Merge branch 'master' into Avogar-patch-3

.github/workflows/master.yml

@@ -149,7 +149,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH"
   SplitBuildSmokeTest:
     needs: [BuilderDebSplitted]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, style-checker]
     steps:
       - name: Set envs
@@ -316,7 +315,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinRelease:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -360,6 +358,51 @@ jobs:
           docker kill "$(docker ps -q)" ||:
           docker rm -f "$(docker ps -a -q)" ||:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
+  BuilderBinGCC:
+    needs: [DockerHubPush]
+    runs-on: [self-hosted, builder]
+    steps:
+      - name: Set envs
+        run: |
+          cat >> "$GITHUB_ENV" << 'EOF'
+          TEMP_PATH=${{runner.temp}}/build_check
+          IMAGES_PATH=${{runner.temp}}/images_path
+          REPO_COPY=${{runner.temp}}/build_check/ClickHouse
+          CACHES_PATH=${{runner.temp}}/../ccaches
+          CHECK_NAME=ClickHouse build check (actions)
+          BUILD_NAME=binary_gcc
+          EOF
+      - name: Download changed images
+        uses: actions/download-artifact@v2
+        with:
+          name: changed_images
+          path: ${{ env.IMAGES_PATH }}
+      - name: Clear repository
+        run: |
+          sudo rm -fr "$GITHUB_WORKSPACE" && mkdir "$GITHUB_WORKSPACE"
+      - name: Check out repository code
+        uses: actions/checkout@v2
+        with:
+          submodules: 'true'
+          fetch-depth: 0 # otherwise we will have no info about contributors
+      - name: Build
+        run: |
+          sudo rm -fr "$TEMP_PATH"
+          mkdir -p "$TEMP_PATH"
+          cp -r "$GITHUB_WORKSPACE" "$TEMP_PATH"
+          cd "$REPO_COPY/tests/ci" && python3 build_check.py "$CHECK_NAME" "$BUILD_NAME"
+      - name: Upload build URLs to artifacts
+        if: ${{ success() || failure() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ env.BUILD_NAME }}
+          path: ${{ env.TEMP_PATH }}/${{ env.BUILD_NAME }}.json
+      - name: Cleanup
+        if: always()
+        run: |
+          docker kill "$(docker ps -q)" ||:
+          docker rm -f "$(docker ps -a -q)" ||:
+          sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderDebAsan:
     needs: [DockerHubPush]
     runs-on: [self-hosted, builder]
@@ -590,7 +633,6 @@ jobs:
 ##########################################################################################
   BuilderDebSplitted:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -636,7 +678,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinTidy:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -682,7 +723,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinDarwin:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -728,7 +768,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinAarch64:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -774,7 +813,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinFreeBSD:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -820,7 +858,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinDarwinAarch64:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -866,7 +903,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinPPC64:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -918,6 +954,7 @@ jobs:
       - BuilderDebRelease
       - BuilderDebAarch64
       - BuilderBinRelease
+      - BuilderBinGCC
       - BuilderDebAsan
       - BuilderDebTsan
       - BuilderDebUBsan
@@ -2608,6 +2645,40 @@ jobs:
           docker kill "$(docker ps -q)" ||:
           docker rm -f "$(docker ps -a -q)" ||:
           sudo rm -fr "$TEMP_PATH"
+  UnitTestsReleaseGCC:
+    needs: [BuilderBinGCC]
+    runs-on: [self-hosted, fuzzer-unit-tester]
+    steps:
+      - name: Set envs
+        run: |
+          cat >> "$GITHUB_ENV" << 'EOF'
+          TEMP_PATH=${{runner.temp}}/unit_tests_asan
+          REPORTS_PATH=${{runner.temp}}/reports_dir
+          CHECK_NAME=Unit tests (release-gcc, actions)
+          REPO_COPY=${{runner.temp}}/unit_tests_asan/ClickHouse
+          EOF
+      - name: Download json reports
+        uses: actions/download-artifact@v2
+        with:
+          path: ${{ env.REPORTS_PATH }}
+      - name: Clear repository
+        run: |
+          sudo rm -fr "$GITHUB_WORKSPACE" && mkdir "$GITHUB_WORKSPACE"
+      - name: Check out repository code
+        uses: actions/checkout@v2
+      - name: Unit test
+        run: |
+          sudo rm -fr "$TEMP_PATH"
+          mkdir -p "$TEMP_PATH"
+          cp -r "$GITHUB_WORKSPACE" "$TEMP_PATH"
+          cd "$REPO_COPY/tests/ci"
+          python3 unit_tests_check.py "$CHECK_NAME"
+      - name: Cleanup
+        if: always()
+        run: |
+          docker kill "$(docker ps -q)" ||:
+          docker rm -f "$(docker ps -a -q)" ||:
+          sudo rm -fr "$TEMP_PATH"
   UnitTestsTsan:
     needs: [BuilderDebTsan]
     runs-on: [self-hosted, fuzzer-unit-tester]

