Merge branch 'master' into fix-union

2024-09-20 00:30:49 +00:00 · 2022-09-29 23:16:22 +02:00 · 2022-09-29 23:16:22 +02:00 · 9f5276f839
commit 9f5276f839
parent 6bc8aa3341 105b210126
120 changed files with 2061 additions and 572 deletions
--- a/.gitmodules
+++ b/.gitmodules
@ -30,9 +30,6 @@
 [submodule "contrib/re2"]
 	path = contrib/re2
 	url = https://github.com/google/re2.git
-[submodule "contrib/llvm"]
-	path = contrib/llvm
-	url = https://github.com/ClickHouse/llvm
 [submodule "contrib/mariadb-connector-c"]
 	path = contrib/mariadb-connector-c
 	url = https://github.com/ClickHouse/mariadb-connector-c.git
@ -284,3 +281,6 @@
 [submodule "contrib/c-ares"]
 	path = contrib/c-ares
 	url = https://github.com/ClickHouse/c-ares
+[submodule "contrib/llvm-project"]
+	path = contrib/llvm-project
+	url = https://github.com/ClickHouse/llvm-project.git
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -18,7 +18,7 @@ include (cmake/target.cmake)
 include (cmake/tools.cmake)
 include (cmake/ccache.cmake)
 include (cmake/clang_tidy.cmake)
-include (cmake/git_status.cmake)
+include (cmake/git.cmake)

 # Ignore export() since we don't use it,
 # but it gets broken with a global targets via link_libraries()
--- a/base/base/ReplxxLineReader.cpp
+++ b/base/base/ReplxxLineReader.cpp
@ -1,6 +1,7 @@
 #include <base/ReplxxLineReader.h>
 #include <base/errnoToString.h>

+#include <stdexcept>
 #include <chrono>
 #include <cerrno>
 #include <cstring>
@ -13,8 +14,10 @@
 #include <dlfcn.h>
 #include <fcntl.h>
 #include <fstream>
+#include <filesystem>
 #include <fmt/format.h>
-
+#include <boost/algorithm/string/split.hpp>
+#include <boost/algorithm/string/classification.hpp> /// is_any_of

 namespace
 {
@ -35,6 +38,166 @@ std::string getEditor()
    return editor;
 }

+std::string getFuzzyFinder()
+{
+    const char * env_path = std::getenv("PATH"); // NOLINT(concurrency-mt-unsafe)
+
+    if (!env_path || !*env_path)
+        return {};
+
+    std::vector<std::string> paths;
+    boost::split(paths, env_path, boost::is_any_of(":"));
+    for (const auto & path_str : paths)
+    {
+        std::filesystem::path path(path_str);
+        std::filesystem::path sk_bin_path = path / "sk";
+        if (!access(sk_bin_path.c_str(), X_OK))
+            return sk_bin_path;
+
+        std::filesystem::path fzf_bin_path = path / "fzf";
+        if (!access(fzf_bin_path.c_str(), X_OK))
+            return fzf_bin_path;
+    }
+
+    return {};
+}
+
+/// See comments in ShellCommand::executeImpl()
+/// (for the vfork via dlsym())
+int executeCommand(char * const argv[])
+{
+#if !defined(USE_MUSL)
+    /** Here it is written that with a normal call `vfork`, there is a chance of deadlock in multithreaded programs,
+      *  because of the resolving of symbols in the shared library
+      * http://www.oracle.com/technetwork/server-storage/solaris10/subprocess-136439.html
+      * Therefore, separate the resolving of the symbol from the call.
+      */
+    static void * real_vfork = dlsym(RTLD_DEFAULT, "vfork");
+#else
+    /// If we use Musl with static linking, there is no dlsym and no issue with vfork.
+    static void * real_vfork = reinterpret_cast<void *>(&vfork);
+#endif
+    if (!real_vfork)
+        throw std::runtime_error("Cannot find vfork symbol");
+
+    pid_t pid = reinterpret_cast<pid_t (*)()>(real_vfork)();
+
+    if (-1 == pid)
+        throw std::runtime_error(fmt::format("Cannot vfork {}: {}", argv[0], errnoToString()));
+
+    /// Child
+    if (0 == pid)
+    {
+        sigset_t mask;
+        sigemptyset(&mask);
+        sigprocmask(0, nullptr, &mask); // NOLINT(concurrency-mt-unsafe) // ok in newly created process
+        sigprocmask(SIG_UNBLOCK, &mask, nullptr); // NOLINT(concurrency-mt-unsafe) // ok in newly created process
+
+        execvp(argv[0], argv);
+        _exit(-1);
+    }
+
+    int status = 0;
+    do
+    {
+        int exited_pid = waitpid(pid, &status, 0);
+        if (exited_pid != -1)
+            break;
+
+        if (errno == EINTR)
+            continue;
+
+        throw std::runtime_error(fmt::format("Cannot waitpid {}: {}", pid, errnoToString()));
+    } while (true);
+
+    return status;
+}
+
+void writeRetry(int fd, const std::string & data)
+{
+    size_t bytes_written = 0;
+    const char * begin = data.c_str();
+    size_t offset = data.size();
+
+    while (bytes_written != offset)
+    {
+        ssize_t res = ::write(fd, begin + bytes_written, offset - bytes_written);
+        if ((-1 == res || 0 == res) && errno != EINTR)
+            throw std::runtime_error(fmt::format("Cannot write to {}: {}", fd, errnoToString()));
+        bytes_written += res;
+    }
+}
+std::string readFile(const std::string & path)
+{
+    std::ifstream t(path);
+    std::string str;
+    t.seekg(0, std::ios::end);
+    str.reserve(t.tellg());
+    t.seekg(0, std::ios::beg);
+    str.assign((std::istreambuf_iterator<char>(t)), std::istreambuf_iterator<char>());
+    return str;
+}
+
+/// Simple wrapper for temporary files.
+class TemporaryFile
+{
+private:
+    std::string path;
+    int fd = -1;
+
+public:
+    explicit TemporaryFile(const char * pattern)
+        : path(pattern)
+    {
+        size_t dot_pos = path.rfind('.');
+        if (dot_pos != std::string::npos)
+            fd = ::mkstemps(path.data(), path.size() - dot_pos);
+        else
+            fd = ::mkstemp(path.data());
+
+        if (-1 == fd)
+            throw std::runtime_error(fmt::format("Cannot create temporary file {}: {}", path, errnoToString()));
+    }
+    ~TemporaryFile()
+    {
+        try
+        {
+            close();
+            unlink();
+        }
+        catch (const std::runtime_error & e)
+        {
+            fmt::print(stderr, "{}", e.what());
+        }
+    }
+
+    void close()
+    {
+        if (fd == -1)
+            return;
+
+        if (0 != ::close(fd))
+            throw std::runtime_error(fmt::format("Cannot close temporary file {}: {}", path, errnoToString()));
+        fd = -1;
+    }
+
+    void write(const std::string & data)
+    {
+        if (fd == -1)
+            throw std::runtime_error(fmt::format("Cannot write to uninitialized file {}", path));
+
+        writeRetry(fd, data);
+    }
+
+    void unlink()
+    {
+        if (0 != ::unlink(path.c_str()))
+            throw std::runtime_error(fmt::format("Cannot remove temporary file {}: {}", path, errnoToString()));
+    }
+
+    std::string & getPath() { return path; }
+};
+
 /// Copied from replxx::src/util.cxx::now_ms_str() under the terms of 3-clause BSD license of Replxx.
 /// Copyright (c) 2017-2018, Marcin Konarski (amok at codestation.org)
 /// Copyright (c) 2010, Salvatore Sanfilippo (antirez at gmail dot com)
@ -142,6 +305,7 @@ ReplxxLineReader::ReplxxLineReader(
    replxx::Replxx::highlighter_callback_t highlighter_)
    : LineReader(history_file_path_, multiline_, std::move(extenders_), std::move(delimiters_)), highlighter(std::move(highlighter_))
    , editor(getEditor())
+    , fuzzy_finder(getFuzzyFinder())
 {
    using namespace std::placeholders;
    using Replxx = replxx::Replxx;
@ -249,6 +413,17 @@ ReplxxLineReader::ReplxxLineReader(
        return rx.invoke(Replxx::ACTION::COMMIT_LINE, code);
    };
    rx.bind_key(Replxx::KEY::meta('#'), insert_comment_action);
+
+    /// interactive search in history (requires fzf/sk)
+    if (!fuzzy_finder.empty())
+    {
+        auto interactive_history_search = [this](char32_t code)
+        {
+            openInteractiveHistorySearch();
+            return rx.invoke(Replxx::ACTION::REPAINT, code);
+        };
+        rx.bind_key(Replxx::KEY::control('R'), interactive_history_search);
+    }
 }

 ReplxxLineReader::~ReplxxLineReader()
@ -293,116 +468,70 @@ void ReplxxLineReader::addToHistory(const String & line)
        rx.print("Unlock of history file failed: %s\n", errnoToString().c_str());
 }

-/// See comments in ShellCommand::executeImpl()
-/// (for the vfork via dlsym())
-int ReplxxLineReader::executeEditor(const std::string & path)
-{
-    std::vector<char> argv0(editor.data(), editor.data() + editor.size() + 1);
-    std::vector<char> argv1(path.data(), path.data() + path.size() + 1);
-    char * const argv[] = {argv0.data(), argv1.data(), nullptr};
-
-    static void * real_vfork = dlsym(RTLD_DEFAULT, "vfork");
-    if (!real_vfork)
-    {
-        rx.print("Cannot find symbol vfork in myself: %s\n", errnoToString().c_str());
-        return -1;
-    }
-
-    pid_t pid = reinterpret_cast<pid_t (*)()>(real_vfork)();
-
-    if (-1 == pid)
-    {
-        rx.print("Cannot vfork: %s\n", errnoToString().c_str());
-        return -1;
-    }
-
-    /// Child
-    if (0 == pid)
-    {
-        sigset_t mask;
-        sigemptyset(&mask);
-        sigprocmask(0, nullptr, &mask); // NOLINT(concurrency-mt-unsafe) // ok in newly created process
-        sigprocmask(SIG_UNBLOCK, &mask, nullptr); // NOLINT(concurrency-mt-unsafe) // ok in newly created process
-
-        execvp(editor.c_str(), argv);
-        rx.print("Cannot execute %s: %s\n", editor.c_str(), errnoToString().c_str());
-        _exit(-1);
-    }
-
-    int status = 0;
-    do
-    {
-        int exited_pid = waitpid(pid, &status, 0);
-        if (exited_pid == -1)
-        {
-            if (errno == EINTR)
-                continue;
-
-            rx.print("Cannot waitpid: %s\n", errnoToString().c_str());
-            return -1;
-        }
-        else
-            break;
-    } while (true);
-    return status;
-}
-
 void ReplxxLineReader::openEditor()
 {
-    char filename[] = "clickhouse_replxx_XXXXXX.sql";
-    int fd = ::mkstemps(filename, 4);
-    if (-1 == fd)
-    {
-        rx.print("Cannot create temporary file to edit query: %s\n", errnoToString().c_str());
-        return;
-    }
+    TemporaryFile editor_file("clickhouse_client_editor_XXXXXX.sql");
+    editor_file.write(rx.get_state().text());
+    editor_file.close();

-    replxx::Replxx::State state(rx.get_state());
-
-    size_t bytes_written = 0;
-    const char * begin = state.text();
-    size_t offset = strlen(state.text());
-    while (bytes_written != offset)
+    char * const argv[] = {editor.data(), editor_file.getPath().data(), nullptr};
+    try
    {
-        ssize_t res = ::write(fd, begin + bytes_written, offset - bytes_written);
-        if ((-1 == res || 0 == res) && errno != EINTR)
+        if (executeCommand(argv) == 0)
        {
-            rx.print("Cannot write to temporary query file %s: %s\n", filename, errnoToString().c_str());
-            break;
+            const std::string & new_query = readFile(editor_file.getPath());
+            rx.set_state(replxx::Replxx::State(new_query.c_str(), new_query.size()));
        }
-        bytes_written += res;
    }
-
-    if (0 != ::close(fd))
+    catch (const std::runtime_error & e)
    {
-        rx.print("Cannot close temporary query file %s: %s\n", filename, errnoToString().c_str());
-        return;
-    }
-
-    if (0 == executeEditor(filename))
-    {
-        try
-        {
-            std::ifstream t(filename);
-            std::string str;
-            t.seekg(0, std::ios::end);
-            str.reserve(t.tellg());
-            t.seekg(0, std::ios::beg);
-            str.assign((std::istreambuf_iterator<char>(t)), std::istreambuf_iterator<char>());
-            rx.set_state(replxx::Replxx::State(str.c_str(), str.size()));
-        }
-        catch (...)
-        {
-            rx.print("Cannot read from temporary query file %s: %s\n", filename, errnoToString().c_str());
-            return;
-        }
+        rx.print(e.what());
    }

    if (bracketed_paste_enabled)
        enableBracketedPaste();
+}

-    if (0 != ::unlink(filename))
-        rx.print("Cannot remove temporary query file %s: %s\n", filename, errnoToString().c_str());
+void ReplxxLineReader::openInteractiveHistorySearch()
+{
+    assert(!fuzzy_finder.empty());
+    TemporaryFile history_file("clickhouse_client_history_in_XXXXXX.bin");
+    auto hs(rx.history_scan());
+    while (hs.next())
+    {
+        history_file.write(hs.get().text());
+        history_file.write(std::string(1, '\0'));
+    }
+    history_file.close();
+
+    TemporaryFile output_file("clickhouse_client_history_out_XXXXXX.sql");
+    output_file.close();
+
+    char sh[] = "sh";
+    char sh_c[] = "-c";
+    /// NOTE: You can use one of the following to configure the behaviour additionally:
+    /// - SKIM_DEFAULT_OPTIONS
+    /// - FZF_DEFAULT_OPTS
+    std::string fuzzy_finder_command = fmt::format(
+        "{} --read0 --tac --no-sort --tiebreak=index --bind=ctrl-r:toggle-sort --height=30% < {} > {}",
+        fuzzy_finder, history_file.getPath(), output_file.getPath());
+    char * const argv[] = {sh, sh_c, fuzzy_finder_command.data(), nullptr};
+
+    try
+    {
+        if (executeCommand(argv) == 0)
+        {
+            const std::string & new_query = readFile(output_file.getPath());
+            rx.set_state(replxx::Replxx::State(new_query.c_str(), new_query.size()));
+        }
+    }
+    catch (const std::runtime_error & e)
+    {
+        rx.print(e.what());
+    }
+
+    if (bracketed_paste_enabled)
+        enableBracketedPaste();
 }

 void ReplxxLineReader::enableBracketedPaste()
--- a/base/base/ReplxxLineReader.h
+++ b/base/base/ReplxxLineReader.h
@ -27,6 +27,7 @@ private:
    void addToHistory(const String & line) override;
    int executeEditor(const std::string & path);
    void openEditor();
+    void openInteractiveHistorySearch();

    replxx::Replxx rx;
    replxx::Replxx::highlighter_callback_t highlighter;
@ -36,4 +37,5 @@ private:
    bool bracketed_paste_enabled = false;

    std::string editor;
+    std::string fuzzy_finder;
 };
--- a/cmake/cpu_features.cmake
+++ b/cmake/cpu_features.cmake
@ -45,6 +45,8 @@ elseif (ARCH_AARCH64)
        # dotprod: Scalar vector product (SDOT and UDOT instructions). Probably the most obscure extra flag with doubtful performance benefits
        #          but it has been activated since always, so why not enable it. It's not 100% clear in which revision this flag was
        #          introduced as optional, either in v8.2 [7] or in v8.4 [8].
+        # ldapr:   Load-Acquire RCpc Register. Better support of release/acquire of atomics. Good for allocators and high contention code.
+        #          Optional in v8.2, mandatory in v8.3 [9]. Supported in Graviton 2+, Azure and GCP instances. Generated from clang 15.
        #
        # [1] https://github.com/aws/aws-graviton-getting-started/blob/main/c-c%2B%2B.md
        # [2] https://community.arm.com/arm-community-blogs/b/tools-software-ides-blog/posts/making-the-most-of-the-arm-architecture-in-gcc-10
@ -54,7 +56,8 @@ elseif (ARCH_AARCH64)
        # [6] https://developer.arm.com/documentation/100067/0612/armclang-Command-line-Options/-mcpu?lang=en
        # [7] https://gcc.gnu.org/onlinedocs/gcc/ARM-Options.html
        # [8] https://developer.arm.com/documentation/102651/a/What-are-dot-product-intructions-
-        set (COMPILER_FLAGS "${COMPILER_FLAGS} -march=armv8.2-a+simd+crypto+dotprod+ssbs")
+        # [9] https://developer.arm.com/documentation/dui0801/g/A64-Data-Transfer-Instructions/LDAPR?lang=en
+        set (COMPILER_FLAGS "${COMPILER_FLAGS} -march=armv8.2-a+simd+crypto+dotprod+ssbs -Xclang=-target-feature -Xclang=+ldapr -Wno-unused-command-line-argument")
    endif ()

 elseif (ARCH_PPC64LE)
--- a/cmake/git.cmake
+++ b/cmake/git.cmake
@ -0,0 +1,42 @@
+find_package(Git)
+
+# Make basic Git information available as variables. Such data will later be embedded into the build, e.g. for view SYSTEM.BUILD_OPTIONS.
+if (Git_FOUND)
+  # Commit hash + whether the building workspace was dirty or not
+  execute_process(COMMAND
+    "${GIT_EXECUTABLE}" rev-parse HEAD
+    WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
+    OUTPUT_VARIABLE GIT_HASH
+    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
+
+  # Branch name
+  execute_process(COMMAND
+    "${GIT_EXECUTABLE}" rev-parse --abbrev-ref HEAD
+    WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
+    OUTPUT_VARIABLE GIT_BRANCH
+    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
+
+  # Date of the commit
+  SET(ENV{TZ} "UTC")
+  execute_process(COMMAND
+    "${GIT_EXECUTABLE}" log -1 --format=%ad --date=iso-local
+    WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
+    OUTPUT_VARIABLE GIT_DATE
+    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
+
+  # Subject of the commit
+  execute_process(COMMAND
+    "${GIT_EXECUTABLE}" log -1 --format=%s
+    WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
+    OUTPUT_VARIABLE GIT_COMMIT_SUBJECT
+    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
+
+  message(STATUS "Git HEAD commit hash: ${GIT_HASH}")
+
+  execute_process(
+    COMMAND ${GIT_EXECUTABLE} status
+    WORKING_DIRECTORY ${CMAKE_SOURCE_DIR} OUTPUT_STRIP_TRAILING_WHITESPACE)
+else()
+  message(STATUS "Git could not be found.")
+endif()
+
--- a/cmake/git_status.cmake
+++ b/cmake/git_status.cmake
@ -1,22 +0,0 @@
-# Print the status of the git repository (if git is available).
-# This is useful for troubleshooting build failure reports
-
-find_package(Git)
-
-if (Git_FOUND)
-
-  execute_process(
-    COMMAND ${GIT_EXECUTABLE} rev-parse HEAD
-    WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
-    OUTPUT_VARIABLE GIT_COMMIT_ID
-    OUTPUT_STRIP_TRAILING_WHITESPACE)
-
-  message(STATUS "HEAD's commit hash ${GIT_COMMIT_ID}")
-
-  execute_process(
-    COMMAND ${GIT_EXECUTABLE} status
-    WORKING_DIRECTORY ${CMAKE_SOURCE_DIR} OUTPUT_STRIP_TRAILING_WHITESPACE)
-
-else()
-  message(STATUS "Git could not be found.")
-endif()
--- a/contrib/CMakeLists.txt
+++ b/contrib/CMakeLists.txt
@ -107,7 +107,7 @@ if (ENABLE_TESTS)
    add_contrib (googletest-cmake googletest)
 endif()

