Merge branch 'master' into data-type-with-dictionary

2024-12-11 17:02:25 +00:00 · 2018-04-17 13:44:19 +03:00 · 2018-04-17 13:44:19 +03:00 · c6a3fec726
commit c6a3fec726
parent 0f0d5b3c0c c9c09a9d13
757 changed files with 12265 additions and 7580 deletions
--- a/.gitmodules
+++ b/.gitmodules
@ -1,6 +1,3 @@
-[submodule "contrib/zookeeper"]
-	path = contrib/zookeeper
-	url = https://github.com/ClickHouse-Extras/zookeeper.git
 [submodule "contrib/poco"]
 	path = contrib/poco
 	url = https://github.com/ClickHouse-Extras/poco
--- a/.travis.yml
+++ b/.travis.yml
@ -13,7 +13,7 @@ matrix:
 #        apt:
 #          sources:
 #            - ubuntu-toolchain-r-test
-#          packages: [ g++-7,            libicu-dev, libreadline-dev, libmysqlclient-dev, unixodbc-dev, libltdl-dev, libssl-dev, libboost-dev, zlib1g-dev, libdouble-conversion-dev, libzookeeper-mt-dev, libsparsehash-dev, librdkafka-dev, libcapnp-dev, libsparsehash-dev, libgoogle-perftools-dev, bash, expect, python, python-lxml, python-termcolor, curl, perl, sudo ]
+#          packages: [ g++-7,            libicu-dev, libreadline-dev, libmysqlclient-dev, unixodbc-dev, libltdl-dev, libssl-dev, libboost-dev, zlib1g-dev, libdouble-conversion-dev, libzookeeper-mt-dev, libsparsehash-dev, librdkafka-dev, libcapnp-dev, libsparsehash-dev, libgoogle-perftools-dev, bash, expect, python, python-lxml, python-termcolor, curl, perl, sudo, openssl ]
 #
 #      env:
 #        - MATRIX_EVAL="export CC=gcc-7 && export CXX=g++-7"
@ -36,7 +36,7 @@ matrix:
          sources:
            - ubuntu-toolchain-r-test
            - llvm-toolchain-trusty-5.0
-          packages: [ g++-7, clang-5.0, lld-5.0, libicu-dev, libreadline-dev, libmysqlclient-dev, unixodbc-dev, libltdl-dev, libssl-dev, libboost-dev, zlib1g-dev, libdouble-conversion-dev, libzookeeper-mt-dev, libsparsehash-dev, librdkafka-dev, libcapnp-dev, libsparsehash-dev, libgoogle-perftools-dev, bash, expect, python, python-lxml, python-termcolor, curl, perl, sudo ]
+          packages: [ g++-7, clang-5.0, lld-5.0, libicu-dev, libreadline-dev, libmysqlclient-dev, unixodbc-dev, libltdl-dev, libssl-dev, libboost-dev, zlib1g-dev, libdouble-conversion-dev, libzookeeper-mt-dev, libsparsehash-dev, librdkafka-dev, libcapnp-dev, libsparsehash-dev, libgoogle-perftools-dev, bash, expect, python, python-lxml, python-termcolor, curl, perl, sudo, openssl]

      env:
        - MATRIX_EVAL="export CC=clang-5.0 && export CXX=clang++-5.0"
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -1,4 +1,31 @@
-# ClickHouse 1.1.54356 Release Candidate, 2018-03-06
+# ClickHouse release 1.1.54370, 2018-03-16
+
+## New features:
+
+* Added the `system.macros` table and auto updating of macros when the config file is changed.
+* Added the `SYSTEM RELOAD CONFIG` query.
+* Added the `maxIntersections(left_col, right_col)` aggregate function, which returns the maximum number of simultaneously intersecting intervals `[left; right]`. The `maxIntersectionsPosition(left, right)` function returns the beginning of the "maximum" interval. ([Michael Furmur](https://github.com/yandex/ClickHouse/pull/2012)).
+
+## Improvements:
+
+* When inserting data in a `Replicated` table, fewer requests are made to `ZooKeeper` (and most of the user-level errors have disappeared from the `ZooKeeper` log).
+* Added the ability to create aliases for sets. Example: `WITH (1, 2, 3) AS set SELECT number IN set FROM system.numbers LIMIT 10`.
+
+## Bug fixes:
+
+* Fixed the `Illegal PREWHERE` error when reading from `Merge` tables over `Distributed` tables.
+* Added fixes that allow you to run `clickhouse-server` in IPv4-only Docker containers.
+* Fixed a race condition when reading from system `system.parts_columns` tables.
+* Removed double buffering during a synchronous insert to a `Distributed` table, which could have caused the connection to timeout.
+* Fixed a bug that caused excessively long waits for an unavailable replica before beginning a `SELECT` query.
+* Fixed incorrect dates in the `system.parts` table.
+* Fixed a bug that made it impossible to insert data in a `Replicated` table if `chroot` was non-empty in the configuration of the `ZooKeeper` cluster.
+* Fixed the vertical merging algorithm for an empty `ORDER BY` table.
+* Restored the ability to use dictionaries in queries to remote tables, even if these dictionaries are not present on the requestor server. This functionality was lost in release 1.1.54362.
+* Restored the behavior for queries like `SELECT * FROM remote('server2', default.table) WHERE col IN (SELECT col2 FROM default.table)` when the right side argument of the `IN` should use a remote `default.table` instead of a local one. This behavior was broken in version 1.1.54358.
+* Removed extraneous error-level logging of `Not found column ... in block`.
+
+# ClickHouse release 1.1.54356, 2018-03-06

 ## New features:

--- a/CHANGELOG_RU.md
+++ b/CHANGELOG_RU.md
@ -1,4 +1,32 @@
-# ClickHouse 1.1.54362 Release Candidate, 2018-03-11
+# ClickHouse release 1.1.54370, 2018-03-16
+
+## Новые возможности:
+
+* Добавлена системная таблица `system.macros` и автоматическое обновление макросов при изменении конфигурационного файла.
+* Добавлен запрос `SYSTEM RELOAD CONFIG`.
+* Добавлена агрегатная функция `maxIntersections(left_col, right_col)`, возвращающая максимальное количество одновременно пересекающихся интервалов `[left; right]`. Функция `maxIntersectionsPosition(left, right)` возвращает начало такого "максимального" интервала. ([Michael Furmur](https://github.com/yandex/ClickHouse/pull/2012)).
+
+## Улучшения:
+
+* При вставке данных в `Replicated`-таблицу делается меньше обращений к `ZooKeeper` (также из лога `ZooKeeper` исчезло большинство user-level ошибок).
+* Добавлена возможность создавать алиасы для множеств. Пример: `WITH (1, 2, 3) AS set SELECT number IN set FROM system.numbers LIMIT 10`.
+
+## Исправление ошибок:
+
+* Исправлена ошибка `Illegal PREWHERE` при чтении из Merge-таблицы над `Distributed`-таблицами.
+* Добавлены исправления, позволяющие запускать clickhouse-server в IPv4-only docker-контейнерах.
+* Исправлен race condition при чтении из системной таблицы `system.parts_columns`
+* Убрана двойная буферизация при синхронной вставке в `Distributed`-таблицу, которая могла приводить к timeout-ам соединений.
+* Исправлена ошибка, вызывающая чрезмерно долгое ожидание недоступной реплики перед началом выполнения `SELECT`.
+* Исправлено некорректное отображение дат в таблице `system.parts`.
+* Исправлена ошибка, приводящая к невозможности вставить данные в `Replicated`-таблицу, если в конфигурации кластера `ZooKeeper` задан непустой `chroot`.
+* Исправлен алгоритм вертикального мержа при пустом ключе `ORDER BY` таблицы.
+* Возвращена возможность использовать словари в запросах к удаленным таблицам, даже если этих словарей нет на сервере-инициаторе. Данная функциональность была потеряна в версии 1.1.54362.
+* Восстановлено поведение, при котором в запросах типа `SELECT * FROM remote('server2', default.table) WHERE col IN (SELECT col2 FROM default.table)` в правой части `IN` должна использоваться удаленная таблица `default.table`, а не локальная. данное поведение было нарушено в версии 1.1.54358.
+* Устранено ненужное Error-level логирование `Not found column ... in block`.
+
+
+# Релиз ClickHouse 1.1.54362, 2018-03-11

 ## Новые возможности:

--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -252,13 +252,13 @@ if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/contrib/poco/cmake/FindODBC.cmake)
 else ()
    include (cmake/find_odbc.cmake)
 endif ()
+message (STATUS "Using odbc: ${ODBC_INCLUDE_DIRECTORIES} : ${ODBC_LIBRARIES}")
 include (cmake/find_poco.cmake)
 include (cmake/find_lz4.cmake)
 include (cmake/find_sparsehash.cmake)
 include (cmake/find_rt.cmake)
 include (cmake/find_execinfo.cmake)
 include (cmake/find_readline_edit.cmake)
-include (cmake/find_zookeeper.cmake)
 include (cmake/find_re2.cmake)
 include (cmake/find_rdkafka.cmake)
 include (cmake/find_capnp.cmake)
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@ -30,3 +30,5 @@ Replace the bracketed text as follows:
 * [link] is the link at the current version of the CLA (you may add here a link https://yandex.ru/legal/cla/?lang=en (in English) or a link https://yandex.ru/legal/cla/?lang=ru (in Russian).

 It is enough to provide us such notification at once.
+
+If you don't agree with the CLA, you still can open a pull request to provide your contributions.
--- a/README.md
+++ b/README.md
@ -1,5 +1,8 @@
+# ClickHouse
 ClickHouse is an open-source column-oriented database management system that allows generating analytical data reports in real time.

-[Read more...](https://clickhouse.yandex/)
+🎤🥂 **ClickHouse Meetup in [Sunnyvale](https://www.meetup.com/San-Francisco-Bay-Area-ClickHouse-Meetup/events/248898966/) & [San Francisco](https://www.meetup.com/San-Francisco-Bay-Area-ClickHouse-Meetup/events/249162518/), April 23-27** 🍰🔥🐻
+
+Learn more about ClickHouse at [https://clickhouse.yandex/](https://clickhouse.yandex/)

 [![Build Status](https://travis-ci.org/yandex/ClickHouse.svg?branch=master)](https://travis-ci.org/yandex/ClickHouse)
--- a/cmake/Modules/FindPoco.cmake
+++ b/cmake/Modules/FindPoco.cmake
@ -51,6 +51,7 @@

 set(Poco_HINTS
    /usr/local
+    /usr/local/include/Poco
    C:/AppliedInformatics
    ${Poco_DIR}
    $ENV{Poco_DIR}
@ -230,5 +231,3 @@ if(${Poco_OSP_FOUND})
 endif()

 message(STATUS "Found Poco: ${Poco_LIBRARIES}")
-
-
--- a/cmake/find_poco.cmake
+++ b/cmake/find_poco.cmake
@ -9,7 +9,7 @@ if (NOT EXISTS "${ClickHouse_SOURCE_DIR}/contrib/poco/CMakeLists.txt")
 endif ()

 if (NOT USE_INTERNAL_POCO_LIBRARY)
-    find_package (Poco COMPONENTS Net NetSSL XML Data Crypto DataODBC MongoDB)
+    find_package (Poco COMPONENTS Net NetSSL XML SQL Data Crypto DataODBC MongoDB)
 endif ()

 if (Poco_INCLUDE_DIRS AND Poco_Foundation_LIBRARY)
@ -24,6 +24,15 @@ elseif (NOT MISSING_INTERNAL_POCO_LIBRARY)
    set (ENABLE_DATA_SQLITE 0 CACHE BOOL "")
    set (ENABLE_DATA_MYSQL 0 CACHE BOOL "")
    set (ENABLE_DATA_POSTGRESQL 0 CACHE BOOL "")
+    # new after 2.0.0:
+    set (POCO_ENABLE_ZIP 0 CACHE BOOL "")
+    set (POCO_ENABLE_PAGECOMPILER 0 CACHE BOOL "")
+    set (POCO_ENABLE_PAGECOMPILER_FILE2PAGE 0 CACHE BOOL "")
+    set (POCO_ENABLE_REDIS 0 CACHE BOOL "")
+    set (POCO_ENABLE_SQL_SQLITE 0 CACHE BOOL "")
+    set (POCO_ENABLE_SQL_MYSQL 0 CACHE BOOL "")
+    set (POCO_ENABLE_SQL_POSTGRESQL 0 CACHE BOOL "")
+
    set (POCO_UNBUNDLED 1 CACHE BOOL "")
    set (POCO_UNBUNDLED_PCRE 0 CACHE BOOL "")
    set (POCO_UNBUNDLED_EXPAT 0 CACHE BOOL "")
@ -43,13 +52,39 @@ elseif (NOT MISSING_INTERNAL_POCO_LIBRARY)
        set (Poco_MongoDB_INCLUDE_DIRS "${ClickHouse_SOURCE_DIR}/contrib/poco/MongoDB/include/")
    endif ()

+
+    if (EXISTS "${ClickHouse_SOURCE_DIR}/contrib/poco/SQL/ODBC/include/")
+        set (Poco_SQL_FOUND 1)
+        set (Poco_SQL_LIBRARY PocoSQL)
+        set (Poco_SQL_INCLUDE_DIRS
+             "${ClickHouse_SOURCE_DIR}/contrib/poco/SQL/include"
+             "${ClickHouse_SOURCE_DIR}/contrib/poco/Data/include"
+             )
+        if (ODBC_FOUND)
+            set (Poco_SQLODBC_FOUND 1)
+            set (Poco_SQLODBC_INCLUDE_DIRS
+                "${ClickHouse_SOURCE_DIR}/contrib/poco/SQL/ODBC/include/"
+                "${ClickHouse_SOURCE_DIR}/contrib/poco/Data/ODBC/include/"
+                ${ODBC_INCLUDE_DIRECTORIES}
+                )
+            set (Poco_SQLODBC_LIBRARY PocoSQLODBC ${ODBC_LIBRARIES} ${LTDL_LIBRARY})
+        endif ()
+    else ()
+        set (Poco_Data_FOUND 1)
+        set (Poco_Data_INCLUDE_DIRS "${ClickHouse_SOURCE_DIR}/contrib/poco/Data/include")
+        set (Poco_Data_LIBRARY PocoData)
        if (ODBC_FOUND)
            set (Poco_DataODBC_FOUND 1)
+            set (Poco_DataODBC_INCLUDE_DIRS
+                "${ClickHouse_SOURCE_DIR}/contrib/poco/Data/ODBC/include/"
+                ${ODBC_INCLUDE_DIRECTORIES}
+            )
            set (Poco_DataODBC_LIBRARY PocoDataODBC ${ODBC_LIBRARIES} ${LTDL_LIBRARY})
-        set (Poco_DataODBC_INCLUDE_DIRS "${ClickHouse_SOURCE_DIR}/contrib/poco/Data/ODBC/include/")
+        endif ()
    endif ()

-    if (OPENSSL_FOUND)
+    # TODO! fix internal ssl
+    if (OPENSSL_FOUND AND NOT USE_INTERNAL_SSL_LIBRARY)
        set (Poco_NetSSL_FOUND 1)
        set (Poco_NetSSL_LIBRARY PocoNetSSL)
        set (Poco_Crypto_LIBRARY PocoCrypto)
@ -65,7 +100,6 @@ elseif (NOT MISSING_INTERNAL_POCO_LIBRARY)
    set (Poco_Foundation_LIBRARY PocoFoundation)
    set (Poco_Util_LIBRARY PocoUtil)
    set (Poco_Net_LIBRARY PocoNet)
-    set (Poco_Data_LIBRARY PocoData)
    set (Poco_XML_LIBRARY PocoXML)
 endif ()

@ -85,3 +119,5 @@ message(STATUS "Using Poco: ${Poco_INCLUDE_DIRS} : ${Poco_Foundation_LIBRARY},${
 # ClickHouse-Extras/clickhouse_warning
 # ClickHouse-Extras/clickhouse-purge-logs-on-no-space
 # ClickHouse-Extras/clickhouse_freebsd
+# ClickHouse-Extras/clikhouse_no_zlib
+# ClickHouse-Extras/clickhouse-fix-atomic
--- a/cmake/find_zookeeper.cmake
+++ b/cmake/find_zookeeper.cmake
@ -1,35 +0,0 @@
-if (ARCH_ARM)
-    # bundled have some asm broken for arm, use package libzookeeper-mt-dev
-    set(USE_INTERNAL_ZOOKEEPER_LIBRARY 0 CACHE BOOL "")
-endif ()
-
-option (USE_INTERNAL_ZOOKEEPER_LIBRARY "Set to FALSE to use system zookeeper library instead of bundled" ${NOT_UNBUNDLED})
-
-if (USE_INTERNAL_ZOOKEEPER_LIBRARY AND NOT EXISTS "${ClickHouse_SOURCE_DIR}/contrib/zookeeper/src/c/CMakeLists.txt")
-   message (WARNING "submodule contrib/zookeeper is missing. to fix try run: \n git submodule update --init --recursive")
-   set (USE_INTERNAL_ZOOKEEPER_LIBRARY 0)
-endif ()
-
-if (NOT USE_INTERNAL_ZOOKEEPER_LIBRARY)
-    find_library (ZOOKEEPER_LIBRARY zookeeper_mt)
-    find_path (ZOOKEEPER_INCLUDE_DIR NAMES zookeeper/zookeeper.h PATHS ${ZOOKEEPER_INCLUDE_PATHS})
-    set(ZOOKEEPER_INCLUDE_DIR "${ZOOKEEPER_INCLUDE_DIR}/zookeeper")
-endif ()
-
-if (ZOOKEEPER_LIBRARY AND ZOOKEEPER_INCLUDE_DIR)
-else ()
-    set (USE_INTERNAL_ZOOKEEPER_LIBRARY 1)
-    set(WANT_CPPUNIT 0 CACHE BOOL "")
-    set (ZOOKEEPER_LIBRARY zookeeper)
-endif ()
-
-message (STATUS "Using zookeeper: ${ZOOKEEPER_INCLUDE_DIR} : ${ZOOKEEPER_LIBRARY}")
-
-
-# how to make clickhouse branch of https://github.com/ClickHouse-Extras/zookeeper.git :
-# git remote add upstream https://github.com/apache/zookeeper.git
-# git checkhout upstream/master
-# git branch -D clickhouse
-# git checkout -b clickhouse
-# git merge clickhouse_misc
-# git merge clickhouse_706
--- a/contrib/CMakeLists.txt
+++ b/contrib/CMakeLists.txt
@ -25,10 +25,6 @@ if (USE_INTERNAL_DOUBLE_CONVERSION_LIBRARY)
    add_subdirectory (double-conversion)
 endif ()

-if (USE_INTERNAL_ZOOKEEPER_LIBRARY)
-    add_subdirectory (zookeeper/src/c)
-endif ()
-
 if (USE_INTERNAL_CITYHASH_LIBRARY)
    add_subdirectory (libcityhash)
 endif ()
@ -100,13 +96,17 @@ if (USE_INTERNAL_RDKAFKA_LIBRARY)
    mark_as_advanced (ZLIB_INCLUDE_DIR)

    if (USE_INTERNAL_SSL_LIBRARY)
+        if (MAKE_STATIC_LIBRARIES)
            add_library(bundled-ssl ALIAS ${OPENSSL_SSL_LIBRARY})
-        set (WITH_BUNDLED_SSL 1)
+            set (WITH_BUNDLED_SSL 1 CACHE INTERNAL "")
+        else ()
+            set (WITH_SSL 0 CACHE INTERNAL "")
+        endif ()
    endif ()

    add_subdirectory (librdkafka)

-    if (USE_INTERNAL_SSL_LIBRARY)
+    if (USE_INTERNAL_SSL_LIBRARY AND MAKE_STATIC_LIBRARIES)
        target_include_directories(rdkafka PRIVATE BEFORE ${OPENSSL_INCLUDE_DIR})
    endif ()
    target_include_directories(rdkafka PRIVATE BEFORE ${ZLIB_INCLUDE_DIR})
@ -127,16 +127,18 @@ endif ()


 if (USE_INTERNAL_POCO_LIBRARY)
-    set (ALLOW_DUPLICATE_CUSTOM_TARGETS 1)
    set (save_CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS})
    set (save_CMAKE_C_FLAGS ${CMAKE_C_FLAGS})
    set (_save ${ENABLE_TESTS})
    set (ENABLE_TESTS 0)
    set (CMAKE_DISABLE_FIND_PACKAGE_ZLIB 1)
    if (USE_INTERNAL_SSL_LIBRARY)
-        set (DISABLE_INTERNAL_OPENSSL 1)
-        set (ENABLE_NETSSL 0) # TODO!
-        set (ENABLE_CRYPTO 0) # TODO!
+        set (DISABLE_INTERNAL_OPENSSL 1 CACHE INTERNAL "")
+        set (ENABLE_NETSSL 0 CACHE INTERNAL "") # TODO!
+        set (ENABLE_CRYPTO 0 CACHE INTERNAL "") # TODO!
+    endif ()
+    if (MSVC)
+        set (ENABLE_DATA_ODBC 0 CACHE INTERNAL "") # TODO (build fail)
    endif ()
    add_subdirectory (poco)
    unset (CMAKE_DISABLE_FIND_PACKAGE_ZLIB)
--- a/contrib/boost
+++ b/contrib/boost
@ -1 +1 @@
-Subproject commit eb5943711e88d1008583e6ae3720a5489313d02e
+Subproject commit 5121cc9d0375c7b81b24b6087a51684e6cd62ded
--- a/contrib/libcityhash/src/config.h
+++ b/contrib/libcityhash/src/config.h
@ -5,7 +5,11 @@
 /* #undef AC_APPLE_UNIVERSAL_BUILD */

 /* Define to 1 if the compiler supports __builtin_expect. */
+#if _MSC_VER
+#define HAVE_BUILTIN_EXPECT 0
+#else
 #define HAVE_BUILTIN_EXPECT 1
+#endif

 /* Define to 1 if you have the <dlfcn.h> header file. */
 #define HAVE_DLFCN_H 1
--- a/contrib/libfarmhash/CMakeLists.txt
+++ b/contrib/libfarmhash/CMakeLists.txt
@ -2,4 +2,8 @@ add_library(farmhash
 	farmhash.cc
 	farmhash.h)

+if (MSVC)
+    target_compile_definitions (farmhash PRIVATE FARMHASH_NO_BUILTIN_EXPECT=1)
+endif ()
+
 target_include_directories (farmhash PUBLIC ${CMAKE_CURRENT_BINARY_DIR})
--- a/contrib/libtcmalloc/src/config.h
+++ b/contrib/libtcmalloc/src/config.h
@ -13,7 +13,11 @@
 /* #undef ENABLE_SIZED_DELETE */

 /* Define to 1 if compiler supports __builtin_expect */
+#if _MSC_VER
+#define HAVE_BUILTIN_EXPECT 0
+#else
 #define HAVE_BUILTIN_EXPECT 1
+#endif

