColumn(const MergeTreeData::DataPartsVector & parts
     return Block{ColumnWithTypeAndName(std::move(column), std::make_shared<DataTypeString>(), "_part")};
 }
 
-/// Check if ORDER BY clause of the query has some expression.
-static bool sortingDescriptionHasExpressions(const SortDescription & sort_description, const StorageMetadataPtr & metadata_snapshot)
-{
-    auto all_columns = metadata_snapshot->getColumns();
-    for (const auto & sort_column : sort_description)
-    {
-        if (!all_columns.has(sort_column.column_name))
-            return true;
-    }
-    return false;
-}
 
 size_t MergeTreeDataSelectExecutor::getApproximateTotalRowsToRead(
     const MergeTreeData::DataPartsVector & parts,
@@ -144,7 +129,7 @@ static RelativeSize convertAbsoluteSampleSizeToRelative(const ASTPtr & node, siz
 }
 
 
-Pipe MergeTreeDataSelectExecutor::read(
+QueryPlanPtr MergeTreeDataSelectExecutor::read(
     const Names & column_names_to_return,
     const StorageMetadataPtr & metadata_snapshot,
     const SelectQueryInfo & query_info,
@@ -159,7 +144,7 @@ Pipe MergeTreeDataSelectExecutor::read(
         max_block_numbers_to_read);
 }
 
-Pipe MergeTreeDataSelectExecutor::readFromParts(
+QueryPlanPtr MergeTreeDataSelectExecutor::readFromParts(
     MergeTreeData::DataPartsVector parts,
     const Names & column_names_to_return,
     const StorageMetadataPtr & metadata_snapshot,
@@ -301,7 +286,7 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
     /// Sampling.
     Names column_names_to_read = real_column_names;
     std::shared_ptr<ASTFunction> filter_function;
-    ExpressionActionsPtr filter_expression;
+    ActionsDAGPtr filter_expression;
 
     RelativeSize relative_sample_size = 0;
     RelativeSize relative_sample_offset = 0;
@@ -537,13 +522,13 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
 
             ASTPtr query = filter_function;
             auto syntax_result = TreeRewriter(context).analyze(query, available_real_columns);
-            filter_expression = ExpressionAnalyzer(filter_function, syntax_result, context).getActions(false);
+            filter_expression = ExpressionAnalyzer(filter_function, syntax_result, context).getActionsDAG(false);
 
             if (!select.final())
             {
                 /// Add columns needed for `sample_by_ast` to `column_names_to_read`.
                 /// Skip this if final was used, because such columns were already added from PK.
-                std::vector<String> add_columns = filter_expression->getRequiredColumns();
+                std::vector<String> add_columns = filter_expression->getRequiredColumns().getNames();
                 column_names_to_read.insert(column_names_to_read.end(), add_columns.begin(), add_columns.end());
                 std::sort(column_names_to_read.begin(), column_names_to_read.end());
                 column_names_to_read.erase(std::unique(column_names_to_read.begin(), column_names_to_read.end()),
@@ -555,7 +540,7 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
     if (no_data)
     {
         LOG_DEBUG(log, "Sampling yields no data.");
-        return {};
+        return std::make_unique<QueryPlan>();
     }
 
     LOG_DEBUG(log, "Key condition: {}", key_condition.toString());
@@ -725,13 +710,13 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
     LOG_DEBUG(log, "Selected {} parts by partition key, {} parts by primary key, {} marks by primary key, {} marks to read from {} ranges", parts.size(), parts_with_ranges.size(), sum_marks_pk.load(std::memory_order_relaxed), sum_marks, sum_ranges);
 
     if (parts_with_ranges.empty())
-        return {};
+        return std::make_unique<QueryPlan>();
 
     ProfileEvents::increment(ProfileEvents::SelectedParts, parts_with_ranges.size());
     ProfileEvents::increment(ProfileEvents::SelectedRanges, sum_ranges);
     ProfileEvents::increment(ProfileEvents::SelectedMarks, sum_marks);
 
-    Pipe res;
+    QueryPlanPtr plan;
 
     /// Projection, that needed to drop columns, which have appeared by execution
     /// of some extra expressions, and to allow execute the same expressions later.
@@ -752,7 +737,7 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
         std::sort(column_names_to_read.begin(), column_names_to_read.end());
         column_names_to_read.erase(std::unique(column_names_to_read.begin(), column_names_to_read.end()), column_names_to_read.end());
 
-        res = spreadMarkRangesAmongStreamsFinal(
+        plan = spreadMarkRangesAmongStreamsFinal(
             std::move(parts_with_ranges),
             num_streams,
             column_names_to_read,
@@ -772,9 +757,9 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
         order_key_prefix_ast->children.resize(prefix_size);
 
         auto syntax_result = TreeRewriter(context).analyze(order_key_prefix_ast, metadata_snapshot->getColumns().getAllPhysical());
-        auto sorting_key_prefix_expr = ExpressionAnalyzer(order_key_prefix_ast, syntax_result, context).getActions(false);
+        auto sorting_key_prefix_expr = ExpressionAnalyzer(order_key_prefix_ast, syntax_result, context).getActionsDAG(false);
 