-add_contrib (llvm-cmake llvm)
+add_contrib (llvm-project-cmake llvm-project)
 add_contrib (libxml2-cmake libxml2)
 add_contrib (aws-s3-cmake
    aws
--- a/contrib/llvm
+++ b/contrib/llvm
@ -1 +0,0 @@
-Subproject commit 0db5bf5bd2452cd8f1283a1fcdc04845af705bfc
--- a/contrib/llvm-cmake/CMakeLists.txt
+++ b/contrib/llvm-cmake/CMakeLists.txt
@ -1,112 +0,0 @@
-if (APPLE OR NOT ARCH_AMD64 OR SANITIZE STREQUAL "undefined")
-   set (ENABLE_EMBEDDED_COMPILER_DEFAULT OFF)
-else()
-   set (ENABLE_EMBEDDED_COMPILER_DEFAULT ON)
-endif()
-
-option (ENABLE_EMBEDDED_COMPILER "Enable support for 'compile_expressions' option for query execution" ${ENABLE_EMBEDDED_COMPILER_DEFAULT})
-
-if (NOT ENABLE_EMBEDDED_COMPILER)
-    message(STATUS "Not using LLVM")
-    return()
-endif()
-
-set (LLVM_FOUND 1)
-set (LLVM_VERSION "12.0.0bundled")
-set (LLVM_INCLUDE_DIRS
-    "${ClickHouse_SOURCE_DIR}/contrib/llvm/llvm/include"
-    "${ClickHouse_BINARY_DIR}/contrib/llvm/llvm/include"
-)
-set (LLVM_LIBRARY_DIRS "${ClickHouse_BINARY_DIR}/contrib/llvm/llvm")
-
-# This list was generated by listing all LLVM libraries, compiling the binary and removing all libraries while it still compiles.
-set (REQUIRED_LLVM_LIBRARIES
-    LLVMExecutionEngine
-    LLVMRuntimeDyld
-    LLVMAsmPrinter
-    LLVMDebugInfoDWARF
-    LLVMGlobalISel
-    LLVMSelectionDAG
-    LLVMMCDisassembler
-    LLVMPasses
-    LLVMCodeGen
-    LLVMipo
-    LLVMBitWriter
-    LLVMInstrumentation
-    LLVMScalarOpts
-    LLVMAggressiveInstCombine
-    LLVMInstCombine
-    LLVMVectorize
-    LLVMTransformUtils
-    LLVMTarget
-    LLVMAnalysis
-    LLVMProfileData
-    LLVMObject
-    LLVMBitReader
-    LLVMCore
-    LLVMRemarks
-    LLVMBitstreamReader
-    LLVMMCParser
-    LLVMMC
-    LLVMBinaryFormat
-    LLVMDebugInfoCodeView
-    LLVMSupport
-    LLVMDemangle
-)
-
-if (ARCH_AMD64)
-    list(APPEND REQUIRED_LLVM_LIBRARIES LLVMX86Info LLVMX86Desc LLVMX86CodeGen)
-elseif (ARCH_AARCH64)
-    list(APPEND REQUIRED_LLVM_LIBRARIES LLVMAArch64Info LLVMAArch64Desc LLVMAArch64CodeGen)
-endif ()
-
-#function(llvm_libs_all REQUIRED_LLVM_LIBRARIES)
-#    llvm_map_components_to_libnames (result all)
-#    if (USE_STATIC_LIBRARIES OR NOT "LLVM" IN_LIST result)
-#        list (REMOVE_ITEM result "LTO" "LLVM")
-#    else()
-#        set (result "LLVM")
-#    endif ()
-#    list (APPEND result ${CMAKE_DL_LIBS} ch_contrib::zlib)
-#    set (${REQUIRED_LLVM_LIBRARIES} ${result} PARENT_SCOPE)
-#endfunction()
-
-message (STATUS "LLVM include Directory: ${LLVM_INCLUDE_DIRS}")
-message (STATUS "LLVM library Directory: ${LLVM_LIBRARY_DIRS}")
-message (STATUS "LLVM C++ compiler flags: ${LLVM_CXXFLAGS}")
-
-# ld: unknown option: --color-diagnostics
-set (LINKER_SUPPORTS_COLOR_DIAGNOSTICS 0 CACHE INTERNAL "")
-
-# Do not adjust RPATH in llvm, since then it will not be able to find libcxx/libcxxabi/libunwind
-set (CMAKE_INSTALL_RPATH "ON")
-set (LLVM_COMPILER_CHECKED 1 CACHE INTERNAL "")
-set (LLVM_ENABLE_EH 1 CACHE INTERNAL "")
-set (LLVM_ENABLE_RTTI 1 CACHE INTERNAL "")
-set (LLVM_ENABLE_PIC 0 CACHE INTERNAL "")
-set (LLVM_TARGETS_TO_BUILD "X86;AArch64" CACHE STRING "")
-
-# Need to use C++17 since the compilation is not possible with C++20 currently, due to ambiguous operator != etc.
-# LLVM project will set its default value for the -std=... but our global setting from CMake will override it.
-set (CMAKE_CXX_STANDARD 17)
-
-set (LLVM_SOURCE_DIR "${ClickHouse_SOURCE_DIR}/contrib/llvm/llvm")
-set (LLVM_BINARY_DIR "${ClickHouse_BINARY_DIR}/contrib/llvm/llvm")
-add_subdirectory ("${LLVM_SOURCE_DIR}" "${LLVM_BINARY_DIR}")
-set_directory_properties (PROPERTIES
-   # due to llvm crosscompile cmake does not know how to clean it, and on clean
-   # will lead to the following error:
-   #
-   #   ninja: error: remove(contrib/llvm/llvm/NATIVE): Directory not empty
-   #
-   ADDITIONAL_CLEAN_FILES "${LLVM_BINARY_DIR}"
-   # llvm's cmake configuring this file only when cmake runs,
-   # and after clean cmake will not know that it should re-run,
-   # add explicitly depends from llvm-config.h
-   CMAKE_CONFIGURE_DEPENDS "${LLVM_BINARY_DIR}/include/llvm/Config/llvm-config.h"
-)
-
-add_library (_llvm INTERFACE)
-target_link_libraries (_llvm INTERFACE ${REQUIRED_LLVM_LIBRARIES})
-target_include_directories (_llvm SYSTEM BEFORE INTERFACE ${LLVM_INCLUDE_DIRS})
-add_library(ch_contrib::llvm ALIAS _llvm)
--- a/contrib/llvm-project
+++ b/contrib/llvm-project
@ -0,0 +1 @@
+Subproject commit 6ca2b5b3927226f6bcf6c656f502ff5d012ad9b6
--- a/contrib/llvm-project-cmake/CMakeLists.txt
+++ b/contrib/llvm-project-cmake/CMakeLists.txt
@ -0,0 +1,122 @@
+if (APPLE OR NOT ARCH_AMD64 OR SANITIZE STREQUAL "undefined" OR NOT USE_STATIC_LIBRARIES)
+   set (ENABLE_EMBEDDED_COMPILER_DEFAULT OFF)
+else()
+   set (ENABLE_EMBEDDED_COMPILER_DEFAULT ON)
+endif()
+
+option (ENABLE_EMBEDDED_COMPILER "Enable support for 'compile_expressions' option for query execution" ${ENABLE_EMBEDDED_COMPILER_DEFAULT})
+
+if (NOT ENABLE_EMBEDDED_COMPILER)
+    message(STATUS "Not using LLVM")
+    return()
+endif()
+
+# TODO: Enable shared library build
+# TODO: Enable compilation on AArch64
+
+set (LLVM_VERSION "13.0.0bundled")
+set (LLVM_INCLUDE_DIRS
+    "${ClickHouse_SOURCE_DIR}/contrib/llvm-project/llvm/include"
+    "${ClickHouse_BINARY_DIR}/contrib/llvm-project/llvm/include"
+)
+set (LLVM_LIBRARY_DIRS "${ClickHouse_BINARY_DIR}/contrib/llvm-project/llvm")
+
+# This list was generated by listing all LLVM libraries, compiling the binary and removing all libraries while it still compiles.
+set (REQUIRED_LLVM_LIBRARIES
+    LLVMExecutionEngine
+    LLVMRuntimeDyld
+    LLVMAsmPrinter
+    LLVMDebugInfoDWARF
+    LLVMGlobalISel
+    LLVMSelectionDAG
+    LLVMMCDisassembler
+    LLVMPasses
+    LLVMCodeGen
+    LLVMipo
+    LLVMBitWriter
+    LLVMInstrumentation
+    LLVMScalarOpts
+    LLVMAggressiveInstCombine
+    LLVMInstCombine
+    LLVMVectorize
+    LLVMTransformUtils
+    LLVMTarget
+    LLVMAnalysis
+    LLVMProfileData
+    LLVMObject
+    LLVMBitReader
+    LLVMCore
+    LLVMRemarks
+    LLVMBitstreamReader
+    LLVMMCParser
+    LLVMMC
+    LLVMBinaryFormat
+    LLVMDebugInfoCodeView
+    LLVMSupport
+    LLVMDemangle
+)
+
+# if (ARCH_AMD64)
+    list(APPEND REQUIRED_LLVM_LIBRARIES LLVMX86Info LLVMX86Desc LLVMX86CodeGen)
+# elseif (ARCH_AARCH64)
+#     list(APPEND REQUIRED_LLVM_LIBRARIES LLVMAArch64Info LLVMAArch64Desc LLVMAArch64CodeGen)
+# endif ()
+
+# ld: unknown option: --color-diagnostics
+# set (LINKER_SUPPORTS_COLOR_DIAGNOSTICS 0 CACHE INTERNAL "")
+
+set (CMAKE_INSTALL_RPATH "ON") # Do not adjust RPATH in llvm, since then it will not be able to find libcxx/libcxxabi/libunwind
+set (LLVM_COMPILER_CHECKED 1 CACHE INTERNAL "") # Skip internal compiler selection
+set (LLVM_ENABLE_EH 1 CACHE INTERNAL "") # With exception handling
+set (LLVM_ENABLE_RTTI 1 CACHE INTERNAL "")
+set (LLVM_ENABLE_PIC 0 CACHE INTERNAL "")
+set (LLVM_TARGETS_TO_BUILD "X86" CACHE STRING "") # for x86 + ARM: "X86;AArch64"
+
+# Omit unnecessary stuff (just the options which are ON by default)
+set(LLVM_ENABLE_BACKTRACES 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_CRASH_OVERRIDES 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_TERMINFO 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_LIBXML2 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_LIBEDIT 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_LIBPFM 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_ZLIB 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_Z3_SOLVER 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_TOOLS 0 CACHE INTERNAL "")
+set(LLVM_BUILD_TOOLS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_UTILS 0 CACHE INTERNAL "")
+set(LLVM_BUILD_UTILS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_RUNTIMES 0 CACHE INTERNAL "")
+set(LLVM_BUILD_RUNTIMES 0 CACHE INTERNAL "")
+set(LLVM_BUILD_RUNTIME 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_EXAMPLES 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_TESTS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_GO_TESTS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_BENCHMARKS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_DOCS 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_OCAMLDOC 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_BINDINGS 0 CACHE INTERNAL "")
+
+# C++20 is currently not supported due to ambiguous operator != etc.
+set (CMAKE_CXX_STANDARD 17)
+
+set (LLVM_SOURCE_DIR "${ClickHouse_SOURCE_DIR}/contrib/llvm-project/llvm")
+set (LLVM_BINARY_DIR "${ClickHouse_BINARY_DIR}/contrib/llvm-project/llvm")
+add_subdirectory ("${LLVM_SOURCE_DIR}" "${LLVM_BINARY_DIR}")
+
+set_directory_properties (PROPERTIES
+   # due to llvm crosscompile cmake does not know how to clean it, and on clean
+   # will lead to the following error:
+   #
+   #   ninja: error: remove(contrib/llvm/llvm/NATIVE): Directory not empty
+   #
+   ADDITIONAL_CLEAN_FILES "${LLVM_BINARY_DIR}"
+   # llvm's cmake configuring this file only when cmake runs,
+   # and after clean cmake will not know that it should re-run,
+   # add explicitly depends from llvm-config.h
+   CMAKE_CONFIGURE_DEPENDS "${LLVM_BINARY_DIR}/include/llvm/Config/llvm-config.h"
+)
+
+add_library (_llvm INTERFACE)
+target_link_libraries (_llvm INTERFACE ${REQUIRED_LLVM_LIBRARIES})
+target_include_directories (_llvm SYSTEM BEFORE INTERFACE ${LLVM_INCLUDE_DIRS})
+add_library(ch_contrib::llvm ALIAS _llvm)
--- a/docs/_includes/install/universal.sh
+++ b/docs/_includes/install/universal.sh
@ -15,7 +15,7 @@ then
        # If the system has >=ARMv8.2 (https://en.wikipedia.org/wiki/AArch64), choose the corresponding build, else fall back to a v8.0
        # compat build. Unfortunately, the ARM ISA level cannot be read directly, we need to guess from the "features" in /proc/cpuinfo.
        # Also, the flags in /proc/cpuinfo are named differently than the flags passed to the compiler (cmake/cpu_features.cmake).
-        ARMV82=$(grep -m 1 'Features' /proc/cpuinfo | awk '/asimd/ && /sha1/ && /aes/ && /atomics/')
+        ARMV82=$(grep -m 1 'Features' /proc/cpuinfo | awk '/asimd/ && /sha1/ && /aes/ && /atomics/ && /lrcpc/')
        if [ "${ARMV82}" ]
        then
            DIR="aarch64"
--- a/docs/en/operations/settings/merge-tree-settings.md
+++ b/docs/en/operations/settings/merge-tree-settings.md
@ -15,7 +15,7 @@ Possible values:

 -   Any positive integer.

-Default value: 10.
+Default value: 100.

 Override example in `config.xml`:

@ -231,7 +231,7 @@ Possible values:

 -   Any positive integer.

-Default value: 1800
+Default value: 10800

 ## try_fetch_recompressed_part_timeout

@ -261,7 +261,7 @@ Possible values:

 -   Any positive integer.

-Default value: 10
+Default value: 100

 ## max_suspicious_broken_parts_bytes

--- a/docs/en/sql-reference/aggregate-functions/reference/anyheavy.md
+++ b/docs/en/sql-reference/aggregate-functions/reference/anyheavy.md
@ -5,7 +5,7 @@ sidebar_position: 103

 # anyHeavy

-Selects a frequently occurring value using the [heavy hitters](http://www.cs.umd.edu/~samir/498/karp.pdf) algorithm. If there is a value that occurs more than in half the cases in each of the query’s execution threads, this value is returned. Normally, the result is nondeterministic.
+Selects a frequently occurring value using the [heavy hitters](https://doi.org/10.1145/762471.762473) algorithm. If there is a value that occurs more than in half the cases in each of the query’s execution threads, this value is returned. Normally, the result is nondeterministic.

 ``` sql
 anyHeavy(column)
--- a/docs/en/sql-reference/aggregate-functions/reference/topk.md
+++ b/docs/en/sql-reference/aggregate-functions/reference/topk.md
@ -7,7 +7,7 @@ sidebar_position: 108

 Returns an array of the approximately most frequent values in the specified column. The resulting array is sorted in descending order of approximate frequency of values (not by the values themselves).

-Implements the [Filtered Space-Saving](http://www.l2f.inesc-id.pt/~fmmb/wiki/uploads/Work/misnis.ref0a.pdf) algorithm for analyzing TopK, based on the reduce-and-combine algorithm from [Parallel Space Saving](https://arxiv.org/pdf/1401.0702.pdf).
+Implements the [Filtered Space-Saving](https://doi.org/10.1016/j.ins.2010.08.024) algorithm for analyzing TopK, based on the reduce-and-combine algorithm from [Parallel Space Saving](https://doi.org/10.1016/j.ins.2015.09.003).

 ``` sql
 topK(N)(column)
--- a/docs/en/sql-reference/functions/other-functions.md
+++ b/docs/en/sql-reference/functions/other-functions.md
@ -1818,11 +1818,6 @@ Result:
 └──────────────────────────────────────────────────┘
 ```

-## modelEvaluate(model_name, …)
-
-Evaluate external model.
-Accepts a model name and model arguments. Returns Float64.
-
 ## catboostEvaluate(path_to_model, feature_1, feature_2, …, feature_n)

 Evaluate external catboost model. [CatBoost](https://catboost.ai) is an open-source gradient boosting library developed by Yandex for machine learing.
--- a/docs/en/sql-reference/functions/string-functions.md
+++ b/docs/en/sql-reference/functions/string-functions.md
@ -565,6 +565,10 @@ Result:
 └────────────────────────────┘
 ```

+## tryBase58Decode(s)
+
+Similar to base58Decode, but returns an empty string in case of error.
+
 ## base64Encode(s)

 Encodes ‘s’ string into base64
@ -579,7 +583,7 @@ Alias: `FROM_BASE64`.

 ## tryBase64Decode(s)

-Similar to base64Decode, but in case of error an empty string would be returned.
+Similar to base64Decode, but returns an empty string in case of error.

 ## endsWith(s, suffix)

--- a/docs/ru/sql-reference/aggregate-functions/reference/anyheavy.md
+++ b/docs/ru/sql-reference/aggregate-functions/reference/anyheavy.md
@ -5,7 +5,7 @@ sidebar_position: 103

 # anyHeavy {#anyheavyx}

-Выбирает часто встречающееся значение с помощью алгоритма «[heavy hitters](http://www.cs.umd.edu/~samir/498/karp.pdf)». Если существует значение, которое встречается чаще, чем в половине случаев, в каждом потоке выполнения запроса, то возвращается данное значение. В общем случае, результат недетерминирован.
+Выбирает часто встречающееся значение с помощью алгоритма «[heavy hitters](https://doi.org/10.1145/762471.762473)». Если существует значение, которое встречается чаще, чем в половине случаев, в каждом потоке выполнения запроса, то возвращается данное значение. В общем случае, результат недетерминирован.

 ``` sql
 anyHeavy(column)
--- a/docs/ru/sql-reference/aggregate-functions/reference/topk.md
+++ b/docs/ru/sql-reference/aggregate-functions/reference/topk.md
@ -7,7 +7,7 @@ sidebar_position: 108

 Возвращает массив наиболее часто встречающихся значений в указанном столбце. Результирующий массив упорядочен по убыванию частоты значения (не по самим значениям).

-Реализует [Filtered Space-Saving](http://www.l2f.inesc-id.pt/~fmmb/wiki/uploads/Work/misnis.ref0a.pdf) алгоритм для анализа TopK, на основе reduce-and-combine алгоритма из методики [Parallel Space Saving](https://arxiv.org/pdf/1401.0702.pdf).
+Реализует [Filtered Space-Saving](https://doi.org/10.1016/j.ins.2010.08.024) алгоритм для анализа TopK, на основе reduce-and-combine алгоритма из методики [Parallel Space Saving](https://doi.org/10.1016/j.ins.2015.09.003).

 ``` sql
 topK(N)(column)
--- a/docs/ru/sql-reference/functions/other-functions.md
+++ b/docs/ru/sql-reference/functions/other-functions.md
@ -1722,12 +1722,6 @@ SELECT joinGet(db_test.id_val,'val',toUInt32(number)) from numbers(4) SETTINGS j
 └──────────────────────────────────────────────────┘
 ```

-## modelEvaluate(model_name, …) {#function-modelevaluate}
-
-Оценивает внешнюю модель.
-
-Принимает на вход имя и аргументы модели. Возвращает Float64.
-
 ## throwIf(x\[, message\[, error_code\]\]) {#throwifx-custom-message}

 Бросает исключение, если аргумент не равен нулю.
--- a/docs/ru/sql-reference/statements/select/from.md
+++ b/docs/ru/sql-reference/statements/select/from.md
@ -32,7 +32,7 @@ sidebar_label: FROM

 Запросы, которые используют `FINAL` выполняются немного медленее, чем аналогичные запросы без него, потому что:

-   Данные мёржатся во время выполнения запроса.
+-   Данные мёржатся во время выполнения запроса в памяти, и это не приводит к физическому мёржу кусков на дисках.
 -   Запросы с модификатором `FINAL` читают столбцы первичного ключа в дополнение к столбцам, используемым в запросе.

 **В большинстве случаев избегайте использования `FINAL`.** Общий подход заключается в использовании агрегирующих запросов, которые предполагают, что фоновые процессы движков семейства `MergeTree` ещё не случились  (например, сами отбрасывают дубликаты). {## TODO: examples ##}
--- a/docs/zh/sql-reference/aggregate-functions/reference/anyheavy.md
+++ b/docs/zh/sql-reference/aggregate-functions/reference/anyheavy.md
@ -5,7 +5,7 @@ sidebar_position: 103

 # anyHeavy {#anyheavyx}

-选择一个频繁出现的值，使用[heavy hitters](http://www.cs.umd.edu/~samir/498/karp.pdf) 算法。 如果某个值在查询的每个执行线程中出现的情况超过一半，则返回此值。 通常情况下，结果是不确定的。
+选择一个频繁出现的值，使用[heavy hitters](https://doi.org/10.1145/762471.762473) 算法。 如果某个值在查询的每个执行线程中出现的情况超过一半，则返回此值。 通常情况下，结果是不确定的。

 ``` sql
 anyHeavy(column)
--- a/docs/zh/sql-reference/aggregate-functions/reference/topk.md
+++ b/docs/zh/sql-reference/aggregate-functions/reference/topk.md
@ -7,7 +7,7 @@ sidebar_position: 108

 返回指定列中近似最常见值的数组。 生成的数组按值的近似频率降序排序（而不是值本身）。

-实现了[过滤节省空间](http://www.l2f.inesc-id.pt/~fmmb/wiki/uploads/Work/misnis.ref0a.pdf)算法， 使用基于reduce-and-combine的算法，借鉴[并行节省空间](https://arxiv.org/pdf/1401.0702.pdf)。
+实现了[过滤节省空间](https://doi.org/10.1016/j.ins.2010.08.024)算法， 使用基于reduce-and-combine的算法，借鉴[并行节省空间](https://doi.org/10.1016/j.ins.2015.09.003)。

 **语法**

--- a/docs/zh/sql-reference/functions/other-functions.md
+++ b/docs/zh/sql-reference/functions/other-functions.md
@ -625,11 +625,6 @@ ORDER BY k ASC

 使用指定的连接键从Join类型引擎的表中获取数据。

-## modelEvaluate(model_name, …) {#function-modelevaluate}
-
-使用外部模型计算。
-接受模型的名称以及模型的参数。返回Float64类型的值。
-
 ## throwIf(x) {#throwifx}

 如果参数不为零则抛出异常。
--- a/programs/client/Client.cpp
+++ b/programs/client/Client.cpp
@ -842,6 +842,7 @@ void Client::addOptions(OptionsDescription & options_description)

        ("no-warnings", "disable warnings when client connects to server")
        ("fake-drop", "Ignore all DROP queries, should be used only for testing")
+        ("accept-invalid-certificate", "Ignore certificate verification errors, equal to config parameters openSSL.client.invalidCertificateHandler.name=AcceptCertificateHandler and openSSL.client.verificationMode=none")
    ;

    /// Commandline options related to external tables.
@ -976,6 +977,13 @@ void Client::processOptions(const OptionsDescription & options_description,
        config().setBool("no-warnings", true);
    if (options.count("fake-drop"))
        fake_drop = true;
+    if (options.count("accept-invalid-certificate"))
+    {
+        config().setString("openSSL.client.invalidCertificateHandler.name", "AcceptCertificateHandler");
+        config().setString("openSSL.client.verificationMode", "none");
+    }
+    else
+        config().setString("openSSL.client.invalidCertificateHandler.name", "RejectCertificateHandler");

    if ((query_fuzzer_runs = options["query-fuzzer-runs"].as<int>()))
    {
--- a/programs/keeper/CMakeLists.txt
+++ b/programs/keeper/CMakeLists.txt
@ -33,7 +33,7 @@ install(FILES keeper_config.xml DESTINATION "${CLICKHOUSE_ETC_DIR}/clickhouse-ke
 add_dependencies(clickhouse-keeper-lib clickhouse_keeper_configs)

 if (BUILD_STANDALONE_KEEPER)
-    # Sraight list of all required sources
+    # Straight list of all required sources
    set(CLICKHOUSE_KEEPER_STANDALONE_SOURCES
        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/ACLMap.cpp
        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/Changelog.cpp
@ -92,6 +92,7 @@ if (BUILD_STANDALONE_KEEPER)
        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Daemon/BaseDaemon.cpp
        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Daemon/SentryWriter.cpp
        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Daemon/GraphiteWriter.cpp
+        ${CMAKE_CURRENT_BINARY_DIR}/../../src/Daemon/GitHash.generated.cpp

        Keeper.cpp
        TinyContext.cpp
--- a/programs/keeper/Keeper.cpp
+++ b/programs/keeper/Keeper.cpp
@ -490,8 +490,9 @@ int Keeper::main(const std::vector<std::string> & /*args*/)
 void Keeper::logRevision() const
 {
    Poco::Logger::root().information("Starting ClickHouse Keeper " + std::string{VERSION_STRING}
-        + " with revision " + std::to_string(ClickHouseRevision::getVersionRevision())
-        + ", " + build_id_info
+        + "(revision : " + std::to_string(ClickHouseRevision::getVersionRevision())
+        + ", git hash: " + (git_hash.empty() ? "<unknown>" : git_hash)
+        + ", build id: " + (build_id.empty() ? "<unknown>" : build_id) + ")"
        + ", PID " + std::to_string(getpid()));
 }