 /* Define to 1 if compiler supports __builtin_stack_pointer */
 /* #undef HAVE_BUILTIN_STACK_POINTER */
--- a/contrib/poco
+++ b/contrib/poco
@ -1 +1 @@
-Subproject commit 8238852d7ab2a4abdf87adff233b3b83686f4fe4
+Subproject commit 930a7ec1154f4f9711edfb4b4a39f9fff2a5bbb5
--- a/contrib/zookeeper
+++ b/contrib/zookeeper
@ -1 +0,0 @@
-Subproject commit 438afae5af36c5be9c82d074f43a9bb19e0797c0
--- a/contrib/zstd
+++ b/contrib/zstd
@ -1 +1 @@
-Subproject commit f4340f46b2387bc8de7d5320c0b83bb1499933ad
+Subproject commit 255597502c3a4ef150abc964e376d4202a8c2929
--- a/copy_headers.sh
+++ b/copy_headers.sh
@ -8,7 +8,7 @@
 # Требуется clang, желательно наиболее свежий (trunk).
 #
 # Используется при сборке пакетов.
-# Заголовочные файлы записываются в пакет clickhouse-server-base, в директорию /usr/share/clickhouse/headers.
+# Заголовочные файлы записываются в пакет clickhouse-common, в директорию /usr/share/clickhouse/headers.
 #
 # Если вы хотите установить их самостоятельно, без сборки пакета,
 #  чтобы clickhouse-server видел их там, где ожидается, выполните:
--- a/dbms/CMakeLists.txt
+++ b/dbms/CMakeLists.txt
@ -131,7 +131,6 @@ target_link_libraries (clickhouse_common_io
    ${LINK_LIBRARIES_ONLY_ON_X86_64}
    ${LZ4_LIBRARY}
    ${ZSTD_LIBRARY}
-    ${ZOOKEEPER_LIBRARY}
    ${DOUBLE_CONVERSION_LIBRARIES}
    ${Poco_Net_LIBRARY}
    ${Poco_Data_LIBRARY}
@ -146,8 +145,6 @@ target_link_libraries (dbms
    clickhouse_common_config
    clickhouse_common_io
    ${MYSQLXX_LIBRARY}
-    ${FARMHASH_LIBRARIES}
-    ${METROHASH_LIBRARIES}
    ${RE2_LIBRARY}
    ${RE2_ST_LIBRARY}
    ${OPENSSL_CRYPTO_LIBRARY}
@ -158,19 +155,33 @@ if (NOT USE_INTERNAL_RE2_LIBRARY)
    target_include_directories (dbms BEFORE PRIVATE ${RE2_INCLUDE_DIR})
 endif ()

-if (NOT USE_INTERNAL_ZOOKEEPER_LIBRARY)
-    target_include_directories (clickhouse_common_io BEFORE PUBLIC ${ZOOKEEPER_INCLUDE_DIR})
-endif ()
-
 if (NOT USE_INTERNAL_BOOST_LIBRARY)
    target_include_directories (clickhouse_common_io BEFORE PUBLIC ${Boost_INCLUDE_DIRS})
 endif ()

-if (Poco_DataODBC_FOUND)
-    target_link_libraries (dbms ${Poco_DataODBC_LIBRARY})
-    target_include_directories (dbms PRIVATE ${ODBC_INCLUDE_DIRECTORIES})
+if (Poco_SQLODBC_FOUND)
+    target_link_libraries (clickhouse_common_io ${Poco_SQL_LIBRARY})
+    target_link_libraries (dbms ${Poco_SQLODBC_LIBRARY} ${Poco_SQL_LIBRARY})
+    if (NOT USE_INTERNAL_POCO_LIBRARY)
+        target_include_directories (clickhouse_common_io PRIVATE ${ODBC_INCLUDE_DIRECTORIES} ${Poco_SQL_INCLUDE_DIRS})
+        target_include_directories (dbms PRIVATE ${ODBC_INCLUDE_DIRECTORIES} ${Poco_SQLODBC_INCLUDE_DIRS} PUBLIC ${Poco_SQL_INCLUDE_DIRS})
+    endif()
 endif()

+if (Poco_Data_FOUND AND NOT USE_INTERNAL_POCO_LIBRARY)
+    target_include_directories (clickhouse_common_io PRIVATE ${Poco_Data_INCLUDE_DIRS})
+    target_include_directories (dbms PRIVATE ${Poco_Data_INCLUDE_DIRS})
+endif()
+
+if (Poco_DataODBC_FOUND)
+    target_link_libraries (clickhouse_common_io ${Poco_Data_LIBRARY})
+    target_link_libraries (dbms ${Poco_DataODBC_LIBRARY})
+    if (NOT USE_INTERNAL_POCO_LIBRARY)
+        target_include_directories (dbms PRIVATE ${ODBC_INCLUDE_DIRECTORIES} ${Poco_DataODBC_INCLUDE_DIRS})
+    endif()
+endif()
+
+
 if (Poco_MongoDB_FOUND)
    target_link_libraries (dbms ${Poco_MongoDB_LIBRARY})
 endif()
@ -217,6 +228,7 @@ endif ()
 target_include_directories (dbms PUBLIC ${DBMS_INCLUDE_DIR})
 target_include_directories (clickhouse_common_io PUBLIC ${DBMS_INCLUDE_DIR})
 target_include_directories (clickhouse_common_io PUBLIC ${PCG_RANDOM_INCLUDE_DIR})
+target_include_directories (clickhouse_common_io PUBLIC ${Poco_DataODBC_INCLUDE_DIRS})
 target_include_directories (clickhouse_common_io BEFORE PUBLIC ${DOUBLE_CONVERSION_INCLUDE_DIR})

 # also for copy_headers.sh:
--- a/dbms/cmake/version.cmake
+++ b/dbms/cmake/version.cmake
@ -1,6 +1,7 @@
 # This strings autochanged from release_lib.sh:
-set(VERSION_DESCRIBE v1.1.54362-testing)
-set(VERSION_REVISION 54362)
+set(VERSION_DESCRIBE v1.1.54378-testing)
+set(VERSION_REVISION 54378)
+set(VERSION_GITHASH 5b19d89133a5ff7c72e40cc8c0226cb00466ba10)
 # end of autochange

 set (VERSION_MAJOR 1)
--- a/dbms/src/AggregateFunctions/AggregateFunctionFactory.cpp
+++ b/dbms/src/AggregateFunctions/AggregateFunctionFactory.cpp
@ -70,7 +70,10 @@ AggregateFunctionPtr AggregateFunctionFactory::get(
        return combinator->transformAggregateFunction(nested_function, argument_types, parameters);
    }

-    return getImpl(name, argument_types, parameters, recursion_level);
+    auto res = getImpl(name, argument_types, parameters, recursion_level);
+    if (!res)
+        throw Exception("Logical error: AggregateFunctionFactory returned nullptr", ErrorCodes::LOGICAL_ERROR);
+    return res;
 }


--- a/dbms/src/AggregateFunctions/AggregateFunctionMaxIntersections.cpp
+++ b/dbms/src/AggregateFunctions/AggregateFunctionMaxIntersections.cpp
@ -0,0 +1,37 @@
+#include <AggregateFunctions/AggregateFunctionFactory.h>
+#include <AggregateFunctions/AggregateFunctionMaxIntersections.h>
+#include <AggregateFunctions/FactoryHelpers.h>
+#include <AggregateFunctions/Helpers.h>
+
+
+namespace DB
+{
+
+namespace
+{
+    AggregateFunctionPtr createAggregateFunctionMaxIntersections(
+        AggregateFunctionIntersectionsKind kind,
+        const std::string & name, const DataTypes & argument_types, const Array & parameters)
+    {
+        assertBinary(name, argument_types);
+        assertNoParameters(name, parameters);
+
+        AggregateFunctionPtr res(createWithNumericType<AggregateFunctionIntersectionsMax>(*argument_types[0], kind, argument_types));
+        if (!res)
+            throw Exception("Illegal types " + argument_types[0]->getName() + " and " + argument_types[1]->getName()
+                + " of argument for aggregate function " + name, ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+
+        return res;
+    }
+}
+
+void registerAggregateFunctionsMaxIntersections(AggregateFunctionFactory & factory)
+{
+    factory.registerFunction("maxIntersections", [](const std::string & name, const DataTypes & argument_types, const Array & parameters)
+        { return createAggregateFunctionMaxIntersections(AggregateFunctionIntersectionsKind::Count, name, argument_types, parameters); });
+
+    factory.registerFunction("maxIntersectionsPosition", [](const std::string & name, const DataTypes & argument_types, const Array & parameters)
+        { return createAggregateFunctionMaxIntersections(AggregateFunctionIntersectionsKind::Position, name, argument_types, parameters); });
+}
+
+}
--- a/dbms/src/AggregateFunctions/AggregateFunctionMaxIntersections.h
+++ b/dbms/src/AggregateFunctions/AggregateFunctionMaxIntersections.h
@ -0,0 +1,170 @@
+#pragma once
+
+#include <common/logger_useful.h>
+
+#include <DataTypes/DataTypesNumber.h>
+#include <Columns/ColumnsNumber.h>
+
+#include <IO/ReadHelpers.h>
+#include <IO/WriteHelpers.h>
+
+#include <Common/ArenaAllocator.h>
+#include <Common/NaNUtils.h>
+
+#include <AggregateFunctions/IAggregateFunction.h>
+
+#define AGGREGATE_FUNCTION_MAX_INTERSECTIONS_MAX_ARRAY_SIZE 0xFFFFFF
+
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int ILLEGAL_TYPE_OF_ARGUMENT;
+    extern const int TOO_LARGE_ARRAY_SIZE;
+}
+
+
+/** maxIntersections: returns maximum count of the intersected intervals defined by start_column and end_column values,
+  * maxIntersectionsPosition: returns leftmost position of maximum intersection of intervals.
+  */
+
+/// Similar to GroupArrayNumericData.
+template <typename T>
+struct MaxIntersectionsData
+{
+    /// Left or right end of the interval and signed weight; with positive sign for begin of interval and negative sign for end of interval.
+    using Value = std::pair<T, Int64>;
+
+    // Switch to ordinary Allocator after 4096 bytes to avoid fragmentation and trash in Arena
+    using Allocator = MixedArenaAllocator<4096>;
+    using Array = PODArray<Value, 32, Allocator>;
+
+    Array value;
+};
+
+enum class AggregateFunctionIntersectionsKind
+{
+    Count,
+    Position
+};
+
+template <typename PointType>
+class AggregateFunctionIntersectionsMax final
+    : public IAggregateFunctionDataHelper<MaxIntersectionsData<PointType>, AggregateFunctionIntersectionsMax<PointType>>
+{
+private:
+    AggregateFunctionIntersectionsKind kind;
+
+public:
+    AggregateFunctionIntersectionsMax(AggregateFunctionIntersectionsKind kind_, const DataTypes & arguments)
+        : kind(kind_)
+    {
+        if (!arguments[0]->isNumber())
+            throw Exception{getName() + ": first argument must be represented by integer", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT};
+
+        if (!arguments[1]->isNumber())
+            throw Exception{getName() + ": second argument must be represented by integer", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT};
+
+        if (!arguments[0]->equals(*arguments[1]))
+            throw Exception{getName() + ": arguments must have the same type", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT};
+    }
+
+    String getName() const override
+    {
+        return kind == AggregateFunctionIntersectionsKind::Count
+            ? "maxIntersections"
+            : "maxIntersectionsPosition";
+    }
+
+    DataTypePtr getReturnType() const override
+    {
+        if (kind == AggregateFunctionIntersectionsKind::Count)
+            return std::make_shared<DataTypeUInt64>();
+        else
+            return std::make_shared<DataTypeNumber<PointType>>();
+    }
+
+    void add(AggregateDataPtr place, const IColumn ** columns, size_t row_num, Arena * arena) const override
+    {
+        PointType left = static_cast<const ColumnVector<PointType> &>(*columns[0]).getData()[row_num];
+        PointType right = static_cast<const ColumnVector<PointType> &>(*columns[1]).getData()[row_num];
+
+        if (!isNaN(left))
+            this->data(place).value.push_back(std::make_pair(left, Int64(1)), arena);
+
+        if (!isNaN(right))
+            this->data(place).value.push_back(std::make_pair(right, Int64(-1)), arena);
+    }
+
+    void merge(AggregateDataPtr place, ConstAggregateDataPtr rhs, Arena * arena) const override
+    {
+        auto & cur_elems = this->data(place);
+        auto & rhs_elems = this->data(rhs);
+
+        cur_elems.value.insert(rhs_elems.value.begin(), rhs_elems.value.end(), arena);
+    }
+
+    void serialize(ConstAggregateDataPtr place, WriteBuffer & buf) const override
+    {
+        const auto & value = this->data(place).value;
+        size_t size = value.size();
+        writeVarUInt(size, buf);
+        buf.write(reinterpret_cast<const char *>(&value[0]), size * sizeof(value[0]));
+    }
+
+    void deserialize(AggregateDataPtr place, ReadBuffer & buf, Arena * arena) const override
+    {
+        size_t size = 0;
+        readVarUInt(size, buf);
+
+        if (unlikely(size > AGGREGATE_FUNCTION_MAX_INTERSECTIONS_MAX_ARRAY_SIZE))
+            throw Exception("Too large array size", ErrorCodes::TOO_LARGE_ARRAY_SIZE);
+
+        auto & value = this->data(place).value;
+
+        value.resize(size, arena);
+        buf.read(reinterpret_cast<char *>(&value[0]), size * sizeof(value[0]));
+    }
+
+    void insertResultInto(ConstAggregateDataPtr place, IColumn & to) const override
+    {
+        Int64 current_intersections = 0;
+        Int64 max_intersections = 0;
+        PointType position_of_max_intersections = 0;
+
+        /// const_cast because we will sort the array
+        auto & array = const_cast<typename MaxIntersectionsData<PointType>::Array &>(this->data(place).value);
+
+        std::sort(array.begin(), array.end(), [](const auto & a, const auto & b) { return a.first < b.first; });
+
+        for (const auto & point_weight : array)
+        {
+            current_intersections += point_weight.second;
+            if (current_intersections > max_intersections)
+            {
+                max_intersections = current_intersections;
+                position_of_max_intersections = point_weight.first;
+            }
+        }
+
+        if (kind == AggregateFunctionIntersectionsKind::Count)
+        {
+            auto & result_column = static_cast<ColumnUInt64 &>(to).getData();
+            result_column.push_back(max_intersections);
+        }
+        else
+        {
+            auto & result_column = static_cast<ColumnVector<PointType> &>(to).getData();
+            result_column.push_back(position_of_max_intersections);
+        }
+    }
+
+    const char * getHeaderFilePath() const override
+    {
+        return __FILE__;
+    }
+};
+
+}
--- a/dbms/src/AggregateFunctions/AggregateFunctionSumMap.cpp
+++ b/dbms/src/AggregateFunctions/AggregateFunctionSumMap.cpp
@ -37,7 +37,11 @@ AggregateFunctionPtr createAggregateFunctionSumMap(const std::string & name, con
        values_types.push_back(array_type->getNestedType());
    }

-    return AggregateFunctionPtr(createWithNumericType<AggregateFunctionSumMap>(*keys_type, keys_type, std::move(values_types)));
+    AggregateFunctionPtr res(createWithNumericType<AggregateFunctionSumMap>(*keys_type, keys_type, std::move(values_types)));
+    if (!res)
+        throw Exception("Illegal type of argument for aggregate function " + name, ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+
+    return res;
 }

 }
--- a/dbms/src/AggregateFunctions/AggregateFunctionUniq.cpp
+++ b/dbms/src/AggregateFunctions/AggregateFunctionUniq.cpp
@ -33,6 +33,10 @@ AggregateFunctionPtr createAggregateFunctionUniq(const std::string & name, const
 {
    assertNoParameters(name, params);

+    if (argument_types.empty())
+        throw Exception("Incorrect number of arguments for aggregate function " + name,
+            ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
    if (argument_types.size() == 1)
    {
        const IDataType & argument_type = *argument_types[0];
@ -51,23 +55,18 @@ AggregateFunctionPtr createAggregateFunctionUniq(const std::string & name, const
            return std::make_shared<AggregateFunctionUniqVariadic<DataForVariadic, true>>(argument_types);
        else if (typeid_cast<const DataTypeUUID *>(&argument_type))
            return std::make_shared<AggregateFunctionUniq<DataTypeUUID::FieldType, Data>>();
-        else
-            throw Exception("Illegal type " + argument_types[0]->getName() + " of argument for aggregate function " + name,
-                ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
    }
-    else if (argument_types.size() > 1)
+    else
    {
        /// If there are several arguments, then no tuples allowed among them.
        for (const auto & type : argument_types)
            if (typeid_cast<const DataTypeTuple *>(type.get()))
                throw Exception("Tuple argument of function " + name + " must be the only argument",
                    ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
-
-        return std::make_shared<AggregateFunctionUniqVariadic<DataForVariadic, false>>(argument_types);
    }
-    else
-        throw Exception("Incorrect number of arguments for aggregate function " + name,
-            ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
+    /// "Variadic" method also works as a fallback generic case for single argument.
+    return std::make_shared<AggregateFunctionUniqVariadic<DataForVariadic, false>>(argument_types);
 }

 template <template <typename> class Data, typename DataForVariadic>
@ -75,6 +74,10 @@ AggregateFunctionPtr createAggregateFunctionUniq(const std::string & name, const
 {
    assertNoParameters(name, params);

+    if (argument_types.empty())
+        throw Exception("Incorrect number of arguments for aggregate function " + name,
+            ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
    if (argument_types.size() == 1)
    {
        const IDataType & argument_type = *argument_types[0];
@ -93,23 +96,18 @@ AggregateFunctionPtr createAggregateFunctionUniq(const std::string & name, const
            return std::make_shared<AggregateFunctionUniqVariadic<DataForVariadic, true>>(argument_types);
        else if (typeid_cast<const DataTypeUUID *>(&argument_type))
            return std::make_shared<AggregateFunctionUniq<DataTypeUUID::FieldType, Data<DataTypeUUID::FieldType>>>();
-        else
-            throw Exception("Illegal type " + argument_types[0]->getName() + " of argument for aggregate function " + name,
-                ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
    }
-    else if (argument_types.size() > 1)
+    else
    {
        /// If there are several arguments, then no tuples allowed among them.
        for (const auto & type : argument_types)
            if (typeid_cast<const DataTypeTuple *>(type.get()))
                throw Exception("Tuple argument of function " + name + " must be the only argument",
                    ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
-
-        return std::make_shared<AggregateFunctionUniqVariadic<DataForVariadic, false>>(argument_types);
    }
-    else
-        throw Exception("Incorrect number of arguments for aggregate function " + name,
-            ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
+    /// "Variadic" method also works as a fallback generic case for single argument.
+    return std::make_shared<AggregateFunctionUniqVariadic<DataForVariadic, false>>(argument_types);
 }

 }
--- a/dbms/src/AggregateFunctions/AggregateFunctionUniq.h
+++ b/dbms/src/AggregateFunctions/AggregateFunctionUniq.h
@ -122,8 +122,8 @@ struct AggregateFunctionUniqExactData<String>
    static String getName() { return "uniqExact"; }
 };

-template <typename T, HyperLogLogMode mode>
-struct BaseUniqCombinedData
+template <typename T>
+struct AggregateFunctionUniqCombinedData
 {
    using Key = UInt32;
    using Set = CombinedCardinalityEstimator<
@ -135,14 +135,15 @@ struct BaseUniqCombinedData
        TrivialHash,
        UInt32,
        HyperLogLogBiasEstimator<UniqCombinedBiasData>,
-        mode
-    >;
+        HyperLogLogMode::FullFeatured>;

    Set set;
+
+    static String getName() { return "uniqCombined"; }
 };

-template <HyperLogLogMode mode>
-struct BaseUniqCombinedData<String, mode>
+template <>
+struct AggregateFunctionUniqCombinedData<String>
 {
    using Key = UInt64;
    using Set = CombinedCardinalityEstimator<
@ -154,44 +155,14 @@ struct BaseUniqCombinedData<String, mode>
        TrivialHash,
        UInt64,
        HyperLogLogBiasEstimator<UniqCombinedBiasData>,
-        mode
-    >;
+        HyperLogLogMode::FullFeatured>;

    Set set;
-};

-/// Aggregate functions uniqCombinedRaw, uniqCombinedLinearCounting, and uniqCombinedBiasCorrected
-///  are intended for development of new versions of the uniqCombined function.
-/// Users should only use uniqCombined.
-
-template <typename T>
-struct AggregateFunctionUniqCombinedRawData
-    : public BaseUniqCombinedData<T, HyperLogLogMode::Raw>
-{
-    static String getName() { return "uniqCombinedRaw"; }
-};
-
-template <typename T>
-struct AggregateFunctionUniqCombinedLinearCountingData
-    : public BaseUniqCombinedData<T, HyperLogLogMode::LinearCounting>
-{
-    static String getName() { return "uniqCombinedLinearCounting"; }
-};
-
-template <typename T>
-struct AggregateFunctionUniqCombinedBiasCorrectedData
-    : public BaseUniqCombinedData<T, HyperLogLogMode::BiasCorrected>
-{
-    static String getName() { return "uniqCombinedBiasCorrected"; }
-};
-
-template <typename T>
-struct AggregateFunctionUniqCombinedData
-    : public BaseUniqCombinedData<T, HyperLogLogMode::FullFeatured>
-{
    static String getName() { return "uniqCombined"; }
 };

+
 namespace detail
 {

@ -288,10 +259,7 @@ struct OneAdder
                data.set.insert(CityHash_v1_0_2::CityHash64(value.data, value.size));
            }
        }
-        else if constexpr (std::is_same_v<Data, AggregateFunctionUniqCombinedRawData<T>>
-            || std::is_same_v<Data, AggregateFunctionUniqCombinedLinearCountingData<T>>
-            || std::is_same_v<Data, AggregateFunctionUniqCombinedBiasCorrectedData<T>>
-            || std::is_same_v<Data, AggregateFunctionUniqCombinedData<T>>)
+        else if constexpr (std::is_same_v<Data, AggregateFunctionUniqCombinedData<T>>)
        {
            if constexpr (!std::is_same_v<T, String>)
            {
@ -371,7 +339,7 @@ public:

 /** For multiple arguments. To compute, hashes them.
  * You can pass multiple arguments as is; You can also pass one argument - a tuple.
-  * But (for the possibility of effective implementation), you can not pass several arguments, among which there are tuples.
+  * But (for the possibility of efficient implementation), you can not pass several arguments, among which there are tuples.
  */
 template <typename Data, bool argument_is_tuple>
 class AggregateFunctionUniqVariadic final : public IAggregateFunctionDataHelper<Data, AggregateFunctionUniqVariadic<Data, argument_is_tuple>>
--- a/dbms/src/AggregateFunctions/AggregateFunctionUniqUpTo.cpp
+++ b/dbms/src/AggregateFunctions/AggregateFunctionUniqUpTo.cpp
@ -42,6 +42,10 @@ AggregateFunctionPtr createAggregateFunctionUniqUpTo(const std::string & name, c
        threshold = threshold_param;
    }