-        res = spreadMarkRangesAmongStreamsWithOrder(
+        plan = spreadMarkRangesAmongStreamsWithOrder(
             std::move(parts_with_ranges),
             num_streams,
             column_names_to_read,
@@ -790,7 +775,7 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
     }
     else
     {
-        res = spreadMarkRangesAmongStreams(
+        plan = spreadMarkRangesAmongStreams(
             std::move(parts_with_ranges),
             num_streams,
             column_names_to_read,
@@ -803,43 +788,52 @@ Pipe MergeTreeDataSelectExecutor::readFromParts(
             reader_settings);
     }
 
+    if (!plan)
+        return std::make_unique<QueryPlan>();
+
     if (use_sampling)
     {
-        res.addSimpleTransform([&filter_expression, &filter_function](const Block & header)
-        {
-            return std::make_shared<FilterTransform>(
-                    header, filter_expression, filter_function->getColumnName(), false);
-        });
+        auto sampling_step = std::make_unique<FilterStep>(
+                plan->getCurrentDataStream(),
+                filter_expression,
+                filter_function->getColumnName(),
+                false);
+
+        sampling_step->setStepDescription("Sampling");
+        plan->addStep(std::move(sampling_step));
     }
 
     if (result_projection)
     {
-        auto result_projection_actions = std::make_shared<ExpressionActions>(result_projection);
-        res.addSimpleTransform([&result_projection_actions](const Block & header)
-        {
-            return std::make_shared<ExpressionTransform>(header, result_projection_actions);
-        });
+        auto projection_step = std::make_unique<ExpressionStep>(plan->getCurrentDataStream(), result_projection);
+        projection_step->setStepDescription("Remove unused columns after reading from storage");
+        plan->addStep(std::move(projection_step));
     }
 
     /// By the way, if a distributed query or query to a Merge table is made, then the `_sample_factor` column can have different values.
     if (sample_factor_column_queried)
     {
-        res.addSimpleTransform([used_sample_factor](const Block & header)
-        {
-            return std::make_shared<AddingConstColumnTransform<Float64>>(
-                    header, std::make_shared<DataTypeFloat64>(), used_sample_factor, "_sample_factor");
-        });
+        ColumnWithTypeAndName column;
+        column.name = "_sample_factor";
+        column.type = std::make_shared<DataTypeFloat64>();
+        column.column = column.type->createColumnConst(0, Field(used_sample_factor));
+
+        auto adding_column = std::make_unique<AddingConstColumnStep>(plan->getCurrentDataStream(), std::move(column));
+        adding_column->setStepDescription("Add _sample_factor column");
+        plan->addStep(std::move(adding_column));
     }
 
     if (query_info.prewhere_info && query_info.prewhere_info->remove_columns_actions)
     {
-        res.addSimpleTransform([&query_info](const Block & header)
-        {
-            return std::make_shared<ExpressionTransform>(header, query_info.prewhere_info->remove_columns_actions);
-        });
+        auto expression_step = std::make_unique<ExpressionStep>(
+                plan->getCurrentDataStream(),
+                query_info.prewhere_info->remove_columns_actions->getActionsDAG().clone());
+
+        expression_step->setStepDescription("Remove unused columns after PREWHERE");
+        plan->addStep(std::move(expression_step));
     }
 
-    return res;
+    return plan;
 }
 
 namespace
@@ -864,8 +858,20 @@ size_t roundRowsOrBytesToMarks(
 
 }
 
+static QueryPlanPtr createPlanFromPipe(Pipe pipe, const std::string & description = "")
+{
+    auto plan = std::make_unique<QueryPlan>();
 
-Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreams(
+    std::string storage_name = "MergeTree";
+    if (!description.empty())
+        storage_name += ' ' + description;
+
+    auto step = std::make_unique<ReadFromStorageStep>(std::move(pipe), storage_name);
+    plan->addStep(std::move(step));
+    return plan;
+}
+
+QueryPlanPtr MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreams(
     RangesInDataParts && parts,
     size_t num_streams,
     const Names & column_names,
@@ -959,7 +965,7 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreams(
             res.emplace_back(std::move(source));
         }
 
-        return Pipe::unitePipes(std::move(res));
+        return createPlanFromPipe(Pipe::unitePipes(std::move(res)));
     }
     else
     {
@@ -983,20 +989,19 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreams(
         if (pipe.numOutputPorts() > 1)
             pipe.addTransform(std::make_shared<ConcatProcessor>(pipe.getHeader(), pipe.numOutputPorts()));
 
-        return pipe;
+        return createPlanFromPipe(std::move(pipe));
     }
 }
 
-static ActionsDAGPtr createProjection(const Pipe & pipe)
+static ActionsDAGPtr createProjection(const Block & header)
 {
-    const auto & header = pipe.getHeader();
     auto projection = std::make_shared<ActionsDAG>(header.getNamesAndTypesList());
     projection->removeUnusedActions(header.getNames());
     projection->projectInput();
     return projection;
 }
 
-Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsWithOrder(
+QueryPlanPtr MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsWithOrder(
     RangesInDataParts && parts,
     size_t num_streams,
     const Names & column_names,
@@ -1004,7 +1009,7 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsWithOrder(
     UInt64 max_block_size,
     bool use_uncompressed_cache,
     const SelectQueryInfo & query_info,
-    const ExpressionActionsPtr & sorting_key_prefix_expr,
+    const ActionsDAGPtr & sorting_key_prefix_expr,
     const Names & virt_columns,
     const Settings & settings,
     const MergeTreeReaderSettings & reader_settings,
@@ -1097,7 +1102,8 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsWithOrder(
 
     const size_t min_marks_per_stream = (sum_marks - 1) / num_streams + 1;
     bool need_preliminary_merge = (parts.size() > settings.read_in_order_two_level_merge_threshold);
-    size_t max_output_ports = 0;
+
+    std::vector<QueryPlanPtr> plans;
 
     for (size_t i = 0; i < num_streams && !parts.empty(); ++i)
     {
@@ -1197,60 +1203,64 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsWithOrder(
             }
         }
 
-        auto pipe = Pipe::unitePipes(std::move(pipes));
+        auto plan = createPlanFromPipe(Pipe::unitePipes(std::move(pipes)), " with order");
 
         if (input_order_info->direction != 1)
         {
-            pipe.addSimpleTransform([](const Block & header)
-            {
-                return std::make_shared<ReverseTransform>(header);
-            });
+            auto reverse_step = std::make_unique<ReverseRowsStep>(plan->getCurrentDataStream());
+            plan->addStep(std::move(reverse_step));
         }
 
-        max_output_ports = std::max(pipe.numOutputPorts(), max_output_ports);
-        res.emplace_back(std::move(pipe));
+        plans.emplace_back(std::move(plan));
     }
 
     if (need_preliminary_merge)
     {
-        /// If ORDER BY clause of the query contains some expression,
-        /// then those new columns should be added for the merge step,
-        /// and this should be done always, if there is at least one pipe that
-        /// has multiple output ports.
-        bool sorting_key_has_expression = sortingDescriptionHasExpressions(input_order_info->order_key_prefix_descr, metadata_snapshot);
-        bool force_sorting_key_transform = res.size() > 1 && max_output_ports > 1 && sorting_key_has_expression;
+        SortDescription sort_description;
+        for (size_t j = 0; j < input_order_info->order_key_prefix_descr.size(); ++j)
+            sort_description.emplace_back(metadata_snapshot->getSortingKey().column_names[j],
+                                          input_order_info->direction, 1);
 
-        for (auto & pipe : res)
+        for (auto & plan : plans)
         {
-            SortDescription sort_description;
+            /// Drop temporary columns, added by 'sorting_key_prefix_expr'
+            out_projection = createProjection(plan->getCurrentDataStream().header);
 
-            if (pipe.numOutputPorts() > 1 || force_sorting_key_transform)
-            {
-                for (size_t j = 0; j < input_order_info->order_key_prefix_descr.size(); ++j)
-                    sort_description.emplace_back(metadata_snapshot->getSortingKey().column_names[j],
-                                                  input_order_info->direction, 1);
+            auto expression_step = std::make_unique<ExpressionStep>(
+                    plan->getCurrentDataStream(),
+                    sorting_key_prefix_expr);
 
-                /// Drop temporary columns, added by 'sorting_key_prefix_expr'
-                out_projection = createProjection(pipe);
-                pipe.addSimpleTransform([sorting_key_prefix_expr](const Block & header)
-                {
-                    return std::make_shared<ExpressionTransform>(header, sorting_key_prefix_expr);
-                });
-            }
+            expression_step->setStepDescription("Calculate sorting key prefix");
+            plan->addStep(std::move(expression_step));
 
-            if (pipe.numOutputPorts() > 1)
-            {
-                pipe.addTransform(std::make_shared<MergingSortedTransform>(
-                        pipe.getHeader(), pipe.numOutputPorts(), sort_description, max_block_size));
-            }
+            auto merging_sorted = std::make_unique<MergingSortedStep>(
+                    plan->getCurrentDataStream(),
+                    sort_description,
+                    max_block_size);
+
+            merging_sorted->setStepDescription("Merge sorting mark ranges");
+            plan->addStep(std::move(merging_sorted));
         }
     }
 
-    return Pipe::unitePipes(std::move(res));
+    if (plans.size() == 1)
+        return std::move(plans.front());
+
+    DataStreams input_streams;
+    for (const auto & plan : plans)
+        input_streams.emplace_back(plan->getCurrentDataStream());
+
+    const auto & common_header = plans.front()->getCurrentDataStream().header;
+    auto union_step = std::make_unique<UnionStep>(std::move(input_streams), common_header);
+
+    auto plan = std::make_unique<QueryPlan>();
+    plan->unitePlans(std::move(union_step), std::move(plans));
+
+    return plan;
 }
 