.github/workflows/pull_request.yml

@@ -370,6 +370,48 @@ jobs:
           docker kill "$(docker ps -q)" ||:
           docker rm -f "$(docker ps -a -q)" ||:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
+  BuilderBinGCC:
+    needs: [DockerHubPush, FastTest]
+    runs-on: [self-hosted, builder]
+    steps:
+      - name: Set envs
+        run: |
+          cat >> "$GITHUB_ENV" << 'EOF'
+          TEMP_PATH=${{runner.temp}}/build_check
+          IMAGES_PATH=${{runner.temp}}/images_path
+          REPO_COPY=${{runner.temp}}/build_check/ClickHouse
+          CACHES_PATH=${{runner.temp}}/../ccaches
+          CHECK_NAME=ClickHouse build check (actions)
+          BUILD_NAME=binary_gcc
+          EOF
+      - name: Download changed images
+        uses: actions/download-artifact@v2
+        with:
+          name: changed_images
+          path: ${{ runner.temp }}/images_path
+      - name: Check out repository code
+        uses: actions/checkout@v2
+        with:
+          submodules: 'true'
+          fetch-depth: 0 # otherwise we will have no info about contributors
+      - name: Build
+        run: |
+          sudo rm -fr "$TEMP_PATH"
+          mkdir -p "$TEMP_PATH"
+          cp -r "$GITHUB_WORKSPACE" "$TEMP_PATH"
+          cd "$REPO_COPY/tests/ci" && python3 build_check.py "$CHECK_NAME" "$BUILD_NAME"
+      - name: Upload build URLs to artifacts
+        if: ${{ success() || failure() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ env.BUILD_NAME }}
+          path: ${{ runner.temp }}/build_check/${{ env.BUILD_NAME }}.json
+      - name: Cleanup
+        if: always()
+        run: |
+          docker kill "$(docker ps -q)" ||:
+          docker rm -f "$(docker ps -a -q)" ||:
+          sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderDebAarch64:
     needs: [DockerHubPush, FastTest]
     runs-on: [self-hosted, builder]
@@ -963,6 +1005,7 @@ jobs:
       - BuilderDebRelease
       - BuilderDebAarch64
       - BuilderBinRelease
+      - BuilderBinGCC
       - BuilderDebAsan
       - BuilderDebTsan
       - BuilderDebUBsan
@@ -2808,6 +2851,40 @@ jobs:
           docker kill "$(docker ps -q)" ||:
           docker rm -f "$(docker ps -a -q)" ||:
           sudo rm -fr "$TEMP_PATH"
+  UnitTestsReleaseGCC:
+    needs: [BuilderBinGCC]
+    runs-on: [self-hosted, fuzzer-unit-tester]
+    steps:
+      - name: Set envs
+        run: |
+          cat >> "$GITHUB_ENV" << 'EOF'
+          TEMP_PATH=${{runner.temp}}/unit_tests_asan
+          REPORTS_PATH=${{runner.temp}}/reports_dir
+          CHECK_NAME=Unit tests (release-gcc, actions)
+          REPO_COPY=${{runner.temp}}/unit_tests_asan/ClickHouse
+          EOF
+      - name: Download json reports
+        uses: actions/download-artifact@v2
+        with:
+          path: ${{ env.REPORTS_PATH }}
+      - name: Clear repository
+        run: |
+          sudo rm -fr "$GITHUB_WORKSPACE" && mkdir "$GITHUB_WORKSPACE"
+      - name: Check out repository code
+        uses: actions/checkout@v2
+      - name: Unit test
+        run: |
+          sudo rm -fr "$TEMP_PATH"
+          mkdir -p "$TEMP_PATH"
+          cp -r "$GITHUB_WORKSPACE" "$TEMP_PATH"
+          cd "$REPO_COPY/tests/ci"
+          python3 unit_tests_check.py "$CHECK_NAME"
+      - name: Cleanup
+        if: always()
+        run: |
+          docker kill "$(docker ps -q)" ||:
+          docker rm -f "$(docker ps -a -q)" ||:
+          sudo rm -fr "$TEMP_PATH"
   UnitTestsTsan:
     needs: [BuilderDebTsan]
     runs-on: [self-hosted, fuzzer-unit-tester]