+    if (argument_types.empty())
+        throw Exception("Incorrect number of arguments for aggregate function " + name,
+            ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
    if (argument_types.size() == 1)
    {
        const IDataType & argument_type = *argument_types[0];
@ -60,22 +64,18 @@ AggregateFunctionPtr createAggregateFunctionUniqUpTo(const std::string & name, c
            return std::make_shared<AggregateFunctionUniqUpToVariadic<true>>(argument_types, threshold);
        else if (typeid_cast<const DataTypeUUID *>(&argument_type))
            return std::make_shared<AggregateFunctionUniqUpTo<DataTypeUUID::FieldType>>(threshold);
-        else
-            throw Exception("Illegal type " + argument_types[0]->getName() + " of argument for aggregate function " + name, ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
    }
-    else if (argument_types.size() > 1)
+    else
    {
        /// If there are several arguments, then no tuples allowed among them.
        for (const auto & type : argument_types)
            if (typeid_cast<const DataTypeTuple *>(type.get()))
                throw Exception("Tuple argument of function " + name + " must be the only argument",
                    ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
-
-        return std::make_shared<AggregateFunctionUniqUpToVariadic<false>>(argument_types, threshold);
    }
-    else
-        throw Exception("Incorrect number of arguments for aggregate function " + name,
-            ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
+    /// "Variadic" method also works as a fallback generic case for single argument.
+    return std::make_shared<AggregateFunctionUniqUpToVariadic<false>>(argument_types, threshold);
 }

 }
--- a/dbms/src/AggregateFunctions/QuantileExact.h
+++ b/dbms/src/AggregateFunctions/QuantileExact.h
@ -1,6 +1,7 @@
 #pragma once

 #include <Common/PODArray.h>
+#include <Common/NaNUtils.h>
 #include <IO/WriteBuffer.h>
 #include <IO/ReadBuffer.h>
 #include <Core/Types.h>
@ -32,6 +33,8 @@ struct QuantileExact

    void add(const Value & x)
    {
+        /// We must skip NaNs as they are not compatible with comparison sorting.
+        if (!isNaN(x))
            array.push_back(x);
    }

--- a/dbms/src/AggregateFunctions/QuantileExactWeighted.h
+++ b/dbms/src/AggregateFunctions/QuantileExactWeighted.h
@ -1,6 +1,7 @@
 #pragma once

 #include <Common/HashTable/HashMap.h>
+#include <Common/NaNUtils.h>


 namespace DB
@ -33,11 +34,14 @@ struct QuantileExactWeighted

    void add(const Value & x)
    {
+        /// We must skip NaNs as they are not compatible with comparison sorting.
+        if (!isNaN(x))
            ++map[x];
    }

    void add(const Value & x, const Weight & weight)
    {
+        if (!isNaN(x))
            map[x] += weight;
    }

--- a/dbms/src/AggregateFunctions/QuantilesCommon.h
+++ b/dbms/src/AggregateFunctions/QuantilesCommon.h
@ -4,6 +4,7 @@

 #include <Core/Field.h>
 #include <Common/FieldVisitors.h>
+#include <Common/NaNUtils.h>


 namespace DB
@ -12,6 +13,7 @@ namespace DB
 namespace ErrorCodes
 {
    extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
+    extern const int PARAMETER_OUT_OF_BOUND;
 }


@ -55,6 +57,10 @@ struct QuantileLevels
        for (size_t i = 0; i < size; ++i)
        {
            levels[i] = applyVisitor(FieldVisitorConvertToNumber<Float64>(), params[i]);
+
+            if (isNaN(levels[i]) || levels[i] < 0 || levels[i] > 1)
+                throw Exception("Quantile level is out of range [0..1]", ErrorCodes::PARAMETER_OUT_OF_BOUND);
+
            permutation[i] = i;
        }

--- a/dbms/src/AggregateFunctions/ReservoirSampler.h
+++ b/dbms/src/AggregateFunctions/ReservoirSampler.h
@ -9,6 +9,7 @@
 #include <IO/ReadHelpers.h>
 #include <IO/WriteHelpers.h>
 #include <Common/PODArray.h>
+#include <Common/NaNUtils.h>
 #include <Poco/Exception.h>
 #include <pcg_random.hpp>

@ -67,6 +68,9 @@ public:

    void insert(const T & v)
    {
+        if (isNaN(v))
+            return;
+
        sorted = false;
        ++total_values;
        if (samples.size() < sample_count)
--- a/dbms/src/AggregateFunctions/ReservoirSamplerDeterministic.h
+++ b/dbms/src/AggregateFunctions/ReservoirSamplerDeterministic.h
@ -11,6 +11,7 @@
 #include <IO/ReadHelpers.h>
 #include <IO/WriteHelpers.h>
 #include <Common/PODArray.h>
+#include <Common/NaNUtils.h>
 #include <Poco/Exception.h>


@ -66,6 +67,9 @@ public:

    void insert(const T & v, const UInt64 determinator)
    {
+        if (isNaN(v))
+            return;
+
        const UInt32 hash = intHash64(determinator);
        if (!good(hash))
            return;
--- a/dbms/src/AggregateFunctions/parseAggregateFunctionParameters.cpp
+++ b/dbms/src/AggregateFunctions/parseAggregateFunctionParameters.cpp
@ -65,7 +65,7 @@ void getAggregateFunctionNameAndParametersArray(
    ParserExpressionList params_parser(false);
    ASTPtr args_ast = parseQuery(params_parser,
        parameters_str.data(), parameters_str.data() + parameters_str.size(),
-        "parameters of aggregate function in " + error_context);
+        "parameters of aggregate function in " + error_context, 0);

    ASTExpressionList & args_list = typeid_cast<ASTExpressionList &>(*args_ast);
    if (args_list.children.empty())
--- a/dbms/src/AggregateFunctions/registerAggregateFunctions.cpp
+++ b/dbms/src/AggregateFunctions/registerAggregateFunctions.cpp
@ -23,6 +23,7 @@ void registerAggregateFunctionsUniq(AggregateFunctionFactory &);
 void registerAggregateFunctionUniqUpTo(AggregateFunctionFactory &);
 void registerAggregateFunctionTopK(AggregateFunctionFactory &);
 void registerAggregateFunctionsBitwise(AggregateFunctionFactory &);
+void registerAggregateFunctionsMaxIntersections(AggregateFunctionFactory &);

 void registerAggregateFunctionCombinatorIf(AggregateFunctionCombinatorFactory &);
 void registerAggregateFunctionCombinatorArray(AggregateFunctionCombinatorFactory &);
@ -53,6 +54,7 @@ void registerAggregateFunctions()
        registerAggregateFunctionUniqUpTo(factory);
        registerAggregateFunctionTopK(factory);
        registerAggregateFunctionsBitwise(factory);
+        registerAggregateFunctionsMaxIntersections(factory);
    }

    {
--- a/dbms/src/Analyzers/AnalyzeColumns.cpp
+++ b/dbms/src/Analyzers/AnalyzeColumns.cpp
@ -173,7 +173,7 @@ ASTPtr createASTIdentifierForColumnInTable(const String & column, const CollectT
 void createASTsForAllColumnsInTable(const CollectTables::TableInfo & table, ASTs & res)
 {
    if (table.storage)
-        for (const auto & name : table.storage->getColumnNamesList())
+        for (const auto & name : table.storage->getColumns().getNamesOfPhysical())
            res.emplace_back(createASTIdentifierForColumnInTable(name, table));
    else
        for (size_t i = 0, size = table.structure_of_subquery.columns(); i < size; ++i)
@ -315,7 +315,7 @@ void processIdentifier(
    }
    else if (table->storage)
    {
-        info.data_type = table->storage->getDataTypeByName(column_name);
+        info.data_type = table->storage->getColumn(column_name).type;
    }
    else
        throw Exception("Logical error: no storage and no structure of subquery is specified for table", ErrorCodes::LOGICAL_ERROR);
--- a/dbms/src/Analyzers/ExecuteTableFunctions.cpp
+++ b/dbms/src/Analyzers/ExecuteTableFunctions.cpp
@ -72,7 +72,7 @@ void ExecuteTableFunctions::dump(WriteBuffer & out) const
    {
        writeString(table.second->getName(), out);
        writeCString("\n\n", out);
-        writeString(table.second->getColumnsList().toString(), out);
+        writeString(table.second->getColumns().getAllPhysical().toString(), out);
        writeCString("\n", out);
    }
 }
--- a/dbms/src/Analyzers/tests/analyze_columns.cpp
+++ b/dbms/src/Analyzers/tests/analyze_columns.cpp
@ -30,7 +30,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    Context context = Context::createGlobal();

--- a/dbms/src/Analyzers/tests/analyze_lambdas.cpp
+++ b/dbms/src/Analyzers/tests/analyze_lambdas.cpp
@ -23,7 +23,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    AnalyzeLambdas analyzer;
    analyzer.process(ast);
--- a/dbms/src/Analyzers/tests/analyze_result_of_query.cpp
+++ b/dbms/src/Analyzers/tests/analyze_result_of_query.cpp
@ -26,7 +26,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    Context context = Context::createGlobal();

--- a/dbms/src/Analyzers/tests/collect_aliases.cpp
+++ b/dbms/src/Analyzers/tests/collect_aliases.cpp
@ -21,7 +21,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    CollectAliases collector;
    collector.process(ast);
--- a/dbms/src/Analyzers/tests/collect_tables.cpp
+++ b/dbms/src/Analyzers/tests/collect_tables.cpp
@ -27,7 +27,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    Context context = Context::createGlobal();

--- a/dbms/src/Analyzers/tests/optimize_group_order_limit_by.cpp
+++ b/dbms/src/Analyzers/tests/optimize_group_order_limit_by.cpp
@ -33,7 +33,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    Context context = Context::createGlobal();

--- a/dbms/src/Analyzers/tests/translate_positional_arguments.cpp
+++ b/dbms/src/Analyzers/tests/translate_positional_arguments.cpp
@ -21,7 +21,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    TranslatePositionalArguments translator;
    translator.process(ast);
--- a/dbms/src/Analyzers/tests/type_and_constant_inference.cpp
+++ b/dbms/src/Analyzers/tests/type_and_constant_inference.cpp
@ -38,7 +38,7 @@ try
    readStringUntilEOF(query, in);

    ParserSelectQuery parser;
-    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query");
+    ASTPtr ast = parseQuery(parser, query.data(), query.data() + query.size(), "query", 0);

    Context context = Context::createGlobal();

--- a/dbms/src/Client/Connection.cpp
+++ b/dbms/src/Client/Connection.cpp
@ -12,6 +12,7 @@
 #include <DataStreams/NativeBlockInputStream.h>
 #include <DataStreams/NativeBlockOutputStream.h>
 #include <Client/Connection.h>
+#include <Client/TimeoutSetter.h>
 #include <Common/ClickHouseRevision.h>
 #include <Common/Exception.h>
 #include <Common/NetException.h>
@ -50,14 +51,15 @@ void Connection::connect()
        if (connected)
            disconnect();

-        LOG_TRACE(log_wrapper.get(), "Connecting. Database: " << (default_database.empty() ? "(not specified)" : default_database) << ". User: " << user);
+        LOG_TRACE(log_wrapper.get(), "Connecting. Database: " << (default_database.empty() ? "(not specified)" : default_database) << ". User: " << user
+        << (static_cast<bool>(secure) ? ". Secure" : "") << (static_cast<bool>(compression) ? "" : ". Uncompressed") );

-        if (static_cast<bool>(encryption))
+        if (static_cast<bool>(secure))
        {
 #if Poco_NetSSL_FOUND
            socket = std::make_unique<Poco::Net::SecureStreamSocket>();
 #else
-            throw Exception{"tcp_ssl protocol is disabled because poco library was built without NetSSL support.", ErrorCodes::SUPPORT_IS_DISABLED};
+            throw Exception{"tcp_secure protocol is disabled because poco library was built without NetSSL support.", ErrorCodes::SUPPORT_IS_DISABLED};
 #endif
        }
        else
@ -148,6 +150,10 @@ void Connection::receiveHello()
        {
            readStringBinary(server_timezone, *in);
        }
+        if (server_revision >= DBMS_MIN_REVISION_WITH_SERVER_DISPLAY_NAME)
+        {
+            readStringBinary(server_display_name, *in);
+        }
    }
    else if (packet_type == Protocol::Server::Exception)
        receiveException()->rethrow();
@ -203,6 +209,14 @@ const String & Connection::getServerTimezone()
    return server_timezone;
 }

+const String & Connection::getServerDisplayName()
+{
+    if (!connected)
+        connect();
+
+    return server_display_name;
+}
+
 void Connection::forceConnected()
 {
    if (!connected)
@ -216,37 +230,11 @@ void Connection::forceConnected()
    }
 }

-struct TimeoutSetter
-{
-    TimeoutSetter(Poco::Net::StreamSocket & socket_, const Poco::Timespan & timeout_)
-        : socket(socket_), timeout(timeout_)
-    {
-        old_send_timeout = socket.getSendTimeout();
-        old_receive_timeout = socket.getReceiveTimeout();
-
-        if (old_send_timeout > timeout)
-            socket.setSendTimeout(timeout);
-        if (old_receive_timeout > timeout)
-            socket.setReceiveTimeout(timeout);
-    }
-
-    ~TimeoutSetter()
-    {
-        socket.setSendTimeout(old_send_timeout);
-        socket.setReceiveTimeout(old_receive_timeout);
-    }
-
-    Poco::Net::StreamSocket & socket;
-    Poco::Timespan timeout;
-    Poco::Timespan old_send_timeout;
-    Poco::Timespan old_receive_timeout;
-};
-
 bool Connection::ping()
 {
    // LOG_TRACE(log_wrapper.get(), "Ping");

-    TimeoutSetter timeout_setter(*socket, sync_request_timeout);
+    TimeoutSetter timeout_setter(*socket, sync_request_timeout, true);
    try
    {
        UInt64 pong = 0;
@ -286,7 +274,7 @@ TablesStatusResponse Connection::getTablesStatus(const TablesStatusRequest & req
    if (!connected)
        connect();

-    TimeoutSetter timeout_setter(*socket, sync_request_timeout);
+    TimeoutSetter timeout_setter(*socket, sync_request_timeout, true);

    writeVarUInt(Protocol::Client::TablesStatusRequest, *out);
    request.write(*out, server_revision);
--- a/dbms/src/Client/Connection.h
+++ b/dbms/src/Client/Connection.h
@ -58,14 +58,14 @@ public:
        const ConnectionTimeouts & timeouts_,
        const String & client_name_ = "client",
        Protocol::Compression compression_ = Protocol::Compression::Enable,
-        Protocol::Encryption encryption_ = Protocol::Encryption::Disable,
+        Protocol::Secure secure_ = Protocol::Secure::Disable,
        Poco::Timespan sync_request_timeout_ = Poco::Timespan(DBMS_DEFAULT_SYNC_REQUEST_TIMEOUT_SEC, 0))
        :
        host(host_), port(port_), default_database(default_database_),
        user(user_), password(password_), resolved_address(host, port),
        client_name(client_name_),
        compression(compression_),
-        encryption(encryption_),
+        secure(secure_),
        timeouts(timeouts_),
        sync_request_timeout(sync_request_timeout_),
        log_wrapper(*this)
@ -84,7 +84,7 @@ public:
        const ConnectionTimeouts & timeouts_,
        const String & client_name_ = "client",
        Protocol::Compression compression_ = Protocol::Compression::Enable,
-        Protocol::Encryption encryption_ = Protocol::Encryption::Disable,
+        Protocol::Secure secure_ = Protocol::Secure::Disable,
        Poco::Timespan sync_request_timeout_ = Poco::Timespan(DBMS_DEFAULT_SYNC_REQUEST_TIMEOUT_SEC, 0))
        :
        host(host_), port(port_),
@ -93,7 +93,7 @@ public:
        resolved_address(resolved_address_),
        client_name(client_name_),
        compression(compression_),
-        encryption(encryption_),
+        secure(secure_),
        timeouts(timeouts_),
        sync_request_timeout(sync_request_timeout_),
        log_wrapper(*this)
@ -134,6 +134,7 @@ public:
    void getServerVersion(String & name, UInt64 & version_major, UInt64 & version_minor, UInt64 & revision);

    const String & getServerTimezone();
+    const String & getServerDisplayName();

    /// For log and exception messages.
    const String & getDescription() const;
@ -213,6 +214,7 @@ private:
    UInt64 server_version_minor = 0;
    UInt64 server_revision = 0;
    String server_timezone;
+    String server_display_name;

    std::unique_ptr<Poco::Net::StreamSocket> socket;
    std::shared_ptr<ReadBuffer> in;
@ -220,7 +222,7 @@ private:

    String query_id;
    Protocol::Compression compression;        /// Enable data compression for communication.
-    Protocol::Encryption encryption;             /// Enable data encryption for communication.
+    Protocol::Secure secure;             /// Enable data encryption for communication.

    /// What compression settings to use while sending data for INSERT queries and external tables.
    CompressionSettings compression_settings;
--- a/dbms/src/Client/ConnectionPool.h
+++ b/dbms/src/Client/ConnectionPool.h
@ -51,11 +51,12 @@ public:
            const ConnectionTimeouts & timeouts,
            const String & client_name_ = "client",
            Protocol::Compression compression_ = Protocol::Compression::Enable,
-            Protocol::Encryption encryption_ = Protocol::Encryption::Disable)
+            Protocol::Secure secure_ = Protocol::Secure::Disable)
       : Base(max_connections_, &Logger::get("ConnectionPool (" + host_ + ":" + toString(port_) + ")")),
        host(host_), port(port_), default_database(default_database_),
        user(user_), password(password_), resolved_address(host_, port_),
-        client_name(client_name_), compression(compression_), encryption(encryption_),
+        client_name(client_name_), compression(compression_),
+        secure{secure_},
        timeouts(timeouts)
    {
    }
@ -67,11 +68,12 @@ public:
            const ConnectionTimeouts & timeouts,
            const String & client_name_ = "client",
            Protocol::Compression compression_ = Protocol::Compression::Enable,
-            Protocol::Encryption encryption_ = Protocol::Encryption::Disable)
+            Protocol::Secure secure_ = Protocol::Secure::Disable)
        : Base(max_connections_, &Logger::get("ConnectionPool (" + host_ + ":" + toString(port_) + ")")),
        host(host_), port(port_), default_database(default_database_),
        user(user_), password(password_), resolved_address(resolved_address_),
-        client_name(client_name_), compression(compression_), encryption(encryption_),
+        client_name(client_name_), compression(compression_),
+        secure{secure_},
        timeouts(timeouts)
    {
    }
@ -102,7 +104,7 @@ protected:
        return std::make_shared<Connection>(
            host, port, resolved_address,
            default_database, user, password, timeouts,
-            client_name, compression, encryption);
+            client_name, compression, secure);
    }

 private:
@ -119,7 +121,7 @@ private:

    String client_name;
    Protocol::Compression compression;        /// Whether to compress data when interacting with the server.
-    Protocol::Encryption encryption;          /// Whether to encrypt data when interacting with the server.
+    Protocol::Secure secure;          /// Whether to encrypt data when interacting with the server.

    ConnectionTimeouts timeouts;
 };
--- a/dbms/src/Client/MultiplexedConnections.cpp
+++ b/dbms/src/Client/MultiplexedConnections.cpp
@ -280,8 +280,7 @@ MultiplexedConnections::ReplicaState & MultiplexedConnections::getReplicaForRead
            read_list.push_back(*connection->socket);
    }

-    /// If no data was found, then we check if there are any connections
-    /// ready for reading.
+    /// If no data was found, then we check if there are any connections ready for reading.
    if (read_list.empty())
    {
        Poco::Net::Socket::SocketList write_list;
@ -300,6 +299,9 @@ MultiplexedConnections::ReplicaState & MultiplexedConnections::getReplicaForRead
            throw Exception("Timeout exceeded while reading from " + dumpAddressesUnlocked(), ErrorCodes::TIMEOUT_EXCEEDED);
    }

+    /// TODO Absolutely wrong code: read_list could be empty; rand() is not thread safe and has low quality; motivation of rand is unclear.
+    /// This code path is disabled by default.
+
    auto & socket = read_list[rand() % read_list.size()];
    if (fd_to_replica_state_idx.empty())
    {
--- a/dbms/src/Client/TimeoutSetter.h
+++ b/dbms/src/Client/TimeoutSetter.h
@ -0,0 +1,47 @@
+#pragma once
+
+#include <Poco/Timespan.h>
+#include <Poco/Net/StreamSocket.h>
+
+
+namespace DB
+{
+
+/// Temporarily overrides socket send/recieve timeouts and reset them back into destructor
+/// If "limit_max_timeout" is true, timeouts could be only decreased (maxed by previous value).
+struct TimeoutSetter
+{
+    TimeoutSetter(Poco::Net::StreamSocket & socket_, const Poco::Timespan & send_timeout_, const Poco::Timespan & recieve_timeout_,
+                  bool limit_max_timeout = false)
+        : socket(socket_), send_timeout(send_timeout_), recieve_timeout(recieve_timeout_)
+    {
+        old_send_timeout = socket.getSendTimeout();
+        old_receive_timeout = socket.getReceiveTimeout();
+
+        if (!limit_max_timeout || old_send_timeout > send_timeout)
+            socket.setSendTimeout(send_timeout);
+
+        if (!limit_max_timeout || old_receive_timeout > recieve_timeout)
+            socket.setReceiveTimeout(recieve_timeout);
+    }
+
+    TimeoutSetter(Poco::Net::StreamSocket & socket_, const Poco::Timespan & timeout_, bool limit_max_timeout = false)
+        : TimeoutSetter(socket_, timeout_, timeout_, limit_max_timeout) {}
+
+    ~TimeoutSetter()
+    {
+        socket.setSendTimeout(old_send_timeout);
+        socket.setReceiveTimeout(old_receive_timeout);
+    }
+
+    Poco::Net::StreamSocket & socket;
+
+    Poco::Timespan send_timeout;
+    Poco::Timespan recieve_timeout;
+
+    Poco::Timespan old_send_timeout;
+    Poco::Timespan old_receive_timeout;
+};
+
+
+}
--- a/dbms/src/Columns/ColumnAggregateFunction.cpp
+++ b/dbms/src/Columns/ColumnAggregateFunction.cpp
@ -113,7 +113,7 @@ void ColumnAggregateFunction::insertRangeFrom(const IColumn & from, size_t start
 }


-MutableColumnPtr ColumnAggregateFunction::filter(const Filter & filter, ssize_t result_size_hint) const
+ColumnPtr ColumnAggregateFunction::filter(const Filter & filter, ssize_t result_size_hint) const
 {
    size_t size = getData().size();
    if (size != filter.size())
@ -140,7 +140,7 @@ MutableColumnPtr ColumnAggregateFunction::filter(const Filter & filter, ssize_t
 }


-MutableColumnPtr ColumnAggregateFunction::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnAggregateFunction::permute(const Permutation & perm, size_t limit) const
 {
    size_t size = getData().size();

@ -325,7 +325,7 @@ void ColumnAggregateFunction::popBack(size_t n)
    data.resize_assume_reserved(new_size);
 }

-MutableColumnPtr ColumnAggregateFunction::replicate(const IColumn::Offsets & offsets) const
+ColumnPtr ColumnAggregateFunction::replicate(const IColumn::Offsets & offsets) const
 {
    size_t size = data.size();
    if (size != offsets.size())
--- a/dbms/src/Columns/ColumnAggregateFunction.h
+++ b/dbms/src/Columns/ColumnAggregateFunction.h
@ -152,11 +152,11 @@ public:

    void popBack(size_t n) override;

-    MutableColumnPtr filter(const Filter & filter, ssize_t result_size_hint) const override;
+    ColumnPtr filter(const Filter & filter, ssize_t result_size_hint) const override;

-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;