 
-Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsFinal(
+QueryPlanPtr MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsFinal(
     RangesInDataParts && parts,
     size_t num_streams,
     const Names & column_names,
@@ -1318,11 +1328,11 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsFinal(
         parts_to_merge_ranges.push_back(parts.end());
     }
 
-    Pipes partition_pipes;
+    std::vector<QueryPlanPtr> partition_plans;
 
     for (size_t range_index = 0; range_index < parts_to_merge_ranges.size() - 1; ++range_index)
     {
-        Pipe pipe;
+        QueryPlanPtr plan;
 
         {
             Pipes pipes;
@@ -1348,12 +1358,17 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsFinal(
                 pipes.emplace_back(std::move(source_processor));
             }
 
-            pipe = Pipe::unitePipes(std::move(pipes));
-        }
+            if (pipes.empty())
+                continue;
 
-        /// Drop temporary columns, added by 'sorting_key_expr'
-        if (!out_projection)
-            out_projection = createProjection(pipe);
+            auto pipe = Pipe::unitePipes(std::move(pipes));
+
+            /// Drop temporary columns, added by 'sorting_key_expr'
+            if (!out_projection)
+                out_projection = createProjection(pipe.getHeader());
+
+            plan = createPlanFromPipe(std::move(pipe), "with final");
+        }
 
         /// If do_not_merge_across_partitions_select_final is true and there is only one part in partition
         /// with level > 0 then we won't postprocess this part
@@ -1361,14 +1376,16 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsFinal(
             std::distance(parts_to_merge_ranges[range_index], parts_to_merge_ranges[range_index + 1]) == 1 &&
             parts_to_merge_ranges[range_index]->data_part->info.level > 0)
         {
-            partition_pipes.emplace_back(std::move(pipe));
+            partition_plans.emplace_back(std::move(plan));
             continue;
         }
 
-        pipe.addSimpleTransform([&metadata_snapshot](const Block & header)
-        {
-            return std::make_shared<ExpressionTransform>(header, metadata_snapshot->getSortingKey().expression);
-        });
+        auto expression_step = std::make_unique<ExpressionStep>(
+                plan->getCurrentDataStream(),
+                metadata_snapshot->getSortingKey().expression->getActionsDAG().clone());
+
+        expression_step->setStepDescription("Calculate sorting key expression");
+        plan->addStep(std::move(expression_step));
 
         Names sort_columns = metadata_snapshot->getSortingKeyColumns();
         SortDescription sort_description;
@@ -1377,111 +1394,40 @@ Pipe MergeTreeDataSelectExecutor::spreadMarkRangesAmongStreamsFinal(
 
         Names partition_key_columns = metadata_snapshot->getPartitionKey().column_names;
 
-        Block header = pipe.getHeader();
+        const auto & header = plan->getCurrentDataStream().header;
         for (size_t i = 0; i < sort_columns_size; ++i)
             sort_description.emplace_back(header.getPositionByName(sort_columns[i]), 1, 1);
 
-        auto get_merging_processor = [&]() -> MergingTransformPtr
-        {
-            switch (data.merging_params.mode)
-            {
-                case MergeTreeData::MergingParams::Ordinary:
-                {
-                    return std::make_shared<MergingSortedTransform>(header, pipe.numOutputPorts(), sort_description, max_block_size);
-                }
+        auto final_step = std::make_unique<MergingFinal>(
+            plan->getCurrentDataStream(),
+            std::min<size_t>(num_streams, settings.max_final_threads),
+            sort_description,
+            data.merging_params,
+            partition_key_columns,
+            max_block_size);
 
-                case MergeTreeData::MergingParams::Collapsing:
-                    return std::make_shared<CollapsingSortedTransform>(
-                        header, pipe.numOutputPorts(), sort_description, data.merging_params.sign_column, true, max_block_size);
+        final_step->setStepDescription("Merge rows for FINAL");
+        plan->addStep(std::move(final_step));
 