CMakeLists.txt

@@ -261,8 +261,8 @@ endif ()
 # Add a section with the hash of the compiled machine code for integrity checks.
 # Only for official builds, because adding a section can be time consuming (rewrite of several GB).
 # And cross compiled binaries are not supported (since you cannot execute clickhouse hash-binary)
-if (OBJCOPY_PATH AND YANDEX_OFFICIAL_BUILD AND (NOT CMAKE_TOOLCHAIN_FILE))
-    set (USE_BINARY_HASH 1)
+if (OBJCOPY_PATH AND CLICKHOUSE_OFFICIAL_BUILD AND (NOT CMAKE_TOOLCHAIN_FILE OR CMAKE_TOOLCHAIN_FILE MATCHES "linux/toolchain-x86_64.cmake$"))
+    set (USE_BINARY_HASH 1 CACHE STRING "Calculate binary hash and store it in the separate section")
 endif ()
 
 # Allows to build stripped binary in a separate directory

base/base/CMakeLists.txt

@@ -2,6 +2,7 @@ set (SRCS
     argsToConfig.cpp
     coverage.cpp
     demangle.cpp
+    getAvailableMemoryAmount.cpp
     getFQDNOrHostName.cpp
     getMemoryAmount.cpp
     getPageSize.cpp

base/base/getAvailableMemoryAmount.cpp

@@ -0,0 +1,44 @@
+#include <stdexcept>
+#include <fstream>
+#include <base/getAvailableMemoryAmount.h>
+#include <base/getPageSize.h>
+
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/param.h>
+#if defined(BSD)
+#include <sys/sysctl.h>
+#include <sys/vmmeter.h>
+#endif
+
+
+uint64_t getAvailableMemoryAmountOrZero()
+{
+#if defined(_SC_AVPHYS_PAGES) // linux
+    return getPageSize() * sysconf(_SC_AVPHYS_PAGES);
+#elif defined(__FreeBSD__)
+    struct vmtotal vmt;
+    size_t vmt_size = sizeof(vmt);
+    if (sysctlbyname("vm.vmtotal", &vmt, &vmt_size, NULL, 0) == 0)
+        return getPageSize() * vmt.t_avm;
+    else
+        return 0;
+#else // darwin
+    unsigned int usermem;
+    size_t len = sizeof(usermem);
+    static int mib[2] = { CTL_HW, HW_USERMEM };
+    if (sysctl(mib, 2, &usermem, &len, nullptr, 0) == 0 && len == sizeof(usermem))
+        return usermem;
+    else
+        return 0;
+#endif
+}
+
+
+uint64_t getAvailableMemoryAmount()
+{
+    auto res = getAvailableMemoryAmountOrZero();
+    if (!res)
+        throw std::runtime_error("Cannot determine available memory amount");
+    return res;
+}

base/base/getAvailableMemoryAmount.h

@@ -0,0 +1,12 @@
+#pragma once
+
+#include <cstdint>
+
+/** Returns the size of currently available physical memory (RAM) in bytes.
+  * Returns 0 on unsupported platform or if it cannot determine the size of physical memory.
+  */
+uint64_t getAvailableMemoryAmountOrZero();
+
+/** Throws exception if it cannot determine the size of physical memory.
+  */
+uint64_t getAvailableMemoryAmount();

base/glibc-compatibility/CMakeLists.txt

@@ -51,6 +51,6 @@ if (GLIBC_COMPATIBILITY)
 
     message (STATUS "Some symbols from glibc will be replaced for compatibility")
 