-    MutableColumnPtr replicate(const Offsets & offsets) const override;
+    ColumnPtr replicate(const Offsets & offsets) const override;

    MutableColumns scatter(ColumnIndex num_columns, const Selector & selector) const override;

--- a/dbms/src/Columns/ColumnArray.cpp
+++ b/dbms/src/Columns/ColumnArray.cpp
@ -31,10 +31,10 @@ namespace ErrorCodes
 }


-ColumnArray::ColumnArray(const ColumnPtr & nested_column, const ColumnPtr & offsets_column)
-    : data(nested_column), offsets(offsets_column)
+ColumnArray::ColumnArray(MutableColumnPtr && nested_column, MutableColumnPtr && offsets_column)
+    : data(std::move(nested_column)), offsets(std::move(offsets_column))
 {
-    if (!typeid_cast<const ColumnOffsets *>(offsets_column.get()))
+    if (!typeid_cast<const ColumnOffsets *>(offsets.get()))
        throw Exception("offsets_column must be a ColumnUInt64", ErrorCodes::ILLEGAL_COLUMN);

    /** NOTE
@ -43,8 +43,8 @@ ColumnArray::ColumnArray(const ColumnPtr & nested_column, const ColumnPtr & offs
      */
 }

-ColumnArray::ColumnArray(const ColumnPtr & nested_column)
-    : data(nested_column)
+ColumnArray::ColumnArray(MutableColumnPtr && nested_column)
+    : data(std::move(nested_column))
 {
    if (!data->empty())
        throw Exception("Not empty data passed to ColumnArray, but no offsets passed", ErrorCodes::ILLEGAL_COLUMN);
@ -317,7 +317,7 @@ bool ColumnArray::hasEqualOffsets(const ColumnArray & other) const
 }


-MutableColumnPtr ColumnArray::convertToFullColumnIfConst() const
+ColumnPtr ColumnArray::convertToFullColumnIfConst() const
 {
    ColumnPtr new_data;

@ -391,7 +391,7 @@ void ColumnArray::insertRangeFrom(const IColumn & src, size_t start, size_t leng
 }


-MutableColumnPtr ColumnArray::filter(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnArray::filter(const Filter & filt, ssize_t result_size_hint) const
 {
    if (typeid_cast<const ColumnUInt8 *>(data.get()))      return filterNumber<UInt8>(filt, result_size_hint);
    if (typeid_cast<const ColumnUInt16 *>(data.get()))     return filterNumber<UInt16>(filt, result_size_hint);
@ -410,7 +410,7 @@ MutableColumnPtr ColumnArray::filter(const Filter & filt, ssize_t result_size_hi
 }

 template <typename T>
-MutableColumnPtr ColumnArray::filterNumber(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnArray::filterNumber(const Filter & filt, ssize_t result_size_hint) const
 {
    if (getOffsets().size() == 0)
        return ColumnArray::create(data);
@ -424,7 +424,7 @@ MutableColumnPtr ColumnArray::filterNumber(const Filter & filt, ssize_t result_s
    return std::move(res);
 }

-MutableColumnPtr ColumnArray::filterString(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnArray::filterString(const Filter & filt, ssize_t result_size_hint) const
 {
    size_t col_size = getOffsets().size();
    if (col_size != filt.size())
@ -492,7 +492,7 @@ MutableColumnPtr ColumnArray::filterString(const Filter & filt, ssize_t result_s
    return std::move(res);
 }

-MutableColumnPtr ColumnArray::filterGeneric(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnArray::filterGeneric(const Filter & filt, ssize_t result_size_hint) const
 {
    size_t size = getOffsets().size();
    if (size != filt.size())
@ -537,7 +537,7 @@ MutableColumnPtr ColumnArray::filterGeneric(const Filter & filt, ssize_t result_
    return std::move(res);
 }

-MutableColumnPtr ColumnArray::filterNullable(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnArray::filterNullable(const Filter & filt, ssize_t result_size_hint) const
 {
    if (getOffsets().size() == 0)
        return ColumnArray::create(data);
@ -560,7 +560,7 @@ MutableColumnPtr ColumnArray::filterNullable(const Filter & filt, ssize_t result
        filtered_offsets);
 }

-MutableColumnPtr ColumnArray::filterTuple(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnArray::filterTuple(const Filter & filt, ssize_t result_size_hint) const
 {
    if (getOffsets().size() == 0)
        return ColumnArray::create(data);
@ -576,7 +576,8 @@ MutableColumnPtr ColumnArray::filterTuple(const Filter & filt, ssize_t result_si

    Columns temporary_arrays(tuple_size);
    for (size_t i = 0; i < tuple_size; ++i)
-        temporary_arrays[i] = ColumnArray(tuple.getColumns()[i], getOffsetsPtr()).filter(filt, result_size_hint);
+        temporary_arrays[i] = ColumnArray(tuple.getColumns()[i]->assumeMutable(), getOffsetsPtr()->assumeMutable())
+                .filter(filt, result_size_hint);

    Columns tuple_columns(tuple_size);
    for (size_t i = 0; i < tuple_size; ++i)
@ -588,7 +589,7 @@ MutableColumnPtr ColumnArray::filterTuple(const Filter & filt, ssize_t result_si
 }


-MutableColumnPtr ColumnArray::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnArray::permute(const Permutation & perm, size_t limit) const
 {
    size_t size = getOffsets().size();

@ -652,7 +653,7 @@ void ColumnArray::getPermutation(bool reverse, size_t limit, int nan_direction_h
 }


-MutableColumnPtr ColumnArray::replicate(const Offsets & replicate_offsets) const
+ColumnPtr ColumnArray::replicate(const Offsets & replicate_offsets) const
 {
    if (typeid_cast<const ColumnUInt8 *>(data.get()))     return replicateNumber<UInt8>(replicate_offsets);
    if (typeid_cast<const ColumnUInt16 *>(data.get()))    return replicateNumber<UInt16>(replicate_offsets);
@ -673,7 +674,7 @@ MutableColumnPtr ColumnArray::replicate(const Offsets & replicate_offsets) const


 template <typename T>
-MutableColumnPtr ColumnArray::replicateNumber(const Offsets & replicate_offsets) const
+ColumnPtr ColumnArray::replicateNumber(const Offsets & replicate_offsets) const
 {
    size_t col_size = size();
    if (col_size != replicate_offsets.size())
@ -721,7 +722,7 @@ MutableColumnPtr ColumnArray::replicateNumber(const Offsets & replicate_offsets)
 }


-MutableColumnPtr ColumnArray::replicateString(const Offsets & replicate_offsets) const
+ColumnPtr ColumnArray::replicateString(const Offsets & replicate_offsets) const
 {
    size_t col_size = size();
    if (col_size != replicate_offsets.size())
@ -796,7 +797,7 @@ MutableColumnPtr ColumnArray::replicateString(const Offsets & replicate_offsets)
 }


-MutableColumnPtr ColumnArray::replicateConst(const Offsets & replicate_offsets) const
+ColumnPtr ColumnArray::replicateConst(const Offsets & replicate_offsets) const
 {
    size_t col_size = size();
    if (col_size != replicate_offsets.size())
@ -834,7 +835,7 @@ MutableColumnPtr ColumnArray::replicateConst(const Offsets & replicate_offsets)
 }


-MutableColumnPtr ColumnArray::replicateGeneric(const Offsets & replicate_offsets) const
+ColumnPtr ColumnArray::replicateGeneric(const Offsets & replicate_offsets) const
 {
    size_t col_size = size();
    if (col_size != replicate_offsets.size())
@ -860,25 +861,27 @@ MutableColumnPtr ColumnArray::replicateGeneric(const Offsets & replicate_offsets
 }


-MutableColumnPtr ColumnArray::replicateNullable(const Offsets & replicate_offsets) const
+ColumnPtr ColumnArray::replicateNullable(const Offsets & replicate_offsets) const
 {
    const ColumnNullable & nullable = static_cast<const ColumnNullable &>(*data);

    /// Make temporary arrays for each components of Nullable. Then replicate them independently and collect back to result.
    /// NOTE Offsets are calculated twice and it is redundant.

-    auto array_of_nested = ColumnArray(nullable.getNestedColumnPtr(), getOffsetsPtr()).replicate(replicate_offsets);
-    auto array_of_null_map = ColumnArray(nullable.getNullMapColumnPtr(), getOffsetsPtr()).replicate(replicate_offsets);
+    auto array_of_nested = ColumnArray(nullable.getNestedColumnPtr()->assumeMutable(), getOffsetsPtr()->assumeMutable())
+            .replicate(replicate_offsets);
+    auto array_of_null_map = ColumnArray(nullable.getNullMapColumnPtr()->assumeMutable(), getOffsetsPtr()->assumeMutable())
+            .replicate(replicate_offsets);

    return ColumnArray::create(
        ColumnNullable::create(
-            static_cast<ColumnArray &>(*array_of_nested).getDataPtr(),
-            static_cast<ColumnArray &>(*array_of_null_map).getDataPtr()),
-        static_cast<ColumnArray &>(*array_of_nested).getOffsetsPtr());
+            static_cast<const ColumnArray &>(*array_of_nested).getDataPtr(),
+            static_cast<const ColumnArray &>(*array_of_null_map).getDataPtr()),
+        static_cast<const ColumnArray &>(*array_of_nested).getOffsetsPtr());
 }


-MutableColumnPtr ColumnArray::replicateTuple(const Offsets & replicate_offsets) const
+ColumnPtr ColumnArray::replicateTuple(const Offsets & replicate_offsets) const
 {
    const ColumnTuple & tuple = static_cast<const ColumnTuple &>(*data);

@ -891,7 +894,8 @@ MutableColumnPtr ColumnArray::replicateTuple(const Offsets & replicate_offsets)

    Columns temporary_arrays(tuple_size);
    for (size_t i = 0; i < tuple_size; ++i)
-        temporary_arrays[i] = ColumnArray(tuple.getColumns()[i], getOffsetsPtr()).replicate(replicate_offsets);
+        temporary_arrays[i] = ColumnArray(tuple.getColumns()[i]->assumeMutable(), getOffsetsPtr()->assumeMutable())
+                .replicate(replicate_offsets);

    Columns tuple_columns(tuple_size);
    for (size_t i = 0; i < tuple_size; ++i)
--- a/dbms/src/Columns/ColumnArray.h
+++ b/dbms/src/Columns/ColumnArray.h
@ -24,14 +24,32 @@ private:
    friend class COWPtrHelper<IColumn, ColumnArray>;

    /** Create an array column with specified values and offsets. */
-    ColumnArray(const ColumnPtr & nested_column, const ColumnPtr & offsets_column);
+    ColumnArray(MutableColumnPtr && nested_column, MutableColumnPtr && offsets_column);

    /** Create an empty column of arrays with the type of values as in the column `nested_column` */
-    ColumnArray(const ColumnPtr & nested_column);
+    explicit ColumnArray(MutableColumnPtr && nested_column);

    ColumnArray(const ColumnArray &) = default;

 public:
+    /** Create immutable column using immutable arguments. This arguments may be shared with other columns.
+      * Use IColumn::mutate in order to make mutable column and mutate shared nested columns.
+      */
+    using Base = COWPtrHelper<IColumn, ColumnArray>;
+
+    static Ptr create(const ColumnPtr & nested_column, const ColumnPtr & offsets_column)
+    {
+        return ColumnArray::create(nested_column->assumeMutable(), offsets_column->assumeMutable());
+    }
+
+    static Ptr create(const ColumnPtr & nested_column)
+    {
+        return ColumnArray::create(nested_column->assumeMutable());
+    }
+
+    template <typename ... Args, typename = typename std::enable_if<IsMutableColumns<Args ...>::value>::type>
+    static MutablePtr create(Args &&... args) { return Base::create(std::forward<Args>(args)...); }
+
    /** On the index i there is an offset to the beginning of the i + 1 -th element. */
    using ColumnOffsets = ColumnVector<Offset>;

@ -51,15 +69,15 @@ public:
    void insertFrom(const IColumn & src_, size_t n) override;
    void insertDefault() override;
    void popBack(size_t n) override;
-    MutableColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;
    int compareAt(size_t n, size_t m, const IColumn & rhs_, int nan_direction_hint) const override;
    void getPermutation(bool reverse, size_t limit, int nan_direction_hint, Permutation & res) const override;
    void reserve(size_t n) override;
    size_t byteSize() const override;
    size_t allocatedBytes() const override;
-    MutableColumnPtr replicate(const Offsets & replicate_offsets) const override;
-    MutableColumnPtr convertToFullColumnIfConst() const override;
+    ColumnPtr replicate(const Offsets & replicate_offsets) const override;
+    ColumnPtr convertToFullColumnIfConst() const override;
    void getExtremes(Field & min, Field & max) const override;

    bool hasEqualOffsets(const ColumnArray & other) const;
@ -110,33 +128,33 @@ private:

    /// Multiply values if the nested column is ColumnVector<T>.
    template <typename T>
-    MutableColumnPtr replicateNumber(const Offsets & replicate_offsets) const;
+    ColumnPtr replicateNumber(const Offsets & replicate_offsets) const;

    /// Multiply the values if the nested column is ColumnString. The code is too complicated.
-    MutableColumnPtr replicateString(const Offsets & replicate_offsets) const;
+    ColumnPtr replicateString(const Offsets & replicate_offsets) const;

    /** Non-constant arrays of constant values are quite rare.
      * Most functions can not work with them, and does not create such columns as a result.
      * An exception is the function `replicate`(see FunctionsMiscellaneous.h), which has service meaning for the implementation of lambda functions.
      * Only for its sake is the implementation of the `replicate` method for ColumnArray(ColumnConst).
      */
-    MutableColumnPtr replicateConst(const Offsets & replicate_offsets) const;
+    ColumnPtr replicateConst(const Offsets & replicate_offsets) const;

    /** The following is done by simply replicating of nested columns.
      */
-    MutableColumnPtr replicateTuple(const Offsets & replicate_offsets) const;
-    MutableColumnPtr replicateNullable(const Offsets & replicate_offsets) const;
-    MutableColumnPtr replicateGeneric(const Offsets & replicate_offsets) const;
+    ColumnPtr replicateTuple(const Offsets & replicate_offsets) const;
+    ColumnPtr replicateNullable(const Offsets & replicate_offsets) const;
+    ColumnPtr replicateGeneric(const Offsets & replicate_offsets) const;


    /// Specializations for the filter function.
    template <typename T>
-    MutableColumnPtr filterNumber(const Filter & filt, ssize_t result_size_hint) const;
+    ColumnPtr filterNumber(const Filter & filt, ssize_t result_size_hint) const;

-    MutableColumnPtr filterString(const Filter & filt, ssize_t result_size_hint) const;
-    MutableColumnPtr filterTuple(const Filter & filt, ssize_t result_size_hint) const;
-    MutableColumnPtr filterNullable(const Filter & filt, ssize_t result_size_hint) const;
-    MutableColumnPtr filterGeneric(const Filter & filt, ssize_t result_size_hint) const;
+    ColumnPtr filterString(const Filter & filt, ssize_t result_size_hint) const;
+    ColumnPtr filterTuple(const Filter & filt, ssize_t result_size_hint) const;
+    ColumnPtr filterNullable(const Filter & filt, ssize_t result_size_hint) const;
+    ColumnPtr filterGeneric(const Filter & filt, ssize_t result_size_hint) const;
 };


--- a/dbms/src/Columns/ColumnConst.cpp
+++ b/dbms/src/Columns/ColumnConst.cpp
@ -25,12 +25,12 @@ ColumnConst::ColumnConst(const ColumnPtr & data_, size_t s)
            ErrorCodes::SIZES_OF_COLUMNS_DOESNT_MATCH);
 }

-MutableColumnPtr ColumnConst::convertToFullColumn() const
+ColumnPtr ColumnConst::convertToFullColumn() const
 {
    return data->replicate(Offsets(1, s));
 }

-MutableColumnPtr ColumnConst::filter(const Filter & filt, ssize_t /*result_size_hint*/) const
+ColumnPtr ColumnConst::filter(const Filter & filt, ssize_t /*result_size_hint*/) const
 {
    if (s != filt.size())
        throw Exception("Size of filter (" + toString(filt.size()) + ") doesn't match size of column (" + toString(s) + ")",
@ -39,7 +39,7 @@ MutableColumnPtr ColumnConst::filter(const Filter & filt, ssize_t /*result_size_
    return ColumnConst::create(data, countBytesInFilter(filt));
 }

-MutableColumnPtr ColumnConst::replicate(const Offsets & offsets) const
+ColumnPtr ColumnConst::replicate(const Offsets & offsets) const
 {
    if (s != offsets.size())
        throw Exception("Size of offsets (" + toString(offsets.size()) + ") doesn't match size of column (" + toString(s) + ")",
@ -49,7 +49,7 @@ MutableColumnPtr ColumnConst::replicate(const Offsets & offsets) const
    return ColumnConst::create(data, replicated_size);
 }

-MutableColumnPtr ColumnConst::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnConst::permute(const Permutation & perm, size_t limit) const
 {
    if (limit == 0)
        limit = s;
--- a/dbms/src/Columns/ColumnConst.h
+++ b/dbms/src/Columns/ColumnConst.h
@ -29,9 +29,9 @@ private:
    ColumnConst(const ColumnConst & src) = default;

 public:
-    MutableColumnPtr convertToFullColumn() const;
+    ColumnPtr convertToFullColumn() const;

-    MutableColumnPtr convertToFullColumnIfConst() const override
+    ColumnPtr convertToFullColumnIfConst() const override
    {
        return convertToFullColumn();
    }
@ -145,9 +145,9 @@ public:
        data->updateHashWithValue(0, hash);
    }

-    MutableColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
-    MutableColumnPtr replicate(const Offsets & offsets) const override;
-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr replicate(const Offsets & offsets) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;
    void getPermutation(bool reverse, size_t limit, int nan_direction_hint, Permutation & res) const override;

    size_t byteSize() const override
--- a/dbms/src/Columns/ColumnFixedString.cpp
+++ b/dbms/src/Columns/ColumnFixedString.cpp
@ -153,7 +153,7 @@ void ColumnFixedString::insertRangeFrom(const IColumn & src, size_t start, size_
    memcpy(&chars[old_size], &src_concrete.chars[start * n], length * n);
 }

-MutableColumnPtr ColumnFixedString::filter(const IColumn::Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnFixedString::filter(const IColumn::Filter & filt, ssize_t result_size_hint) const
 {
    size_t col_size = size();
    if (col_size != filt.size())
@ -230,7 +230,7 @@ MutableColumnPtr ColumnFixedString::filter(const IColumn::Filter & filt, ssize_t
    return std::move(res);
 }

-MutableColumnPtr ColumnFixedString::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnFixedString::permute(const Permutation & perm, size_t limit) const
 {
    size_t col_size = size();

@ -258,7 +258,7 @@ MutableColumnPtr ColumnFixedString::permute(const Permutation & perm, size_t lim
    return std::move(res);
 }

-MutableColumnPtr ColumnFixedString::replicate(const Offsets & offsets) const
+ColumnPtr ColumnFixedString::replicate(const Offsets & offsets) const
 {
    size_t col_size = size();
    if (col_size != offsets.size())
--- a/dbms/src/Columns/ColumnFixedString.h
+++ b/dbms/src/Columns/ColumnFixedString.h
@ -104,11 +104,11 @@ public:

    void insertRangeFrom(const IColumn & src, size_t start, size_t length) override;

-    MutableColumnPtr filter(const IColumn::Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr filter(const IColumn::Filter & filt, ssize_t result_size_hint) const override;

-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;

-    MutableColumnPtr replicate(const Offsets & offsets) const override;
+    ColumnPtr replicate(const Offsets & offsets) const override;

    MutableColumns scatter(ColumnIndex num_columns, const Selector & selector) const override
    {
--- a/dbms/src/Columns/ColumnFunction.cpp
+++ b/dbms/src/Columns/ColumnFunction.cpp
@ -26,7 +26,7 @@ MutableColumnPtr ColumnFunction::cloneResized(size_t size) const
    return ColumnFunction::create(size, function, capture);
 }

-MutableColumnPtr ColumnFunction::replicate(const Offsets & offsets) const
+ColumnPtr ColumnFunction::replicate(const Offsets & offsets) const
 {
    if (size_ != offsets.size())
        throw Exception("Size of offsets (" + toString(offsets.size()) + ") doesn't match size of column ("
@ -40,7 +40,7 @@ MutableColumnPtr ColumnFunction::replicate(const Offsets & offsets) const
    return ColumnFunction::create(replicated_size, function, capture);
 }

-MutableColumnPtr ColumnFunction::cut(size_t start, size_t length) const
+ColumnPtr ColumnFunction::cut(size_t start, size_t length) const
 {
    ColumnsWithTypeAndName capture = captured_columns;
    for (auto & column : capture)
@ -49,7 +49,7 @@ MutableColumnPtr ColumnFunction::cut(size_t start, size_t length) const
    return ColumnFunction::create(length, function, capture);
 }

-MutableColumnPtr ColumnFunction::filter(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnFunction::filter(const Filter & filt, ssize_t result_size_hint) const
 {
    if (size_ != filt.size())
        throw Exception("Size of filter (" + toString(filt.size()) + ") doesn't match size of column ("
@ -68,7 +68,7 @@ MutableColumnPtr ColumnFunction::filter(const Filter & filt, ssize_t result_size
    return ColumnFunction::create(filtered_size, function, capture);
 }

-MutableColumnPtr ColumnFunction::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnFunction::permute(const Permutation & perm, size_t limit) const
 {
    if (limit == 0)
        limit = size_;
--- a/dbms/src/Columns/ColumnFunction.h
+++ b/dbms/src/Columns/ColumnFunction.h
@ -29,10 +29,10 @@ public:

    size_t size() const override { return size_; }

-    MutableColumnPtr cut(size_t start, size_t length) const override;
-    MutableColumnPtr replicate(const Offsets & offsets) const override;
-    MutableColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr cut(size_t start, size_t length) const override;
+    ColumnPtr replicate(const Offsets & offsets) const override;
+    ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;
    void insertDefault() override;
    void popBack(size_t n) override;
    std::vector<MutableColumnPtr> scatter(IColumn::ColumnIndex num_columns,
--- a/dbms/src/Columns/ColumnNullable.cpp
+++ b/dbms/src/Columns/ColumnNullable.cpp
@ -18,8 +18,8 @@ namespace ErrorCodes
 }


-ColumnNullable::ColumnNullable(const ColumnPtr & nested_column_, const ColumnPtr & null_map_)
-    : nested_column{nested_column_}, null_map{null_map_}
+ColumnNullable::ColumnNullable(MutableColumnPtr && nested_column_, MutableColumnPtr && null_map_)
+    : nested_column(std::move(nested_column_)), null_map(std::move(null_map_))
 {
    /// ColumnNullable cannot have constant nested column. But constant argument could be passed. Materialize it.
    if (ColumnPtr nested_column_materialized = getNestedColumn().convertToFullColumnIfConst())
@ -44,7 +44,7 @@ void ColumnNullable::updateHashWithValue(size_t n, SipHash & hash) const

 MutableColumnPtr ColumnNullable::cloneResized(size_t new_size) const
 {
-    ColumnPtr new_nested_col = getNestedColumn().cloneResized(new_size);
+    MutableColumnPtr new_nested_col = getNestedColumn().cloneResized(new_size);
    auto new_null_map = ColumnUInt8::create();

    if (new_size > 0)
@ -59,7 +59,7 @@ MutableColumnPtr ColumnNullable::cloneResized(size_t new_size) const
            memset(&new_null_map->getData()[count], 1, new_size - count);
    }