--- a/programs/server/config.d/legacy_geobase.xml
+++ b/programs/server/config.d/legacy_geobase.xml
@ -0,0 +1,4 @@
+<clickhouse>
+    <path_to_regions_hierarchy_file>config.d/regions_hierarchy.txt</path_to_regions_hierarchy_file>
+    <path_to_regions_names_files>config.d/</path_to_regions_names_files>
+</clickhouse>
--- a/programs/server/config.d/regions_hierarchy.txt
+++ b/programs/server/config.d/regions_hierarchy.txt
@ -0,0 +1 @@
+../../../tests/config/regions_hierarchy.txt
--- a/programs/server/config.d/regions_names_en.txt
+++ b/programs/server/config.d/regions_names_en.txt
@ -0,0 +1 @@
+../../../tests/config/regions_names_en.txt
--- a/src/AggregateFunctions/AggregateFunctionAvg.h
+++ b/src/AggregateFunctions/AggregateFunctionAvg.h
@ -164,8 +164,10 @@ public:

        auto * denominator_type = toNativeType<Denominator>(b);
        static constexpr size_t denominator_offset = offsetof(Fraction, denominator);
-        auto * denominator_dst_ptr = b.CreatePointerCast(b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_dst_ptr, denominator_offset), denominator_type->getPointerTo());
-        auto * denominator_src_ptr = b.CreatePointerCast(b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_src_ptr, denominator_offset), denominator_type->getPointerTo());
+        auto * ty_aggregate_data_dst_ptr = llvm::cast<llvm::PointerType>(aggregate_data_dst_ptr->getType()->getScalarType())->getElementType();
+        auto * denominator_dst_ptr = b.CreatePointerCast(b.CreateConstInBoundsGEP1_64(ty_aggregate_data_dst_ptr, aggregate_data_dst_ptr, denominator_offset), denominator_type->getPointerTo());
+        auto * ty_aggregate_data_src_ptr = llvm::cast<llvm::PointerType>(aggregate_data_src_ptr->getType()->getScalarType())->getElementType();
+        auto * denominator_src_ptr = b.CreatePointerCast(b.CreateConstInBoundsGEP1_64(ty_aggregate_data_src_ptr, aggregate_data_src_ptr, denominator_offset), denominator_type->getPointerTo());

        auto * denominator_dst_value = b.CreateLoad(denominator_type, denominator_dst_ptr);
        auto * denominator_src_value = b.CreateLoad(denominator_type, denominator_src_ptr);
@ -184,7 +186,8 @@ public:

        auto * denominator_type = toNativeType<Denominator>(b);
        static constexpr size_t denominator_offset = offsetof(Fraction, denominator);
-        auto * denominator_ptr = b.CreatePointerCast(b.CreateConstGEP1_32(nullptr, aggregate_data_ptr, denominator_offset), denominator_type->getPointerTo());
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * denominator_ptr = b.CreatePointerCast(b.CreateConstGEP1_32(ty_aggregate_data_ptr, aggregate_data_ptr, denominator_offset), denominator_type->getPointerTo());
        auto * denominator_value = b.CreateLoad(denominator_type, denominator_ptr);

        auto * double_numerator = nativeCast<Numerator>(b, numerator_value, b.getDoubleTy());
@ -311,7 +314,8 @@ public:

        auto * denominator_type = toNativeType<Denominator>(b);
        static constexpr size_t denominator_offset = offsetof(Fraction, denominator);
-        auto * denominator_ptr = b.CreatePointerCast(b.CreateConstGEP1_32(nullptr, aggregate_data_ptr, denominator_offset), denominator_type->getPointerTo());
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * denominator_ptr = b.CreatePointerCast(b.CreateConstGEP1_32(ty_aggregate_data_ptr, aggregate_data_ptr, denominator_offset), denominator_type->getPointerTo());
        auto * denominator_value_updated = b.CreateAdd(b.CreateLoad(denominator_type, denominator_ptr), llvm::ConstantInt::get(denominator_type, 1));
        b.CreateStore(denominator_value_updated, denominator_ptr);
    }
--- a/src/AggregateFunctions/AggregateFunctionAvgWeighted.h
+++ b/src/AggregateFunctions/AggregateFunctionAvgWeighted.h
@ -74,7 +74,8 @@ public:
        auto * denominator_type = toNativeType<Denominator>(b);

        static constexpr size_t denominator_offset = offsetof(Fraction, denominator);
-        auto * denominator_offset_ptr = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, denominator_offset);
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * denominator_offset_ptr = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, denominator_offset);
        auto * denominator_ptr = b.CreatePointerCast(denominator_offset_ptr, denominator_type->getPointerTo());

        auto * weight_cast_to_denominator = nativeCast(b, arguments_types[1], argument_values[1], denominator_type);
--- a/src/AggregateFunctions/AggregateFunctionIf.cpp
+++ b/src/AggregateFunctions/AggregateFunctionIf.cpp
@ -207,7 +207,8 @@ public:
        if constexpr (result_is_nullable)
            b.CreateStore(llvm::ConstantInt::get(b.getInt8Ty(), 1), aggregate_data_ptr);

-        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, this->prefix_size);
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, this->prefix_size);
        this->nested_function->compileAdd(b, aggregate_data_ptr_with_prefix_size_offset, { removeNullable(nullable_type) }, { wrapped_value });
        b.CreateBr(join_block);

@ -419,7 +420,8 @@ public:
        if constexpr (result_is_nullable)
            b.CreateStore(llvm::ConstantInt::get(b.getInt8Ty(), 1), aggregate_data_ptr);

-        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, this->prefix_size);
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, this->prefix_size);
        this->nested_function->compileAdd(b, aggregate_data_ptr_with_prefix_size_offset, non_nullable_types, wrapped_values);
        b.CreateBr(join_block);

--- a/src/AggregateFunctions/AggregateFunctionMinMaxAny.h
+++ b/src/AggregateFunctions/AggregateFunctionMinMaxAny.h
@ -201,7 +201,8 @@ public:
        static constexpr size_t value_offset_from_structure = offsetof(SingleValueDataFixed<T>, value);

        auto * type = toNativeType<T>(builder);
-        auto * value_ptr_with_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, value_offset_from_structure);
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * value_ptr_with_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, value_offset_from_structure);
        auto * value_ptr = b.CreatePointerCast(value_ptr_with_offset, type->getPointerTo());

        return value_ptr;
--- a/src/AggregateFunctions/AggregateFunctionNull.h
+++ b/src/AggregateFunctions/AggregateFunctionNull.h
@ -225,7 +225,8 @@ public:
        if constexpr (result_is_nullable)
            b.CreateMemSet(aggregate_data_ptr, llvm::ConstantInt::get(b.getInt8Ty(), 0), this->prefix_size, llvm::assumeAligned(this->alignOfData()));

-        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, this->prefix_size);
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, this->prefix_size);
        this->nested_function->compileCreate(b, aggregate_data_ptr_with_prefix_size_offset);
    }

@ -235,16 +236,25 @@ public:

        if constexpr (result_is_nullable)
        {
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
            auto * aggregate_data_is_null_dst_value = b.CreateLoad(aggregate_data_dst_ptr);
            auto * aggregate_data_is_null_src_value = b.CreateLoad(aggregate_data_src_ptr);
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif

            auto * is_src_null = nativeBoolCast(b, std::make_shared<DataTypeUInt8>(), aggregate_data_is_null_src_value);
            auto * is_null_result_value = b.CreateSelect(is_src_null, llvm::ConstantInt::get(b.getInt8Ty(), 1), aggregate_data_is_null_dst_value);
            b.CreateStore(is_null_result_value, aggregate_data_dst_ptr);
        }

-        auto * aggregate_data_dst_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_dst_ptr, this->prefix_size);
-        auto * aggregate_data_src_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_src_ptr, this->prefix_size);
+        auto * ty_aggregate_data_dst_ptr = llvm::cast<llvm::PointerType>(aggregate_data_dst_ptr->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_dst_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_dst_ptr, aggregate_data_dst_ptr, this->prefix_size);
+        auto * ty_aggregate_data_src_ptr = llvm::cast<llvm::PointerType>(aggregate_data_src_ptr->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_src_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_src_ptr, aggregate_data_src_ptr, this->prefix_size);

        this->nested_function->compileMerge(b, aggregate_data_dst_ptr_with_prefix_size_offset, aggregate_data_src_ptr_with_prefix_size_offset);
    }
@ -278,7 +288,8 @@ public:
            b.CreateBr(join_block);

            b.SetInsertPoint(if_not_null);
-            auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, this->prefix_size);
+            auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+            auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, this->prefix_size);
            auto * nested_result = this->nested_function->compileGetResult(builder, aggregate_data_ptr_with_prefix_size_offset);
            b.CreateStore(b.CreateInsertValue(nullable_value, nested_result, {0}), nullable_value_ptr);
            b.CreateBr(join_block);
@ -374,7 +385,8 @@ public:
        if constexpr (result_is_nullable)
            b.CreateStore(llvm::ConstantInt::get(b.getInt8Ty(), 1), aggregate_data_ptr);

-        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, this->prefix_size);
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, this->prefix_size);
        this->nested_function->compileAdd(b, aggregate_data_ptr_with_prefix_size_offset, { removeNullable(nullable_type) }, { wrapped_value });
        b.CreateBr(join_block);

@ -598,7 +610,8 @@ public:
        if constexpr (result_is_nullable)
            b.CreateStore(llvm::ConstantInt::get(b.getInt8Ty(), 1), aggregate_data_ptr);

-        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_ptr, this->prefix_size);
+        auto * ty_aggregate_data_ptr = llvm::cast<llvm::PointerType>(aggregate_data_ptr->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_ptr_with_prefix_size_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_ptr, aggregate_data_ptr, this->prefix_size);
        this->nested_function->compileAdd(b, aggregate_data_ptr_with_prefix_size_offset, arguments_types, wrapped_values);
        b.CreateBr(join_block);

--- a/src/Common/Base58.cpp
+++ b/src/Common/Base58.cpp
@ -1,4 +1,4 @@
-#include <Common/base58.h>
+#include <Common/Base58.h>


 namespace DB
--- a/src/Common/Base58.h
+++ b/src/Common/Base58.h
@ -1,6 +1,6 @@
 #pragma once

-#include <Core/Types.h>
+#include <base/types.h>
 #include <optional>


--- a/src/Daemon/BaseDaemon.cpp
+++ b/src/Daemon/BaseDaemon.cpp
@ -266,8 +266,8 @@ private:
    {
        size_t pos = message.find('\n');

-        LOG_FATAL(log, "(version {}{}, {}) (from thread {}) {}",
-            VERSION_STRING, VERSION_OFFICIAL, daemon.build_id_info, thread_num, message.substr(0, pos));
+        LOG_FATAL(log, "(version {}{}, build id: {}) (from thread {}) {}",
+            VERSION_STRING, VERSION_OFFICIAL, daemon.build_id, thread_num, message.substr(0, pos));

        /// Print trace from std::terminate exception line-by-line to make it easy for grep.
        while (pos != std::string_view::npos)
@ -315,14 +315,14 @@ private:

        if (query_id.empty())
        {
-            LOG_FATAL(log, "(version {}{}, {}) (from thread {}) (no query) Received signal {} ({})",
-                VERSION_STRING, VERSION_OFFICIAL, daemon.build_id_info,
+            LOG_FATAL(log, "(version {}{}, build id: {}) (from thread {}) (no query) Received signal {} ({})",
+                VERSION_STRING, VERSION_OFFICIAL, daemon.build_id,
                thread_num, strsignal(sig), sig); // NOLINT(concurrency-mt-unsafe) // it is not thread-safe but ok in this context
        }
        else
        {
-            LOG_FATAL(log, "(version {}{}, {}) (from thread {}) (query_id: {}) (query: {}) Received signal {} ({})",
-                VERSION_STRING, VERSION_OFFICIAL, daemon.build_id_info,
+            LOG_FATAL(log, "(version {}{}, build id: {}) (from thread {}) (query_id: {}) (query: {}) Received signal {} ({})",
+                VERSION_STRING, VERSION_OFFICIAL, daemon.build_id,
                thread_num, query_id, query, strsignal(sig), sig); // NOLINT(concurrency-mt-unsafe) // it is not thread-safe but ok in this context)
        }

@ -838,6 +838,7 @@ static void blockSignals(const std::vector<int> & signals)
        throw Poco::Exception("Cannot block signal.");
 }

+extern String getGitHash();

 void BaseDaemon::initializeTerminationAndSignalProcessing()
 {
@ -870,13 +871,15 @@ void BaseDaemon::initializeTerminationAndSignalProcessing()
 #if defined(__ELF__) && !defined(OS_FREEBSD)
    String build_id_hex = DB::SymbolIndex::instance()->getBuildIDHex();
    if (build_id_hex.empty())
-        build_id_info = "no build id";
+        build_id = "";
    else
-        build_id_info = "build id: " + build_id_hex;
+        build_id = build_id_hex;
 #else
-    build_id_info = "no build id";
+    build_id = "";
 #endif

+    git_hash = getGitHash();
+
 #if defined(OS_LINUX)
    std::string executable_path = getExecutablePath();

@ -888,8 +891,9 @@ void BaseDaemon::initializeTerminationAndSignalProcessing()
 void BaseDaemon::logRevision() const
 {
    Poco::Logger::root().information("Starting " + std::string{VERSION_FULL}
-        + " with revision " + std::to_string(ClickHouseRevision::getVersionRevision())
-        + ", " + build_id_info
+        + " (revision: " + std::to_string(ClickHouseRevision::getVersionRevision())
+        + ", git hash: " + (git_hash.empty() ? "<unknown>" : git_hash)
+        + ", build id: " + (build_id.empty() ? "<unknown>" : build_id) + ")"
        + ", PID " + std::to_string(getpid()));
 }

--- a/src/Daemon/BaseDaemon.h
+++ b/src/Daemon/BaseDaemon.h
@ -172,7 +172,8 @@ protected:
    DB::ConfigProcessor::LoadedConfig loaded_config;
    Poco::Util::AbstractConfiguration * last_configuration = nullptr;

-    String build_id_info;
+    String build_id;
+    String git_hash;
    String stored_binary_hash;

    std::vector<int> handled_signals;
--- a/src/Daemon/CMakeLists.txt
+++ b/src/Daemon/CMakeLists.txt
@ -1,7 +1,10 @@
+configure_file(GitHash.cpp.in GitHash.generated.cpp)
+
 add_library (daemon
    BaseDaemon.cpp
    GraphiteWriter.cpp
    SentryWriter.cpp
+    GitHash.generated.cpp
 )

 if (OS_DARWIN AND NOT USE_STATIC_LIBRARIES)
--- a/src/Daemon/GitHash.cpp.in
+++ b/src/Daemon/GitHash.cpp.in
@ -0,0 +1,8 @@
+// File was generated by CMake
+
+#include <base/types.h>
+
+String getGitHash()
+{
+    return "@GIT_HASH@";
+}
--- a/src/Databases/DatabaseReplicated.cpp
+++ b/src/Databases/DatabaseReplicated.cpp
@ -30,6 +30,8 @@
 #include <Common/Macros.h>
 #include <base/chrono_io.h>

+#include <utility>
+
 namespace DB
 {
 namespace ErrorCodes
@ -367,7 +369,7 @@ bool DatabaseReplicated::looksLikeReplicatedDatabasePath(const ZooKeeperPtr & cu
        return false;
    if (maybe_database_mark.starts_with(REPLICATED_DATABASE_MARK))
        return true;
-    if (maybe_database_mark.empty())
+    if (!maybe_database_mark.empty())
        return false;

    /// Old versions did not have REPLICATED_DATABASE_MARK. Check specific nodes exist and add mark.
@ -862,7 +864,19 @@ void DatabaseReplicated::recoverLostReplica(const ZooKeeperPtr & current_zookeep
    for (const auto & id : dropped_tables)
        DatabaseCatalog::instance().waitTableFinallyDropped(id);

-    for (const auto & name_and_meta : table_name_to_metadata)
+    /// FIXME: Use proper dependency calculation instead of just moving MV to the end
+    using NameToMetadata = std::pair<String, String>;
+    std::vector<NameToMetadata> table_name_to_metadata_sorted;
+    table_name_to_metadata_sorted.reserve(table_name_to_metadata.size());
+    std::move(table_name_to_metadata.begin(), table_name_to_metadata.end(), std::back_inserter(table_name_to_metadata_sorted));
+    std::sort(table_name_to_metadata_sorted.begin(), table_name_to_metadata_sorted.end(), [](const NameToMetadata & lhs, const NameToMetadata & rhs) -> bool
+    {
+        const bool is_materialized_view_lhs = lhs.second.find("MATERIALIZED VIEW") != std::string::npos;
+        const bool is_materialized_view_rhs = rhs.second.find("MATERIALIZED VIEW") != std::string::npos;
+        return is_materialized_view_lhs < is_materialized_view_rhs;
+    });
+
+    for (const auto & name_and_meta : table_name_to_metadata_sorted)
    {
        if (isTableExist(name_and_meta.first, getContext()))
        {
--- a/src/Disks/IO/CachedOnDiskReadBufferFromFile.cpp
+++ b/src/Disks/IO/CachedOnDiskReadBufferFromFile.cpp
@ -615,10 +615,7 @@ void CachedOnDiskReadBufferFromFile::predownload(FileSegmentPtr & file_segment)
                }
                else
                {
-                    LOG_TEST(log, "Bypassing cache because writeCache method failed");
-                    read_type = ReadType::REMOTE_FS_READ_BYPASS_CACHE;
-                    file_segment->completeWithState(FileSegment::State::PARTIALLY_DOWNLOADED_NO_CONTINUATION);
-
+                    LOG_TEST(log, "Bypassing cache because writeCache (in predownload) method failed");
                    continue_predownload = false;
                }
            }
--- a/src/Disks/ObjectStorages/DiskObjectStorageTransaction.cpp
+++ b/src/Disks/ObjectStorages/DiskObjectStorageTransaction.cpp
@ -490,7 +490,7 @@ void DiskObjectStorageTransaction::moveFile(const String & from_path, const Stri
                throw Exception("File already exists: " + to_path, ErrorCodes::FILE_ALREADY_EXISTS);

            if (!metadata_storage.exists(from_path))
-                throw Exception(ErrorCodes::FILE_DOESNT_EXIST, "File {} doesn't exist, cannot move", to_path);
+                throw Exception(ErrorCodes::FILE_DOESNT_EXIST, "File {} doesn't exist, cannot move", from_path);

            tx->moveFile(from_path, to_path);
        }));
--- a/src/Formats/FormatFactory.cpp
+++ b/src/Formats/FormatFactory.cpp
@ -100,6 +100,7 @@ FormatSettings getFormatSettings(ContextPtr context, const Settings & settings)
    format_settings.json.try_infer_numbers_from_strings = settings.input_format_json_try_infer_numbers_from_strings;
    format_settings.json.validate_types_from_metadata = settings.input_format_json_validate_types_from_metadata;
    format_settings.json.validate_utf8 = settings.output_format_json_validate_utf8;
+    format_settings.json.try_infer_objects = context->getSettingsRef().allow_experimental_object_type;
    format_settings.null_as_default = settings.input_format_null_as_default;
    format_settings.decimal_trailing_zeros = settings.output_format_decimal_trailing_zeros;
    format_settings.parquet.row_group_size = settings.output_format_parquet_row_group_size;
--- a/src/Formats/FormatSettings.h
+++ b/src/Formats/FormatSettings.h
@ -153,6 +153,7 @@ struct FormatSettings
        bool try_infer_numbers_from_strings = false;
        bool validate_types_from_metadata = true;
        bool validate_utf8 = false;
+        bool try_infer_objects = false;
    } json;

    struct
--- a/src/Formats/JSONUtils.cpp
+++ b/src/Formats/JSONUtils.cpp
@ -225,7 +225,7 @@ namespace JSONUtils
                if (!type)
                    continue;

-                if (isObject(type))
+                if (settings.json.try_infer_objects && isObject(type))
                    return std::make_shared<DataTypeObject>("json", true);

                value_types.push_back(type);
@ -240,7 +240,11 @@ namespace JSONUtils
                are_types_equal &= value_types[i]->equals(*value_types[0]);

            if (!are_types_equal)
+            {
+                if (!settings.json.try_infer_objects)
+                    return nullptr;
                return std::make_shared<DataTypeObject>("json", true);
+            }

            return std::make_shared<DataTypeMap>(std::make_shared<DataTypeString>(), value_types[0]);
        }
--- a/src/Functions/FunctionBase58Conversion.h
+++ b/src/Functions/FunctionBase58Conversion.h
@ -6,7 +6,7 @@
 #include <Functions/FunctionFactory.h>
 #include <Functions/FunctionHelpers.h>
 #include <IO/WriteHelpers.h>
-#include <Common/base58.h>
+#include <Common/Base58.h>
 #include <cstring>


@ -62,9 +62,16 @@ struct Base58Encode
    }
 };

+enum class Base58DecodeErrorHandling
+{
+    ThrowException,
+    ReturnEmptyString
+};
+
+template <typename Name, Base58DecodeErrorHandling ErrorHandling>
 struct Base58Decode
 {
-    static constexpr auto name = "base58Decode";
+    static constexpr auto name = Name::name;

    static void process(const ColumnString & src_column, ColumnString::MutablePtr & dst_column, size_t input_rows_count)
    {
@ -93,7 +100,12 @@ struct Base58Decode
            size_t src_length = current_src_offset - prev_src_offset - 1;
            std::optional<size_t> decoded_size = decodeBase58(&src[prev_src_offset], src_length, &dst[current_dst_offset]);
            if (!decoded_size)
-                throw Exception("Invalid Base58 value, cannot be decoded", ErrorCodes::BAD_ARGUMENTS);
+            {
+                if constexpr (ErrorHandling == Base58DecodeErrorHandling::ThrowException)
+                    throw Exception(ErrorCodes::BAD_ARGUMENTS, "Invalid Base58 value, cannot be decoded");
+                else
+                    decoded_size = 0;
+            }

            prev_src_offset = current_src_offset;
            current_dst_offset += *decoded_size;
@ -113,33 +125,23 @@ class FunctionBase58Conversion : public IFunction
 public:
    static constexpr auto name = Func::name;

-    static FunctionPtr create(ContextPtr)
-    {
-        return std::make_shared<FunctionBase58Conversion>();
-    }
-
-    String getName() const override
-    {
-        return Func::name;
-    }
-
+    static FunctionPtr create(ContextPtr) { return std::make_shared<FunctionBase58Conversion>(); }
+    String getName() const override { return Func::name; }
    size_t getNumberOfArguments() const override { return 1; }
-
    bool isSuitableForShortCircuitArgumentsExecution(const DataTypesWithConstInfo & /*arguments*/) const override { return true; }
-
    bool useDefaultImplementationForConstants() const override { return true; }
-
    ColumnNumbers getArgumentsThatAreAlwaysConstant() const override { return {1}; }

    DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
    {
        if (arguments.size() != 1)
-            throw Exception("Wrong number of arguments for function " + getName() + ":  1 expected.", ErrorCodes::BAD_ARGUMENTS);
+            throw Exception(ErrorCodes::BAD_ARGUMENTS, "Wrong number of arguments for function {}: 1 expected.", getName());

        if (!isString(arguments[0].type))
            throw Exception(
-                "Illegal type " + arguments[0].type->getName() + " of first argument of function " + getName() + ". Must be String.",
-                ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+                ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
+                "Illegal type {} of first argument of function {}. Must be String.",
+                arguments[0].type->getName(), getName());

        return std::make_shared<DataTypeString>();
    }
@ -150,8 +152,9 @@ public:
        const ColumnString * input = checkAndGetColumn<ColumnString>(column_string.get());
        if (!input)
            throw Exception(
-                "Illegal column " + arguments[0].column->getName() + " of first argument of function " + getName() + ", must be String",
-                ErrorCodes::ILLEGAL_COLUMN);
+                ErrorCodes::ILLEGAL_COLUMN,
+                "Illegal column {} of first argument of function {}, must be String",
+                arguments[0].column->getName(), getName());

        auto dst_column = ColumnString::create();

--- a/src/Functions/FunctionBase64Conversion.h
+++ b/src/Functions/FunctionBase64Conversion.h
@ -19,6 +19,7 @@ using namespace GatherUtils;

 namespace ErrorCodes
 {
+    extern const int BAD_ARGUMENTS;
    extern const int ILLEGAL_COLUMN;
    extern const int ILLEGAL_TYPE_OF_ARGUMENT;
    extern const int INCORRECT_DATA;
@ -59,34 +60,22 @@ class FunctionBase64Conversion : public IFunction
 public:
    static constexpr auto name = Func::name;

-    static FunctionPtr create(ContextPtr)
-    {
-        return std::make_shared<FunctionBase64Conversion>();
-    }
-
-    String getName() const override
-    {
-        return Func::name;
-    }
-
-    size_t getNumberOfArguments() const override
-    {
-        return 1;
-    }
-
+    static FunctionPtr create(ContextPtr) { return std::make_shared<FunctionBase64Conversion>(); }
+    String getName() const override { return Func::name; }
+    size_t getNumberOfArguments() const override { return 1; }
    bool isSuitableForShortCircuitArgumentsExecution(const DataTypesWithConstInfo & /*arguments*/) const override { return true; }
-
-    bool useDefaultImplementationForConstants() const override
-    {
-        return true;
-    }
+    bool useDefaultImplementationForConstants() const override { return true; }

    DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
    {
+        if (arguments.size() != 1)
+            throw Exception(ErrorCodes::BAD_ARGUMENTS, "Wrong number of arguments for function {}: 1 expected.", getName());
+
        if (!WhichDataType(arguments[0].type).isString())
            throw Exception(
-                "Illegal type " + arguments[0].type->getName() + " of 1st argument of function " + getName() + ". Must be String.",
-                ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+                ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
+                "Illegal type {} of 1st argument of function {}. Must be String.",
+                arguments[0].type->getName(), getName());

        return std::make_shared<DataTypeString>();
    }
@ -98,8 +87,9 @@ public:

        if (!input)
            throw Exception(
-                "Illegal column " + arguments[0].column->getName() + " of first argument of function " + getName() + ", must be of type String",
-                ErrorCodes::ILLEGAL_COLUMN);
+                ErrorCodes::ILLEGAL_COLUMN,
+                "Illegal column {} of first argument of function {}, must be of type String",
+                arguments[0].column->getName(), getName());

        auto dst_column = ColumnString::create();
        auto & dst_data = dst_column->getChars();
@ -145,7 +135,10 @@ public:
 #endif

                    if (!outlen)
-                        throw Exception("Failed to " + getName() + " input '" + String(reinterpret_cast<const char *>(source), srclen) + "'", ErrorCodes::INCORRECT_DATA);
+                        throw Exception(
+                                ErrorCodes::INCORRECT_DATA,
+                                "Failed to {} input '{}'",
+                                getName(), String(reinterpret_cast<const char *>(source), srclen));
                }
            }
            else
--- a/src/Functions/FunctionsComparison.h
+++ b/src/Functions/FunctionsComparison.h
@ -387,15 +387,38 @@ struct StringEqualsImpl
        size_t size = a_offsets.size();
        ColumnString::Offset prev_a_offset = 0;

-        for (size_t i = 0; i < size; ++i)
+        if (b_size == 0)
        {
-            auto a_size = a_offsets[i] - prev_a_offset - 1;
+            /*
+             * Add the fast path of string comparison if the string constant is empty
+             * and b_size is 0. If a_size is also 0, both of string a and b are empty
+             * string. There is no need to call memequalSmallAllowOverflow15() for
+             * string comparison.
+             */
+            for (size_t i = 0; i < size; ++i)
+            {
+                auto a_size = a_offsets[i] - prev_a_offset - 1;

-            c[i] = positive == memequalSmallAllowOverflow15(
-                a_data.data() + prev_a_offset, a_size,
-                b_data.data(), b_size);
+                if (a_size == 0)
+                    c[i] = positive;
+                else
+                    c[i] = !positive;

-            prev_a_offset = a_offsets[i];
+                prev_a_offset = a_offsets[i];
+            }
+        }
+        else
+        {
+            for (size_t i = 0; i < size; ++i)
+            {
+                auto a_size = a_offsets[i] - prev_a_offset - 1;
+
+                c[i] = positive == memequalSmallAllowOverflow15(
+                    a_data.data() + prev_a_offset, a_size,
+                    b_data.data(), b_size);
+
+                prev_a_offset = a_offsets[i];
+            }
        }
    }

--- a/src/Functions/base58Decode.cpp
+++ b/src/Functions/base58Decode.cpp
@ -0,0 +1,24 @@
+#include <Functions/FunctionBase58Conversion.h>
+#include <Functions/FunctionFactory.h>
+
+namespace DB
+{
+namespace
+{
+
+struct NameBase58Decode
+{
+    static constexpr auto name = "base58Decode";
+};
+
+using Base58DecodeImpl = Base58Decode<NameBase58Decode, Base58DecodeErrorHandling::ThrowException>;
+using FunctionBase58Decode = FunctionBase58Conversion<Base58DecodeImpl>;
+
+}
+
+REGISTER_FUNCTION(Base58Decode)
+{
+    factory.registerFunction<FunctionBase58Decode>();
+}
+
+}
--- a/src/Functions/FunctionsBase58.cpp
+++ b/src/Functions/FunctionsBase58.cpp
@ -7,9 +7,4 @@ REGISTER_FUNCTION(Base58Encode)
 {
    factory.registerFunction<FunctionBase58Conversion<Base58Encode>>();
 }
-
-REGISTER_FUNCTION(Base58Decode)
-{
-    factory.registerFunction<FunctionBase58Conversion<Base58Decode>>();
-}
 }
--- a/src/Functions/tryBase58Decode.cpp
+++ b/src/Functions/tryBase58Decode.cpp
@ -0,0 +1,24 @@
+#include <Functions/FunctionBase58Conversion.h>
+#include <Functions/FunctionFactory.h>
+
+namespace DB
+{
+namespace
+{
+
+struct NameTryBase58Decode
+{
+    static constexpr auto name = "tryBase58Decode";
+};
+
+using TryBase58DecodeImpl = Base58Decode<NameTryBase58Decode, Base58DecodeErrorHandling::ReturnEmptyString>;
+using FunctionTryBase58Decode = FunctionBase58Conversion<TryBase58DecodeImpl>;
+
+}
+
+REGISTER_FUNCTION(TryBase58Decode)
+{
+    factory.registerFunction<FunctionTryBase58Decode>();
+}
+
+}
--- a/src/Interpreters/InterpreterInsertQuery.cpp
+++ b/src/Interpreters/InterpreterInsertQuery.cpp
@ -214,20 +214,25 @@ Chain InterpreterInsertQuery::buildChain(
    const StoragePtr & table,
    const StorageMetadataPtr & metadata_snapshot,
    const Names & columns,
-    ThreadStatus * thread_status,
+    ThreadStatusesHolderPtr thread_status_holder,
    std::atomic_uint64_t * elapsed_counter_ms)
 {
    auto sample = getSampleBlock(columns, table, metadata_snapshot);
-    return buildChainImpl(table, metadata_snapshot, sample, thread_status, elapsed_counter_ms);
+    return buildChainImpl(table, metadata_snapshot, sample, thread_status_holder, elapsed_counter_ms);
 }

 Chain InterpreterInsertQuery::buildChainImpl(
    const StoragePtr & table,
    const StorageMetadataPtr & metadata_snapshot,
    const Block & query_sample_block,
-    ThreadStatus * thread_status,
+    ThreadStatusesHolderPtr thread_status_holder,
    std::atomic_uint64_t * elapsed_counter_ms)
 {
+    ThreadStatus * thread_status = current_thread;
+
+    if (!thread_status_holder)
+        thread_status = nullptr;
+
    auto context_ptr = getContext();
    const ASTInsertQuery * query = nullptr;
    if (query_ptr)
@ -252,7 +257,7 @@ Chain InterpreterInsertQuery::buildChainImpl(
    }
    else
    {
-        out = buildPushingToViewsChain(table, metadata_snapshot, context_ptr, query_ptr, no_destination, thread_status, elapsed_counter_ms);
+        out = buildPushingToViewsChain(table, metadata_snapshot, context_ptr, query_ptr, no_destination, thread_status_holder, elapsed_counter_ms);
    }

    /// Note that we wrap transforms one on top of another, so we write them in reverse of data processing order.
--- a/src/Interpreters/InterpreterInsertQuery.h
+++ b/src/Interpreters/InterpreterInsertQuery.h
@ -11,6 +11,9 @@ namespace DB
 class Chain;
 class ThreadStatus;

+struct ThreadStatusesHolder;
+using ThreadStatusesHolderPtr = std::shared_ptr<ThreadStatusesHolder>;
+
 /** Interprets the INSERT query.
  */
 class InterpreterInsertQuery : public IInterpreter, WithContext
@ -37,7 +40,7 @@ public:
        const StoragePtr & table,
        const StorageMetadataPtr & metadata_snapshot,
        const Names & columns,
-        ThreadStatus * thread_status = nullptr,
+        ThreadStatusesHolderPtr thread_status_holder = {},
        std::atomic_uint64_t * elapsed_counter_ms = nullptr);

    static void extendQueryLogElemImpl(QueryLogElement & elem, ContextPtr context_);
@ -61,7 +64,7 @@ private:
        const StoragePtr & table,
        const StorageMetadataPtr & metadata_snapshot,
        const Block & query_sample_block,
-        ThreadStatus * thread_status,
+        ThreadStatusesHolderPtr thread_status_holder,
        std::atomic_uint64_t * elapsed_counter_ms);
 };

--- a/src/Interpreters/JIT/compileFunction.cpp
+++ b/src/Interpreters/JIT/compileFunction.cpp
@ -234,9 +234,13 @@ static void compileFunction(llvm::Module & module, const IFunctionBase & functio
    auto * cur_block = b.GetInsertBlock();
    for (auto & col : columns)
    {
-        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.data, 1), cur_block);
+        auto * ty_data = llvm::cast<llvm::PointerType>(col.data->getType()->getScalarType())->getElementType();
+        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(ty_data, col.data, 1), cur_block);
        if (col.null)
-            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.null, 1), cur_block);
+        {
+            auto * ty_null = llvm::cast<llvm::PointerType>(col.null->getType()->getScalarType())->getElementType();
+            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(ty_null, col.null, 1), cur_block);
+        }
    }

    auto * value = b.CreateAdd(counter_phi, llvm::ConstantInt::get(size_type, 1));
@ -293,7 +297,8 @@ static void compileCreateAggregateStatesFunctions(llvm::Module & module, const s
    {
        size_t aggregate_function_offset = function_to_compile.aggregate_data_offset;
        const auto * aggregate_function = function_to_compile.function;
-        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_place_arg, aggregate_function_offset);
+        auto * ty_aggregate_data_place_arg = llvm::cast<llvm::PointerType>(aggregate_data_place_arg->getType()->getScalarType())->getElementType();
+        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_place_arg, aggregate_data_place_arg, aggregate_function_offset);
        aggregate_function->compileCreate(b, aggregation_place_with_offset);
    }

@ -324,7 +329,8 @@ static void compileAddIntoAggregateStatesFunctions(llvm::Module & module, const
    b.SetInsertPoint(entry);

    llvm::IRBuilder<> entry_builder(entry);
-    auto * places_start_arg = entry_builder.CreateInBoundsGEP(nullptr, places_arg, row_start_arg);
+    auto * ty_places_arg = llvm::cast<llvm::PointerType>(places_arg->getType()->getScalarType())->getElementType();
+    auto * places_start_arg = entry_builder.CreateInBoundsGEP(ty_places_arg, places_arg, row_start_arg);

    std::vector<ColumnDataPlaceholder> columns;
    size_t previous_columns_size = 0;
@ -342,11 +348,13 @@ static void compileAddIntoAggregateStatesFunctions(llvm::Module & module, const
            const auto & argument_type = argument_types[column_argument_index];
            auto * data = b.CreateLoad(column_data_type, b.CreateConstInBoundsGEP1_64(column_data_type, columns_arg, previous_columns_size + column_argument_index));
            data_placeholder.data_init = b.CreatePointerCast(b.CreateExtractValue(data, {0}), toNativeType(b, removeNullable(argument_type))->getPointerTo());
-            data_placeholder.data_init = entry_builder.CreateInBoundsGEP(nullptr, data_placeholder.data_init, row_start_arg);
+            auto * ty_data_init = llvm::cast<llvm::PointerType>(data_placeholder.data_init->getType()->getScalarType())->getElementType();
+            data_placeholder.data_init = entry_builder.CreateInBoundsGEP(ty_data_init, data_placeholder.data_init, row_start_arg);
            if (argument_type->isNullable())
            {
                data_placeholder.null_init = b.CreateExtractValue(data, {1});
-                data_placeholder.null_init = entry_builder.CreateInBoundsGEP(nullptr, data_placeholder.null_init, row_start_arg);
+                auto * ty_null_init = llvm::cast<llvm::PointerType>(data_placeholder.null_init->getType()->getScalarType())->getElementType();
+                data_placeholder.null_init = entry_builder.CreateInBoundsGEP(ty_null_init, data_placeholder.null_init, row_start_arg);
            }
            else
            {
@ -419,7 +427,8 @@ static void compileAddIntoAggregateStatesFunctions(llvm::Module & module, const
            arguments_values[column_argument_index] = nullable_value;
        }

-        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregation_place, aggregate_function_offset);
+        auto * ty_aggregation_place = llvm::cast<llvm::PointerType>(aggregation_place->getType()->getScalarType())->getElementType();
+        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(ty_aggregation_place, aggregation_place, aggregate_function_offset);
        aggregate_function_ptr->compileAdd(b, aggregation_place_with_offset, arguments_types, arguments_values);

        previous_columns_size += function_arguments_size;
@ -430,13 +439,18 @@ static void compileAddIntoAggregateStatesFunctions(llvm::Module & module, const
    auto * cur_block = b.GetInsertBlock();
    for (auto & col : columns)
    {
-        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.data, 1), cur_block);
+        auto * ty_data = llvm::cast<llvm::PointerType>(col.data->getType()->getScalarType())->getElementType();
+        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(ty_data, col.data, 1), cur_block);

        if (col.null)
-            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.null, 1), cur_block);
+        {
+            auto * ty_null = llvm::cast<llvm::PointerType>(col.null->getType()->getScalarType())->getElementType();
+            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(ty_null, col.null, 1), cur_block);
+        }
    }

-    places_phi->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, places_phi, 1), cur_block);
+    auto * ty_places_phi = llvm::cast<llvm::PointerType>(places_phi->getType()->getScalarType())->getElementType();
+    places_phi->addIncoming(b.CreateConstInBoundsGEP1_64(ty_places_phi, places_phi, 1), cur_block);

    auto * value = b.CreateAdd(counter_phi, llvm::ConstantInt::get(size_type, 1));
    counter_phi->addIncoming(value, cur_block);
@ -488,11 +502,13 @@ static void compileAddIntoAggregateStatesFunctionsSinglePlace(llvm::Module & mod
            const auto & argument_type = argument_types[column_argument_index];
            auto * data = b.CreateLoad(column_data_type, b.CreateConstInBoundsGEP1_64(column_data_type, columns_arg, previous_columns_size + column_argument_index));
            data_placeholder.data_init = b.CreatePointerCast(b.CreateExtractValue(data, {0}), toNativeType(b, removeNullable(argument_type))->getPointerTo());
-            data_placeholder.data_init = entry_builder.CreateInBoundsGEP(nullptr, data_placeholder.data_init, row_start_arg);
+            auto * ty_data_init = llvm::cast<llvm::PointerType>(data_placeholder.data_init->getType()->getScalarType())->getElementType();
+            data_placeholder.data_init = entry_builder.CreateInBoundsGEP(ty_data_init, data_placeholder.data_init, row_start_arg);
            if (argument_type->isNullable())
            {
                data_placeholder.null_init = b.CreateExtractValue(data, {1});
-                data_placeholder.null_init = entry_builder.CreateInBoundsGEP(nullptr, data_placeholder.null_init, row_start_arg);
+                auto * ty_null_init = llvm::cast<llvm::PointerType>(data_placeholder.null_init->getType()->getScalarType())->getElementType();
+                data_placeholder.null_init = entry_builder.CreateInBoundsGEP(ty_null_init, data_placeholder.null_init, row_start_arg);
            }
            else
            {
@ -560,7 +576,8 @@ static void compileAddIntoAggregateStatesFunctionsSinglePlace(llvm::Module & mod
            arguments_values[column_argument_index] = nullable_value;
        }

-        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(nullptr, place_arg, aggregate_function_offset);
+        auto * ty_place_arg = llvm::cast<llvm::PointerType>(place_arg->getType()->getScalarType())->getElementType();
+        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(ty_place_arg, place_arg, aggregate_function_offset);
        aggregate_function_ptr->compileAdd(b, aggregation_place_with_offset, arguments_types, arguments_values);

        previous_columns_size += function_arguments_size;
@ -571,10 +588,14 @@ static void compileAddIntoAggregateStatesFunctionsSinglePlace(llvm::Module & mod
    auto * cur_block = b.GetInsertBlock();
    for (auto & col : columns)
    {
-        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.data, 1), cur_block);
+        auto * ty_data = llvm::cast<llvm::PointerType>(col.data->getType()->getScalarType())->getElementType();
+        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(ty_data, col.data, 1), cur_block);

        if (col.null)
-            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.null, 1), cur_block);
+        {
+            auto * ty_null = llvm::cast<llvm::PointerType>(col.null->getType()->getScalarType())->getElementType();
+            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(ty_null, col.null, 1), cur_block);
+        }
    }

    auto * value = b.CreateAdd(counter_phi, llvm::ConstantInt::get(size_type, 1));
@ -607,8 +628,10 @@ static void compileMergeAggregatesStates(llvm::Module & module, const std::vecto
        size_t aggregate_function_offset = function_to_compile.aggregate_data_offset;
        const auto * aggregate_function_ptr = function_to_compile.function;

-        auto * aggregate_data_place_merge_dst_with_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_place_dst_arg, aggregate_function_offset);
-        auto * aggregate_data_place_merge_src_with_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_place_src_arg, aggregate_function_offset);
+        auto * ty_aggregate_data_place_dst_arg = llvm::cast<llvm::PointerType>(aggregate_data_place_dst_arg->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_place_merge_dst_with_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_place_dst_arg, aggregate_data_place_dst_arg, aggregate_function_offset);
+        auto * ty_aggregate_data_place_src_arg = llvm::cast<llvm::PointerType>(aggregate_data_place_src_arg->getType()->getScalarType())->getElementType();
+        auto * aggregate_data_place_merge_src_with_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_place_src_arg, aggregate_data_place_src_arg, aggregate_function_offset);

        aggregate_function_ptr->compileMerge(b, aggregate_data_place_merge_dst_with_offset, aggregate_data_place_merge_src_with_offset);
    }
@ -645,11 +668,13 @@ static void compileInsertAggregatesIntoResultColumns(llvm::Module & module, cons
        auto return_type = functions[i].function->getReturnType();
        auto * data = b.CreateLoad(column_data_type, b.CreateConstInBoundsGEP1_64(column_data_type, columns_arg, i));
        columns[i].data_init = b.CreatePointerCast(b.CreateExtractValue(data, {0}), toNativeType(b, removeNullable(return_type))->getPointerTo());
-        columns[i].data_init = entry_builder.CreateInBoundsGEP(nullptr, columns[i].data_init, row_start_arg);
+        auto * ty_data_init = llvm::cast<llvm::PointerType>(columns[i].data_init->getType()->getScalarType())->getElementType();
+        columns[i].data_init = entry_builder.CreateInBoundsGEP(ty_data_init, columns[i].data_init, row_start_arg);
        if (return_type->isNullable())
        {
            columns[i].null_init = b.CreateExtractValue(data, {1});
-            columns[i].null_init = entry_builder.CreateInBoundsGEP(nullptr, columns[i].null_init, row_start_arg);
+            auto * ty_null_init = llvm::cast<llvm::PointerType>(columns[i].null_init->getType()->getScalarType())->getElementType();
+            columns[i].null_init = entry_builder.CreateInBoundsGEP(ty_null_init, columns[i].null_init, row_start_arg);
        }
        else
        {
@ -688,7 +713,8 @@ static void compileInsertAggregatesIntoResultColumns(llvm::Module & module, cons
        const auto * aggregate_function_ptr = functions[i].function;

        auto * aggregate_data_place = b.CreateLoad(b.getInt8Ty()->getPointerTo(), aggregate_data_place_phi);
-        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_place, aggregate_function_offset);
+        auto * ty_aggregate_data_place = llvm::cast<llvm::PointerType>(aggregate_data_place->getType()->getScalarType())->getElementType();
+        auto * aggregation_place_with_offset = b.CreateConstInBoundsGEP1_64(ty_aggregate_data_place, aggregate_data_place, aggregate_function_offset);

        auto * final_value = aggregate_function_ptr->compileGetResult(b, aggregation_place_with_offset);

@ -708,16 +734,21 @@ static void compileInsertAggregatesIntoResultColumns(llvm::Module & module, cons
    auto * cur_block = b.GetInsertBlock();
    for (auto & col : columns)
    {
-        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.data, 1), cur_block);
+        auto * ty_col_data = llvm::cast<llvm::PointerType>(col.data->getType()->getScalarType())->getElementType();
+        col.data->addIncoming(b.CreateConstInBoundsGEP1_64(ty_col_data, col.data, 1), cur_block);

        if (col.null)
-            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, col.null, 1), cur_block);
+        {
+            auto * ty_col_null = llvm::cast<llvm::PointerType>(col.null->getType()->getScalarType())->getElementType();
+            col.null->addIncoming(b.CreateConstInBoundsGEP1_64(ty_col_null, col.null, 1), cur_block);
+        }
    }

    auto * value = b.CreateAdd(counter_phi, llvm::ConstantInt::get(size_type, 1), "", true, true);
    counter_phi->addIncoming(value, cur_block);