-elseif (YANDEX_OFFICIAL_BUILD)
+elseif (CLICKHOUSE_OFFICIAL_BUILD)
     message (WARNING "Option GLIBC_COMPATIBILITY must be turned on for production builds.")
 endif ()

cmake/version.cmake

@@ -18,6 +18,6 @@ set (VERSION_STRING_SHORT "${VERSION_MAJOR}.${VERSION_MINOR}")
 
 math (EXPR VERSION_INTEGER "${VERSION_PATCH} + ${VERSION_MINOR}*1000 + ${VERSION_MAJOR}*1000000")
 
-if(YANDEX_OFFICIAL_BUILD)
+if(CLICKHOUSE_OFFICIAL_BUILD)
     set(VERSION_OFFICIAL " (official build)")
 endif()

contrib/avro-cmake/CMakeLists.txt

@@ -69,9 +69,10 @@ endif ()
 target_compile_options(_avrocpp PRIVATE ${SUPPRESS_WARNINGS})
 
 # create a symlink to include headers with <avro/...>
+set(AVRO_INCLUDE_DIR "${CMAKE_CURRENT_BINARY_DIR}/include")
 ADD_CUSTOM_TARGET(avro_symlink_headers ALL
-    COMMAND ${CMAKE_COMMAND} -E make_directory "${AVROCPP_ROOT_DIR}/include"
-    COMMAND ${CMAKE_COMMAND} -E create_symlink "${AVROCPP_ROOT_DIR}/api" "${AVROCPP_ROOT_DIR}/include/avro"
+    COMMAND ${CMAKE_COMMAND} -E make_directory "${AVRO_INCLUDE_DIR}"
+    COMMAND ${CMAKE_COMMAND} -E create_symlink "${AVROCPP_ROOT_DIR}/api" "${AVRO_INCLUDE_DIR}/avro"
 )
 add_dependencies(_avrocpp avro_symlink_headers)
-target_include_directories(_avrocpp SYSTEM BEFORE PUBLIC "${AVROCPP_ROOT_DIR}/include")
+target_include_directories(_avrocpp SYSTEM BEFORE PUBLIC "${AVRO_INCLUDE_DIR}")

contrib/boost-cmake/CMakeLists.txt

@@ -27,7 +27,11 @@ target_include_directories (_boost_headers_only SYSTEM BEFORE INTERFACE ${LIBRAR
 
 # asio
 
-target_compile_definitions (_boost_headers_only INTERFACE BOOST_ASIO_STANDALONE=1)
+target_compile_definitions (_boost_headers_only INTERFACE
+  BOOST_ASIO_STANDALONE=1
+  # Avoid using of deprecated in c++ > 17 std::result_of
+  BOOST_ASIO_HAS_STD_INVOKE_RESULT=1
+)
 
 # iostreams
 

docker/packager/packager

@@ -163,6 +163,7 @@ def parse_env_variables(
         cmake_flags.append("-DCMAKE_INSTALL_PREFIX=/usr")
         cmake_flags.append("-DCMAKE_INSTALL_SYSCONFDIR=/etc")
         cmake_flags.append("-DCMAKE_INSTALL_LOCALSTATEDIR=/var")
+        cmake_flags.append("-DBUILD_STANDALONE_KEEPER=ON")
         if is_release_build(build_type, package_type, sanitizer, split_binary):
             cmake_flags.append("-DINSTALL_STRIPPED_BINARIES=ON")
 