-    return ColumnNullable::create(new_nested_col, std::move(new_null_map));
+    return ColumnNullable::create(std::move(new_nested_col), std::move(new_null_map));
 }


@ -152,14 +152,14 @@ void ColumnNullable::popBack(size_t n)
    getNullMapColumn().popBack(n);
 }

-MutableColumnPtr ColumnNullable::filter(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnNullable::filter(const Filter & filt, ssize_t result_size_hint) const
 {
    ColumnPtr filtered_data = getNestedColumn().filter(filt, result_size_hint);
    ColumnPtr filtered_null_map = getNullMapColumn().filter(filt, result_size_hint);
    return ColumnNullable::create(filtered_data, filtered_null_map);
 }

-MutableColumnPtr ColumnNullable::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnNullable::permute(const Permutation & perm, size_t limit) const
 {
    ColumnPtr permuted_data = getNestedColumn().permute(perm, limit);
    ColumnPtr permuted_null_map = getNullMapColumn().permute(perm, limit);
@ -384,7 +384,7 @@ void ColumnNullable::getExtremes(Field & min, Field & max) const
 }


-MutableColumnPtr ColumnNullable::replicate(const Offsets & offsets) const
+ColumnPtr ColumnNullable::replicate(const Offsets & offsets) const
 {
    ColumnPtr replicated_data = getNestedColumn().replicate(offsets);
    ColumnPtr replicated_null_map = getNullMapColumn().replicate(offsets);
--- a/dbms/src/Columns/ColumnNullable.h
+++ b/dbms/src/Columns/ColumnNullable.h
@ -23,10 +23,22 @@ class ColumnNullable final : public COWPtrHelper<IColumn, ColumnNullable>
 private:
    friend class COWPtrHelper<IColumn, ColumnNullable>;

-    ColumnNullable(const ColumnPtr & nested_column_, const ColumnPtr & null_map_);
+    ColumnNullable(MutableColumnPtr && nested_column_, MutableColumnPtr && null_map_);
    ColumnNullable(const ColumnNullable &) = default;

 public:
+    /** Create immutable column using immutable arguments. This arguments may be shared with other columns.
+      * Use IColumn::mutate in order to make mutable column and mutate shared nested columns.
+      */
+    using Base = COWPtrHelper<IColumn, ColumnNullable>;
+    static Ptr create(const ColumnPtr & nested_column_, const ColumnPtr & null_map_)
+    {
+        return ColumnNullable::create(nested_column_->assumeMutable(), null_map_->assumeMutable());
+    }
+
+    template <typename ... Args, typename = typename std::enable_if<IsMutableColumns<Args ...>::value>::type>
+    static MutablePtr create(Args &&... args) { return Base::create(std::forward<Args>(args)...); }
+
    const char * getFamilyName() const override { return "Nullable"; }
    std::string getName() const override { return "Nullable(" + nested_column->getName() + ")"; }
    MutableColumnPtr cloneResized(size_t size) const override;
@ -50,14 +62,14 @@ public:
    }

    void popBack(size_t n) override;
-    MutableColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;
    int compareAt(size_t n, size_t m, const IColumn & rhs_, int null_direction_hint) const override;
    void getPermutation(bool reverse, size_t limit, int null_direction_hint, Permutation & res) const override;
    void reserve(size_t n) override;
    size_t byteSize() const override;
    size_t allocatedBytes() const override;
-    MutableColumnPtr replicate(const Offsets & replicate_offsets) const override;
+    ColumnPtr replicate(const Offsets & replicate_offsets) const override;
    void updateHashWithValue(size_t n, SipHash & hash) const override;
    void getExtremes(Field & min, Field & max) const override;

--- a/dbms/src/Columns/ColumnString.cpp
+++ b/dbms/src/Columns/ColumnString.cpp
@ -97,7 +97,7 @@ void ColumnString::insertRangeFrom(const IColumn & src, size_t start, size_t len
 }


-MutableColumnPtr ColumnString::filter(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnString::filter(const Filter & filt, ssize_t result_size_hint) const
 {
    if (offsets.size() == 0)
        return ColumnString::create();
@ -112,7 +112,7 @@ MutableColumnPtr ColumnString::filter(const Filter & filt, ssize_t result_size_h
 }


-MutableColumnPtr ColumnString::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnString::permute(const Permutation & perm, size_t limit) const
 {
    size_t size = offsets.size();

@ -208,7 +208,7 @@ void ColumnString::getPermutation(bool reverse, size_t limit, int /*nan_directio
 }


-MutableColumnPtr ColumnString::replicate(const Offsets & replicate_offsets) const
+ColumnPtr ColumnString::replicate(const Offsets & replicate_offsets) const
 {
    size_t col_size = size();
    if (col_size != replicate_offsets.size())
--- a/dbms/src/Columns/ColumnString.h
+++ b/dbms/src/Columns/ColumnString.h
@ -89,6 +89,12 @@ public:
        return StringRef(&chars[offsetAt(n)], sizeAt(n));
    }

+/// Suppress gcc 7.3.1 warning: '*((void*)&<anonymous> +8)' may be used uninitialized in this function
+#if !__clang__
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wmaybe-uninitialized"
+#endif
+
    void insert(const Field & x) override
    {
        const String & s = DB::get<const String &>(x);
@ -101,6 +107,10 @@ public:
        offsets.push_back(new_size);
    }

+#if !__clang__
+#pragma GCC diagnostic pop
+#endif
+
        void insertFrom(const IColumn & src_, size_t n) override
    {
        const ColumnString & src = static_cast<const ColumnString &>(src_);
@ -206,9 +216,9 @@ public:

    void insertRangeFrom(const IColumn & src, size_t start, size_t length) override;

-    MutableColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;

-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;

    void insertDefault() override
    {
@ -239,7 +249,7 @@ public:
    /// Sorting with respect of collation.
    void getPermutationWithCollation(const Collator & collator, bool reverse, size_t limit, Permutation & res) const;

-    MutableColumnPtr replicate(const Offsets & replicate_offsets) const override;
+    ColumnPtr replicate(const Offsets & replicate_offsets) const override;

    MutableColumns scatter(ColumnIndex num_columns, const Selector & selector) const override
    {
--- a/dbms/src/Columns/ColumnTuple.cpp
+++ b/dbms/src/Columns/ColumnTuple.cpp
@ -31,21 +31,38 @@ std::string ColumnTuple::getName() const
    return res.str();
 }

-ColumnTuple::ColumnTuple(const Columns & columns) : columns(columns)
+ColumnTuple::ColumnTuple(MutableColumns && mutable_columns)
+{
+    columns.reserve(mutable_columns.size());
+    for (auto & column : mutable_columns)
+    {
+        if (column->isColumnConst())
+            throw Exception{"ColumnTuple cannot have ColumnConst as its element", ErrorCodes::ILLEGAL_COLUMN};
+
+        columns.push_back(std::move(column));
+    }
+}
+
+ColumnTuple::Ptr ColumnTuple::create(const Columns & columns)
 {
    for (const auto & column : columns)
        if (column->isColumnConst())
            throw Exception{"ColumnTuple cannot have ColumnConst as its element", ErrorCodes::ILLEGAL_COLUMN};
+
+    auto column_tuple = ColumnTuple::create(MutableColumns());
+    column_tuple->columns = columns;
+
+    return std::move(column_tuple);
 }

 MutableColumnPtr ColumnTuple::cloneEmpty() const
 {
    const size_t tuple_size = columns.size();
-    Columns new_columns(tuple_size);
+    MutableColumns new_columns(tuple_size);
    for (size_t i = 0; i < tuple_size; ++i)
        new_columns[i] = columns[i]->cloneEmpty();

-    return ColumnTuple::create(new_columns);
+    return ColumnTuple::create(std::move(new_columns));
 }

 Field ColumnTuple::operator[](size_t n) const
@ -140,7 +157,7 @@ void ColumnTuple::insertRangeFrom(const IColumn & src, size_t start, size_t leng
            start, length);
 }

-MutableColumnPtr ColumnTuple::filter(const Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnTuple::filter(const Filter & filt, ssize_t result_size_hint) const
 {
    const size_t tuple_size = columns.size();
    Columns new_columns(tuple_size);
@ -151,7 +168,7 @@ MutableColumnPtr ColumnTuple::filter(const Filter & filt, ssize_t result_size_hi
    return ColumnTuple::create(new_columns);
 }

-MutableColumnPtr ColumnTuple::permute(const Permutation & perm, size_t limit) const
+ColumnPtr ColumnTuple::permute(const Permutation & perm, size_t limit) const
 {
    const size_t tuple_size = columns.size();
    Columns new_columns(tuple_size);
@ -162,7 +179,7 @@ MutableColumnPtr ColumnTuple::permute(const Permutation & perm, size_t limit) co
    return ColumnTuple::create(new_columns);
 }

-MutableColumnPtr ColumnTuple::replicate(const Offsets & offsets) const
+ColumnPtr ColumnTuple::replicate(const Offsets & offsets) const
 {
    const size_t tuple_size = columns.size();
    Columns new_columns(tuple_size);
@ -185,10 +202,10 @@ MutableColumns ColumnTuple::scatter(ColumnIndex num_columns, const Selector & se

    for (size_t scattered_idx = 0; scattered_idx < num_columns; ++scattered_idx)
    {
-        Columns new_columns(tuple_size);
+        MutableColumns new_columns(tuple_size);
        for (size_t tuple_element_idx = 0; tuple_element_idx < tuple_size; ++tuple_element_idx)
            new_columns[tuple_element_idx] = std::move(scattered_tuple_elements[tuple_element_idx][scattered_idx]);
-        res[scattered_idx] = ColumnTuple::create(new_columns);
+        res[scattered_idx] = ColumnTuple::create(std::move(new_columns));
    }

    return res;
--- a/dbms/src/Columns/ColumnTuple.h
+++ b/dbms/src/Columns/ColumnTuple.h
@ -22,10 +22,19 @@ private:
    template <bool positive>
    struct Less;

-    ColumnTuple(const Columns & columns);
+    explicit ColumnTuple(MutableColumns && columns);
    ColumnTuple(const ColumnTuple &) = default;

 public:
+    /** Create immutable column using immutable arguments. This arguments may be shared with other columns.
+      * Use IColumn::mutate in order to make mutable column and mutate shared nested columns.
+      */
+    using Base = COWPtrHelper<IColumn, ColumnTuple>;
+    static Ptr create(const Columns & columns);
+
+    template <typename Arg, typename = typename std::enable_if<std::is_rvalue_reference<Arg &&>::value>::type>
+    static MutablePtr create(Arg && arg) { return Base::create(std::forward<Arg>(arg)); }
+
    std::string getName() const override;
    const char * getFamilyName() const override { return "Tuple"; }

@ -49,9 +58,9 @@ public:
    const char * deserializeAndInsertFromArena(const char * pos) override;
    void updateHashWithValue(size_t n, SipHash & hash) const override;
    void insertRangeFrom(const IColumn & src, size_t start, size_t length) override;
-    MutableColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override;
-    MutableColumnPtr replicate(const Offsets & offsets) const override;
+    ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr replicate(const Offsets & offsets) const override;
    MutableColumns scatter(ColumnIndex num_columns, const Selector & selector) const override;
    void gather(ColumnGathererStream & gatherer_stream) override;
    int compareAt(size_t n, size_t m, const IColumn & rhs, int nan_direction_hint) const override;
--- a/dbms/src/Columns/ColumnVector.cpp
+++ b/dbms/src/Columns/ColumnVector.cpp
@ -146,7 +146,7 @@ void ColumnVector<T>::insertRangeFrom(const IColumn & src, size_t start, size_t
 }

 template <typename T>
-MutableColumnPtr ColumnVector<T>::filter(const IColumn::Filter & filt, ssize_t result_size_hint) const
+ColumnPtr ColumnVector<T>::filter(const IColumn::Filter & filt, ssize_t result_size_hint) const
 {
    size_t size = data.size();
    if (size != filt.size())
@ -210,7 +210,7 @@ MutableColumnPtr ColumnVector<T>::filter(const IColumn::Filter & filt, ssize_t r
 }

 template <typename T>
-MutableColumnPtr ColumnVector<T>::permute(const IColumn::Permutation & perm, size_t limit) const
+ColumnPtr ColumnVector<T>::permute(const IColumn::Permutation & perm, size_t limit) const
 {
    size_t size = data.size();

@ -231,7 +231,7 @@ MutableColumnPtr ColumnVector<T>::permute(const IColumn::Permutation & perm, siz
 }

 template <typename T>
-MutableColumnPtr ColumnVector<T>::replicate(const IColumn::Offsets & offsets) const
+ColumnPtr ColumnVector<T>::replicate(const IColumn::Offsets & offsets) const
 {
    size_t size = data.size();
    if (size != offsets.size())
--- a/dbms/src/Columns/ColumnVector.h
+++ b/dbms/src/Columns/ColumnVector.h
@ -243,11 +243,11 @@ public:

    void insertRangeFrom(const IColumn & src, size_t start, size_t length) override;

-    MutableColumnPtr filter(const IColumn::Filter & filt, ssize_t result_size_hint) const override;
+    ColumnPtr filter(const IColumn::Filter & filt, ssize_t result_size_hint) const override;

-    MutableColumnPtr permute(const IColumn::Permutation & perm, size_t limit) const override;
+    ColumnPtr permute(const IColumn::Permutation & perm, size_t limit) const override;

-    MutableColumnPtr replicate(const IColumn::Offsets & offsets) const override;
+    ColumnPtr replicate(const IColumn::Offsets & offsets) const override;

    void getExtremes(Field & min, Field & max) const override;

--- a/dbms/src/Columns/FilterDescription.cpp
+++ b/dbms/src/Columns/FilterDescription.cpp
@ -57,7 +57,8 @@ FilterDescription::FilterDescription(const IColumn & column)

    if (const ColumnNullable * nullable_column = typeid_cast<const ColumnNullable *>(&column))
    {
-        MutableColumnPtr mutable_holder = nullable_column->getNestedColumn().mutate();
+        ColumnPtr nested_column = nullable_column->getNestedColumnPtr();
+        MutableColumnPtr mutable_holder = (*std::move(nested_column)).mutate();

        ColumnUInt8 * concrete_column = typeid_cast<ColumnUInt8 *>(mutable_holder.get());
        if (!concrete_column)
--- a/dbms/src/Columns/IColumn.h
+++ b/dbms/src/Columns/IColumn.h
@ -45,7 +45,7 @@ public:
    /** If column isn't constant, returns nullptr (or itself).
      * If column is constant, transforms constant to full column (if column type allows such tranform) and return it.
      */
-    virtual MutablePtr convertToFullColumnIfConst() const { return {}; }
+    virtual Ptr convertToFullColumnIfConst() const { return {}; }

    /// Creates empty column with the same type.
    virtual MutablePtr cloneEmpty() const { return cloneResized(0); }
@ -104,11 +104,11 @@ public:

    /// Removes all elements outside of specified range.
    /// Is used in LIMIT operation, for example.
-    virtual MutablePtr cut(size_t start, size_t length) const
+    virtual Ptr cut(size_t start, size_t length) const
    {
        MutablePtr res = cloneEmpty();
        res->insertRangeFrom(*this, start, length);
-        return res;
+        return std::move(res);
    }

    /// Appends new value at the end of column (column's size is increased by 1).
@ -171,12 +171,12 @@ public:
      *  otherwise (i.e. < 0), makes reserve() using size of source column.
      */
    using Filter = PaddedPODArray<UInt8>;
-    virtual MutablePtr filter(const Filter & filt, ssize_t result_size_hint) const = 0;
+    virtual Ptr filter(const Filter & filt, ssize_t result_size_hint) const = 0;

    /// Permutes elements using specified permutation. Is used in sortings.
    /// limit - if it isn't 0, puts only first limit elements in the result.
    using Permutation = PaddedPODArray<size_t>;
-    virtual MutablePtr permute(const Permutation & perm, size_t limit) const = 0;
+    virtual Ptr permute(const Permutation & perm, size_t limit) const = 0;

    /** Compares (*this)[n] and rhs[m].
      * Returns negative number, 0, or positive number (*this)[n] is less, equal, greater than rhs[m] respectively.
@ -205,7 +205,7 @@ public:
      */
    using Offset = UInt64;
    using Offsets = PaddedPODArray<Offset>;
-    virtual MutablePtr replicate(const Offsets & offsets) const = 0;
+    virtual Ptr replicate(const Offsets & offsets) const = 0;

    /** Split column to smaller columns. Each value goes to column index, selected by corresponding element of 'selector'.
      * Selector must contain values from 0 to num_columns - 1.
@ -247,10 +247,10 @@ public:
    virtual void forEachSubcolumn(ColumnCallback) {}


-    MutablePtr mutate() const
+    MutablePtr mutate() const &&
    {
        MutablePtr res = COWPtr<IColumn>::mutate();
-        res->forEachSubcolumn([](Ptr & subcolumn) { subcolumn = subcolumn->mutate(); });
+        res->forEachSubcolumn([](Ptr & subcolumn) { subcolumn = (*std::move(subcolumn)).mutate(); });
        return res;
    }

@ -360,4 +360,16 @@ using MutableColumns = std::vector<MutableColumnPtr>;
 using ColumnRawPtrs = std::vector<const IColumn *>;
 //using MutableColumnRawPtrs = std::vector<IColumn *>;

+template <typename ... Args>
+struct IsMutableColumns;
+
+template <typename Arg, typename ... Args>
+struct IsMutableColumns<Arg, Args ...>
+{
+    static const bool value = std::is_assignable<MutableColumnPtr &&, Arg>::value && IsMutableColumns<Args ...>::value;
+};
+
+template <>
+struct IsMutableColumns<> { static const bool value = true; };
+
 }
--- a/dbms/src/Columns/IColumnDummy.h
+++ b/dbms/src/Columns/IColumnDummy.h
@ -74,12 +74,12 @@ public:
        s += length;
    }

-    MutableColumnPtr filter(const Filter & filt, ssize_t /*result_size_hint*/) const override
+    ColumnPtr filter(const Filter & filt, ssize_t /*result_size_hint*/) const override
    {
        return cloneDummy(countBytesInFilter(filt));
    }

-    MutableColumnPtr permute(const Permutation & perm, size_t limit) const override
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override
    {
        if (s != perm.size())
            throw Exception("Size of permutation doesn't match size of column.", ErrorCodes::SIZES_OF_COLUMNS_DOESNT_MATCH);
@ -94,7 +94,7 @@ public:
            res[i] = i;
    }

-    MutableColumnPtr replicate(const Offsets & offsets) const override
+    ColumnPtr replicate(const Offsets & offsets) const override
    {
        if (s != offsets.size())
            throw Exception("Size of offsets doesn't match size of column.", ErrorCodes::SIZES_OF_COLUMNS_DOESNT_MATCH);
--- a/dbms/src/Common/AIO.h
+++ b/dbms/src/Common/AIO.h
@ -1,6 +1,6 @@
 #pragma once

-#if !(defined(__FreeBSD__) || defined(__APPLE__))
+#if !(defined(__FreeBSD__) || defined(__APPLE__) || defined(_MSC_VER))

 #include <Common/Exception.h>
 #include <common/logger_useful.h>
--- a/dbms/src/Common/COWPtr.h
+++ b/dbms/src/Common/COWPtr.h
@ -68,10 +68,9 @@
  * of this shared state.
  *
  * Caveats:
-  * - after a call to 'mutate' method, you can still have a reference to immutable ptr somewhere
-  *   and it can still become shared. Also it would be better to make 'mutate' method rvalue-qualified.
+  * - after a call to 'mutate' method, you can still have a reference to immutable ptr somewhere.
  * - as 'mutable_ptr' should be unique, it's refcount is redundant - probably it would be better
-  *   to use std::unique_ptr for it, but see above.
+  *   to use std::unique_ptr for it somehow.
  */
 template <typename Derived>
 class COWPtr : public boost::intrusive_ref_counter<Derived>
@ -80,12 +79,22 @@ private:
    Derived * derived() { return static_cast<Derived *>(this); }
    const Derived * derived() const { return static_cast<const Derived *>(this); }

+    template <typename T>
+    class IntrusivePtr : public boost::intrusive_ptr<T>
+    {
+    public:
+        using boost::intrusive_ptr<T>::intrusive_ptr;
+
+        T & operator*() const & { return boost::intrusive_ptr<T>::operator*(); }
+        T && operator*() const && { return const_cast<typename std::remove_const<T>::type &&>(*boost::intrusive_ptr<T>::get()); }
+    };
+
 protected:
    template <typename T>
-    class mutable_ptr : public boost::intrusive_ptr<T>
+    class mutable_ptr : public IntrusivePtr<T>
    {
    private:
-        using Base = boost::intrusive_ptr<T>;
+        using Base = IntrusivePtr<T>;

        template <typename> friend class COWPtr;
        template <typename, typename> friend class COWPtrHelper;
@ -114,10 +123,10 @@ public:

 protected:
    template <typename T>
-    class immutable_ptr : public boost::intrusive_ptr<const T>
+    class immutable_ptr : public IntrusivePtr<const T>
    {
    private:
-        using Base = boost::intrusive_ptr<const T>;
+        using Base = IntrusivePtr<const T>;

        template <typename> friend class COWPtr;
        template <typename, typename> friend class COWPtrHelper;
--- a/dbms/src/Common/Config/ConfigProcessor.cpp
+++ b/dbms/src/Common/Config/ConfigProcessor.cpp
@ -11,6 +11,7 @@
 #include <Poco/DOM/Comment.h>
 #include <Poco/Util/XMLConfiguration.h>
 #include <Common/ZooKeeper/ZooKeeperNodeCache.h>
+#include <Common/ZooKeeper/KeeperException.h>
 #include <Common/StringUtils/StringUtils.h>

 #define PREPROCESSED_SUFFIX "-preprocessed"
@ -276,6 +277,8 @@ void ConfigProcessor::doIncludesRecursive(
    /// Replace the original contents, not add to it.
    bool replace = attributes->getNamedItem("replace");

+    bool included_something = false;
+
    auto process_include = [&](const Node * include_attr, const std::function<const Node * (const std::string &)> & get_node, const char * error_msg)
    {
        std::string name = include_attr->getNodeValue();
@ -316,6 +319,8 @@ void ConfigProcessor::doIncludesRecursive(
            {
                element->setAttributeNode(dynamic_cast<Attr *>(config->importNode(from_attrs->item(i), true)));
            }
+
+            included_something = true;
        }
    };