-    aggregate_data_place_phi->addIncoming(b.CreateConstInBoundsGEP1_64(nullptr, aggregate_data_place_phi, 1), cur_block);
+    auto * ty_aggregate_data_place_phi = llvm::cast<llvm::PointerType>(aggregate_data_place_phi->getType()->getScalarType())->getElementType();
+    aggregate_data_place_phi->addIncoming(b.CreateConstInBoundsGEP1_64(ty_aggregate_data_place_phi, aggregate_data_place_phi, 1), cur_block);

    b.CreateCondBr(b.CreateICmpEQ(value, row_end_arg), end, loop);

@ -842,11 +873,20 @@ CompiledSortDescriptionFunction compileSortDescription(
            auto * lhs_column_data = b.CreatePointerCast(b.CreateExtractValue(lhs_column, {0}), column_native_type_pointer);
            auto * lhs_column_null_data = column_type_is_nullable ? b.CreateExtractValue(lhs_column, {1}) : nullptr;

-            llvm::Value * lhs_value = b.CreateLoad(b.CreateInBoundsGEP(nullptr, lhs_column_data, lhs_index_arg));
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
+            auto * ty_lhs_column_data = llvm::cast<llvm::PointerType>(lhs_column_data->getType()->getScalarType())->getElementType();
+            llvm::Value * lhs_value = b.CreateLoad(b.CreateInBoundsGEP(ty_lhs_column_data, lhs_column_data, lhs_index_arg));
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif

            if (lhs_column_null_data)
            {
-                auto * is_null_value_pointer = b.CreateInBoundsGEP(nullptr, lhs_column_null_data, lhs_index_arg);
+                auto * ty_lhs_column_null_data = llvm::cast<llvm::PointerType>(lhs_column_null_data->getType()->getScalarType())->getElementType();
+                auto * is_null_value_pointer = b.CreateInBoundsGEP(ty_lhs_column_null_data, lhs_column_null_data, lhs_index_arg);
                auto * is_null = b.CreateICmpNE(b.CreateLoad(b.getInt8Ty(), is_null_value_pointer), b.getInt8(0));
                auto * lhs_nullable_value = b.CreateInsertValue(b.CreateInsertValue(nullable_unitilized, lhs_value, {0}), is_null, {1});
                lhs_value = lhs_nullable_value;
@ -856,10 +896,19 @@ CompiledSortDescriptionFunction compileSortDescription(
            auto * rhs_column_data = b.CreatePointerCast(b.CreateExtractValue(rhs_column, {0}), column_native_type_pointer);
            auto * rhs_column_null_data = column_type_is_nullable ? b.CreateExtractValue(rhs_column, {1}) : nullptr;

-            llvm::Value * rhs_value = b.CreateLoad(b.CreateInBoundsGEP(nullptr, rhs_column_data, rhs_index_arg));
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
+            auto * ty_rhs_column_data = llvm::cast<llvm::PointerType>(rhs_column_data->getType()->getScalarType())->getElementType();
+            llvm::Value * rhs_value = b.CreateLoad(b.CreateInBoundsGEP(ty_rhs_column_data, rhs_column_data, rhs_index_arg));
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
            if (rhs_column_null_data)
            {
-                auto * is_null_value_pointer = b.CreateInBoundsGEP(nullptr, rhs_column_null_data, rhs_index_arg);
+                auto * ty_rhs_column_null_data = llvm::cast<llvm::PointerType>(rhs_column_null_data->getType()->getScalarType())->getElementType();
+                auto * is_null_value_pointer = b.CreateInBoundsGEP(ty_rhs_column_null_data, rhs_column_null_data, rhs_index_arg);
                auto * is_null = b.CreateICmpNE(b.CreateLoad(b.getInt8Ty(), is_null_value_pointer), b.getInt8(0));
                auto * rhs_nullable_value = b.CreateInsertValue(b.CreateInsertValue(nullable_unitilized, rhs_value, {0}), is_null, {1});
                rhs_value = rhs_nullable_value;
--- a/src/Interpreters/MonotonicityCheckVisitor.h
+++ b/src/Interpreters/MonotonicityCheckVisitor.h
@ -70,6 +70,12 @@ public:
            if (!pos)
                return false;

+            /// It is possible that tables list is empty.
+            /// IdentifierSemantic get the position from AST, and it can be not valid to use it.
+            /// Example is re-analysing a part of AST for storage Merge, see 02147_order_by_optimizations.sql
+            if (*pos >= tables.size())
+                return false;
+
            if (auto data_type_and_name = tables[*pos].columns.tryGetByName(identifier->shortName()))
            {
                arg_data_type = data_type_and_name->type;
--- a/src/Interpreters/QueryViewsLog.h
+++ b/src/Interpreters/QueryViewsLog.h
@ -39,7 +39,7 @@ struct QueryViewsLogElement
    {
        String target_name;
        ViewType type = ViewType::DEFAULT;
-        std::unique_ptr<ThreadStatus> thread_status = nullptr;
+        ThreadStatus * thread_status = nullptr;
        std::atomic_uint64_t elapsed_ms = 0;
        std::chrono::time_point<std::chrono::system_clock> event_time;
        ViewStatus event_status = ViewStatus::QUERY_START;
--- a/src/Processors/Transforms/buildPushingToViewsChain.cpp
+++ b/src/Processors/Transforms/buildPushingToViewsChain.cpp
@ -40,8 +40,22 @@ namespace ErrorCodes
    extern const int LOGICAL_ERROR;
 }

+ThreadStatusesHolder::~ThreadStatusesHolder()
+{
+    auto * original_thread = current_thread;
+    SCOPE_EXIT({ current_thread = original_thread; });
+
+    while (!thread_statuses.empty())
+    {
+        current_thread = thread_statuses.front().get();
+        thread_statuses.pop_front();
+    }
+}
+
 struct ViewsData
 {
+    /// A separate holder for thread statuses, needed for proper destruction order.
+    ThreadStatusesHolderPtr thread_status_holder;
    /// Separate information for every view.
    std::list<ViewRuntimeData> views;
    /// Some common info about source storage.
@ -57,8 +71,9 @@ struct ViewsData
    std::atomic_bool has_exception = false;
    std::exception_ptr first_exception;

-    ViewsData(ContextPtr context_, StorageID source_storage_id_, StorageMetadataPtr source_metadata_snapshot_ , StoragePtr source_storage_)
-        : context(std::move(context_))
+    ViewsData(ThreadStatusesHolderPtr thread_status_holder_, ContextPtr context_, StorageID source_storage_id_, StorageMetadataPtr source_metadata_snapshot_ , StoragePtr source_storage_)
+        : thread_status_holder(std::move(thread_status_holder_))
+        , context(std::move(context_))
        , source_storage_id(std::move(source_storage_id_))
        , source_metadata_snapshot(std::move(source_metadata_snapshot_))
        , source_storage(std::move(source_storage_))
@ -177,13 +192,21 @@ Chain buildPushingToViewsChain(
    ContextPtr context,
    const ASTPtr & query_ptr,
    bool no_destination,
-    ThreadStatus * thread_status,
+    ThreadStatusesHolderPtr thread_status_holder,
    std::atomic_uint64_t * elapsed_counter_ms,
    const Block & live_view_header)
 {
    checkStackSize();
    Chain result_chain;

+    ThreadStatus * thread_status = current_thread;
+
+    if (!thread_status_holder)
+    {
+        thread_status_holder = std::make_shared<ThreadStatusesHolder>();
+        thread_status = nullptr;
+    }
+
    /// If we don't write directly to the destination
    /// then expect that we're inserting with precalculated virtual columns
    auto storage_header = no_destination ? metadata_snapshot->getSampleBlockWithVirtuals(storage->getVirtuals())
@ -225,7 +248,7 @@ Chain buildPushingToViewsChain(
        if (insert_settings.min_insert_block_size_bytes_for_materialized_views)
            insert_context->setSetting("min_insert_block_size_bytes", insert_settings.min_insert_block_size_bytes_for_materialized_views.value);

-        views_data = std::make_shared<ViewsData>(select_context, table_id, metadata_snapshot, storage);
+        views_data = std::make_shared<ViewsData>(thread_status_holder, select_context, table_id, metadata_snapshot, storage);
    }

    std::vector<Chain> chains;
@ -263,15 +286,17 @@ Chain buildPushingToViewsChain(
        };
        view_thread_status_ptr->attachQuery(running_group);

+        auto * view_thread_status = view_thread_status_ptr.get();
+        views_data->thread_status_holder->thread_statuses.push_front(std::move(view_thread_status_ptr));
+
        auto runtime_stats = std::make_unique<QueryViewsLogElement::ViewRuntimeStats>();
        runtime_stats->target_name = database_table.getFullTableName();
-        runtime_stats->thread_status = std::move(view_thread_status_ptr);
+        runtime_stats->thread_status = view_thread_status;
        runtime_stats->event_time = std::chrono::system_clock::now();
        runtime_stats->event_status = QueryViewsLogElement::ViewStatus::EXCEPTION_BEFORE_START;

        auto & type = runtime_stats->type;
        auto & target_name = runtime_stats->target_name;
-        auto * view_thread_status = runtime_stats->thread_status.get();
        auto * view_counter_ms = &runtime_stats->elapsed_ms;

        if (auto * materialized_view = dynamic_cast<StorageMaterializedView *>(dependent_table.get()))
@ -300,7 +325,7 @@ Chain buildPushingToViewsChain(
            }

            InterpreterInsertQuery interpreter(nullptr, insert_context, false, false, false);
-            out = interpreter.buildChain(inner_table, inner_metadata_snapshot, insert_columns, view_thread_status, view_counter_ms);
+            out = interpreter.buildChain(inner_table, inner_metadata_snapshot, insert_columns, thread_status_holder, view_counter_ms);
            out.addStorageHolder(dependent_table);
            out.addStorageHolder(inner_table);
        }
@ -309,18 +334,18 @@ Chain buildPushingToViewsChain(
            runtime_stats->type = QueryViewsLogElement::ViewType::LIVE;
            query = live_view->getInnerQuery(); // Used only to log in system.query_views_log
            out = buildPushingToViewsChain(
-                dependent_table, dependent_metadata_snapshot, insert_context, ASTPtr(), true, view_thread_status, view_counter_ms, storage_header);
+                dependent_table, dependent_metadata_snapshot, insert_context, ASTPtr(), true, thread_status_holder, view_counter_ms, storage_header);
        }
        else if (auto * window_view = dynamic_cast<StorageWindowView *>(dependent_table.get()))
        {
            runtime_stats->type = QueryViewsLogElement::ViewType::WINDOW;
            query = window_view->getMergeableQuery(); // Used only to log in system.query_views_log
            out = buildPushingToViewsChain(
-                dependent_table, dependent_metadata_snapshot, insert_context, ASTPtr(), true, view_thread_status, view_counter_ms);
+                dependent_table, dependent_metadata_snapshot, insert_context, ASTPtr(), true, thread_status_holder, view_counter_ms);
        }
        else
            out = buildPushingToViewsChain(
-                dependent_table, dependent_metadata_snapshot, insert_context, ASTPtr(), false, view_thread_status, view_counter_ms);
+                dependent_table, dependent_metadata_snapshot, insert_context, ASTPtr(), false, thread_status_holder, view_counter_ms);

        views_data->views.emplace_back(ViewRuntimeData{ //-V614
            std::move(query),
--- a/src/Processors/Transforms/buildPushingToViewsChain.h
+++ b/src/Processors/Transforms/buildPushingToViewsChain.h
@ -38,6 +38,16 @@ struct ViewRuntimeData
    }
 };

+/// A special holder for view's thread statuses.
+/// The goal is to control a destruction order
+struct ThreadStatusesHolder
+{
+    std::list<std::unique_ptr<ThreadStatus>> thread_statuses;
+    ~ThreadStatusesHolder();
+};
+
+using ThreadStatusesHolderPtr = std::shared_ptr<ThreadStatusesHolder>;
+
 /** Writes data to the specified table and to all dependent materialized views.
  */
 Chain buildPushingToViewsChain(
@ -50,7 +60,7 @@ Chain buildPushingToViewsChain(
    bool no_destination,
    /// We could specify separate thread_status for each view.
    /// Needed mainly to collect counters separately. Should be improved.
-    ThreadStatus * thread_status,
+    ThreadStatusesHolderPtr thread_status_holder,
    /// Counter to measure time spent separately per view. Should be improved.
    std::atomic_uint64_t * elapsed_counter_ms,
    /// LiveView executes query itself, it needs source block structure.
--- a/src/Storages/MergeTree/DataPartStorageOnDisk.cpp
+++ b/src/Storages/MergeTree/DataPartStorageOnDisk.cpp
@ -21,6 +21,7 @@ namespace ErrorCodes
    extern const int DIRECTORY_ALREADY_EXISTS;
    extern const int NOT_ENOUGH_SPACE;
    extern const int LOGICAL_ERROR;
+    extern const int FILE_DOESNT_EXIST;
 }

 DataPartStorageOnDisk::DataPartStorageOnDisk(VolumePtr volume_, std::string root_path_, std::string part_dir_)
@ -274,12 +275,21 @@ void DataPartStorageOnDisk::remove(
            disk->moveDirectory(from, to);
            onRename(root_path, part_dir_without_slash);
        }
+        catch (const Exception & e)
+        {
+            if (e.code() == ErrorCodes::FILE_DOESNT_EXIST)
+            {
+                LOG_ERROR(log, "Directory {} (part to remove) doesn't exist or one of nested files has gone. Most likely this is due to manual removing. This should be discouraged. Ignoring.", fullPath(disk, from));
+                return;
+            }
+            throw;
+        }
        catch (const fs::filesystem_error & e)
        {
            if (e.code() == std::errc::no_such_file_or_directory)
            {
                LOG_ERROR(log, "Directory {} (part to remove) doesn't exist or one of nested files has gone. "
-                          "Most likely this is due to manual removing. This should be discouraged. Ignoring.", fullPath(disk, to));
+                          "Most likely this is due to manual removing. This should be discouraged. Ignoring.", fullPath(disk, from));
                return;
            }
            throw;
@ -732,12 +742,14 @@ DataPartStoragePtr DataPartStorageOnDisk::freeze(
    const std::string & dir_path,
    bool make_source_readonly,
    std::function<void(const DiskPtr &)> save_metadata_callback,
-    bool copy_instead_of_hardlink) const
+    bool copy_instead_of_hardlink,
+    const NameSet & files_to_copy_instead_of_hardlinks) const
+
 {
    auto disk = volume->getDisk();
    disk->createDirectories(to);

-    localBackup(disk, getRelativePath(), fs::path(to) / dir_path, make_source_readonly, {}, copy_instead_of_hardlink);
+    localBackup(disk, getRelativePath(), fs::path(to) / dir_path, make_source_readonly, {}, copy_instead_of_hardlink, files_to_copy_instead_of_hardlinks);

    if (save_metadata_callback)
        save_metadata_callback(disk);
--- a/src/Storages/MergeTree/DataPartStorageOnDisk.h
+++ b/src/Storages/MergeTree/DataPartStorageOnDisk.h
@ -100,7 +100,8 @@ public:
        const std::string & dir_path,
        bool make_source_readonly,
        std::function<void(const DiskPtr &)> save_metadata_callback,
-        bool copy_instead_of_hardlink) const override;
+        bool copy_instead_of_hardlink,
+        const NameSet & files_to_copy_instead_of_hardlinks) const override;

    DataPartStoragePtr clone(
        const std::string & to,
--- a/src/Storages/MergeTree/IDataPartStorage.h
+++ b/src/Storages/MergeTree/IDataPartStorage.h
@ -192,12 +192,17 @@ public:

    /// Creates hardlinks into 'to/dir_path' for every file in data part.
    /// Callback is called after hardlinks are created, but before 'delete-on-destroy.txt' marker is removed.
+    /// Some files can be copied instead of hardlinks. It's because of details of zero copy replication
+    /// implementation which relies on paths of some blobs in S3. For example if we want to hardlink
+    /// the whole part during mutation we shouldn't hardlink checksums.txt, because otherwise
+    /// zero-copy locks for different parts will be on the same path in zookeeper.
    virtual std::shared_ptr<IDataPartStorage> freeze(
        const std::string & to,
        const std::string & dir_path,
        bool make_source_readonly,
        std::function<void(const DiskPtr &)> save_metadata_callback,
-        bool copy_instead_of_hardlink) const = 0;
+        bool copy_instead_of_hardlink,
+        const NameSet & files_to_copy_instead_of_hardlinks) const = 0;

    /// Make a full copy of a data part into 'to/dir_path' (possibly to a different disk).
    virtual std::shared_ptr<IDataPartStorage> clone(
--- a/src/Storages/MergeTree/IMergeTreeDataPart.cpp
+++ b/src/Storages/MergeTree/IMergeTreeDataPart.cpp
@ -1516,12 +1516,19 @@ void IMergeTreeDataPart::renameToDetached(const String & prefix, DataPartStorage

 void IMergeTreeDataPart::makeCloneInDetached(const String & prefix, const StorageMetadataPtr & /*metadata_snapshot*/) const
 {
+    auto storage_settings = storage.getSettings();
+
+    /// In case of zero-copy replication we copy directory instead of hardlinks
+    /// because hardlinks tracking doesn't work for detached parts.
+    bool copy_instead_of_hardlink = isStoredOnRemoteDiskWithZeroCopySupport() && storage.supportsReplication() && storage_settings->allow_remote_fs_zero_copy_replication;
+
    data_part_storage->freeze(
        storage.relative_data_path,
        getRelativePathForDetachedPart(prefix),
        /*make_source_readonly*/ true,
        {},
-        /*copy_instead_of_hardlink*/ false);
+        copy_instead_of_hardlink,
+        {});
 }

 DataPartStoragePtr IMergeTreeDataPart::makeCloneOnDisk(const DiskPtr & disk, const String & directory_name) const
--- a/src/Storages/MergeTree/MergeTreeData.cpp
+++ b/src/Storages/MergeTree/MergeTreeData.cpp
@ -6144,6 +6144,9 @@ MergeTreeData & MergeTreeData::checkStructureAndGetMergeTreeData(IStorage & sour
    if (format_version != src_data->format_version)
        throw Exception("Tables have different format_version", ErrorCodes::BAD_ARGUMENTS);

+    if (query_to_string(my_snapshot->getPrimaryKeyAST()) != query_to_string(src_snapshot->getPrimaryKeyAST()))
+        throw Exception("Tables have different primary key", ErrorCodes::BAD_ARGUMENTS);
+
    return *src_data;
 }

@ -6160,7 +6163,8 @@ std::pair<MergeTreeData::MutableDataPartPtr, scope_guard> MergeTreeData::cloneAn
    const StorageMetadataPtr & metadata_snapshot,
    const MergeTreeTransactionPtr & txn,
    HardlinkedFiles * hardlinked_files,
-    bool copy_instead_of_hardlink)
+    bool copy_instead_of_hardlink,
+    const NameSet & files_to_copy_instead_of_hardlinks)
 {
    /// Check that the storage policy contains the disk where the src_part is located.
    bool does_storage_policy_allow_same_disk = false;
@ -6204,7 +6208,7 @@ std::pair<MergeTreeData::MutableDataPartPtr, scope_guard> MergeTreeData::cloneAn
              std::string(fs::path(src_part_storage->getFullRootPath()) / tmp_dst_part_name),
              with_copy);

-    auto dst_part_storage = src_part_storage->freeze(relative_data_path, tmp_dst_part_name, /* make_source_readonly */ false, {}, /* copy_instead_of_hardlinks */ copy_instead_of_hardlink);
+    auto dst_part_storage = src_part_storage->freeze(relative_data_path, tmp_dst_part_name, /* make_source_readonly */ false, {}, copy_instead_of_hardlink, files_to_copy_instead_of_hardlinks);

    auto dst_data_part = createPart(dst_part_name, dst_part_info, dst_part_storage);

@ -6215,7 +6219,9 @@ std::pair<MergeTreeData::MutableDataPartPtr, scope_guard> MergeTreeData::cloneAn

        for (auto it = src_part->data_part_storage->iterate(); it->isValid(); it->next())
        {
-            if (it->name() != IMergeTreeDataPart::DELETE_ON_DESTROY_MARKER_FILE_NAME && it->name() != IMergeTreeDataPart::TXN_VERSION_METADATA_FILE_NAME)
+            if (!files_to_copy_instead_of_hardlinks.contains(it->name())
+                && it->name() != IMergeTreeDataPart::DELETE_ON_DESTROY_MARKER_FILE_NAME
+                && it->name() != IMergeTreeDataPart::TXN_VERSION_METADATA_FILE_NAME)
                hardlinked_files->hardlinks_from_source_part.insert(it->name());
        }
    }
@ -6390,7 +6396,8 @@ PartitionCommandsResultInfo MergeTreeData::freezePartitionsByMatcher(
            part->data_part_storage->getPartDirectory(),
            /*make_source_readonly*/ true,
            callback,
-            /*copy_instead_of_hardlink*/ false);
+            /*copy_instead_of_hardlink*/ false,
+            {});

        part->is_frozen.store(true, std::memory_order_relaxed);
        result.push_back(PartitionCommandResultInfo{
--- a/src/Storages/MergeTree/MergeTreeData.h
+++ b/src/Storages/MergeTree/MergeTreeData.h
@ -785,7 +785,7 @@ public:
        const MergeTreeData::DataPartPtr & src_part, const String & tmp_part_prefix,
        const MergeTreePartInfo & dst_part_info, const StorageMetadataPtr & metadata_snapshot,
        const MergeTreeTransactionPtr & txn, HardlinkedFiles * hardlinked_files,
-        bool copy_instead_of_hardlink);
+        bool copy_instead_of_hardlink, const NameSet & files_to_copy_instead_of_hardlinks);

    virtual std::vector<MergeTreeMutationStatus> getMutationsStatus() const = 0;

--- a/src/Storages/MergeTree/MergeTreeDataPartCompact.cpp
+++ b/src/Storages/MergeTree/MergeTreeDataPartCompact.cpp
@ -4,6 +4,7 @@
 #include <Storages/MergeTree/MergeTreeDataPartWriterCompact.h>
 #include <Interpreters/Context.h>
 #include <Storages/MergeTree/LoadedMergeTreeDataPartInfoForReader.h>
+#include <Compression/CompressedReadBufferFromFile.h>


 namespace DB
@ -110,18 +111,26 @@ void MergeTreeDataPartCompact::loadIndexGranularityImpl(

    size_t marks_file_size = data_part_storage_->getFileSize(marks_file_path);