-                case MergeTreeData::MergingParams::Summing:
-                    return std::make_shared<SummingSortedTransform>(
-                        header,
-                        pipe.numOutputPorts(),
-                        sort_description,
-                        data.merging_params.columns_to_sum,
-                        partition_key_columns,
-                        max_block_size);
-
-                case MergeTreeData::MergingParams::Aggregating:
-                    return std::make_shared<AggregatingSortedTransform>(header, pipe.numOutputPorts(), sort_description, max_block_size);
-
-                case MergeTreeData::MergingParams::Replacing:
-                    return std::make_shared<ReplacingSortedTransform>(
-                        header, pipe.numOutputPorts(), sort_description, data.merging_params.version_column, max_block_size);
-
-                case MergeTreeData::MergingParams::VersionedCollapsing:
-                    return std::make_shared<VersionedCollapsingTransform>(
-                        header, pipe.numOutputPorts(), sort_description, data.merging_params.sign_column, max_block_size);
-
-                case MergeTreeData::MergingParams::Graphite:
-                    throw Exception("GraphiteMergeTree doesn't support FINAL", ErrorCodes::LOGICAL_ERROR);
-            }
-
-            __builtin_unreachable();
-        };
-
-        if (num_streams <= 1 || sort_description.empty())
-        {
-            pipe.addTransform(get_merging_processor());
-            partition_pipes.emplace_back(std::move(pipe));
-            continue;
-        }
-
-        ColumnNumbers key_columns;
-        key_columns.reserve(sort_description.size());
-
-        for (auto & desc : sort_description)
-        {
-            if (!desc.column_name.empty())
-                key_columns.push_back(header.getPositionByName(desc.column_name));
-            else
-                key_columns.emplace_back(desc.column_number);
-        }
-
-        pipe.addSimpleTransform([&](const Block & stream_header)
-        {
-            return std::make_shared<AddingSelectorTransform>(stream_header, num_streams, key_columns);
-        });
-
-        pipe.transform([&](OutputPortRawPtrs ports)
-        {
-            Processors processors;
-            std::vector<OutputPorts::iterator> output_ports;
-            processors.reserve(ports.size() + num_streams);
-            output_ports.reserve(ports.size());
-
-            for (auto & port : ports)
-            {
-                auto copier = std::make_shared<CopyTransform>(header, num_streams);
-                connect(*port, copier->getInputPort());
-                output_ports.emplace_back(copier->getOutputs().begin());
-                processors.emplace_back(std::move(copier));
-            }
-
-            for (size_t i = 0; i < num_streams; ++i)
-            {
-                auto merge = get_merging_processor();
-                merge->setSelectorPosition(i);
-                auto input = merge->getInputs().begin();
-
-                /// Connect i-th merge with i-th input port of every copier.
-                for (size_t j = 0; j < ports.size(); ++j)
-                {
-                    connect(*output_ports[j], *input);
-                    ++output_ports[j];
-                    ++input;
-                }
-
-                processors.emplace_back(std::move(merge));
-            }
-
-            return processors;
-        });
-        partition_pipes.emplace_back(std::move(pipe));
+        partition_plans.emplace_back(std::move(plan));
     }
 
-    return Pipe::unitePipes(std::move(partition_pipes));
+    if (partition_plans.empty())
+        return {};
+
+    if (partition_plans.size() == 1)
+        return std::move(partition_plans.front());
+
+    auto result_header = partition_plans.front()->getCurrentDataStream().header;
+    DataStreams input_streams;
+    for (const auto & partition_plan : partition_plans)
+        input_streams.push_back(partition_plan->getCurrentDataStream());
+
+    auto union_step = std::make_unique<UnionStep>(std::move(input_streams), result_header);
+    union_step->setStepDescription("Unite sources after FINAL");
+    QueryPlanPtr plan = std::make_unique<QueryPlan>();
+    plan->unitePlans(std::move(union_step), std::move(partition_plans));
+    return plan;
 }
 
 /// Calculates a set of mark ranges, that could possibly contain keys, required by condition.

src/Storages/MergeTree/MergeTreeDataSelectExecutor.h

@@ -24,7 +24,7 @@ public:
       */
     using PartitionIdToMaxBlock = std::unordered_map<String, Int64>;
 
-    Pipe read(
+    QueryPlanPtr read(
         const Names & column_names,
         const StorageMetadataPtr & metadata_snapshot,
         const SelectQueryInfo & query_info,
@@ -33,7 +33,7 @@ public:
         unsigned num_streams,
         const PartitionIdToMaxBlock * max_block_numbers_to_read = nullptr) const;
 
-    Pipe readFromParts(
+    QueryPlanPtr readFromParts(
         MergeTreeData::DataPartsVector parts,
         const Names & column_names,
         const StorageMetadataPtr & metadata_snapshot,
@@ -48,7 +48,7 @@ private:
 
     Poco::Logger * log;
 
-    Pipe spreadMarkRangesAmongStreams(
+    QueryPlanPtr spreadMarkRangesAmongStreams(
         RangesInDataParts && parts,
         size_t num_streams,
         const Names & column_names,
@@ -61,7 +61,7 @@ private:
         const MergeTreeReaderSettings & reader_settings) const;
 
     /// out_projection - save projection only with columns, requested to read
-    Pipe spreadMarkRangesAmongStreamsWithOrder(
+    QueryPlanPtr spreadMarkRangesAmongStreamsWithOrder(
         RangesInDataParts && parts,
         size_t num_streams,
         const Names & column_names,
@@ -69,13 +69,13 @@ private:
         UInt64 max_block_size,
         bool use_uncompressed_cache,
         const SelectQueryInfo & query_info,
-        const ExpressionActionsPtr & sorting_key_prefix_expr,
+        const ActionsDAGPtr & sorting_key_prefix_expr,
         const Names & virt_columns,
         const Settings & settings,
         const MergeTreeReaderSettings & reader_settings,
         ActionsDAGPtr & out_projection) const;
 