@ -348,10 +353,13 @@ void ConfigProcessor::doIncludesRecursive(
        }
    }

+    if (included_something)
+        doIncludesRecursive(config, include_from, node, zk_node_cache, contributing_zk_paths);
+    else
+    {
        NodeListPtr children = node->childNodes();
        Node * child = nullptr;
        for (size_t i = 0; (child = children->item(i)); ++i)
-    {
            doIncludesRecursive(config, include_from, child, zk_node_cache, contributing_zk_paths);
    }
 }
@ -375,17 +383,19 @@ ConfigProcessor::Files ConfigProcessor::getConfigMergeFiles(const std::string &
        Poco::File merge_dir(merge_dir_name);
        if (!merge_dir.exists() || !merge_dir.isDirectory())
            continue;
+
        for (Poco::DirectoryIterator it(merge_dir_name); it != Poco::DirectoryIterator(); ++it)
        {
            Poco::File & file = *it;
-            if (file.isFile()
-                && (endsWith(file.path(), ".xml") || endsWith(file.path(), ".conf"))
-                && !startsWith(file.path(), ".")) // skip temporary files
-            {
+            Poco::Path path(file.path());
+            std::string extension = path.getExtension();
+            std::string base_name = path.getBaseName();
+
+            // Skip non-config and temporary files
+            if (file.isFile() && (extension == "xml" || extension == "conf") && !startsWith(base_name, "."))
                files.push_back(file.path());
        }
    }
-    }

    std::sort(files.begin(), files.end());

--- a/dbms/src/Common/Config/ConfigReloader.cpp
+++ b/dbms/src/Common/Config/ConfigReloader.cpp
@ -67,12 +67,15 @@ void ConfigReloader::run()
        catch (...)
        {
            tryLogCurrentException(log, __PRETTY_FUNCTION__);
+            std::this_thread::sleep_for(reload_interval);
        }
    }
 }

 void ConfigReloader::reloadIfNewer(bool force, bool throw_on_error, bool fallback_to_preprocessed)
 {
+    std::lock_guard<std::mutex> lock(reload_mutex);
+
    FilesChangesTracker new_files = getNewFileList();
    if (force || new_files.isDifferOrNewerThan(files))
    {
--- a/dbms/src/Common/Config/ConfigReloader.h
+++ b/dbms/src/Common/Config/ConfigReloader.h
@ -42,6 +42,9 @@ public:
    /// Call this method to run the backround thread.
    void start();

+    /// Reload immediately. For SYSTEM RELOAD CONFIG query.
+    void reload() { reloadIfNewer(/* force */ true, /* throw_on_error */ true, /* fallback_to_preprocessed */ false); }
+
 private:
    void run();

@ -74,6 +77,9 @@ private:

    std::atomic<bool> quit{false};
    std::thread thread;
+
+    /// Locked inside reloadIfNewer.
+    std::mutex reload_mutex;
 };

 }
--- a/dbms/src/Common/CurrentMetrics.cpp
+++ b/dbms/src/Common/CurrentMetrics.cpp
@ -28,7 +28,9 @@
    M(MemoryTrackingForMerges) \
    M(LeaderElection) \
    M(EphemeralNode) \
+    M(ZooKeeperSession) \
    M(ZooKeeperWatch) \
+    M(ZooKeeperRequest) \
    M(DelayedInserts) \
    M(ContextLockWait) \
    M(StorageBufferRows) \
--- a/dbms/src/Common/ErrorCodes.cpp
+++ b/dbms/src/Common/ErrorCodes.cpp
@ -125,7 +125,6 @@ namespace ErrorCodes
    extern const int UNKNOWN_SETTING = 115;
    extern const int THERE_IS_NO_DEFAULT_VALUE = 116;
    extern const int INCORRECT_DATA = 117;
-    extern const int TABLE_METADATA_DOESNT_EXIST = 118;
    extern const int ENGINE_REQUIRED = 119;
    extern const int CANNOT_INSERT_VALUE_OF_DIFFERENT_SIZE_INTO_TUPLE = 120;
    extern const int UNKNOWN_SET_DATA_VARIANT = 121;
@ -372,6 +371,7 @@ namespace ErrorCodes
    extern const int QUERY_WAS_CANCELLED = 394;
    extern const int FUNCTION_THROW_IF_VALUE_IS_NON_ZERO = 395;
    extern const int TOO_MANY_ROWS_OR_BYTES = 396;
+    extern const int QUERY_IS_NOT_SUPPORTED_IN_MATERIALIZED_VIEW = 397;


    extern const int KEEPER_EXCEPTION = 999;
--- a/dbms/src/Common/Exception.cpp
+++ b/dbms/src/Common/Exception.cpp
@ -11,7 +11,7 @@
 #include <IO/ReadBufferFromString.h>

 #include <Common/Exception.h>
-#include <Common/demangle.h>
+#include <common/demangle.h>


 namespace DB
--- a/dbms/src/Common/ExternalTable.h
+++ b/dbms/src/Common/ExternalTable.h
@ -198,7 +198,7 @@ public:

        /// Create table
        NamesAndTypesList columns = sample_block.getNamesAndTypesList();
-        StoragePtr storage = StorageMemory::create(data.second, columns, NamesAndTypesList{}, NamesAndTypesList{}, ColumnDefaults{});
+        StoragePtr storage = StorageMemory::create(data.second, ColumnsDescription{columns});
        storage->startup();
        context.addExternalTable(data.second, storage);
        BlockOutputStreamPtr output = storage->write(ASTPtr(), context.getSettingsRef());
--- a/dbms/src/Common/FieldVisitors.h
+++ b/dbms/src/Common/FieldVisitors.h
@ -3,7 +3,7 @@
 #include <Core/Field.h>
 #include <Core/AccurateComparison.h>
 #include <common/DateLUT.h>
-#include <Common/demangle.h>
+#include <common/demangle.h>


 class SipHash;
--- a/dbms/src/Common/Macros.cpp
+++ b/dbms/src/Common/Macros.cpp
@ -1,6 +1,8 @@
+#include <Poco/Util/AbstractConfiguration.h>
 #include <Common/Macros.h>
 #include <Common/Exception.h>

+
 namespace DB
 {

--- a/dbms/src/Common/Macros.h
+++ b/dbms/src/Common/Macros.h
@ -1,10 +1,18 @@
 #pragma once

 #include <Core/Types.h>
-#include <Poco/Util/AbstractConfiguration.h>
 #include <map>


+namespace Poco
+{
+    namespace Util
+    {
+        class AbstractConfiguration;
+    }
+}
+
+
 namespace DB
 {

@ -21,10 +29,12 @@ public:
      */
    String expand(const String & s, size_t level = 0) const;

-private:
    using MacroMap = std::map<String, String>;
+    const MacroMap getMacroMap() const { return macros; }

+private:
    MacroMap macros;
 };

+
 }
--- a/dbms/src/Common/ProfileEvents.cpp
+++ b/dbms/src/Common/ProfileEvents.cpp
@ -56,14 +56,20 @@
    \
    M(ZooKeeperInit) \
    M(ZooKeeperTransactions) \
-    M(ZooKeeperGetChildren) \
+    M(ZooKeeperList) \
    M(ZooKeeperCreate) \
    M(ZooKeeperRemove) \
    M(ZooKeeperExists) \
    M(ZooKeeperGet) \
    M(ZooKeeperSet) \
    M(ZooKeeperMulti) \
+    M(ZooKeeperCheck) \
+    M(ZooKeeperClose) \
+    M(ZooKeeperWatchResponse) \
    M(ZooKeeperExceptions) \
+    M(ZooKeeperWaitMicroseconds) \
+    M(ZooKeeperBytesSent) \
+    M(ZooKeeperBytesReceived) \
    \
    M(DistributedConnectionFailTry) \
    M(DistributedConnectionMissingTable) \
--- a/dbms/src/Common/ShellCommand.cpp
+++ b/dbms/src/Common/ShellCommand.cpp
@ -1,13 +1,12 @@
 #include <sys/types.h>
 #include <sys/wait.h>
-#include <unistd.h>
 #include <fcntl.h>
 #include <dlfcn.h>
-
 #include <Common/Exception.h>
 #include <Common/ShellCommand.h>
 #include <IO/WriteBufferFromVector.h>
 #include <IO/WriteHelpers.h>
+#include <port/unistd.h>


 namespace DB
--- a/dbms/src/Common/StackTrace.cpp
+++ b/dbms/src/Common/StackTrace.cpp
@ -7,7 +7,7 @@
 #include <sstream>

 #include <Common/StackTrace.h>
-#include <Common/demangle.h>
+#include <common/demangle.h>


 StackTrace::StackTrace()
--- a/dbms/src/Common/Throttler.h
+++ b/dbms/src/Common/Throttler.h
@ -27,6 +27,9 @@ namespace ErrorCodes
 class Throttler
 {
 public:
+    Throttler(size_t max_speed_, const std::shared_ptr<Throttler> & parent = nullptr)
+            : max_speed(max_speed_), limit_exceeded_exception_message(""), parent(parent) {}
+
    Throttler(size_t max_speed_, size_t limit_, const char * limit_exceeded_exception_message_,
              const std::shared_ptr<Throttler> & parent = nullptr)
        : max_speed(max_speed_), limit(limit_), limit_exceeded_exception_message(limit_exceeded_exception_message_), parent(parent) {}
@ -76,6 +79,12 @@ public:
            parent->add(amount);
    }

+    /// Not thread safe
+    void setParent(const std::shared_ptr<Throttler> & parent_)
+    {
+        parent = parent_;
+    }
+
    void reset()
    {
        std::lock_guard lock(mutex);
--- a/dbms/src/Common/UTF8Helpers.h
+++ b/dbms/src/Common/UTF8Helpers.h
@ -3,6 +3,9 @@
 #include <Core/Types.h>
 #include <Common/BitHelpers.h>

+#if __SSE2__
+#include <emmintrin.h>
+#endif

 namespace DB
 {
@ -49,9 +52,29 @@ inline size_t seqLength(const UInt8 first_octet)
 inline size_t countCodePoints(const UInt8 * data, size_t size)
 {
    size_t res = 0;
+    const auto end = data + size;

-    /// TODO SIMD implementation looks quite simple.
-    for (auto end = data + size; data < end; ++data) /// Skip UTF-8 continuation bytes.
+#if __SSE2__
+    const auto bytes_sse = sizeof(__m128i);
+    const auto src_end_sse = (data + size) - (size % bytes_sse);
+
+    const auto align_sse = _mm_set1_epi8(0x40);
+    const auto upper_bound = _mm_set1_epi8(0xBF);
+
+    for (; data < src_end_sse; data += bytes_sse)
+    {
+        const auto chars = _mm_loadu_si128(reinterpret_cast<const __m128i *>(data));
+
+        ///Align to zero for the solve two case
+        const auto align_res = _mm_adds_epu8(chars, align_sse);
+        const auto less_than_and_equals = _mm_cmpeq_epi8(_mm_min_epu8(align_res, upper_bound), align_res);
+
+        res += __builtin_popcount(_mm_movemask_epi8(less_than_and_equals));
+    }
+
+#endif
+
+    for (; data < end; ++data) /// Skip UTF-8 continuation bytes.
        res += (*data <= 0x7F || *data >= 0xC0);

    return res;
--- a/dbms/src/Common/Visitor.h
+++ b/dbms/src/Common/Visitor.h
@ -1,6 +1,6 @@
 #pragma once

-#include <Common/demangle.h>
+#include <common/demangle.h>
 #include <Common/TypeList.h>
 #include <Common/Exception.h>

--- a/dbms/src/Common/ZooKeeper/CMakeLists.txt
+++ b/dbms/src/Common/ZooKeeper/CMakeLists.txt
@ -1,4 +1,3 @@
-
 include(${ClickHouse_SOURCE_DIR}/cmake/dbms_glob_sources.cmake)

 add_headers_and_sources(clickhouse_common_zookeeper .)
--- a/dbms/src/Common/ZooKeeper/Increment.h
+++ b/dbms/src/Common/ZooKeeper/Increment.h
@ -29,11 +29,11 @@ public:
            if (zookeeper->tryGet(path, result_str, &stat))
            {
                result = std::stol(result_str) + 1;
-                success = zookeeper->trySet(path, std::to_string(result), stat.version) == ZOK;
+                success = zookeeper->trySet(path, std::to_string(result), stat.version) == ZooKeeperImpl::ZooKeeper::ZOK;
            }
            else
            {
-                success = zookeeper->tryCreate(path, std::to_string(result), zkutil::CreateMode::Persistent) == ZOK;
+                success = zookeeper->tryCreate(path, std::to_string(result), zkutil::CreateMode::Persistent) == ZooKeeperImpl::ZooKeeper::ZOK;
            }
        }
        while (!success);
--- a/dbms/src/Common/ZooKeeper/KeeperException.h
+++ b/dbms/src/Common/ZooKeeper/KeeperException.h
@ -1,21 +1,6 @@
 #pragma once
-#include <Common/Exception.h>
+
 #include "Types.h"
-#include <Common/ProfileEvents.h>
-
-
-namespace DB
-{
-    namespace ErrorCodes
-    {
-        extern const int KEEPER_EXCEPTION;
-    }
-}
-
-namespace ProfileEvents
-{
-    extern const Event ZooKeeperExceptions;
-}


 namespace zkutil
@ -23,83 +8,46 @@ namespace zkutil


 /// You should reinitialize ZooKeeper session in case of these errors
-inline bool isUnrecoverableErrorCode(int32_t zk_return_code)
+inline bool isHardwareError(int32_t zk_return_code)
 {
-    return zk_return_code == ZINVALIDSTATE || zk_return_code == ZSESSIONEXPIRED || zk_return_code == ZSESSIONMOVED;
-}
-
-/// Errors related with temporary network problems
-inline bool isTemporaryErrorCode(int32_t zk_return_code)
-{
-    return zk_return_code == ZCONNECTIONLOSS || zk_return_code == ZOPERATIONTIMEOUT;
-}
-
-/// Any error related with network or master election
-/// In case of these errors you should retry the query or reinitialize ZooKeeper session (see isUnrecoverable())
-inline bool isHardwareErrorCode(int32_t zk_return_code)
-{
-    return isUnrecoverableErrorCode(zk_return_code) || isTemporaryErrorCode(zk_return_code);
+    return zk_return_code == ZooKeeperImpl::ZooKeeper::ZINVALIDSTATE
+        || zk_return_code == ZooKeeperImpl::ZooKeeper::ZSESSIONEXPIRED
+        || zk_return_code == ZooKeeperImpl::ZooKeeper::ZSESSIONMOVED
+        || zk_return_code == ZooKeeperImpl::ZooKeeper::ZCONNECTIONLOSS
+        || zk_return_code == ZooKeeperImpl::ZooKeeper::ZOPERATIONTIMEOUT;
 }

 /// Valid errors sent from server
 inline bool isUserError(int32_t zk_return_code)
 {
-    return zk_return_code == ZNONODE
-           || zk_return_code == ZBADVERSION
-           || zk_return_code == ZNOCHILDRENFOREPHEMERALS
-           || zk_return_code == ZNODEEXISTS
-           || zk_return_code == ZNOTEMPTY;
+    return zk_return_code == ZooKeeperImpl::ZooKeeper::ZNONODE
+           || zk_return_code == ZooKeeperImpl::ZooKeeper::ZBADVERSION
+           || zk_return_code == ZooKeeperImpl::ZooKeeper::ZNOCHILDRENFOREPHEMERALS
+           || zk_return_code == ZooKeeperImpl::ZooKeeper::ZNODEEXISTS
+           || zk_return_code == ZooKeeperImpl::ZooKeeper::ZNOTEMPTY;
 }


-class KeeperException : public DB::Exception
+using KeeperException = ZooKeeperImpl::Exception;
+
+
+class KeeperMultiException : public KeeperException
 {
-private:
-    /// delegate constructor, used to minimize repetition; last parameter used for overload resolution
-    KeeperException(const std::string & msg, const int32_t code, int)
-        : DB::Exception(msg, DB::ErrorCodes::KEEPER_EXCEPTION), code(code) { incrementEventCounter(); }
-
 public:
-    explicit KeeperException(const std::string & msg) : KeeperException(msg, ZOK, 0) {}
-    KeeperException(const std::string & msg, const int32_t code)
-        : KeeperException(msg + " (" + zerror(code) + ")", code, 0) {}
-    explicit KeeperException(const int32_t code) : KeeperException(zerror(code), code, 0) {}
-    KeeperException(const int32_t code, const std::string & path)
-        : KeeperException(std::string{zerror(code)} + ", path: " + path, code, 0) {}
+    Requests requests;
+    Responses responses;
+    size_t failed_op_index = 0;

-    KeeperException(const KeeperException & exc) : DB::Exception(exc), code(exc.code) { incrementEventCounter(); }
+    std::string getPathForFirstFailedOp() const;

-    const char * name() const throw() override { return "zkutil::KeeperException"; }
-    const char * className() const throw() override { return "zkutil::KeeperException"; }
-    KeeperException * clone() const override { return new KeeperException(*this); }
+    /// If it is user error throws KeeperMultiException else throws ordinary KeeperException
+    /// If it is ZOK does nothing
+    static void check(int32_t code, const Requests & requests, const Responses & responses);

-    /// You should reinitialize ZooKeeper session in case of these errors
-    bool isUnrecoverable() const
-    {
-        return isUnrecoverableErrorCode(code);
-    }
-
-    /// Errors related with temporary network problems
-    bool isTemporaryError() const
-    {
-        return isTemporaryErrorCode(code);
-    }
-
-    /// Any error related with network or master election
-    /// In case of these errors you should retry the query or reinitialize ZooKeeper session (see isUnrecoverable())
-    bool isHardwareError() const
-    {
-        return isHardwareErrorCode(code);
-    }
-
-    const int32_t code;
+    KeeperMultiException(int32_t code, const Requests & requests, const Responses & responses);

 private:
-    static void incrementEventCounter()
-    {
-        ProfileEvents::increment(ProfileEvents::ZooKeeperExceptions);
-    }
-
+    size_t getFailedOpIndex(int32_t code, const Responses & responses) const;
 };

 };
--- a/dbms/src/Common/ZooKeeper/LeaderElection.h
+++ b/dbms/src/Common/ZooKeeper/LeaderElection.h
@ -33,8 +33,7 @@ public:
      *
      * identifier - if not empty, must uniquely (within same path) identify participant of leader election.
      * It means that different participants of leader election have different identifiers
-      *  and existence of more than one ephemeral node with same identifier indicates an error
-      *  (see cleanOldEphemeralNodes).
+      *  and existence of more than one ephemeral node with same identifier indicates an error.
      */
    LeaderElection(const std::string & path_, ZooKeeper & zookeeper_, LeadershipHandler handler_, const std::string & identifier_ = "")
        : path(path_), zookeeper(zookeeper_), handler(handler_), identifier(identifier_)
@ -42,10 +41,15 @@ public:
        createNode();
    }

-    void yield()
+    void shutdown()
    {
-        releaseNode();
-        createNode();
+        if (shutdown_called)
+            return;
+
+        shutdown_called = true;
+        event->set();
+        if (thread.joinable())
+            thread.join();
    }

    ~LeaderElection()
@ -63,66 +67,31 @@ private:
    std::string node_name;

    std::thread thread;
-    std::atomic<bool> shutdown {false};
+    std::atomic<bool> shutdown_called {false};
    zkutil::EventPtr event = std::make_shared<Poco::Event>();

    CurrentMetrics::Increment metric_increment{CurrentMetrics::LeaderElection};

    void createNode()
    {
-        shutdown = false;
+        shutdown_called = false;
        node = EphemeralNodeHolder::createSequential(path + "/leader_election-", zookeeper, identifier);

        std::string node_path = node->getPath();
        node_name = node_path.substr(node_path.find_last_of('/') + 1);

-        cleanOldEphemeralNodes();
-
        thread = std::thread(&LeaderElection::threadFunction, this);
    }

-    void cleanOldEphemeralNodes()
-    {
-        if (identifier.empty())
-            return;
-
-        /** If there are nodes with same identifier, remove them.
-          * Such nodes could still be alive after failed attempt of removal,
-          *  if it was temporary communication failure, that was continued for more than session timeout,
-          *  but ZK session is still alive for unknown reason, and someone still holds that ZK session.
-          * See comments in destructor of EphemeralNodeHolder.
-          */
-        Strings brothers = zookeeper.getChildren(path);
-        for (const auto & brother : brothers)
-        {
-            if (brother == node_name)
-                continue;
-
-            std::string brother_path = path + "/" + brother;
-            std::string brother_identifier = zookeeper.get(brother_path);
-
-            if (brother_identifier == identifier)
-            {
-                ProfileEvents::increment(ProfileEvents::ObsoleteEphemeralNode);
-                LOG_WARNING(&Logger::get("LeaderElection"), "Found obsolete ephemeral node for identifier "
-                    + identifier + ", removing: " + brother_path);
-                zookeeper.tryRemoveWithRetries(brother_path);
-            }
-        }
-    }
-
    void releaseNode()
    {
-        shutdown = true;
-        event->set();
-        if (thread.joinable())
-            thread.join();
+        shutdown();
        node = nullptr;
    }

    void threadFunction()
    {
-        while (!shutdown)
+        while (!shutdown_called)
        {
            bool success = false;

--- a/dbms/src/Common/ZooKeeper/Lock.cpp
+++ b/dbms/src/Common/ZooKeeper/Lock.cpp
@ -1,5 +1,7 @@
+#include "KeeperException.h"
 #include "Lock.h"

+
 using namespace zkutil;

 bool Lock::tryLock()
@ -11,31 +13,16 @@ bool Lock::tryLock()
        if (tryCheck() != Status::LOCKED_BY_ME)
            locked.reset(nullptr);
    }
-
-    if (!locked)
+    else
    {
-        size_t attempt;
        std::string dummy;
+        int32_t code = zookeeper->tryCreate(lock_path, lock_message, zkutil::CreateMode::Ephemeral, dummy);