-    auto buffer = data_part_storage_->readFile(marks_file_path, ReadSettings().adjustBufferSize(marks_file_size), marks_file_size, std::nullopt);
-    while (!buffer->eof())
+    std::unique_ptr<ReadBufferFromFileBase> buffer = data_part_storage_->readFile(
+        marks_file_path, ReadSettings().adjustBufferSize(marks_file_size), marks_file_size, std::nullopt);
+
+    std::unique_ptr<ReadBuffer> marks_reader;
+    bool marks_compressed = index_granularity_info_.mark_type.compressed;
+    if (marks_compressed)
+        marks_reader = std::make_unique<CompressedReadBufferFromFile>(std::move(buffer));
+    else
+        marks_reader = std::move(buffer);
+
+    while (!marks_reader->eof())
    {
-        /// Skip offsets for columns
-        buffer->seek(columns_count * sizeof(MarkInCompressedFile), SEEK_CUR);
+        marks_reader->ignore(columns_count * sizeof(MarkInCompressedFile));
        size_t granularity;
-        readIntBinary(granularity, *buffer);
+        readIntBinary(granularity, *marks_reader);
        index_granularity_.appendMark(granularity);
    }

-    if (index_granularity_.getMarksCount() * index_granularity_info_.getMarkSizeInBytes(columns_count) != marks_file_size)
-        throw Exception("Cannot read all marks from file " + marks_file_path, ErrorCodes::CANNOT_READ_ALL_DATA);
+    if (!marks_compressed && index_granularity_.getMarksCount() * index_granularity_info_.getMarkSizeInBytes(columns_count) != marks_file_size)
+        throw Exception(ErrorCodes::CANNOT_READ_ALL_DATA, "Cannot read all marks from file {}", marks_file_path);

    index_granularity_.setInitialized();
 }
--- a/src/Storages/MergeTree/MutateTask.cpp
+++ b/src/Storages/MergeTree/MutateTask.cpp
@ -1507,8 +1507,24 @@ bool MutateTask::prepare()
    if (ctx->source_part->isStoredOnDisk() && !isStorageTouchedByMutations(
        ctx->storage_from_source_part, ctx->metadata_snapshot, ctx->commands_for_part, Context::createCopy(context_for_reading)))
    {
+        NameSet files_to_copy_instead_of_hardlinks;
+        auto settings_ptr = ctx->data->getSettings();
+        /// In zero-copy replication checksums file path in s3 (blob path) is used for zero copy locks in ZooKeeper. If we will hardlink checksums file, we will have the same blob path
+        /// and two different parts (source and new mutated part) will use the same locks in ZooKeeper. To avoid this we copy checksums.txt to generate new blob path.
+        /// Example:
+        ///     part: all_0_0_0/checksums.txt -> /s3/blobs/shjfgsaasdasdasdasdasdas
+        ///     locks path in zk: /zero_copy/tbl_id/s3_blobs_shjfgsaasdasdasdasdasdas/replica_name
+        ///                                         ^ part name don't participate in lock path
+        /// In case of full hardlink we will have:
+        ///     part: all_0_0_0_1/checksums.txt -> /s3/blobs/shjfgsaasdasdasdasdasdas
+        ///     locks path in zk: /zero_copy/tbl_id/s3_blobs_shjfgsaasdasdasdasdasdas/replica_name
+        /// So we need to copy to have a new name
+        bool copy_checksumns = ctx->data->supportsReplication() && settings_ptr->allow_remote_fs_zero_copy_replication && ctx->source_part->isStoredOnRemoteDiskWithZeroCopySupport();
+        if (copy_checksumns)
+            files_to_copy_instead_of_hardlinks.insert(IMergeTreeDataPart::FILE_FOR_REFERENCES_CHECK);
+
        LOG_TRACE(ctx->log, "Part {} doesn't change up to mutation version {}", ctx->source_part->name, ctx->future_part->part_info.mutation);
-        auto [part, lock] = ctx->data->cloneAndLoadDataPartOnSameDisk(ctx->source_part, "tmp_clone_", ctx->future_part->part_info, ctx->metadata_snapshot, ctx->txn, &ctx->hardlinked_files, false);
+        auto [part, lock] = ctx->data->cloneAndLoadDataPartOnSameDisk(ctx->source_part, "tmp_clone_", ctx->future_part->part_info, ctx->metadata_snapshot, ctx->txn, &ctx->hardlinked_files, false, files_to_copy_instead_of_hardlinks);
        ctx->temporary_directory_lock = std::move(lock);
        promise.set_value(std::move(part));
        return false;
@ -1621,7 +1637,24 @@ bool MutateTask::prepare()
            LOG_TRACE(ctx->log, "Part {} doesn't change up to mutation version {} (optimized)", ctx->source_part->name, ctx->future_part->part_info.mutation);
            /// new_data_part is not used here, another part is created instead (see the comment above)
            ctx->temporary_directory_lock = {};
-            auto [part, lock] = ctx->data->cloneAndLoadDataPartOnSameDisk(ctx->source_part, "tmp_mut_", ctx->future_part->part_info, ctx->metadata_snapshot, ctx->txn, &ctx->hardlinked_files, false);
+
+            /// In zero-copy replication checksums file path in s3 (blob path) is used for zero copy locks in ZooKeeper. If we will hardlink checksums file, we will have the same blob path
+            /// and two different parts (source and new mutated part) will use the same locks in ZooKeeper. To avoid this we copy checksums.txt to generate new blob path.
+            /// Example:
+            ///     part: all_0_0_0/checksums.txt -> /s3/blobs/shjfgsaasdasdasdasdasdas
+            ///     locks path in zk: /zero_copy/tbl_id/s3_blobs_shjfgsaasdasdasdasdasdas/replica_name
+            ///                                         ^ part name don't participate in lock path
+            /// In case of full hardlink we will have:
+            ///     part: all_0_0_0_1/checksums.txt -> /s3/blobs/shjfgsaasdasdasdasdasdas
+            ///     locks path in zk: /zero_copy/tbl_id/s3_blobs_shjfgsaasdasdasdasdasdas/replica_name
+            /// So we need to copy to have a new name
+            NameSet files_to_copy_instead_of_hardlinks;
+            auto settings_ptr = ctx->data->getSettings();
+            bool copy_checksumns = ctx->data->supportsReplication() && settings_ptr->allow_remote_fs_zero_copy_replication && ctx->source_part->isStoredOnRemoteDiskWithZeroCopySupport();
+            if (copy_checksumns)
+                files_to_copy_instead_of_hardlinks.insert(IMergeTreeDataPart::FILE_FOR_REFERENCES_CHECK);
+
+            auto [part, lock] = ctx->data->cloneAndLoadDataPartOnSameDisk(ctx->source_part, "tmp_mut_", ctx->future_part->part_info, ctx->metadata_snapshot, ctx->txn, &ctx->hardlinked_files, false, files_to_copy_instead_of_hardlinks);
            ctx->temporary_directory_lock = std::move(lock);
            promise.set_value(std::move(part));
            return false;
--- a/src/Storages/MergeTree/ReplicatedMergeTreePartCheckThread.cpp
+++ b/src/Storages/MergeTree/ReplicatedMergeTreePartCheckThread.cpp
@ -149,6 +149,7 @@ ReplicatedMergeTreePartCheckThread::MissingPartSearchResult ReplicatedMergeTreeP
        String replica_path = storage.zookeeper_path + "/replicas/" + replica;

        Strings parts = zookeeper->getChildren(replica_path + "/parts");
+        Strings parts_found;
        for (const String & part_on_replica : parts)
        {
            auto part_on_replica_info = MergeTreePartInfo::fromPartName(part_on_replica, storage.format_version);
@ -174,14 +175,22 @@ ReplicatedMergeTreePartCheckThread::MissingPartSearchResult ReplicatedMergeTreeP
            if (part_info.contains(part_on_replica_info))
            {
                if (part_on_replica_info.min_block == part_info.min_block)
+                {
                    found_part_with_the_same_min_block = true;
+                    parts_found.push_back(part_on_replica);
+                }
                if (part_on_replica_info.max_block == part_info.max_block)
+                {
                    found_part_with_the_same_max_block = true;
+                    parts_found.push_back(part_on_replica);
+                }

                if (found_part_with_the_same_min_block && found_part_with_the_same_max_block)
                {
                    /// FIXME It may never appear
-                    LOG_INFO(log, "Found parts with the same min block and with the same max block as the missing part {} on replica {}. Hoping that it will eventually appear as a result of a merge.", part_name, replica);
+                    LOG_INFO(log, "Found parts with the same min block and with the same max block as the missing part {} on replica {}. "
+                             "Hoping that it will eventually appear as a result of a merge. Parts: {}",
+                             part_name, replica, fmt::join(parts_found, ", "));
                    return MissingPartSearchResult::FoundAndDontNeedFetch;
                }
            }
@ -209,17 +218,19 @@ void ReplicatedMergeTreePartCheckThread::searchForMissingPartAndFetchIfPossible(
    /// If the part is in ZooKeeper, remove it from there and add the task to download it to the queue.
    if (exists_in_zookeeper)
    {
-        /// If part found on some other replica
        if (missing_part_search_result == MissingPartSearchResult::FoundAndNeedFetch)
        {
            LOG_WARNING(log, "Part {} exists in ZooKeeper but not locally and found on other replica. Removing from ZooKeeper and queueing a fetch.", part_name);
-            storage.removePartAndEnqueueFetch(part_name);
        }
-        else /// If we have covering part on other replica or part is lost forever we don't need to fetch anything
+        else
        {
            LOG_WARNING(log, "Part {} exists in ZooKeeper but not locally and not found on other replica. Removing it from ZooKeeper.", part_name);
-            storage.removePartFromZooKeeper(part_name);
        }
+
+        /// We cannot simply remove part from ZooKeeper, because it may be removed from virtual_part,
+        /// so we have to create some entry in the queue. Maybe we will execute it (by fetching part or covering part from somewhere),
+        /// maybe will simply replace with empty part.
+        storage.removePartAndEnqueueFetch(part_name);
    }

    ProfileEvents::increment(ProfileEvents::ReplicatedPartChecksFailed);
--- a/src/Storages/MergeTree/localBackup.cpp
+++ b/src/Storages/MergeTree/localBackup.cpp
@ -19,7 +19,7 @@ namespace
 void localBackupImpl(
    const DiskPtr & disk, const String & source_path,
    const String & destination_path, bool make_source_readonly, size_t level,
-    std::optional<size_t> max_level)
+    std::optional<size_t> max_level, const NameSet & files_to_copy_instead_of_hardlinks)
 {
    if (max_level && level > *max_level)
        return;
@ -38,11 +38,14 @@ void localBackupImpl(
        {
            if (make_source_readonly)
                disk->setReadOnly(source);
-            disk->createHardLink(source, destination);
+            if (files_to_copy_instead_of_hardlinks.contains(it->name()))
+                disk->copyFile(source, *disk, destination);
+            else
+                disk->createHardLink(source, destination);
        }
        else
        {
-            localBackupImpl(disk, source, destination, make_source_readonly, level + 1, max_level);
+            localBackupImpl(disk, source, destination, make_source_readonly, level + 1, max_level, files_to_copy_instead_of_hardlinks);
        }
    }
 }
@ -86,7 +89,7 @@ private:
 void localBackup(
    const DiskPtr & disk, const String & source_path,
    const String & destination_path, bool make_source_readonly,
-    std::optional<size_t> max_level, bool copy_instead_of_hardlinks)
+    std::optional<size_t> max_level, bool copy_instead_of_hardlinks, const NameSet & files_to_copy_intead_of_hardlinks)
 {
    if (disk->exists(destination_path) && !disk->isDirectoryEmpty(destination_path))
    {
@ -109,7 +112,7 @@ void localBackup(
            if (copy_instead_of_hardlinks)
                disk->copyDirectoryContent(source_path, disk, destination_path);
            else
-                localBackupImpl(disk, source_path, destination_path, make_source_readonly, 0, max_level);
+                localBackupImpl(disk, source_path, destination_path, make_source_readonly, 0, max_level, files_to_copy_intead_of_hardlinks);
        }
        catch (const DB::ErrnoException & e)
        {
--- a/src/Storages/MergeTree/localBackup.h
+++ b/src/Storages/MergeTree/localBackup.h
@ -20,6 +20,6 @@ namespace DB
  *  If max_level is specified, than only files which depth relative source_path less or equal max_level will be copied.
  *  So, if max_level=0 than only direct file child are copied.
  */
-void localBackup(const DiskPtr & disk, const String & source_path, const String & destination_path, bool make_source_readonly = true, std::optional<size_t> max_level = {}, bool copy_instead_of_hardlinks = false);
+    void localBackup(const DiskPtr & disk, const String & source_path, const String & destination_path, bool make_source_readonly = true, std::optional<size_t> max_level = {}, bool copy_instead_of_hardlinks = false, const NameSet & files_to_copy_intead_of_hardlinks = {});

 }
--- a/src/Storages/MergeTree/registerStorageMergeTree.cpp
+++ b/src/Storages/MergeTree/registerStorageMergeTree.cpp
@ -683,8 +683,6 @@ static StoragePtr create(const StorageFactory::Arguments & args)

    if (replicated)
    {
-        auto storage_policy = args.getContext()->getStoragePolicy(storage_settings->storage_policy);
-
        return std::make_shared<StorageReplicatedMergeTree>(
            zookeeper_path,
            replica_name,
--- a/src/Storages/StorageKeeperMap.cpp
+++ b/src/Storages/StorageKeeperMap.cpp
@ -316,6 +316,36 @@ StorageKeeperMap::StorageKeeperMap(

    for (size_t i = 0; i < 1000; ++i)
    {
+        std::string stored_metadata_string;
+        auto exists = client->tryGet(metadata_path, stored_metadata_string);
+
+        if (exists)
+        {
+            // this requires same name for columns
+            // maybe we can do a smarter comparison for columns and primary key expression
+            if (stored_metadata_string != metadata_string)
+                throw Exception(
+                    ErrorCodes::BAD_ARGUMENTS,
+                    "Path {} is already used but the stored table definition doesn't match. Stored metadata: {}",
+                    root_path,
+                    stored_metadata_string);
+
+            auto code = client->tryCreate(table_path, "", zkutil::CreateMode::Persistent);
+
+            // tables_path was removed with drop
+            if (code == Coordination::Error::ZNONODE)
+            {
+                LOG_INFO(log, "Metadata nodes were removed by another server, will retry");
+                continue;
+            }
+            else if (code != Coordination::Error::ZOK)
+            {
+                throw zkutil::KeeperException(code, "Failed to create table on path {} because a table with same UUID already exists", root_path);
+            }
+
+            return;
+        }
+
        if (client->exists(dropped_path))
        {
            LOG_INFO(log, "Removing leftover nodes");
@ -342,45 +372,29 @@ StorageKeeperMap::StorageKeeperMap(
            }
        }

-        std::string stored_metadata_string;
-        auto exists = client->tryGet(metadata_path, stored_metadata_string);
+        Coordination::Requests create_requests
+        {
+            zkutil::makeCreateRequest(metadata_path, metadata_string, zkutil::CreateMode::Persistent),
+            zkutil::makeCreateRequest(data_path, metadata_string, zkutil::CreateMode::Persistent),
+            zkutil::makeCreateRequest(tables_path, "", zkutil::CreateMode::Persistent),
+            zkutil::makeCreateRequest(table_path, "", zkutil::CreateMode::Persistent),
+        };

-        if (exists)
+        Coordination::Responses create_responses;
+        auto code = client->tryMulti(create_requests, create_responses);
+        if (code == Coordination::Error::ZNODEEXISTS)
        {
-            // this requires same name for columns
-            // maybe we can do a smarter comparison for columns and primary key expression
-            if (stored_metadata_string != metadata_string)
-                throw Exception(
-                    ErrorCodes::BAD_ARGUMENTS,
-                    "Path {} is already used but the stored table definition doesn't match. Stored metadata: {}",
-                    root_path,
-                    stored_metadata_string);
+            LOG_INFO(log, "It looks like a table on path {} was created by another server at the same moment, will retry", root_path);
+            continue;
        }
-        else
+        else if (code != Coordination::Error::ZOK)
        {
-            auto code = client->tryCreate(metadata_path, metadata_string, zkutil::CreateMode::Persistent);
-            if (code == Coordination::Error::ZNODEEXISTS)
-                continue;
-            else if (code != Coordination::Error::ZOK)
-                throw Coordination::Exception(code, metadata_path);
+            zkutil::KeeperMultiException::check(code, create_requests, create_responses);
        }

-        client->createIfNotExists(tables_path, "");

-        auto code = client->tryCreate(table_path, "", zkutil::CreateMode::Persistent);
-
-        if (code == Coordination::Error::ZOK)
-        {
-            // metadata now should be guaranteed to exist because we added our UUID to the tables_path
-            client->createIfNotExists(data_path, "");
-            table_is_valid = true;
-            return;
-        }
-
-        if (code == Coordination::Error::ZNONODE)
-            LOG_INFO(log, "Metadata nodes were deleted in background, will retry");
-        else
-            throw Coordination::Exception(code, table_path);
+        table_is_valid = true;
+        return;
    }

    throw Exception(ErrorCodes::BAD_ARGUMENTS, "Cannot create metadata for table, because it is removed concurrently or because of wrong root_path ({})", root_path);
--- a/src/Storages/StorageMergeTree.cpp
+++ b/src/Storages/StorageMergeTree.cpp
@ -1569,7 +1569,7 @@ void StorageMergeTree::replacePartitionFrom(const StoragePtr & source_table, con
        Int64 temp_index = insert_increment.get();
        MergeTreePartInfo dst_part_info(partition_id, temp_index, temp_index, src_part->info.level);

-        auto [dst_part, part_lock] = cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, my_metadata_snapshot, local_context->getCurrentTransaction(), {}, false);
+        auto [dst_part, part_lock] = cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, my_metadata_snapshot, local_context->getCurrentTransaction(), {}, false, {});
        dst_parts.emplace_back(std::move(dst_part));
        dst_parts_locks.emplace_back(std::move(part_lock));
    }
@ -1664,7 +1664,7 @@ void StorageMergeTree::movePartitionToTable(const StoragePtr & dest_table, const
        Int64 temp_index = insert_increment.get();
        MergeTreePartInfo dst_part_info(partition_id, temp_index, temp_index, src_part->info.level);

-        auto [dst_part, part_lock] = dest_table_storage->cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, dest_metadata_snapshot, local_context->getCurrentTransaction(), {}, false);
+        auto [dst_part, part_lock] = dest_table_storage->cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, dest_metadata_snapshot, local_context->getCurrentTransaction(), {}, false, {});
        dst_parts.emplace_back(std::move(dst_part));
        dst_parts_locks.emplace_back(std::move(part_lock));
    }
--- a/src/Storages/StorageReplicatedMergeTree.cpp
+++ b/src/Storages/StorageReplicatedMergeTree.cpp
@ -1408,31 +1408,23 @@ MergeTreeData::DataPartsVector StorageReplicatedMergeTree::checkPartChecksumsAnd
            ops = std::move(new_ops);
        }

-        try
-        {
-            Coordination::Responses responses;
-            Coordination::Error e = zookeeper->tryMulti(ops, responses);
-            if (e == Coordination::Error::ZOK)
-                return transaction.commit();
+        Coordination::Responses responses;
+        Coordination::Error e = zookeeper->tryMulti(ops, responses);
+        if (e == Coordination::Error::ZOK)
+            return transaction.commit();

-            if (e == Coordination::Error::ZNODEEXISTS)
+        if (e == Coordination::Error::ZNODEEXISTS)
+        {
+            size_t num_check_ops = 2 * absent_part_paths_on_replicas.size();
+            size_t failed_op_index = zkutil::getFailedOpIndex(e, responses);
+            if (failed_op_index < num_check_ops)
            {
-                size_t num_check_ops = 2 * absent_part_paths_on_replicas.size();
-                size_t failed_op_index = zkutil::getFailedOpIndex(e, responses);
-                if (failed_op_index < num_check_ops)
-                {
-                    LOG_INFO(log, "The part {} on a replica suddenly appeared, will recheck checksums", ops[failed_op_index]->getPath());
-                    continue;
-                }
+                LOG_INFO(log, "The part {} on a replica suddenly appeared, will recheck checksums", ops[failed_op_index]->getPath());
+                continue;
            }
+        }

-            throw zkutil::KeeperException(e);
-        }
-        catch (const std::exception &)
-        {
-            unlockSharedData(*part);
-            throw;
-        }
+        throw zkutil::KeeperException(e);
    }
 }

@ -2172,7 +2164,7 @@ bool StorageReplicatedMergeTree::executeReplaceRange(const LogEntry & entry)
                throw Exception("Checksums of " + part_desc->src_table_part->name + " is suddenly changed", ErrorCodes::UNFINISHED);

            auto [res_part, temporary_part_lock] = cloneAndLoadDataPartOnSameDisk(
-                part_desc->src_table_part, TMP_PREFIX + "clone_", part_desc->new_part_info, metadata_snapshot, NO_TRANSACTION_PTR, &part_desc->hardlinked_files, false);
+                part_desc->src_table_part, TMP_PREFIX + "clone_", part_desc->new_part_info, metadata_snapshot, NO_TRANSACTION_PTR, &part_desc->hardlinked_files, false, {});
            part_desc->res_part = std::move(res_part);
            part_desc->temporary_part_lock = std::move(temporary_part_lock);
        }
@ -3922,7 +3914,7 @@ bool StorageReplicatedMergeTree::fetchPart(const String & part_name, const Stora
    {
        get_part = [&, part_to_clone]()
        {
-            auto [cloned_part, lock] = cloneAndLoadDataPartOnSameDisk(part_to_clone, "tmp_clone_", part_info, metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, false);
+            auto [cloned_part, lock] = cloneAndLoadDataPartOnSameDisk(part_to_clone, "tmp_clone_", part_info, metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, false, {});
            part_to_clone_lock = std::move(lock);
            return cloned_part;
        };
@ -6503,8 +6495,7 @@ void StorageReplicatedMergeTree::replacePartitionFrom(
            bool copy_instead_of_hardlink = storage_settings_ptr->allow_remote_fs_zero_copy_replication
                                            && src_part->isStoredOnRemoteDiskWithZeroCopySupport();

-            auto [dst_part, part_lock] = cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, copy_instead_of_hardlink);
-
+            auto [dst_part, part_lock] = cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, copy_instead_of_hardlink, {});
            src_parts.emplace_back(src_part);
            dst_parts.emplace_back(dst_part);
            dst_parts_locks.emplace_back(std::move(part_lock));
@ -6734,7 +6725,7 @@ void StorageReplicatedMergeTree::movePartitionToTable(const StoragePtr & dest_ta
            bool copy_instead_of_hardlink = storage_settings_ptr->allow_remote_fs_zero_copy_replication
                                            && src_part->isStoredOnRemoteDiskWithZeroCopySupport();