-    Pipe spreadMarkRangesAmongStreamsFinal(
+    QueryPlanPtr spreadMarkRangesAmongStreamsFinal(
         RangesInDataParts && parts,
         size_t num_streams,
         const Names & column_names,

src/Storages/MergeTree/MergeTreeIndexConditionBloomFilter.cpp

@@ -1,16 +1,17 @@
-#include <Storages/MergeTree/MergeTreeIndexConditionBloomFilter.h>
-#include <Interpreters/misc.h>
-#include <Interpreters/BloomFilterHash.h>
 #include <Common/HashTable/ClearableHashMap.h>
-#include <Storages/MergeTree/RPNBuilder.h>
-#include <Storages/MergeTree/MergeTreeIndexGranuleBloomFilter.h>
+#include <Common/FieldVisitorsAccurateComparison.h>
 #include <DataTypes/DataTypeArray.h>
 #include <DataTypes/DataTypeTuple.h>
 #include <Columns/ColumnConst.h>
-#include <ext/bit_cast.h>
+#include <Columns/ColumnTuple.h>
+#include <Storages/MergeTree/RPNBuilder.h>
+#include <Storages/MergeTree/MergeTreeIndexGranuleBloomFilter.h>
+#include <Storages/MergeTree/MergeTreeIndexConditionBloomFilter.h>
 #include <Parsers/ASTSubquery.h>
 #include <Parsers/ASTIdentifier.h>
-#include <Columns/ColumnTuple.h>
+#include <Parsers/ASTLiteral.h>
+#include <Interpreters/misc.h>
+#include <Interpreters/BloomFilterHash.h>
 #include <Interpreters/castColumn.h>
 #include <Interpreters/convertFieldToType.h>
 
@@ -105,11 +106,11 @@ bool MergeTreeIndexConditionBloomFilter::alwaysUnknownOrTrue() const
             rpn_stack.push_back(true);
         }
         else if (element.function == RPNElement::FUNCTION_EQUALS
-                 || element.function == RPNElement::FUNCTION_NOT_EQUALS
-                 || element.function == RPNElement::FUNCTION_HAS
-                 || element.function == RPNElement::FUNCTION_IN
-                 || element.function == RPNElement::FUNCTION_NOT_IN
-                 || element.function == RPNElement::ALWAYS_FALSE)
+            || element.function == RPNElement::FUNCTION_NOT_EQUALS
+            || element.function == RPNElement::FUNCTION_HAS
+            || element.function == RPNElement::FUNCTION_IN
+            || element.function == RPNElement::FUNCTION_NOT_IN
+            || element.function == RPNElement::ALWAYS_FALSE)
         {
             rpn_stack.push_back(false);
         }
@@ -222,9 +223,21 @@ bool MergeTreeIndexConditionBloomFilter::traverseAtomAST(const ASTPtr & node, Bl
         }
     }
 
+    return traverseFunction(node, block_with_constants, out, nullptr);
+}
+
+bool MergeTreeIndexConditionBloomFilter::traverseFunction(const ASTPtr & node, Block & block_with_constants, RPNElement & out, const ASTPtr & parent)
+{
+    bool maybe_useful = false;
+
     if (const auto * function = node->as<ASTFunction>())
     {
         const ASTs & arguments = function->arguments->children;
+        for (const auto & arg : arguments)
+        {
+            if (traverseFunction(arg, block_with_constants, out, node))
+                maybe_useful = true;
+        }
 
         if (arguments.size() != 2)
             return false;
@@ -232,20 +245,29 @@ bool MergeTreeIndexConditionBloomFilter::traverseAtomAST(const ASTPtr & node, Bl
         if (functionIsInOrGlobalInOperator(function->name))
         {
             if (const auto & prepared_set = getPreparedSet(arguments[1]))
-                return traverseASTIn(function->name, arguments[0], prepared_set, out);
+            {
+                if (traverseASTIn(function->name, arguments[0], prepared_set, out))
+                    maybe_useful = true;
+            }
         }
-        else if (function->name == "equals" || function->name  == "notEquals" || function->name == "has")
+        else if (function->name == "equals" || function->name  == "notEquals" || function->name == "has" || function->name == "indexOf")
         {
             Field const_value;
             DataTypePtr const_type;
             if (KeyCondition::getConstant(arguments[1], block_with_constants, const_value, const_type))
-                return traverseASTEquals(function->name, arguments[0], const_type, const_value, out);
+            {
+                if (traverseASTEquals(function->name, arguments[0], const_type, const_value, out, parent))
+                    maybe_useful = true;
+            }
             else if (KeyCondition::getConstant(arguments[0], block_with_constants, const_value, const_type))
-                return traverseASTEquals(function->name, arguments[1], const_type, const_value, out);
+            {
+                if (traverseASTEquals(function->name, arguments[1], const_type, const_value, out, parent))
+                    maybe_useful = true;
+            }
         }
     }
 
-    return false;
+    return maybe_useful;
 }
 
 bool MergeTreeIndexConditionBloomFilter::traverseASTIn(
@@ -302,8 +324,66 @@ bool MergeTreeIndexConditionBloomFilter::traverseASTIn(
     return false;
 }
 