-        /// TODO: ошибка. можно создать эфемерную ноду, но при этом не получить подтверждения даже после нескольких попыток.
-        /// тогда все последующие локи будут неуспешные из-за существования ноды.
-        int32_t code = zookeeper->tryCreateWithRetries(lock_path, lock_message, zkutil::CreateMode::Ephemeral, dummy, &attempt);
-
-        if (code == ZNODEEXISTS)
+        if (code == ZooKeeperImpl::ZooKeeper::ZNODEEXISTS)
        {
-            if (attempt == 0)
-                locked.reset(nullptr);
-            else
-            {
-                zkutil::Stat stat;
-                zookeeper->get(lock_path, &stat);
-                if (stat.ephemeralOwner == zookeeper->getClientID())
-                    locked.reset(new ZooKeeperHandler(zookeeper));
-                else
            locked.reset(nullptr);
        }
-        }
-        else if (code == ZOK)
+        else if (code == ZooKeeperImpl::ZooKeeper::ZOK)
        {
            locked.reset(new ZooKeeperHandler(zookeeper));
        }
@ -52,34 +39,8 @@ void Lock::unlock()
    if (locked)
    {
        auto zookeeper = zookeeper_holder->getZooKeeper();
-        try
-        {
        if (tryCheck() == Status::LOCKED_BY_ME)
-            {
-                size_t attempt;
-                int32_t code = zookeeper->tryRemoveEphemeralNodeWithRetries(lock_path, -1, &attempt);
-
-                if (attempt)
-                {
-                    if (code != ZOK)
-                        throw zkutil::KeeperException(code);
-                }
-                else
-                {
-                    if (code == ZNONODE)
-                        LOG_ERROR(log, "Node " << lock_path << " has been already removed. Probably due to network error.");
-                    else if (code != ZOK)
-                        throw zkutil::KeeperException(code);
-                }
-            }
-        }
-        catch (const zkutil::KeeperException & e)
-        {
-            /// если сессия находится в невостанавливаемом состоянии, то эфемерные ноды нам больше не принадлежат
-            /// и лок через таймаут будет отпущен
-            if (!e.isUnrecoverable())
-                throw;
-        }
+            zookeeper->remove(lock_path, -1);
        locked.reset(nullptr);
    }
 }
@ -97,47 +58,17 @@ Lock::Status Lock::tryCheck() const
    else
    {
        if (stat.ephemeralOwner == zookeeper->getClientID())
-        {
            lock_status = LOCKED_BY_ME;
-        }
        else
-        {
            lock_status = LOCKED_BY_OTHER;
    }
-    }

    if (locked && lock_status != LOCKED_BY_ME)
-        LOG_WARNING(log, "Lock is lost. It is normal if session was reinitialized. Path: " << lock_path << "/" << lock_message);
+        LOG_WARNING(log, "Lock is lost. It is normal if session was expired. Path: " << lock_path << "/" << lock_message);

    return lock_status;
 }

-std::string Lock::status2String(Status status)
-{
-    if (status >= END)
-        throw zkutil::KeeperException("Wrong status code: " + std::to_string(status));
-    static const char * names[] = {"Unlocked", "Locked by me", "Locked by other"};
-    return names[status];
-}
-
-void Lock::unlockOrMoveIfFailed(std::vector<zkutil::Lock> & failed_to_unlock_locks)
-{
-    try
-    {
-        unlock();
-    }
-    catch (const zkutil::KeeperException & e)
-    {
-        if (e.isTemporaryError())
-        {
-            LOG_WARNING(log, "Fail to unlock lock. Move lock to vector to remove later. Path: " << getPath());
-            failed_to_unlock_locks.emplace_back(std::move(*this));
-        }
-        else
-            throw;
-    }
-}
-
 void Lock::unlockAssumeLockNodeRemovedManually()
 {
    locked.reset(nullptr);
--- a/dbms/src/Common/ZooKeeper/Lock.h
+++ b/dbms/src/Common/ZooKeeper/Lock.h
@ -40,7 +40,7 @@ namespace zkutil
            {
                unlock();
            }
-            catch (const zkutil::KeeperException & e)
+            catch (...)
            {
                DB::tryLogCurrentException(__PRETTY_FUNCTION__);
            }
@ -51,15 +51,12 @@ namespace zkutil
            UNLOCKED,
            LOCKED_BY_ME,
            LOCKED_BY_OTHER,
-            END
        };
-        std::string status2String(Status status);

        /// проверяет создана ли эфемерная нода и кто ее владелец.
        Status tryCheck() const;

        void unlock();
-        void unlockOrMoveIfFailed(std::vector<zkutil::Lock> & failed_to_unlock_locks);
        void unlockAssumeLockNodeRemovedManually();

        bool tryLock();
--- a/dbms/src/Common/ZooKeeper/Types.h
+++ b/dbms/src/Common/ZooKeeper/Types.h
@ -3,151 +3,17 @@
 #include <future>
 #include <memory>
 #include <vector>
-#include <zookeeper.h>
+#include <Common/ZooKeeper/ZooKeeperImpl.h>
 #include <Poco/Event.h>


 namespace zkutil
 {

-using ACLPtr = const ACL_vector *;
-using Stat = ::Stat;
-
-struct Op
-{
-public:
-    Op() : data(new zoo_op_t) {}
-    virtual ~Op() {}
-
-    virtual std::unique_ptr<Op> clone() const = 0;
-
-    virtual std::string getPath() const = 0;
-
-    virtual std::string describe() const = 0;
-
-    std::unique_ptr<zoo_op_t> data;
-
-    struct Remove;
-    struct Create;
-    struct SetData;
-    struct Check;
-};
-
-struct Op::Remove : public Op
-{
-    Remove(const std::string & path_, int32_t version_) :
-        path(path_), version(version_)
-    {
-        zoo_delete_op_init(data.get(), path.c_str(), version);
-    }
-
-    std::unique_ptr<Op> clone() const override
-    {
-        return std::unique_ptr<zkutil::Op>(new Remove(path, version));
-    }
-
-    std::string getPath() const override { return path; }
-
-    std::string describe() const override { return "command: remove, path: " + path; }
-
-private:
-    std::string path;
-    int32_t version;
-};
-
-struct Op::Create : public Op
-{
-    Create(const std::string & path_pattern_, const std::string & value_, ACLPtr acl_, int32_t flags_);
-
-    std::unique_ptr<Op> clone() const override
-    {
-        return std::unique_ptr<zkutil::Op>(new Create(path_pattern, value, acl, flags));
-    }
-
-    std::string getPathCreated() { return created_path.data(); }
-
-    std::string getPath() const override { return path_pattern; }
-
-    std::string describe() const override
-    {
-        return     "command: create"
-                ", path: " + path_pattern +
-                ", value: " + value;
-    }
-
-private:
-    std::string path_pattern;
-    std::string value;
-    ACLPtr acl;
-    int32_t flags;
-    std::vector<char> created_path;
-};
-
-struct Op::SetData : public Op
-{
-    SetData(const std::string & path_, const std::string & value_, int32_t version_) :
-        path(path_), value(value_), version(version_)
-    {
-        zoo_set_op_init(data.get(), path.c_str(), value.c_str(), value.size(), version, &stat);
-    }
-
-    std::unique_ptr<Op> clone() const override
-    {
-        return std::unique_ptr<zkutil::Op>(new SetData(path, value, version));
-    }
-
-    std::string getPath() const override { return path; }
-
-    std::string describe() const override
-    {
-        return
-            "command: set"
-            ", path: " + path +
-            ", value: " + value +
-            ", version: " + std::to_string(data->set_op.version);
-    }
-
-private:
-    std::string path;
-    std::string value;
-    int32_t version;
-    Stat stat;
-};
-
-struct Op::Check : public Op
-{
-    Check(const std::string & path_, int32_t version_) :
-        path(path_), version(version_)
-    {
-        zoo_check_op_init(data.get(), path.c_str(), version);
-    }
-
-    std::unique_ptr<Op> clone() const override
-    {
-        return std::unique_ptr<zkutil::Op>(new Check(path, version));
-    }
-
-    std::string getPath() const override { return path; }
-
-    std::string describe() const override { return "command: check, path: " + path; }
-
-private:
-    std::string path;
-    int32_t version;
-};
-
-struct OpResult : public zoo_op_result_t
-{
-    /// Pointers in this class point to fields of class Op.
-    /// Op instances have the same (or longer lifetime), therefore destructor is not required.
-};
-
-using OpPtr = std::unique_ptr<Op>;
-using Ops = std::vector<OpPtr>;
-using OpResults = std::vector<OpResult>;
-using OpResultsPtr = std::shared_ptr<OpResults>;
+using Stat = ZooKeeperImpl::ZooKeeper::Stat;
 using Strings = std::vector<std::string>;

+
 namespace CreateMode
 {
    extern const int Persistent;
@ -158,23 +24,39 @@ namespace CreateMode

 using EventPtr = std::shared_ptr<Poco::Event>;

-class ZooKeeper;
-
 /// Callback to call when the watch fires.
 /// Because callbacks are called in the single "completion" thread internal to libzookeeper,
 /// they must execute as quickly as possible (preferably just set some notification).
-/// Parameters:
-/// zookeeper - zookeeper session to which the fired watch belongs
-/// type - event type, one of the *_EVENT constants from zookeeper.h
-/// state - session connection state, one of the *_STATE constants from zookeeper.h
-/// path - znode path to which the change happened. if event == ZOO_SESSION_EVENT it is either NULL or empty string.
-using WatchCallback = std::function<void(ZooKeeper & zookeeper, int type, int state, const char * path)>;
+using WatchCallback = ZooKeeperImpl::ZooKeeper::WatchCallback;

+using Request = ZooKeeperImpl::ZooKeeper::Request;
+using Response = ZooKeeperImpl::ZooKeeper::Response;

-/// Returns first op which code != ZOK or throws an exception
-/// ZooKeeper client sets correct OP codes if the transaction fails because of logical (user) errors like ZNODEEXISTS
-/// If it is failed because of network error, for example, OP codes is not set.
-/// Therefore you should make zkutil::isUserError() check before the function invocation.
-size_t getFailedOpIndex(const OpResultsPtr & op_results, int32_t transaction_return_code);
+using RequestPtr = ZooKeeperImpl::ZooKeeper::RequestPtr;
+using ResponsePtr = ZooKeeperImpl::ZooKeeper::ResponsePtr;
+
+using Requests = ZooKeeperImpl::ZooKeeper::Requests;
+using Responses = ZooKeeperImpl::ZooKeeper::Responses;
+
+using CreateRequest = ZooKeeperImpl::ZooKeeper::CreateRequest;
+using RemoveRequest = ZooKeeperImpl::ZooKeeper::RemoveRequest;
+using ExistsRequest = ZooKeeperImpl::ZooKeeper::ExistsRequest;
+using GetRequest = ZooKeeperImpl::ZooKeeper::GetRequest;
+using SetRequest = ZooKeeperImpl::ZooKeeper::SetRequest;
+using ListRequest = ZooKeeperImpl::ZooKeeper::ListRequest;
+using CheckRequest = ZooKeeperImpl::ZooKeeper::CheckRequest;
+
+using CreateResponse = ZooKeeperImpl::ZooKeeper::CreateResponse;
+using RemoveResponse = ZooKeeperImpl::ZooKeeper::RemoveResponse;
+using ExistsResponse = ZooKeeperImpl::ZooKeeper::ExistsResponse;
+using GetResponse = ZooKeeperImpl::ZooKeeper::GetResponse;
+using SetResponse = ZooKeeperImpl::ZooKeeper::SetResponse;
+using ListResponse = ZooKeeperImpl::ZooKeeper::ListResponse;
+using CheckResponse = ZooKeeperImpl::ZooKeeper::CheckResponse;
+
+RequestPtr makeCreateRequest(const std::string & path, const std::string & data, int create_mode);
+RequestPtr makeRemoveRequest(const std::string & path, int version);
+RequestPtr makeSetRequest(const std::string & path, const std::string & data, int version);
+RequestPtr makeCheckRequest(const std::string & path, int version);

 }
--- a/dbms/src/Common/ZooKeeper/ZooKeeper.cpp
+++ b/dbms/src/Common/ZooKeeper/ZooKeeper.cpp
--- a/dbms/src/Common/ZooKeeper/ZooKeeper.h
+++ b/dbms/src/Common/ZooKeeper/ZooKeeper.h
@ -1,17 +1,16 @@
 #pragma once

 #include "Types.h"
-#include "KeeperException.h"
 #include <Poco/Util/LayeredConfiguration.h>
 #include <unordered_set>
 #include <future>
 #include <memory>
 #include <mutex>
 #include <string>
-#include <unistd.h>
 #include <common/logger_useful.h>
 #include <Common/ProfileEvents.h>
 #include <Common/CurrentMetrics.h>
+#include <port/unistd.h>


 namespace ProfileEvents
@ -29,14 +28,10 @@ namespace zkutil
 {

 const UInt32 DEFAULT_SESSION_TIMEOUT = 30000;
-const UInt32 MEDIUM_SESSION_TIMEOUT = 120000;
-const UInt32 BIG_SESSION_TIMEOUT = 600000;

 /// Preferred size of multi() command (in number of ops)
 constexpr size_t MULTI_BATCH_SIZE = 100;

-struct WatchContext;
-struct MultiTransactionInfo;

 /// ZooKeeper session. The interface is substantially different from the usual libzookeeper API.
 ///
@ -56,7 +51,7 @@ public:
    using Ptr = std::shared_ptr<ZooKeeper>;

    ZooKeeper(const std::string & hosts, const std::string & identity = "",
-              int32_t session_timeout_ms = DEFAULT_SESSION_TIMEOUT, bool check_root_exists = false);
+              int32_t session_timeout_ms = DEFAULT_SESSION_TIMEOUT, const std::string & chroot = "");

    /** Config of the form:
        <zookeeper>
@ -77,27 +72,15 @@ public:
    */
    ZooKeeper(const Poco::Util::AbstractConfiguration & config, const std::string & config_name);

-    ~ZooKeeper();
-
    /// Creates a new session with the same parameters. This method can be used for reconnecting
    /// after the session has expired.
    /// This object remains unchanged, and the new session is returned.
    Ptr startNewSession() const;

-    /// Returns true, if the session has expired forever.
-    /// This is possible only if the connection has been established, then lost and re-established
-    /// again, but too late.
-    /// In contrast, if, for instance, the server name or port is misconfigured, connection
-    /// attempts will continue indefinitely, expired() will return false and all method calls
-    /// will raise ConnectionLoss exception.
-    /// Also returns true if is_dirty flag is set - a request to close the session ASAP.
+    /// Returns true, if the session has expired.
    bool expired();

-    ACLPtr getDefaultACL();
-
-    void setDefaultACL(ACLPtr new_acl);
-
-    /// Create a znode. ACL set by setDefaultACL is used (full access to everybody by default).
+    /// Create a znode.
    /// Throw an exception if something went wrong.
    std::string create(const std::string & path, const std::string & data, int32_t mode);

@ -108,12 +91,9 @@ public:
    /// In case of other errors throws an exception.
    int32_t tryCreate(const std::string & path, const std::string & data, int32_t mode, std::string & path_created);
    int32_t tryCreate(const std::string & path, const std::string & data, int32_t mode);
-    int32_t tryCreateWithRetries(const std::string & path, const std::string & data, int32_t mode,
-                                 std::string & path_created, size_t * attempt = nullptr);

    /// Create a Persistent node.
    /// Does nothing if the node already exists.
-    /// Retries on ConnectionLoss or OperationTimeout.
    void createIfNotExists(const std::string & path, const std::string & data);

    /// Creates all non-existent ancestors of the given path with empty contents.
@ -123,45 +103,14 @@ public:
    /// Remove the node if the version matches. (if version == -1, remove any version).
    void remove(const std::string & path, int32_t version = -1);

-    /// Removes the node. In case of network errors tries to remove again.
-    /// ZNONODE error for the second and the following tries is ignored.
-    void removeWithRetries(const std::string & path, int32_t version = -1);
-
    /// Doesn't throw in the following cases:
    /// * The node doesn't exist
    /// * Versions don't match
    /// * The node has children.
    int32_t tryRemove(const std::string & path, int32_t version = -1);
-    /// Retries in case of network errors, returns ZNONODE if the node is already removed.
-    int32_t tryRemoveWithRetries(const std::string & path, int32_t version = -1, size_t * attempt = nullptr);
-
-    /// The same, but sets is_dirty flag if all removal attempts were unsuccessful.
-    /// This is needed because the session might still exist after all retries,
-    /// even if more time than session_timeout has passed.
-    /// So we do not rely on the ephemeral node being deleted and set is_dirty to
-    /// try and close the session ASAP.
-    /**     Ridiculously Long Delay to Expire
-            When disconnects do happen, the common case should be a very* quick
-            reconnect to another server, but an extended network outage may
-            introduce a long delay before a client can reconnect to the ZooKeep‐
-            er service. Some developers wonder why the ZooKeeper client li‐
-            brary doesn’t simply decide at some point (perhaps twice the session
-            timeout) that enough is enough and kill the session itself.
-            There are two answers to this. First, ZooKeeper leaves this kind of
-            policy decision up to the developer. Developers can easily implement
-            such a policy by closing the handle themselves. Second, when a Zoo‐
-            Keeper ensemble goes down, time freezes. Thus, when the ensemble is
-            brought back up, session timeouts are restarted. If processes using
-            ZooKeeper hang in there, they may find out that the long timeout was
-            due to an extended ensemble failure that has recovered and pick right
-            up where they left off without any additional startup delay.
-
-            ZooKeeper: Distributed Process Coordination p118
-      */
-    int32_t tryRemoveEphemeralNodeWithRetries(const std::string & path, int32_t version = -1, size_t * attempt = nullptr);

    bool exists(const std::string & path, Stat * stat = nullptr, const EventPtr & watch = nullptr);
-    bool existsWatch(const std::string & path, Stat * stat, const WatchCallback & watch_callback);
+    bool existsWatch(const std::string & path, Stat * stat, WatchCallback watch_callback);

    std::string get(const std::string & path, Stat * stat = nullptr, const EventPtr & watch = nullptr);

@ -169,7 +118,7 @@ public:
    /// * The node doesn't exist. Returns false in this case.
    bool tryGet(const std::string & path, std::string & res, Stat * stat = nullptr, const EventPtr & watch = nullptr, int * code = nullptr);

-    bool tryGetWatch(const std::string & path, std::string & res, Stat * stat, const WatchCallback & watch_callback, int * code = nullptr);
+    bool tryGetWatch(const std::string & path, std::string & res, Stat * stat, WatchCallback watch_callback, int * code = nullptr);

    void set(const std::string & path, const std::string & data,
            int32_t version = -1, Stat * stat = nullptr);
@ -195,15 +144,15 @@ public:

    /// Performs several operations in a transaction.
    /// Throws on every error.
-    OpResultsPtr multi(const Ops & ops);
-
+    Responses multi(const Requests & requests);
    /// Throws only if some operation has returned an "unexpected" error
    /// - an error that would cause the corresponding try- method to throw.
-    int32_t tryMulti(const Ops & ops, OpResultsPtr * out_results = nullptr);
-    /// Like previous one, but does not throw any ZooKeeper exceptions
-    int32_t tryMultiUnsafe(const Ops & ops, MultiTransactionInfo & info);
-    /// Use only with read-only operations.
-    int32_t tryMultiWithRetries(const Ops & ops, OpResultsPtr * out_results = nullptr, size_t * attempt = nullptr);
+    int32_t tryMulti(const Requests & requests, Responses & responses);
+    /// Throws nothing, just alias of multiImpl
+    int32_t tryMultiNoThrow(const Requests & requests, Responses & responses)
+    {
+        return multiImpl(requests, responses);
+    }

    Int64 getClientID();

@ -235,200 +184,56 @@ public:
    ///
    /// Future should not be destroyed before the result is gotten.

-    template <typename Result, typename... TaskParams>
-    class Future
-    {
-    friend class ZooKeeper;
-    private:
-        using Task = std::packaged_task<Result (TaskParams...)>;
-        using TaskPtr = std::unique_ptr<Task>;
-        using TaskPtrPtr = std::unique_ptr<TaskPtr>;
+    std::future<ZooKeeperImpl::ZooKeeper::GetResponse> asyncGet(const std::string & path);

-        /// Everything is complicated.
-        ///
-        /// In libzookeeper async interface a function (e.g. zoo_aget)
-        /// accepts a pointer to a standalone callback function and void* pointer to the context
-        /// which is then passed to the callback.
-        /// The caller is responsible for ensuring that the context lives until the callback
-        /// is finished and we can't simply pass ownership of the context into function object.
-        /// Instead, we save the context in a Future object and return it to the caller.
-        /// The context will live until the Future lives.
-        /// Context data is wrapped in an unique_ptr so that its address (which is passed to
-        /// libzookeeper) remains unchanged after the Future is returned from the function.
-        ///
-        /// The second problem is that after std::promise has been fulfilled, and the user
-        /// has gotten the result from std::future, the Future object can be destroyed
-        /// before the std::promise::set_value() call that fulfils the promise completes in another
-        /// thread.
-        /// See http://stackoverflow.com/questions/10843304/race-condition-in-pthread-once
-        /// To combat this we use the second unique_ptr. Inside the callback, the void* context
-        /// is cast to unique_ptr and moved into the local unique_ptr to prolong the lifetime of
-        /// the context data.
+    std::future<ZooKeeperImpl::ZooKeeper::GetResponse> asyncTryGet(const std::string & path);

-        TaskPtrPtr task;
-        std::future<Result> future;
+    std::future<ZooKeeperImpl::ZooKeeper::ExistsResponse> asyncExists(const std::string & path);

-        template <typename... Args>
-        Future(Args &&... args) :
-            task(std::make_unique<TaskPtr>(std::make_unique<Task>(std::forward<Args>(args)...))),
-            future((*task)->get_future()) {}
+    std::future<ZooKeeperImpl::ZooKeeper::ListResponse> asyncGetChildren(const std::string & path);

-    public:
-        Result get()
-        {
-            return future.get();
-        }
-
-        Future(Future &&) = default;
-        Future & operator= (Future &&) = default;
-
-        ~Future()
-        {
-            /// If nobody has waited for the result, we must wait for it before the object is
-            /// destroyed, because the object contents can still be used in the callback.
-            if (future.valid())
-                future.wait();
-        }
-    };
-
-
-    struct ValueAndStat
-    {
-        std::string value;
-        Stat stat;
-    };
-
-    using GetFuture = Future<ValueAndStat, int, const char *, int, const Stat *>;
-    GetFuture asyncGet(const std::string & path);
-
-
-    struct ValueAndStatAndExists
-    {
-        std::string value;
-        Stat stat;
-        bool exists;
-    };
-
-    using TryGetFuture = Future<ValueAndStatAndExists, int, const char *, int, const Stat *>;
-    TryGetFuture asyncTryGet(const std::string & path);
-
-
-    struct StatAndExists
-    {
-        Stat stat;
-        bool exists;
-    };
-
-    using ExistsFuture = Future<StatAndExists, int, const Stat *>;
-    ExistsFuture asyncExists(const std::string & path);
-
-
-    using GetChildrenFuture = Future<Strings, int, const String_vector *>;
-    GetChildrenFuture asyncGetChildren(const std::string & path);
-
-
-    using RemoveFuture = Future<void, int>;
-    RemoveFuture asyncRemove(const std::string & path, int32_t version = -1);
+    std::future<ZooKeeperImpl::ZooKeeper::RemoveResponse> asyncRemove(const std::string & path, int32_t version = -1);

    /// Doesn't throw in the following cases:
    /// * The node doesn't exist
    /// * The versions do not match
    /// * The node has children
-    using TryRemoveFuture = Future<int32_t, int>;
-    TryRemoveFuture asyncTryRemove(const std::string & path, int32_t version = -1);
+    std::future<ZooKeeperImpl::ZooKeeper::RemoveResponse> asyncTryRemove(const std::string & path, int32_t version = -1);

-    struct OpResultsAndCode
-    {
-        OpResultsPtr results;
-        std::shared_ptr<Ops> ops_ptr;
-        int code;
-    };
+    std::future<ZooKeeperImpl::ZooKeeper::MultiResponse> asyncMulti(const Requests & ops);

-    using MultiFuture = Future<OpResultsAndCode, int>;
-    MultiFuture asyncMulti(const Ops & ops);
    /// Like the previous one but don't throw any exceptions on future.get()
-    MultiFuture tryAsyncMulti(const Ops & ops);
-
+    std::future<ZooKeeperImpl::ZooKeeper::MultiResponse> tryAsyncMulti(const Requests & ops);

    static std::string error2string(int32_t code);

-    /// Max size of node contents in bytes.
-    /// In 3.4.5 max node size is 1Mb.
-    static const size_t MAX_NODE_SIZE = 1048576;
-
-    /// Length of the suffix that ZooKeeper adds to sequential nodes.
-    /// In fact it is smaller, but round it up for convenience.
-    static const size_t SEQUENTIAL_SUFFIX_SIZE = 64;
-
-
-    zhandle_t * getHandle() { return impl; }
-
 private:
-    friend struct WatchContext;
    friend class EphemeralNodeHolder;