-            auto [dst_part, dst_part_lock] = dest_table_storage->cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, dest_metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, copy_instead_of_hardlink);
+            auto [dst_part, dst_part_lock] = dest_table_storage->cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, dest_metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, copy_instead_of_hardlink, {});

            src_parts.emplace_back(src_part);
            dst_parts.emplace_back(dst_part);
@ -7682,7 +7673,7 @@ std::pair<bool, NameSet> StorageReplicatedMergeTree::unlockSharedDataByID(

        if (!children.empty())
        {
-            LOG_TRACE(logger, "Found {} ({}) zookeper locks for {}", children.size(), fmt::join(children, ", "), zookeeper_part_uniq_node);
+            LOG_TRACE(logger, "Found {} ({}) zookeeper locks for {}", children.size(), fmt::join(children, ", "), zookeeper_part_uniq_node);
            part_has_no_more_locks = false;
            continue;
        }
@ -8153,7 +8144,6 @@ bool StorageReplicatedMergeTree::createEmptyPartInsteadOfLost(zkutil::ZooKeeperP
    }
    catch (const Exception & ex)
    {
-        unlockSharedData(*new_data_part);
        LOG_WARNING(log, "Cannot commit empty part {} with error {}", lost_part_name, ex.displayText());
        return false;
    }
--- a/src/Storages/System/CMakeLists.txt
+++ b/src/Storages/System/CMakeLists.txt
@ -2,54 +2,21 @@
 # You can also regenerate it manually this way:
 # execute_process(COMMAND "${CMAKE_CURRENT_SOURCE_DIR}/StorageSystemContributors.sh")

-include(${ClickHouse_SOURCE_DIR}/cmake/embed_binary.cmake)
-
-set (CONFIG_BUILD "${CMAKE_CURRENT_BINARY_DIR}/StorageSystemBuildOptions.generated.cpp")
 get_property (BUILD_COMPILE_DEFINITIONS DIRECTORY ${ClickHouse_SOURCE_DIR} PROPERTY COMPILE_DEFINITIONS)
-
 get_property(TZDATA_VERSION GLOBAL PROPERTY TZDATA_VERSION_PROP)

-
-find_package(Git)
-if(Git_FOUND)
-    # The commit's git hash, and whether the building workspace was dirty or not
-    execute_process(COMMAND
-      "${GIT_EXECUTABLE}" rev-parse HEAD
-      WORKING_DIRECTORY "${ClickHouse_SOURCE_DIR}"
-      OUTPUT_VARIABLE GIT_HASH
-      ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
-    # Git branch name
-    execute_process(COMMAND
-      "${GIT_EXECUTABLE}" rev-parse --abbrev-ref HEAD
-      WORKING_DIRECTORY "${ClickHouse_SOURCE_DIR}"
-      OUTPUT_VARIABLE GIT_BRANCH
-      ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
-    # The date of the commit
-    SET(ENV{TZ} "UTC")
-    execute_process(COMMAND
-      "${GIT_EXECUTABLE}" log -1 --format=%ad --date=iso-local
-      WORKING_DIRECTORY "${ClickHouse_SOURCE_DIR}"
-      OUTPUT_VARIABLE GIT_DATE
-      ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
-    # The subject of the commit
-    execute_process(COMMAND
-      "${GIT_EXECUTABLE}" log -1 --format=%s
-      WORKING_DIRECTORY "${ClickHouse_SOURCE_DIR}"
-      OUTPUT_VARIABLE GIT_COMMIT_SUBJECT
-      ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
-endif()
-
 function(generate_system_build_options)
    include(${ClickHouse_SOURCE_DIR}/src/configure_config.cmake)
    include(${ClickHouse_SOURCE_DIR}/src/Functions/configure_config.cmake)
    include(${ClickHouse_SOURCE_DIR}/src/Formats/configure_config.cmake)
-    configure_file(StorageSystemBuildOptions.generated.cpp.in ${CONFIG_BUILD})
+    configure_file(StorageSystemBuildOptions.cpp.in StorageSystemBuildOptions.generated.cpp)
 endfunction()
+
 generate_system_build_options()

 include("${ClickHouse_SOURCE_DIR}/cmake/dbms_glob_sources.cmake")
 add_headers_and_sources(storages_system .)
-list (APPEND storages_system_sources ${CONFIG_BUILD})
+list (APPEND storages_system_sources StorageSystemBuildOptions.generated.cpp)

 add_custom_target(generate-contributors
    ./StorageSystemContributors.sh
@ -78,6 +45,7 @@ list (APPEND storages_system_sources ${GENERATED_TIMEZONES_SRC})
 # Overlength strings
 set_source_files_properties(${GENERATED_LICENSES_SRC} PROPERTIES COMPILE_FLAGS -w)

+include(${ClickHouse_SOURCE_DIR}/cmake/embed_binary.cmake)
 clickhouse_embed_binaries(
        TARGET information_schema_metadata
        RESOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/InformationSchema/"
--- a/src/Storages/System/StorageSystemBuildOptions.generated.cpp.in
+++ b/src/Storages/System/StorageSystemBuildOptions.generated.cpp.in
@ -1,4 +1,4 @@
-// .cpp autogenerated by cmake
+// File was generated by CMake

 const char * auto_config_build[]
 {
--- a/tests/ci/compatibility_check.py
+++ b/tests/ci/compatibility_check.py
@ -99,6 +99,7 @@ def get_run_commands(
    return [
        f"readelf -s {build_path}/usr/bin/clickhouse | grep '@GLIBC_' > {result_folder}/glibc.log",
        f"readelf -s {build_path}/usr/bin/clickhouse-odbc-bridge | grep '@GLIBC_' >> {result_folder}/glibc.log",
+        f"readelf -s {build_path}/usr/bin/clickhouse-library-bridge | grep '@GLIBC_' >> {result_folder}/glibc.log",
        f"docker run --network=host --volume={build_path}/usr/bin/clickhouse:/clickhouse "
        f"--volume={build_path}/etc/clickhouse-server:/config "
        f"--volume={server_log_folder}:/var/log/clickhouse-server {image_ubuntu} > {result_folder}/ubuntu:12.04",
--- a/tests/ci/worker/ubuntu_ami_for_ci.sh
+++ b/tests/ci/worker/ubuntu_ami_for_ci.sh
@ -24,6 +24,11 @@ runner_arch() {
  esac
 }

+# We have test for cgroups, and it's broken with cgroups v2
+# Ubuntu 22.04 has it enabled by default
+sed -r '/GRUB_CMDLINE_LINUX=/ s/"(.*)"/"\1 systemd.unified_cgroup_hierarchy=0"/' -i /etc/default/grub
+update-grub
+
 apt-get update

 apt-get install --yes --no-install-recommends \
--- a/tests/config/config.d/legacy_geobase.xml
+++ b/tests/config/config.d/legacy_geobase.xml
@ -0,0 +1,4 @@
+<clickhouse>
+    <path_to_regions_hierarchy_file>/etc/clickhouse-server/config.d/regions_hierarchy.txt</path_to_regions_hierarchy_file>
+    <path_to_regions_names_files>/etc/clickhouse-server/config.d/</path_to_regions_names_files>
+</clickhouse>
--- a/tests/config/config.d/storage_conf.xml
+++ b/tests/config/config.d/storage_conf.xml
@ -8,6 +8,7 @@
                <endpoint>http://localhost:11111/test/00170_test/</endpoint>
                <access_key_id>clickhouse</access_key_id>
                <secret_access_key>clickhouse</secret_access_key>
+                <request_timeout_ms>20000</request_timeout_ms>
            </s3_disk>
            <s3_disk_2>
                <type>s3</type>
@ -15,6 +16,7 @@
                <endpoint>http://localhost:11111/test/00170_test/</endpoint>
                <access_key_id>clickhouse</access_key_id>
                <secret_access_key>clickhouse</secret_access_key>
+                <request_timeout_ms>20000</request_timeout_ms>
            </s3_disk_2>
            <s3_disk_3>
                <type>s3</type>
@ -22,6 +24,7 @@
                <endpoint>http://localhost:11111/test/00170_test/</endpoint>
                <access_key_id>clickhouse</access_key_id>
                <secret_access_key>clickhouse</secret_access_key>
+                <request_timeout_ms>20000</request_timeout_ms>
            </s3_disk_3>
            <s3_disk_4>
                <type>s3</type>
@ -29,6 +32,7 @@
                <endpoint>http://localhost:11111/test/00170_test/</endpoint>
                <access_key_id>clickhouse</access_key_id>
                <secret_access_key>clickhouse</secret_access_key>
+                <request_timeout_ms>20000</request_timeout_ms>
            </s3_disk_4>
            <s3_disk_5>
                <type>s3</type>
@ -36,6 +40,7 @@
                <endpoint>http://localhost:11111/test/00170_test/</endpoint>
                <access_key_id>clickhouse</access_key_id>
                <secret_access_key>clickhouse</secret_access_key>
+                <request_timeout_ms>20000</request_timeout_ms>
            </s3_disk_5>
            <s3_disk_6>
                <type>s3</type>
@ -43,6 +48,7 @@
                <endpoint>http://localhost:11111/test/00170_test/</endpoint>
                <access_key_id>clickhouse</access_key_id>
                <secret_access_key>clickhouse</secret_access_key>
+                <request_timeout_ms>20000</request_timeout_ms>
            </s3_disk_6>
            <!-- cache for s3 disks -->
            <s3_cache>
--- a/tests/config/install.sh
+++ b/tests/config/install.sh
@ -52,6 +52,12 @@ ln -sf $SRC_PATH/config.d/enable_zero_copy_replication.xml $DEST_SERVER_PATH/con
 ln -sf $SRC_PATH/config.d/nlp.xml $DEST_SERVER_PATH/config.d/
 ln -sf $SRC_PATH/config.d/enable_keeper_map.xml $DEST_SERVER_PATH/config.d/

+# Not supported with fasttest.
+if [ "${DEST_SERVER_PATH}" = "/etc/clickhouse-server" ]
+then
+   ln -sf $SRC_PATH/config.d/legacy_geobase.xml $DEST_SERVER_PATH/config.d/
+fi
+
 ln -sf $SRC_PATH/users.d/log_queries.xml $DEST_SERVER_PATH/users.d/
 ln -sf $SRC_PATH/users.d/readonly.xml $DEST_SERVER_PATH/users.d/
 ln -sf $SRC_PATH/users.d/access_management.xml $DEST_SERVER_PATH/users.d/
@ -78,6 +84,8 @@ ln -sf $SRC_PATH/executable_pool_dictionary.xml $DEST_SERVER_PATH/
 ln -sf $SRC_PATH/test_function.xml $DEST_SERVER_PATH/

 ln -sf $SRC_PATH/top_level_domains $DEST_SERVER_PATH/
+ln -sf $SRC_PATH/regions_hierarchy.txt $DEST_SERVER_PATH/config.d/
+ln -sf $SRC_PATH/regions_names_en.txt $DEST_SERVER_PATH/config.d/

 ln -sf $SRC_PATH/ext-en.txt $DEST_SERVER_PATH/config.d/
 ln -sf $SRC_PATH/ext-ru.txt $DEST_SERVER_PATH/config.d/
@ -110,9 +118,6 @@ fi

 if [[ -n "$USE_S3_STORAGE_FOR_MERGE_TREE" ]] && [[ "$USE_S3_STORAGE_FOR_MERGE_TREE" -eq 1 ]]; then
    ln -sf $SRC_PATH/config.d/s3_storage_policy_by_default.xml $DEST_SERVER_PATH/config.d/
-    # Too verbose logging in S3 tests
-    rm -f $DEST_SERVER_PATH/config.d/logger_test.xml
-    ln -sf $SRC_PATH/config.d/logger_trace.xml $DEST_SERVER_PATH/config.d/
 fi

 ARM="aarch64"
--- a/tests/config/regions_hierarchy.txt
+++ b/tests/config/regions_hierarchy.txt
@ -0,0 +1,12 @@
+1	0	0	7000000000
+2	10	3	330000000
+3	2	4	5700000
+4	3	5	330000
+5	4	6	100000
+6	12	3	1500000000
+7	6	4	83000000
+8	7	6	20000000
+9	1	1	1000000000
+10	9	1	600000000
+11	1	1	5300000000
+12	11	1	4700000000
--- a/tests/config/regions_names_en.txt
+++ b/tests/config/regions_names_en.txt
@ -0,0 +1,12 @@
+1	World
+2	USA
+3	Colorado
+4	Boulder County
+5	Boulder
+6	China
+7	Sichuan
+8	Chengdu
+9	America
+10	North America
+11	Eurasia
+12	Asia
--- a/tests/integration/helpers/cluster.py
+++ b/tests/integration/helpers/cluster.py
@ -816,7 +816,6 @@ class ClickHouseCluster:
            env_variables[f"keeper_config_dir{i}"] = configs_dir
            env_variables[f"keeper_db_dir{i}"] = coordination_dir
            self.zookeeper_dirs_to_create += [logs_dir, configs_dir, coordination_dir]
-        logging.debug(f"DEBUG KEEPER: {self.zookeeper_dirs_to_create}")

        self.with_zookeeper = True
        self.base_cmd.extend(["--file", keeper_docker_compose_path])
@ -4108,6 +4107,9 @@ class ClickHouseInstance:
    def get_backuped_s3_objects(self, disk, backup_name):
        path = f"/var/lib/clickhouse/disks/{disk}/shadow/{backup_name}/store"
        self.wait_for_path_exists(path, 10)
+        return self.get_s3_objects(path)
+
+    def get_s3_objects(self, path):
        command = [
            "find",
            path,
@ -4120,8 +4122,45 @@ class ClickHouseInstance:
            "{}",
            ";",
        ]
+
        return self.exec_in_container(command).split("\n")

+    def get_s3_data_objects(self, path):
+        command = [
+            "find",
+            path,
+            "-type",
+            "f",
+            "-name",
+            "*.bin",
+            "-exec",
+            "grep",
+            "-o",
+            "r[01]\\{64\\}-file-[[:lower:]]\\{32\\}",
+            "{}",
+            ";",
+        ]
+        return self.exec_in_container(command).split("\n")
+
+    def get_table_objects(self, table, database=None):
+        objects = []
+        database_query = ""
+        if database:
+            database_query = f"AND database='{database}'"
+        data_paths = self.query(
+            f"""
+            SELECT arrayJoin(data_paths)
+            FROM system.tables
+            WHERE name='{table}'
+            {database_query}
+            """
+        )
+        paths = data_paths.split("\n")
+        for path in paths:
+            if path:
+                objects = objects + self.get_s3_data_objects(path)
+        return objects
+

 class ClickHouseKiller(object):
    def __init__(self, clickhouse_node):
--- a/tests/integration/test_compressed_marks_restart/init.py
+++ b/tests/integration/test_compressed_marks_restart/init.py
@ -0,0 +1 @@
+#!/usr/bin/env python3
--- a/tests/integration/test_compressed_marks_restart/test.py
+++ b/tests/integration/test_compressed_marks_restart/test.py
@ -0,0 +1,58 @@
+#!/usr/bin/env python3
+import pytest
+from helpers.cluster import ClickHouseCluster
+
+cluster = ClickHouseCluster(__file__)
+node = cluster.add_instance("node", stay_alive=True)
+
+
+@pytest.fixture(scope="module", autouse=True)
+def started_cluster():
+    try:
+        cluster.start()
+        yield cluster
+
+    finally:
+        cluster.shutdown()
+
+
+def test_compressed_marks_restart_compact():
+    node.query(
+        "create table test_02381_compact (a UInt64, b String) ENGINE = MergeTree order by (a, b)"
+    )
+    node.query("insert into test_02381_compact values (1, 'Hello')")
+    node.query(
+        "alter table test_02381_compact modify setting compress_marks=true, compress_primary_key=true"
+    )
+    node.query("insert into test_02381_compact values (2, 'World')")
+    node.query("optimize table test_02381_compact final")
+
+    assert (
+        node.query("SELECT count() FROM test_02381_compact WHERE not ignore(*)")
+        == "2\n"
+    )
+    node.restart_clickhouse()
+    assert (
+        node.query("SELECT count() FROM test_02381_compact WHERE not ignore(*)")
+        == "2\n"
+    )
+
+
+def test_compressed_marks_restart_wide():
+    node.query(
+        "create table test_02381_wide (a UInt64, b String) ENGINE = MergeTree order by (a, b) SETTINGS min_bytes_for_wide_part=0"
+    )
+    node.query("insert into test_02381_wide values (1, 'Hello')")
+    node.query(
+        "alter table test_02381_wide modify setting compress_marks=true, compress_primary_key=true"
+    )
+    node.query("insert into test_02381_wide values (2, 'World')")
+    node.query("optimize table test_02381_wide final")
+
+    assert (
+        node.query("SELECT count() FROM test_02381_wide WHERE not ignore(*)") == "2\n"
+    )
+    node.restart_clickhouse()
+    assert (
+        node.query("SELECT count() FROM test_02381_wide WHERE not ignore(*)") == "2\n"
+    )
--- a/tests/integration/test_lost_part/test.py
+++ b/tests/integration/test_lost_part/test.py
@ -73,6 +73,9 @@ def test_lost_part_same_replica(start_cluster):
    node1.query("ATTACH TABLE mt0")

    node1.query("SYSTEM START MERGES mt0")
+    res, err = node1.http_query_and_get_answer_with_error("SYSTEM SYNC REPLICA mt0")
+    print("result: ", res)
+    print("error: ", res)

    for i in range(10):
        result = node1.query("SELECT count() FROM system.replication_queue")
@ -133,6 +136,9 @@ def test_lost_part_other_replica(start_cluster):
    node1.query("CHECK TABLE mt1")

    node2.query("SYSTEM START REPLICATION QUEUES")
+    res, err = node1.http_query_and_get_answer_with_error("SYSTEM SYNC REPLICA mt1")
+    print("result: ", res)
+    print("error: ", res)

    for i in range(10):
        result = node2.query("SELECT count() FROM system.replication_queue")
@ -190,6 +196,9 @@ def test_lost_part_mutation(start_cluster):
    node1.query("CHECK TABLE mt2")

    node1.query("SYSTEM START MERGES mt2")
+    res, err = node1.http_query_and_get_answer_with_error("SYSTEM SYNC REPLICA mt2")
+    print("result: ", res)
+    print("error: ", res)

    for i in range(10):
        result = node1.query("SELECT count() FROM system.replication_queue")
@ -237,10 +246,13 @@ def test_lost_last_part(start_cluster):
    node1.query("CHECK TABLE mt3")

    node1.query("SYSTEM START MERGES mt3")
+    res, err = node1.http_query_and_get_answer_with_error("SYSTEM SYNC REPLICA mt3")
+    print("result: ", res)
+    print("error: ", res)

    for i in range(10):
        result = node1.query("SELECT count() FROM system.replication_queue")
-        assert int(result) <= 1, "Have a lot of entries in queue {}".format(
+        assert int(result) <= 2, "Have a lot of entries in queue {}".format(
            node1.query("SELECT * FROM system.replication_queue FORMAT Vertical")
        )
        if node1.contains_in_log("Cannot create empty part") and node1.contains_in_log(
--- a/tests/integration/test_recovery_replica/test.py
+++ b/tests/integration/test_recovery_replica/test.py
@ -196,6 +196,7 @@ def test_update_metadata(start_cluster):

    node1.query("ALTER TABLE update_metadata MODIFY COLUMN col1 String")
    node1.query("ALTER TABLE update_metadata ADD COLUMN col2 INT")
+    node3.query("SYSTEM SYNC REPLICA update_metadata")
    for i in range(1, 11):
        node3.query(
            "INSERT INTO update_metadata VALUES ({}, '{}', {})".format(
--- a/tests/integration/test_s3_zero_copy_replication/test.py
+++ b/tests/integration/test_s3_zero_copy_replication/test.py
@ -8,11 +8,12 @@ from helpers.cluster import ClickHouseCluster
 logging.getLogger().setLevel(logging.INFO)
 logging.getLogger().addHandler(logging.StreamHandler())

+cluster = ClickHouseCluster(__file__)
+

@pytest.fixture(scope="module")
-def cluster():
+def started_cluster():
    try:
-        cluster = ClickHouseCluster(__file__)
        cluster.add_instance(
            "node1",
            main_configs=["configs/config.d/s3.xml"],
@ -96,7 +97,7 @@ def wait_for_active_parts(node, num_expected_parts, table_name, timeout=30):
 # Result of `get_large_objects_count` can be changed in other tests, so run this case at the beginning
@pytest.mark.order(0)
@pytest.mark.parametrize("policy", ["s3"])
-def test_s3_zero_copy_replication(cluster, policy):
+def test_s3_zero_copy_replication(started_cluster, policy):
    node1 = cluster.instances["node1"]
    node2 = cluster.instances["node2"]

@ -153,7 +154,7 @@ def test_s3_zero_copy_replication(cluster, policy):


@pytest.mark.skip(reason="Test is flaky (and never was stable)")
-def test_s3_zero_copy_on_hybrid_storage(cluster):
+def test_s3_zero_copy_on_hybrid_storage(started_cluster):
    node1 = cluster.instances["node1"]
    node2 = cluster.instances["node2"]

@ -268,7 +269,9 @@ def insert_large_data(node, table):
        ("tiered_copy", True, 3),
    ],
 )
-def test_s3_zero_copy_with_ttl_move(cluster, storage_policy, large_data, iterations):
+def test_s3_zero_copy_with_ttl_move(
+    started_cluster, storage_policy, large_data, iterations
+):
    node1 = cluster.instances["node1"]
    node2 = cluster.instances["node2"]

@ -333,7 +336,7 @@ def test_s3_zero_copy_with_ttl_move(cluster, storage_policy, large_data, iterati
        (True, 3),
    ],
 )
-def test_s3_zero_copy_with_ttl_delete(cluster, large_data, iterations):
+def test_s3_zero_copy_with_ttl_delete(started_cluster, large_data, iterations):
    node1 = cluster.instances["node1"]
    node2 = cluster.instances["node2"]

@ -415,6 +418,22 @@ def wait_mutations(node, table, seconds):
    assert mutations == "0\n"


+def wait_for_clean_old_parts(node, table, seconds):
+    time.sleep(1)
+    while seconds > 0:
+        seconds -= 1
+        parts = node.query(
+            f"SELECT count() FROM system.parts WHERE table='{table}' AND active=0"
+        )
+        if parts == "0\n":
+            return
+        time.sleep(1)
+    parts = node.query(
+        f"SELECT count() FROM system.parts WHERE table='{table}' AND active=0"
+    )
+    assert parts == "0\n"
+
+
 def s3_zero_copy_unfreeze_base(cluster, unfreeze_query_template):
    node1 = cluster.instances["node1"]
    node2 = cluster.instances["node2"]
@ -435,6 +454,8 @@ def s3_zero_copy_unfreeze_base(cluster, unfreeze_query_template):

    node1.query("INSERT INTO unfreeze_test VALUES (0)")