+
+static bool indexOfCanUseBloomFilter(const ASTPtr & parent)
+{
+    if (!parent)
+        return true;
+
+    /// `parent` is a function where `indexOf` is located.
+    /// Example: `indexOf(arr, x) = 1`, parent is a function named `equals`.
+    if (const auto * function = parent->as<ASTFunction>())
+    {
+        if (function->name == "and")
+        {
+            return true;
+        }
+        else if (function->name == "equals" /// notEquals is not applicable
+            || function->name == "greater" || function->name == "greaterOrEquals"
+            || function->name == "less" || function->name == "lessOrEquals")
+        {
+            if (function->arguments->children.size() != 2)
+                return false;
+
+            /// We don't allow constant expressions like `indexOf(arr, x) = 1 + 0` but it's neglible.
+
+            /// We should return true when the corresponding expression implies that the array contains the element.
+            /// Example: when `indexOf(arr, x)` > 10 is written, it means that arr definitely should contain the element
+            /// (at least at 11th position but it does not matter).
+
+            bool reversed = false;
+            const ASTLiteral * constant = nullptr;
+
+            if (const ASTLiteral * left = function->arguments->children[0]->as<ASTLiteral>())
+            {
+                constant = left;
+                reversed = true;
+            }
+            else if (const ASTLiteral * right = function->arguments->children[1]->as<ASTLiteral>())
+            {
+                constant = right;
+            }
+            else
+                return false;
+
+            Field zero(0);
+            return (function->name == "equals"  /// indexOf(...) = c, c != 0
+                    && !applyVisitor(FieldVisitorAccurateEquals(), constant->value, zero))
+                || (function->name == "notEquals"  /// indexOf(...) != c, c = 0
+                    && applyVisitor(FieldVisitorAccurateEquals(), constant->value, zero))
+                || (function->name == (reversed ? "less" : "greater")   /// indexOf(...) > c, c >= 0
+                    && !applyVisitor(FieldVisitorAccurateLess(), constant->value, zero))
+                || (function->name == (reversed ? "lessOrEquals" : "greaterOrEquals")   /// indexOf(...) >= c, c > 0
+                    && applyVisitor(FieldVisitorAccurateLess(), zero, constant->value));
+        }
+    }
+
+    return false;
+}
+
+
 bool MergeTreeIndexConditionBloomFilter::traverseASTEquals(
-    const String & function_name, const ASTPtr & key_ast, const DataTypePtr & value_type, const Field & value_field, RPNElement & out)
+    const String & function_name, const ASTPtr & key_ast, const DataTypePtr & value_type, const Field & value_field, RPNElement & out, const ASTPtr & parent)
 {
     if (header.has(key_ast->getColumnName()))
     {
@@ -311,21 +391,26 @@ bool MergeTreeIndexConditionBloomFilter::traverseASTEquals(
         const DataTypePtr & index_type = header.getByPosition(position).type;
         const auto * array_type = typeid_cast<const DataTypeArray *>(index_type.get());
 
-        if (function_name == "has")
+        if (function_name == "has" || function_name == "indexOf")
         {
-            out.function = RPNElement::FUNCTION_HAS;
-
             if (!array_type)
-                throw Exception("First argument for function has must be an array.", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+                throw Exception("First argument for function " + function_name + " must be an array.", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
 
-            const DataTypePtr actual_type = BloomFilter::getPrimitiveType(array_type->getNestedType());
-            Field converted_field = convertFieldToType(value_field, *actual_type, value_type.get());
-            out.predicate.emplace_back(std::make_pair(position, BloomFilterHash::hashWithField(actual_type.get(), converted_field)));
+            /// We can treat `indexOf` function similar to `has`.
+            /// But it is little more cumbersome, compare: `has(arr, elem)` and `indexOf(arr, elem) != 0`.
+            /// The `parent` in this context is expected to be function `!=` (`notEquals`).
+            if (function_name == "has" || indexOfCanUseBloomFilter(parent))
+            {
+                out.function = RPNElement::FUNCTION_HAS;
+                const DataTypePtr actual_type = BloomFilter::getPrimitiveType(array_type->getNestedType());
+                Field converted_field = convertFieldToType(value_field, *actual_type, value_type.get());
+                out.predicate.emplace_back(std::make_pair(position, BloomFilterHash::hashWithField(actual_type.get(), converted_field)));
+            }
         }
         else
         {
             if (array_type)
-                throw Exception("An array type of bloom_filter supports only has() function.", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+                throw Exception("An array type of bloom_filter supports only has() and indexOf() function.", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
 
             out.function = function_name == "equals" ? RPNElement::FUNCTION_EQUALS : RPNElement::FUNCTION_NOT_EQUALS;
             const DataTypePtr actual_type = BloomFilter::getPrimitiveType(index_type);
@@ -353,7 +438,7 @@ bool MergeTreeIndexConditionBloomFilter::traverseASTEquals(
             const DataTypes & subtypes = value_tuple_data_type->getElements();
 
             for (size_t index = 0; index < tuple.size(); ++index)
-                match_with_subtype |= traverseASTEquals(function_name, arguments[index], subtypes[index], tuple[index], out);
+                match_with_subtype |= traverseASTEquals(function_name, arguments[index], subtypes[index], tuple[index], out, key_ast);
 
             return match_with_subtype;
         }

src/Storages/MergeTree/MergeTreeIndexConditionBloomFilter.h

@@ -67,13 +67,15 @@ private:
 
     bool traverseAtomAST(const ASTPtr & node, Block & block_with_constants, RPNElement & out);
 