-    void init(const std::string & hosts, const std::string & identity,
-              int32_t session_timeout_ms, bool check_root_exists);
+    void init(const std::string & hosts_, const std::string & identity_, int32_t session_timeout_ms_, const std::string & chroot_);
+
    void removeChildrenRecursive(const std::string & path);
    void tryRemoveChildrenRecursive(const std::string & path);

-    static WatchCallback callbackForEvent(const EventPtr & event);
-    WatchContext * createContext(WatchCallback && callback);
-    static void destroyContext(WatchContext * context);
-    static void processCallback(zhandle_t * zh, int type, int state, const char * path, void * watcher_ctx);
-
-    template <typename T>
-    int32_t retry(T && operation, size_t * attempt = nullptr)
-    {
-        int32_t code = operation();
-        if (attempt)
-            *attempt = 0;
-        for (size_t i = 0; (i < retry_num) && (code == ZOPERATIONTIMEOUT || code == ZCONNECTIONLOSS); ++i)
-        {
-            if (attempt)
-                *attempt = i;
-
-            /// If the connection has been lost, wait timeout/3 hoping for connection re-establishment.
-            static const int MAX_SLEEP_TIME = 10;
-            if (code == ZCONNECTIONLOSS)
-                usleep(std::min(session_timeout_ms * 1000 / 3, MAX_SLEEP_TIME * 1000 * 1000));
-
-            LOG_WARNING(log, "Error on attempt " << i << ": " << error2string(code)  << ". Retry");
-            code = operation();
-        }
-
-        return code;
-    }
-
    /// The following methods don't throw exceptions but return error codes.
    int32_t createImpl(const std::string & path, const std::string & data, int32_t mode, std::string & path_created);
-    int32_t removeImpl(const std::string & path, int32_t version = -1);
+    int32_t removeImpl(const std::string & path, int32_t version);
    int32_t getImpl(const std::string & path, std::string & res, Stat * stat, WatchCallback watch_callback);
-    int32_t setImpl(const std::string & path, const std::string & data, int32_t version = -1, Stat * stat = nullptr);
+    int32_t setImpl(const std::string & path, const std::string & data, int32_t version, Stat * stat);
    int32_t getChildrenImpl(const std::string & path, Strings & res, Stat * stat, WatchCallback watch_callback);
-    int32_t multiImpl(const Ops & ops, OpResultsPtr * out_results = nullptr);
+    int32_t multiImpl(const Requests & requests, Responses & responses);
    int32_t existsImpl(const std::string & path, Stat * stat_, WatchCallback watch_callback);

-    MultiFuture asyncMultiImpl(const zkutil::Ops & ops_, bool throw_exception);
+    std::unique_ptr<ZooKeeperImpl::ZooKeeper> impl;

    std::string hosts;
    std::string identity;
    int32_t session_timeout_ms;
+    std::string chroot;

    std::mutex mutex;
-    ACLPtr default_acl;
-    zhandle_t * impl;

-    std::unordered_set<WatchContext *> watch_context_store;
-
-    /// Retries number in case of OperationTimeout or ConnectionLoss errors.
-    static constexpr size_t retry_num = 3;
    Logger * log = nullptr;
-
-    /// If true, there were unsuccessfull attempts to remove ephemeral nodes.
-    /// It is better to close the session to remove ephemeral nodes with certainty
-    /// instead of continuing to use re-established session.
-    bool is_dirty = false;
 };


@ -472,13 +277,9 @@ public:
    {
        try
        {
-            /// Important: if the ZooKeeper is temporarily unavailable, repeated attempts to
-            /// delete the node are made.
-            /// Otherwise it is possible that EphemeralNodeHolder is destroyed,
-            /// but the session has recovered and the node in ZooKeeper remains for the long time.
-            zookeeper.tryRemoveEphemeralNodeWithRetries(path);
+            zookeeper.tryRemove(path);
        }
-        catch (const KeeperException & e)
+        catch (...)
        {
            ProfileEvents::increment(ProfileEvents::CannotRemoveEphemeralNode);
            DB::tryLogCurrentException(__PRETTY_FUNCTION__);
@ -493,41 +294,4 @@ private:

 using EphemeralNodeHolderPtr = EphemeralNodeHolder::Ptr;

-
-/// Simple structure to handle transaction execution results
-struct MultiTransactionInfo
-{
-    MultiTransactionInfo() = default;
-
-    Ops ops;
-    int32_t code = ZOK;
-    OpResultsPtr op_results;
-
-    bool empty() const
-    {
-        return ops.empty();
-    }
-
-    bool hasFailedOp() const
-    {
-        return zkutil::isUserError(code);
-    }
-
-    const Op & getFailedOp() const
-    {
-        return *ops.at(getFailedOpIndex(op_results, code));
-    }
-
-    KeeperException getException() const
-    {
-        if (hasFailedOp())
-        {
-            size_t i = getFailedOpIndex(op_results, code);
-            return KeeperException("Transaction failed at op #" + std::to_string(i) + ": " + ops.at(i)->describe(), code);
-        }
-        else
-            return KeeperException(code);
-    }
-};
-
 }
--- a/dbms/src/Common/ZooKeeper/ZooKeeperImpl.cpp
+++ b/dbms/src/Common/ZooKeeper/ZooKeeperImpl.cpp
--- a/dbms/src/Common/ZooKeeper/ZooKeeperImpl.h
+++ b/dbms/src/Common/ZooKeeper/ZooKeeperImpl.h
@ -0,0 +1,635 @@
+#pragma once
+
+#include <Core/Types.h>
+#include <Common/ConcurrentBoundedQueue.h>
+#include <Common/CurrentMetrics.h>
+
+#include <IO/ReadBuffer.h>
+#include <IO/WriteBuffer.h>
+#include <IO/ReadBufferFromPocoSocket.h>
+#include <IO/WriteBufferFromPocoSocket.h>
+
+#include <Poco/Net/StreamSocket.h>
+#include <Poco/Net/SocketAddress.h>
+
+#include <map>
+#include <mutex>
+#include <chrono>
+#include <vector>
+#include <memory>
+#include <thread>
+#include <atomic>
+#include <cstdint>
+#include <optional>
+#include <functional>
+
+
+/** ZooKeeper C++ library, a replacement for libzookeeper.
+  *
+  * Motivation.
+  *
+  * libzookeeper has many bugs:
+  * - segfaults: for example, if zookeeper connection was interrupted while reading result of multi response;
+  * - memory corruption: for example, as a result of double free inside libzookeeper;
+  * - no timeouts for synchronous operations: they may stuck forever under simple Jepsen-like tests;
+  * - logical errors: for example, chroot prefix is not removed from the results of multi responses.
+  * - data races;
+  *
+  * The code of libzookeeper is over complicated:
+  * - memory ownership is unclear and bugs are very difficult to track and fix.
+  * - extremely creepy code for implementation of "chroot" feature.
+  *
+  * As of 2018, there are no active maintainers of libzookeeper:
+  * - bugs in JIRA are fixed only occasionaly with ad-hoc patches by library users.
+  *
+  * libzookeeper is a classical example of bad code written in C.
+  *
+  * In Go, Python and Rust programming languages,
+  *  there are separate libraries for ZooKeeper, not based on libzookeeper.
+  * Motivation is almost the same. Example:
+  * https://github.com/python-zk/kazoo/blob/master/docs/implementation.rst
+  *
+  * About "session restore" feature.
+  *
+  * libzookeeper has the feature of session restore. Client receives session id and session token from the server,
+  *  and when connection is lost, it can quickly reconnect to any server with the same session id and token,
+  *  to continue with existing session.
+  * libzookeeper performs this reconnection automatically.
+  *
+  * This feature is proven to be harmful.
+  * For example, it makes very difficult to correctly remove ephemeral nodes.
+  * This may lead to weird bugs in application code.
+  * For example, our developers have found that type of bugs in Curator Java library.
+  *
+  * On the other side, session restore feature has no advantages,
+  *  because every application should be able to establish new session and reinitialize internal state,
+  *  when the session is lost and cannot be restored.
+  *
+  * This library never restores the session. In case of any error, the session is considered as expired
+  *  and you should create a new instance of ZooKeeper object and reinitialize the application state.
+  *
+  * This library is not intended to be CPU efficient. Hundreds of thousands operations per second is usually enough.
+  */
+
+
+namespace CurrentMetrics
+{
+    extern const Metric ZooKeeperSession;
+}
+
+
+namespace ZooKeeperImpl
+{
+
+using namespace DB;
+
+
+class Exception : public DB::Exception
+{
+private:
+    /// Delegate constructor, used to minimize repetition; last parameter used for overload resolution.
+    Exception(const std::string & msg, const int32_t code, int);
+
+public:
+    explicit Exception(const int32_t code);
+    Exception(const std::string & msg, const int32_t code);
+    Exception(const int32_t code, const std::string & path);
+    Exception(const Exception & exc);
+
+    const char * name() const throw() override { return "ZooKeeperImpl::Exception"; }
+    const char * className() const throw() override { return "ZooKeeperImpl::Exception"; }
+    Exception * clone() const override { return new Exception(*this); }
+
+    const int32_t code;
+};
+
+
+/** Usage scenario:
+  * - create an object and issue commands;
+  * - you provide callbacks for your commands; callbacks are invoked in internal thread and must be cheap:
+  *   for example, just signal a condvar / fulfull a promise.
+  * - you also may provide callbacks for watches; they are also invoked in internal thread and must be cheap.
+  * - whenever you receive exception with ZSESSIONEXPIRED code or method isExpired returns true,
+  *   the ZooKeeper instance is no longer usable - you may only destroy it and probably create another.
+  * - whenever session is expired or ZooKeeper instance is destroying, all callbacks are notified with special event.
+  * - data for callbacks must be alive when ZooKeeper instance is alive.
+  */
+class ZooKeeper
+{
+public:
+    using Addresses = std::vector<Poco::Net::SocketAddress>;
+
+    struct ACL
+    {
+        static constexpr int32_t Read = 1;
+        static constexpr int32_t Write = 2;
+        static constexpr int32_t Create = 4;
+        static constexpr int32_t Delete = 8;
+        static constexpr int32_t Admin = 16;
+        static constexpr int32_t All = 0x1F;
+
+        int32_t permissions;
+        String scheme;
+        String id;
+
+        void write(WriteBuffer & out) const;
+    };
+    using ACLs = std::vector<ACL>;
+
+    struct Stat
+    {
+        int64_t czxid;
+        int64_t mzxid;
+        int64_t ctime;
+        int64_t mtime;
+        int32_t version;
+        int32_t cversion;
+        int32_t aversion;
+        int64_t ephemeralOwner;
+        int32_t dataLength;
+        int32_t numChildren;
+        int64_t pzxid;
+
+        void read(ReadBuffer & in);
+    };
+
+    using XID = int32_t;
+    using OpNum = int32_t;
+
+
+    struct Response
+    {
+        int32_t error = 0;
+        virtual ~Response() {}
+        virtual void readImpl(ReadBuffer &) = 0;
+
+        virtual void removeRootPath(const String & /* root_path */) {}
+    };
+
+    using ResponsePtr = std::shared_ptr<Response>;
+    using Responses = std::vector<ResponsePtr>;
+    using ResponseCallback = std::function<void(const Response &)>;
+
+    struct Request
+    {
+        XID xid = 0;
+        bool has_watch = false;
+
+        virtual ~Request() {};
+        virtual OpNum getOpNum() const = 0;
+
+        /// Writes length, xid, op_num, then the rest.
+        void write(WriteBuffer & out) const;
+        virtual void writeImpl(WriteBuffer &) const = 0;
+
+        virtual ResponsePtr makeResponse() const = 0;
+
+        virtual void addRootPath(const String & /* root_path */) {};
+        virtual String getPath() const = 0;
+    };
+
+    using RequestPtr = std::shared_ptr<Request>;
+    using Requests = std::vector<RequestPtr>;
+
+    struct HeartbeatRequest final : Request
+    {
+        OpNum getOpNum() const override { return 11; }
+        void writeImpl(WriteBuffer &) const override {}
+        ResponsePtr makeResponse() const override;
+        String getPath() const override { return {}; }
+    };
+
+    struct HeartbeatResponse final : Response
+    {
+        void readImpl(ReadBuffer &) override {}
+    };
+
+    struct WatchResponse final : Response
+    {
+        int32_t type = 0;
+        int32_t state = 0;
+        String path;
+
+        void readImpl(ReadBuffer &) override;
+        void removeRootPath(const String & root_path) override;
+    };
+
+    using WatchCallback = std::function<void(const WatchResponse &)>;
+
+    struct AuthRequest final : Request
+    {
+        int32_t type = 0;   /// ignored by the server
+        String scheme;
+        String data;
+
+        OpNum getOpNum() const override { return 100; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        String getPath() const override { return {}; }
+    };
+
+    struct AuthResponse final : Response
+    {
+        void readImpl(ReadBuffer &) override {};
+    };
+
+    struct CloseRequest final : Request
+    {
+        OpNum getOpNum() const override { return -11; }
+        void writeImpl(WriteBuffer &) const override {}
+        ResponsePtr makeResponse() const override;
+        String getPath() const override { return {}; }
+    };
+
+    struct CloseResponse final : Response
+    {
+        void readImpl(ReadBuffer &) override;
+    };
+
+    struct CreateRequest final : Request
+    {
+        String path;
+        String data;
+        bool is_ephemeral = false;
+        bool is_sequential = false;
+        ACLs acls;
+
+        OpNum getOpNum() const override { return 1; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return path; }
+    };
+
+    struct CreateResponse final : Response
+    {
+        String path_created;
+
+        void readImpl(ReadBuffer &) override;
+        void removeRootPath(const String & root_path) override;
+    };
+
+    struct RemoveRequest final : Request
+    {
+        String path;
+        int32_t version = -1;
+
+        OpNum getOpNum() const override { return 2; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return path; }
+    };
+
+    struct RemoveResponse final : Response
+    {
+        void readImpl(ReadBuffer &) override {}
+    };
+
+    struct ExistsRequest final : Request
+    {
+        String path;
+
+        OpNum getOpNum() const override { return 3; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return path; }
+    };
+
+    struct ExistsResponse final : Response
+    {
+        Stat stat;
+
+        void readImpl(ReadBuffer &) override;
+    };
+
+    struct GetRequest final : Request
+    {
+        String path;
+
+        OpNum getOpNum() const override { return 4; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return path; }
+    };
+
+    struct GetResponse final : Response
+    {
+        String data;
+        Stat stat;
+
+        void readImpl(ReadBuffer &) override;
+    };
+
+    struct SetRequest final : Request
+    {
+        String path;
+        String data;
+        int32_t version = -1;
+
+        OpNum getOpNum() const override { return 5; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return path; }
+    };
+
+    struct SetResponse final : Response
+    {
+        Stat stat;
+
+        void readImpl(ReadBuffer &) override;
+    };
+
+    struct ListRequest final : Request
+    {
+        String path;
+
+        OpNum getOpNum() const override { return 12; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return path; }
+    };
+
+    struct ListResponse final : Response
+    {
+        std::vector<String> names;
+        Stat stat;
+
+        void readImpl(ReadBuffer &) override;
+    };
+
+    struct CheckRequest final : Request
+    {
+        String path;
+        int32_t version = -1;
+
+        OpNum getOpNum() const override { return 13; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return path; }
+    };
+
+    struct CheckResponse final : Response
+    {
+        void readImpl(ReadBuffer &) override {};
+    };
+
+    struct MultiRequest final : Request
+    {
+        Requests requests;
+
+        OpNum getOpNum() const override { return 14; }
+        void writeImpl(WriteBuffer &) const override;
+        ResponsePtr makeResponse() const override;
+        void addRootPath(const String & root_path) override;
+        String getPath() const override { return {}; }
+    };
+
+    struct MultiResponse final : Response
+    {
+        Responses responses;
+
+        MultiResponse(const Requests & requests);
+
+        void readImpl(ReadBuffer &) override;
+        void removeRootPath(const String & root_path) override;
+    };
+
+    /// This response may be received only as an element of responses in MultiResponse.
+    struct ErrorResponse final : Response
+    {
+        void readImpl(ReadBuffer &) override;
+    };
+
+
+    /** Connection to addresses is performed in order. If you want, shuffle them manually.
+      * Operation timeout couldn't be greater than session timeout.
+      * Operation timeout applies independently for network read, network write, waiting for events and synchronization.
+      */
+    ZooKeeper(
+        const Addresses & addresses,
+        const String & root_path,
+        const String & auth_scheme,
+        const String & auth_data,
+        Poco::Timespan session_timeout,
+        Poco::Timespan connection_timeout,
+        Poco::Timespan operation_timeout);
+
+    ~ZooKeeper();
+
+
+    /// If expired, you can only destroy the object. All other methods will throw exception.
+    bool isExpired() const { return expired; }
+
+    /// Useful to check owner of ephemeral node.
+    int64_t getSessionID() const { return session_id; }
+
+
+    using CreateCallback = std::function<void(const CreateResponse &)>;
+    using RemoveCallback = std::function<void(const RemoveResponse &)>;
+    using ExistsCallback = std::function<void(const ExistsResponse &)>;
+    using GetCallback = std::function<void(const GetResponse &)>;
+    using SetCallback = std::function<void(const SetResponse &)>;
+    using ListCallback = std::function<void(const ListResponse &)>;
+    using CheckCallback = std::function<void(const CheckResponse &)>;
+    using MultiCallback = std::function<void(const MultiResponse &)>;
+
+    /// If the method will throw an exception, callbacks won't be called.
+    ///
+    /// After the method is executed successfully, you must wait for callbacks
+    ///  (don't destroy callback data before it will be called).
+    ///
+    /// All callbacks are executed sequentially (the execution of callbacks is serialized).
+    ///
+    /// If an exception is thrown inside the callback, the session will expire,
+    ///  and all other callbacks will be called with "Session expired" error.
+
+    void create(
+        const String & path,
+        const String & data,
+        bool is_ephemeral,
+        bool is_sequential,
+        const ACLs & acls,
+        CreateCallback callback);
+
+    void remove(
+        const String & path,
+        int32_t version,
+        RemoveCallback callback);
+
+    void exists(
+        const String & path,
+        ExistsCallback callback,
+        WatchCallback watch);
+
+    void get(
+        const String & path,
+        GetCallback callback,
+        WatchCallback watch);
+
+    void set(
+        const String & path,
+        const String & data,
+        int32_t version,
+        SetCallback callback);
+
+    void list(
+        const String & path,
+        ListCallback callback,
+        WatchCallback watch);
+
+    void check(
+        const String & path,
+        int32_t version,
+        CheckCallback callback);
+
+    void multi(
+        const Requests & requests,
+        MultiCallback callback);
+
+
+    enum Error
+    {
+        ZOK = 0,
+
+        /** System and server-side errors.
+          * This is never thrown by the server, it shouldn't be used other than
+          * to indicate a range. Specifically error codes greater than this
+          * value, but lesser than ZAPIERROR, are system errors.
+          */
+        ZSYSTEMERROR = -1,
+
+        ZRUNTIMEINCONSISTENCY = -2, /// A runtime inconsistency was found
+        ZDATAINCONSISTENCY = -3,    /// A data inconsistency was found
+        ZCONNECTIONLOSS = -4,       /// Connection to the server has been lost
+        ZMARSHALLINGERROR = -5,     /// Error while marshalling or unmarshalling data
+        ZUNIMPLEMENTED = -6,        /// Operation is unimplemented
+        ZOPERATIONTIMEOUT = -7,     /// Operation timeout
+        ZBADARGUMENTS = -8,         /// Invalid arguments
+        ZINVALIDSTATE = -9,         /// Invliad zhandle state
+
+        /** API errors.
+          * This is never thrown by the server, it shouldn't be used other than
+          * to indicate a range. Specifically error codes greater than this
+          * value are API errors.
+          */
+        ZAPIERROR = -100,
+
+        ZNONODE = -101,                     /// Node does not exist
+        ZNOAUTH = -102,                     /// Not authenticated
+        ZBADVERSION = -103,                 /// Version conflict
+        ZNOCHILDRENFOREPHEMERALS = -108,    /// Ephemeral nodes may not have children
+        ZNODEEXISTS = -110,                 /// The node already exists
+        ZNOTEMPTY = -111,                   /// The node has children
+        ZSESSIONEXPIRED = -112,             /// The session has been expired by the server
+        ZINVALIDCALLBACK = -113,            /// Invalid callback specified
+        ZINVALIDACL = -114,                 /// Invalid ACL specified
+        ZAUTHFAILED = -115,                 /// Client authentication failed
+        ZCLOSING = -116,                    /// ZooKeeper is closing
+        ZNOTHING = -117,                    /// (not error) no server responses to process
+        ZSESSIONMOVED = -118                /// Session moved to another server, so operation is ignored
+    };
+
+    static const char * errorMessage(int32_t code);
+
+    /// For watches.
+    enum State
+    {
+        EXPIRED_SESSION = -112,
+        AUTH_FAILED = -113,
+        CONNECTING = 1,
+        ASSOCIATING = 2,
+        CONNECTED = 3,
+        NOTCONNECTED = 999
+    };
+
+    enum Event
+    {
+        CREATED = 1,
+        DELETED = 2,
+        CHANGED = 3,
+        CHILD = 4,
+        SESSION = -1,
+        NOTWATCHING = -2
+    };
+
+private:
+    String root_path;
+    ACLs default_acls;
+
+    Poco::Timespan session_timeout;
+    Poco::Timespan operation_timeout;
+
+    Poco::Net::StreamSocket socket;
+    std::optional<ReadBufferFromPocoSocket> in;
+    std::optional<WriteBufferFromPocoSocket> out;
+
+    int64_t session_id = 0;
+
+    std::atomic<XID> xid {1};
+    std::atomic<bool> expired {false};
+    std::mutex finalize_mutex;
+
+    using clock = std::chrono::steady_clock;
+
+    struct RequestInfo
+    {
+        RequestPtr request;
+        ResponseCallback callback;
+        WatchCallback watch;
+        clock::time_point time;
+    };
+
+    using RequestsQueue = ConcurrentBoundedQueue<RequestInfo>;
+
+    RequestsQueue requests_queue{1};
+    void pushRequest(RequestInfo && request);
+
+    using Operations = std::map<XID, RequestInfo>;
+
+    Operations operations;
+    std::mutex operations_mutex;
+
+    using WatchCallbacks = std::vector<WatchCallback>;
+    using Watches = std::map<String /* path */, WatchCallbacks>;
+
+    Watches watches;
+    std::mutex watches_mutex;
+
+    std::thread send_thread;
+    std::thread receive_thread;
+
+    void connect(
+        const Addresses & addresses,
+        Poco::Timespan connection_timeout);
+
+    void sendHandshake();
+    void receiveHandshake();
+
+    void sendAuth(const String & scheme, const String & data);
+
+    void receiveEvent();
+
+    void sendThread();
+    void receiveThread();
+
+    void close();
+
+    /// Call all remaining callbacks and watches, passing errors to them.
+    void finalize(bool error_send, bool error_receive);
+
+    template <typename T>
+    void write(const T &);
+
+    template <typename T>
+    void read(T &);
+
+    CurrentMetrics::Increment active_session_metric_increment{CurrentMetrics::ZooKeeperSession};
+};
+
+};
--- a/Show More
+++ b/Show More
				`@ -1 +0,0 @@`
				`Subproject commit 438afae5af36c5be9c82d074f43a9bb19e0797c0`