+    wait_for_active_parts(node2, 1, "unfreeze_test")
+
    node1.query("ALTER TABLE unfreeze_test FREEZE WITH NAME 'freeze_backup1'")
    node2.query("ALTER TABLE unfreeze_test FREEZE WITH NAME 'freeze_backup2'")
    wait_mutations(node1, "unfreeze_test", 10)
@ -472,11 +493,11 @@ def s3_zero_copy_unfreeze_base(cluster, unfreeze_query_template):
    node2.query("DROP TABLE IF EXISTS unfreeze_test NO DELAY")


-def test_s3_zero_copy_unfreeze_alter(cluster):
+def test_s3_zero_copy_unfreeze_alter(started_cluster):
    s3_zero_copy_unfreeze_base(cluster, "ALTER TABLE unfreeze_test UNFREEZE WITH NAME")


-def test_s3_zero_copy_unfreeze_system(cluster):
+def test_s3_zero_copy_unfreeze_system(started_cluster):
    s3_zero_copy_unfreeze_base(cluster, "SYSTEM UNFREEZE WITH NAME")


@ -565,17 +586,17 @@ def s3_zero_copy_drop_detached(cluster, unfreeze_query_template):
    check_objects_not_exisis(cluster, objects1)


-def test_s3_zero_copy_drop_detached_alter(cluster):
+def test_s3_zero_copy_drop_detached_alter(started_cluster):
    s3_zero_copy_drop_detached(
        cluster, "ALTER TABLE drop_detached_test UNFREEZE WITH NAME"
    )


-def test_s3_zero_copy_drop_detached_system(cluster):
+def test_s3_zero_copy_drop_detached_system(started_cluster):
    s3_zero_copy_drop_detached(cluster, "SYSTEM UNFREEZE WITH NAME")


-def test_s3_zero_copy_concurrent_merge(cluster):
+def test_s3_zero_copy_concurrent_merge(started_cluster):
    node1 = cluster.instances["node1"]
    node2 = cluster.instances["node2"]

@ -620,3 +641,119 @@ def test_s3_zero_copy_concurrent_merge(cluster):

    for node in (node1, node2):
        assert node.query("select sum(id) from concurrent_merge").strip() == "1600"
+
+
+def test_s3_zero_copy_keeps_data_after_mutation(started_cluster):
+    node1 = cluster.instances["node1"]
+    node2 = cluster.instances["node2"]
+
+    node1.query("DROP TABLE IF EXISTS zero_copy_mutation NO DELAY")
+    node2.query("DROP TABLE IF EXISTS zero_copy_mutation NO DELAY")
+
+    node1.query(
+        """
+        CREATE TABLE zero_copy_mutation (id UInt64, value1 String, value2 String, value3 String)
+        ENGINE=ReplicatedMergeTree('/clickhouse/tables/zero_copy_mutation', '{replica}')
+        ORDER BY id
+        PARTITION BY (id % 4)
+        SETTINGS storage_policy='s3',
+        old_parts_lifetime=1000
+        """
+    )
+
+    node2.query(
+        """
+        CREATE TABLE zero_copy_mutation (id UInt64, value1 String, value2 String, value3 String)
+        ENGINE=ReplicatedMergeTree('/clickhouse/tables/zero_copy_mutation', '{replica}')
+        ORDER BY id
+        PARTITION BY (id % 4)
+        SETTINGS storage_policy='s3',
+        old_parts_lifetime=1000
+        """
+    )
+
+    node1.query(
+        """
+        INSERT INTO zero_copy_mutation
+        SELECT * FROM generateRandom('id UInt64, value1 String, value2 String, value3 String') limit 1000000
+        """
+    )
+
+    wait_for_active_parts(node2, 4, "zero_copy_mutation")
+
+    objects1 = node1.get_table_objects("zero_copy_mutation")
+    check_objects_exisis(cluster, objects1)
+
+    node1.query(
+        """
+        ALTER TABLE zero_copy_mutation
+        ADD COLUMN valueX String MATERIALIZED value1
+        """
+    )
+
+    node1.query(
+        """
+        ALTER TABLE zero_copy_mutation
+        MATERIALIZE COLUMN valueX
+        """
+    )
+
+    wait_mutations(node1, "zero_copy_mutation", 10)
+    wait_mutations(node2, "zero_copy_mutation", 10)
+
+    # If bug present at least one node has metadata with incorrect ref_count values.
+    # But it may be any node depends on mutation execution order.
+    # We can try to find one, but this required knowledge about internal metadata structure.
+    # It can be change in future, so we do not find this node here.
+    # And with the bug test may be success sometimes.
+    nodeX = node1
+    nodeY = node2
+
+    objectsY = nodeY.get_table_objects("zero_copy_mutation")
+    check_objects_exisis(cluster, objectsY)
+
+    nodeX.query(
+        """
+        ALTER TABLE zero_copy_mutation
+        DETACH PARTITION '0'
+        """
+    )
+
+    nodeX.query(
+        """
+        ALTER TABLE zero_copy_mutation
+        ATTACH PARTITION '0'
+        """
+    )
+
+    wait_mutations(node1, "zero_copy_mutation", 10)
+    wait_mutations(node2, "zero_copy_mutation", 10)
+
+    nodeX.query(
+        """
+        DROP TABLE zero_copy_mutation SYNC
+        """
+    )
+
+    # time to remove objects
+    time.sleep(10)
+
+    nodeY.query(
+        """
+        SELECT count() FROM zero_copy_mutation
+        WHERE value3 LIKE '%ab%'
+        """
+    )
+
+    check_objects_exisis(cluster, objectsY)
+
+    nodeY.query(
+        """
+        DROP TABLE zero_copy_mutation SYNC
+        """
+    )
+
+    # time to remove objects
+    time.sleep(10)
+
+    check_objects_not_exisis(cluster, objectsY)
--- a/tests/jepsen.clickhouse-keeper/project.clj
+++ b/tests/jepsen.clickhouse-keeper/project.clj
@ -7,7 +7,8 @@
  :main jepsen.clickhouse-keeper.main
  :plugins [[lein-cljfmt "0.7.0"]]
  :dependencies [[org.clojure/clojure "1.10.1"]
-                 [jepsen "0.2.6"]
+                 [jepsen "0.2.7"]
                 [zookeeper-clj "0.9.4"]
+                 [com.hierynomus/sshj "0.34.0"]
                 [org.apache.zookeeper/zookeeper "3.6.1" :exclusions [org.slf4j/slf4j-log4j12]]]
  :repl-options {:init-ns jepsen.clickhouse-keeper.main})
--- a/tests/jepsen.clickhouse-keeper/src/jepsen/clickhouse_keeper/main.clj
+++ b/tests/jepsen.clickhouse-keeper/src/jepsen/clickhouse_keeper/main.clj
@ -1,3 +1,98 @@
+(ns jepsen.control.sshj
+  (:require [jepsen.control [core :as core]
+                            [sshj :as sshj]]
+            [slingshot.slingshot :refer [try+ throw+]])
+  (:import (net.schmizz.sshj SSHClient
+                            DefaultConfig)
+           (net.schmizz.sshj.transport.verification PromiscuousVerifier)
+           (java.util.concurrent Semaphore)))
+
+(defrecord SSHJRemote [concurrency-limit
+                       conn-spec
+                       ^SSHClient client
+                       ^Semaphore semaphore]
+  core/Remote
+  (connect [this conn-spec]
+    (if (:dummy conn-spec)
+      (assoc this :conn-spec conn-spec)
+      (try+ (let [c (as-> (SSHClient.) client
+                      (do
+                        (if (:strict-host-key-checking conn-spec)
+                          (.loadKnownHosts client)
+                          (.addHostKeyVerifier client (PromiscuousVerifier.)))
+                        (.connect client (:host conn-spec) (:port conn-spec))
+                        (auth! client conn-spec)
+                        client))]
+              (assoc this
+                     :conn-spec conn-spec
+                     :client c
+                     :semaphore (Semaphore. concurrency-limit true)))
+            (catch Exception e
+              ; SSHJ wraps InterruptedException in its own exceptions, so we
+              ; have to see through that and rethrow properly.
+              (let [cause (util/ex-root-cause e)]
+                (when (instance? InterruptedException cause)
+                  (throw cause)))
+              (throw+ (assoc conn-spec
+                             :type    :jepsen.control/session-error
+                             :message "Error opening SSH session. Verify username, password, and node hostnames are correct."))))))
+
+  (disconnect! [this]
+    (when-let [c client]
+      (.close c)))
+
+  (execute! [this ctx action]
+    ;  (info :permits (.availablePermits semaphore))
+    (when (:dummy conn-spec)
+      (throw+ {:type :jepsen.control/dummy}))
+    (.acquire semaphore)
+    (sshj/with-errors conn-spec ctx
+      (try
+        (with-open [session (.startSession client)]
+          (let [cmd (.exec session (:cmd action))
+                ; Feed it input
+                _ (when-let [input (:in action)]
+                    (let [stream (.getOutputStream cmd)]
+                      (bs/transfer input stream)
+                      (send-eof! client session)
+                      (.close stream)))
+                ; Read output
+                out (.toString (IOUtils/readFully (.getInputStream cmd)))
+                err (.toString (IOUtils/readFully (.getErrorStream cmd)))
+                ; Wait on command
+                _ (.join cmd)]
+            ; Return completion
+            (assoc action
+                   :out   out
+                   :err   err
+                   ; There's also a .getExitErrorMessage that might be
+                   ; interesting here?
+                   :exit  (.getExitStatus cmd))))
+        (finally
+          (.release semaphore)))))
+
+  (upload! [this ctx local-paths remote-path _opts]
+    (when (:dummy conn-spec)
+      (throw+ {:type :jepsen.control/dummy}))
+    (with-errors conn-spec ctx
+      (with-open [sftp (.newSFTPClient client)]
+        (.put sftp (FileSystemFile. local-paths) remote-path))))
+
+  (download! [this ctx remote-paths local-path _opts]
+    (when (:dummy conn-spec)
+      (throw+ {:type :jepsen.control/dummy}))
+    (with-errors conn-spec ctx
+      (with-open [sftp (.newSFTPClient client)]
+        (.get sftp remote-paths (FileSystemFile. local-path))))))
+
+(defn remote
+  "Constructs an SSHJ remote."
+  []
+  (-> (SSHJRemote. concurrency-limit nil nil nil)
+      ; We *can* use our own SCP, but shelling out is faster.
+      scp/remote
+      retry/remote))
+
 (ns jepsen.clickhouse-keeper.main
  (:require [clojure.tools.logging :refer :all]
            [jepsen.clickhouse-keeper.utils :refer :all]
@ -17,7 +112,6 @@
             [checker :as checker]
             [cli :as cli]
             [client :as client]
-             [control :as c]
             [db :as db]
             [nemesis :as nemesis]
             [generator :as gen]
--- a/tests/queries/0_stateless/00732_base64_functions.sql
+++ b/tests/queries/0_stateless/00732_base64_functions.sql
@ -1,8 +1,16 @@
 -- Tags: no-fasttest

 SET send_logs_level = 'fatal';
+
 SELECT base64Encode(val) FROM (select arrayJoin(['', 'f', 'fo', 'foo', 'foob', 'fooba', 'foobar']) val);
+
 SELECT base64Decode(val) FROM (select arrayJoin(['', 'Zg==', 'Zm8=', 'Zm9v', 'Zm9vYg==', 'Zm9vYmE=', 'Zm9vYmFy']) val);
 SELECT base64Decode(base64Encode('foo')) = 'foo', base64Encode(base64Decode('Zm9v')) == 'Zm9v';
+
 SELECT tryBase64Decode('Zm9vYmF=Zm9v');
-SELECT base64Decode('Zm9vYmF=Zm9v'); -- { serverError 117 }
+
+SELECT base64Encode(val, 'excess argument') FROM (select arrayJoin(['', 'f', 'fo', 'foo', 'foob', 'fooba', 'foobar']) val); -- { serverError 42 }
+SELECT base64Decode(val, 'excess argument') FROM (select arrayJoin(['', 'Zg==', 'Zm8=', 'Zm9v', 'Zm9vYg==', 'Zm9vYmE=', 'Zm9vYmFy']) val); -- { serverError 42 }
+SELECT tryBase64Decode('Zm9vYmF=Zm9v', 'excess argument'); -- { serverError 42 }
+
+SELECT base64Decode('Zm9vYmF=Zm9v'); -- { serverError 117 }
--- a/tests/queries/0_stateless/01111_create_drop_replicated_db_stress.sh
+++ b/tests/queries/0_stateless/01111_create_drop_replicated_db_stress.sh
@ -26,8 +26,8 @@ function drop_db()
 {
    while true; do
        database=$($CLICKHOUSE_CLIENT -q "select name from system.databases where name like '${CLICKHOUSE_DATABASE}%' order by rand() limit 1")
-        if [[ "$database" == "$CLICKHOUSE_DATABASE" ]]; then return; fi
-        if [ -z "$database" ]; then return; fi
+        if [[ "$database" == "$CLICKHOUSE_DATABASE" ]]; then continue; fi
+        if [ -z "$database" ]; then continue; fi
        $CLICKHOUSE_CLIENT -n --query \
        "drop database if exists $database" 2>&1| grep -Fa "Exception: "
        sleep 0.$RANDOM
@ -38,7 +38,7 @@ function sync_db()
 {
    while true; do
        database=$($CLICKHOUSE_CLIENT -q "select name from system.databases where name like '${CLICKHOUSE_DATABASE}%' order by rand() limit 1")
-        if [ -z "$database" ]; then return; fi
+        if [ -z "$database" ]; then continue; fi
        $CLICKHOUSE_CLIENT --receive_timeout=1 -q \
        "system sync database replica $database" 2>&1| grep -Fa "Exception: " | grep -Fv TIMEOUT_EXCEEDED | grep -Fv "only with Replicated engine" | grep -Fv UNKNOWN_DATABASE
        sleep 0.$RANDOM
@ -49,7 +49,7 @@ function create_table()
 {
    while true; do
        database=$($CLICKHOUSE_CLIENT -q "select name from system.databases where name like '${CLICKHOUSE_DATABASE}%' order by rand() limit 1")
-        if [ -z "$database" ]; then return; fi
+        if [ -z "$database" ]; then continue; fi
        $CLICKHOUSE_CLIENT --distributed_ddl_task_timeout=0 -q \
        "create table $database.rmt_${RANDOM}_${RANDOM}_${RANDOM} (n int) engine=ReplicatedMergeTree order by tuple() -- suppress $CLICKHOUSE_TEST_ZOOKEEPER_PREFIX" \
        2>&1| grep -Fa "Exception: " | grep -Fv "Macro 'uuid' and empty arguments" | grep -Fv "Cannot enqueue query" | grep -Fv "ZooKeeper session expired" | grep -Fv UNKNOWN_DATABASE
@ -61,9 +61,9 @@ function alter_table()
 {
    while true; do
        table=$($CLICKHOUSE_CLIENT -q "select database || '.' || name from system.tables where database like '${CLICKHOUSE_DATABASE}%' order by rand() limit 1")
-        if [ -z "$table" ]; then return; fi
+        if [ -z "$table" ]; then continue; fi
        $CLICKHOUSE_CLIENT --distributed_ddl_task_timeout=0 -q \
-        "alter table $table on cluster $database update n = n + (select max(n) from merge(REGEXP('${CLICKHOUSE_DATABASE}.*'), '.*')) where 1 settings allow_nondeterministic_mutations=1" \
+        "alter table $table update n = n + (select max(n) from merge(REGEXP('${CLICKHOUSE_DATABASE}.*'), '.*')) where 1 settings allow_nondeterministic_mutations=1" \
        2>&1| grep -Fa "Exception: " | grep -Fv "Cannot enqueue query" | grep -Fv "ZooKeeper session expired" | grep -Fv UNKNOWN_DATABASE | grep -Fv UNKNOWN_TABLE | grep -Fv TABLE_IS_READ_ONLY
        sleep 0.$RANDOM
    done
@ -73,7 +73,7 @@ function insert()
 {
    while true; do
        table=$($CLICKHOUSE_CLIENT -q "select database || '.' || name from system.tables where database like '${CLICKHOUSE_DATABASE}%' order by rand() limit 1")
-        if [ -z "$table" ]; then return; fi
+        if [ -z "$table" ]; then continue; fi
        $CLICKHOUSE_CLIENT -q \
        "insert into $table values ($RANDOM)" 2>&1| grep -Fa "Exception: " | grep -Fv UNKNOWN_DATABASE | grep -Fv UNKNOWN_TABLE | grep -Fv TABLE_IS_READ_ONLY
    done
--- a/tests/queries/0_stateless/01825_type_json_schema_inference.sh
+++ b/tests/queries/0_stateless/01825_type_json_schema_inference.sh
@ -19,7 +19,8 @@ filename="${user_files_path}/${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json"
 echo '{"id": 1, "obj": {"k1": 1, "k2": {"k3": 2, "k4": [{"k5": 3}, {"k5": 4}]}}, "s": "foo"}' > $filename
 echo '{"id": 2, "obj": {"k2": {"k3": "str", "k4": [{"k6": 55}]}, "some": 42}, "s": "bar"}' >> $filename

-${CLICKHOUSE_CLIENT} -q "INSERT INTO t_json_inference SELECT * FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow')"
+${CLICKHOUSE_CLIENT} -q "INSERT INTO t_json_inference SELECT * FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow')" --allow_experimental_object_type 1
+
 ${CLICKHOUSE_CLIENT} -q "SELECT * FROM t_json_inference FORMAT JSONEachRow" --output_format_json_named_tuples_as_objects 1
 ${CLICKHOUSE_CLIENT} -q "SELECT toTypeName(obj) FROM t_json_inference LIMIT 1"

@ -30,7 +31,7 @@ ${CLICKHOUSE_CLIENT} -q "CREATE TABLE t_json_inference (id UInt64, obj String, s
 echo '{"obj": "aaa", "id": 1, "s": "foo"}' > $filename
 echo '{"id": 2, "obj": "bbb", "s": "bar"}' >> $filename

-${CLICKHOUSE_CLIENT} -q "INSERT INTO t_json_inference SELECT * FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow')"
+${CLICKHOUSE_CLIENT} -q "INSERT INTO t_json_inference SELECT * FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow')" --allow_experimental_object_type 1
 ${CLICKHOUSE_CLIENT} -q "SELECT * FROM t_json_inference FORMAT JSONEachRow" --output_format_json_named_tuples_as_objects 1

 ${CLICKHOUSE_CLIENT} -q "DROP TABLE IF EXISTS t_json_inference"
@ -38,14 +39,14 @@ ${CLICKHOUSE_CLIENT} -q "DROP TABLE IF EXISTS t_json_inference"
 echo '{"map": {"k1": 1, "k2": 2}, "obj": {"k1": 1, "k2": {"k3": 2}}}' > $filename

 ${CLICKHOUSE_CLIENT} -q "SELECT map, obj, toTypeName(map) AS map_type, toTypeName(obj) AS obj_type \
-    FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow') FORMAT JSONEachRow" --output_format_json_named_tuples_as_objects 1
+    FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow') FORMAT JSONEachRow" --output_format_json_named_tuples_as_objects 1 --allow_experimental_object_type 1

 ${CLICKHOUSE_CLIENT} -q "CREATE TABLE t_json_inference (obj JSON, map Map(String, UInt64)) \
    ENGINE = MergeTree ORDER BY tuple()" --allow_experimental_object_type 1

 echo '{"map": {"k1": 1, "k2": 2}, "obj": {"k1": 1, "k2": 2}}' > $filename

-${CLICKHOUSE_CLIENT} -q "INSERT INTO t_json_inference SELECT * FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow')"
+${CLICKHOUSE_CLIENT} -q "INSERT INTO t_json_inference SELECT * FROM file('${CLICKHOUSE_TEST_UNIQUE_NAME}/data.json', 'JSONEachRow')" --allow_experimental_object_type 1
 ${CLICKHOUSE_CLIENT} -q "SELECT * FROM t_json_inference FORMAT JSONEachRow" --output_format_json_named_tuples_as_objects 1
 ${CLICKHOUSE_CLIENT} -q "SELECT toTypeName(obj) FROM t_json_inference LIMIT 1"

--- a/tests/queries/0_stateless/02147_order_by_optimizations.sql
+++ b/tests/queries/0_stateless/02147_order_by_optimizations.sql
@ -13,3 +13,7 @@ SET optimize_monotonous_functions_in_order_by = 1;
 EXPLAIN SYNTAX SELECT * FROM t_02147 ORDER BY toStartOfHour(date), v;
 EXPLAIN SYNTAX SELECT * FROM t_02147_dist ORDER BY toStartOfHour(date), v;
 EXPLAIN SYNTAX SELECT * FROM t_02147_merge ORDER BY toStartOfHour(date), v;
+
+drop table t_02147;
+CREATE TABLE t_02147 (date DateTime, v UInt32) ENGINE = MergeTree ORDER BY date;
+select *, toString(t.v) as s from t_02147_merge as t order by date, s;
--- a/tests/queries/0_stateless/02268_json_maps_and_objects.sql
+++ b/tests/queries/0_stateless/02268_json_maps_and_objects.sql
@ -1,4 +1,5 @@
 -- Tags: no-fasttest
+set allow_experimental_object_type=1;
 desc format(JSONEachRow, '{"x" : {"a" : "Some string"}}, {"x" : {"b" : [1, 2, 3]}}, {"x" : {"c" : {"d" : 10}}}');
 desc format(JSONEachRow, '{"x" : {"a" : "Some string"}}, {"x" : {"b" : [1, 2, 3], "c" : {"42" : 42}}}');
 desc format(JSONEachRow, '{"x" : [{"a" : "Some string"}]}, {"x" : [{"b" : [1, 2, 3]}]}');
--- a/tests/queries/0_stateless/02326_numbers_from_json_strings_schema_inference.sql
+++ b/tests/queries/0_stateless/02326_numbers_from_json_strings_schema_inference.sql
@ -1,6 +1,7 @@
 -- Tags: no-fasttest

 set input_format_json_try_infer_numbers_from_strings=1;
+set allow_experimental_object_type=1;

 desc format(JSONEachRow, '{"x" : "123"}');
 desc format(JSONEachRow, '{"x" : ["123", 123, 12.3]}');
--- a/tests/queries/0_stateless/02337_base58.reference
+++ b/tests/queries/0_stateless/02337_base58.reference
@ -8,6 +8,22 @@ fooba
 foobar
 Hello world!

+f
+fo
+foo
+foob
+fooba
+foobar
+Hello world!
+
+
+foob
+
+
+
+foobar
+
+
 2m
 8o8
 bQbp
--- a/Show More
+++ b/Show More
				`@ -1 +0,0 @@`
				`Subproject commit 0db5bf5bd2452cd8f1283a1fcdc04845af705bfc`
				`@ -0,0 +1 @@`
				`Subproject commit 6ca2b5b3927226f6bcf6c656f502ff5d012ad9b6`