+    bool traverseFunction(const ASTPtr & node, Block & block_with_constants, RPNElement & out, const ASTPtr & parent);
+
     bool traverseASTIn(const String & function_name, const ASTPtr & key_ast, const SetPtr & prepared_set, RPNElement & out);
 
     bool traverseASTIn(
         const String & function_name, const ASTPtr & key_ast, const DataTypePtr & type, const ColumnPtr & column, RPNElement & out);
 
     bool traverseASTEquals(
-        const String & function_name, const ASTPtr & key_ast, const DataTypePtr & value_type, const Field & value_field, RPNElement & out);
+        const String & function_name, const ASTPtr & key_ast, const DataTypePtr & value_type, const Field & value_field, RPNElement & out, const ASTPtr & parent);
 };
 
 }

src/Storages/MergeTree/ReplicatedMergeTreeMutationEntry.h

@@ -35,9 +35,10 @@ struct ReplicatedMergeTreeMutationEntry
     /// Replica which initiated mutation
     String source_replica;
 
-    /// Accured numbers of blocks
+    /// Acquired block numbers
     /// partition_id -> block_number
-    std::map<String, Int64> block_numbers;
+    using BlockNumbersType = std::map<String, Int64>;
+    BlockNumbersType block_numbers;
 
     /// Mutation commands which will give to MUTATE_PART entries
     MutationCommands commands;

src/Storages/MergeTree/StorageFromMergeTreeDataPart.h

@@ -3,6 +3,8 @@
 #include <Storages/IStorage.h>
 #include <Storages/MergeTree/IMergeTreeDataPart.h>
 #include <Storages/MergeTree/MergeTreeDataSelectExecutor.h>
+#include <Processors/QueryPlan/QueryPlan.h>
+#include <Processors/QueryPipeline.h>
 #include <Core/Defines.h>
 
 #include <ext/shared_ptr_helper.h>
@@ -27,8 +29,11 @@ public:
         size_t max_block_size,
         unsigned num_streams) override
     {
-        return MergeTreeDataSelectExecutor(part->storage)
-            .readFromParts({part}, column_names, metadata_snapshot, query_info, context, max_block_size, num_streams);
+        QueryPlan query_plan =
+            std::move(*MergeTreeDataSelectExecutor(part->storage)
+                      .readFromParts({part}, column_names, metadata_snapshot, query_info, context, max_block_size, num_streams));
+
+        return query_plan.convertToPipe();
     }
 
 
@@ -45,6 +50,16 @@ public:
         return part->storage.getVirtuals();
     }
 
+    String getPartitionId() const
+    {
+        return part->info.partition_id;
+    }
+
+    String getPartitionIDFromQuery(const ASTPtr & ast, const Context & context) const
+    {
+        return part->storage.getPartitionIDFromQuery(ast, context);
+    }
+
 protected:
     StorageFromMergeTreeDataPart(const MergeTreeData::DataPartPtr & part_)
         : IStorage(getIDFromPart(part_))

src/Storages/MutationCommands.cpp

@@ -2,11 +2,13 @@
 #include <IO/Operators.h>
 #include <Parsers/formatAST.h>
 #include <Parsers/ExpressionListParsers.h>
-#include <Parsers/ASTColumnDeclaration.h>
 #include <Parsers/ParserAlterQuery.h>
 #include <Parsers/parseQuery.h>
 #include <Parsers/ASTAssignment.h>
+#include <Parsers/ASTColumnDeclaration.h>
+#include <Parsers/ASTFunction.h>
 #include <Parsers/ASTIdentifier.h>
+#include <Parsers/ASTLiteral.h>
 #include <Common/typeid_cast.h>
 #include <Common/quoteString.h>
 #include <Core/Defines.h>
@@ -32,6 +34,7 @@ std::optional<MutationCommand> MutationCommand::parse(ASTAlterCommand * command,
         res.ast = command->ptr();
         res.type = DELETE;
         res.predicate = command->predicate;
+        res.partition = command->partition;
         return res;
     }
     else if (command->type == ASTAlterCommand::UPDATE)
@@ -40,6 +43,7 @@ std::optional<MutationCommand> MutationCommand::parse(ASTAlterCommand * command,
         res.ast = command->ptr();
         res.type = UPDATE;
         res.predicate = command->predicate;
+        res.partition = command->partition;
         for (const ASTPtr & assignment_ast : command->update_assignments->children)
         {
             const auto & assignment = assignment_ast->as<ASTAssignment &>();
@@ -124,6 +128,7 @@ std::shared_ptr<ASTAlterCommandList> MutationCommands::ast() const
     return res;
 }
 
+
 void MutationCommands::writeText(WriteBuffer & out) const
 {
     std::stringstream commands_ss;