@@ -244,7 +245,7 @@ def parse_env_variables(
         result.append(f"AUTHOR='{author}'")
 
     if official:
-        cmake_flags.append("-DYANDEX_OFFICIAL_BUILD=1")
+        cmake_flags.append("-DCLICKHOUSE_OFFICIAL_BUILD=1")
 
     result.append('CMAKE_FLAGS="' + " ".join(cmake_flags) + '"')
 

docker/test/fasttest/run.sh

@@ -267,6 +267,7 @@ function run_tests
     local test_opts=(
         --hung-check
         --fast-tests-only
+        --no-random-settings
         --no-long
         --testname
         --shard

docker/test/fuzzer/run-fuzzer.sh

@@ -13,7 +13,7 @@ script_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
 echo "$script_dir"
 repo_dir=ch
 BINARY_TO_DOWNLOAD=${BINARY_TO_DOWNLOAD:="clang-13_debug_none_bundled_unsplitted_disable_False_binary"}
-BINARY_URL_TO_DOWNLOAD=${BINARY_URL_TO_DOWNLOAD:="https://clickhouse-builds.s3.yandex.net/$PR_TO_TEST/$SHA_TO_TEST/clickhouse_build_check/$BINARY_TO_DOWNLOAD/clickhouse"}
+BINARY_URL_TO_DOWNLOAD=${BINARY_URL_TO_DOWNLOAD:="https://clickhouse-builds.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/clickhouse_build_check/$BINARY_TO_DOWNLOAD/clickhouse"}
 
 function clone
 {

docker/test/keeper-jepsen/run.sh

@@ -2,7 +2,7 @@
 set -euo pipefail
 
 
-CLICKHOUSE_PACKAGE=${CLICKHOUSE_PACKAGE:="https://clickhouse-builds.s3.yandex.net/$PR_TO_TEST/$SHA_TO_TEST/clickhouse_build_check/clang-13_relwithdebuginfo_none_bundled_unsplitted_disable_False_binary/clickhouse"}
+CLICKHOUSE_PACKAGE=${CLICKHOUSE_PACKAGE:="https://clickhouse-builds.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/clickhouse_build_check/clang-13_relwithdebuginfo_none_bundled_unsplitted_disable_False_binary/clickhouse"}
 CLICKHOUSE_REPO_PATH=${CLICKHOUSE_REPO_PATH:=""}
 
 
@@ -10,7 +10,7 @@ if [ -z "$CLICKHOUSE_REPO_PATH" ]; then
     CLICKHOUSE_REPO_PATH=ch
     rm -rf ch ||:
     mkdir ch ||:
-    wget -nv -nd -c "https://clickhouse-test-reports.s3.yandex.net/$PR_TO_TEST/$SHA_TO_TEST/repo/clickhouse_no_subs.tar.gz"
+    wget -nv -nd -c "https://clickhouse-test-reports.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/repo/clickhouse_no_subs.tar.gz"
     tar -C ch --strip-components=1 -xf clickhouse_no_subs.tar.gz
     ls -lath ||:
 fi

docker/test/performance-comparison/compare.sh

@@ -1294,15 +1294,15 @@ create table ci_checks engine File(TSVWithNamesAndTypes, 'ci-checks.tsv')
         select '' test_name,
             '$(sed -n 's/.*<!--message: \(.*\)-->/\1/p' report.html)' test_status,
             0 test_duration_ms,
-            'https://clickhouse-test-reports.s3.yandex.net/$PR_TO_TEST/$SHA_TO_TEST/performance_comparison/report.html#fail1' report_url
+            'https://clickhouse-test-reports.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/performance_comparison/report.html#fail1' report_url
         union all
             select test || ' #' || toString(query_index), 'slower' test_status, 0 test_duration_ms,
-                'https://clickhouse-test-reports.s3.yandex.net/$PR_TO_TEST/$SHA_TO_TEST/performance_comparison/report.html#changes-in-performance.'
+                'https://clickhouse-test-reports.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/performance_comparison/report.html#changes-in-performance.'
                     || test || '.' || toString(query_index) report_url
             from queries where changed_fail != 0 and diff > 0
         union all
             select test || ' #' || toString(query_index), 'unstable' test_status, 0 test_duration_ms,
-                'https://clickhouse-test-reports.s3.yandex.net/$PR_TO_TEST/$SHA_TO_TEST/performance_comparison/report.html#unstable-queries.'
+                'https://clickhouse-test-reports.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/performance_comparison/report.html#unstable-queries.'
                     || test || '.' || toString(query_index) report_url
             from queries where unstable_fail != 0
     )
@@ -1378,7 +1378,7 @@ $REF_SHA	$SHA_TO_TEST	$(numactl --hardware | sed -n 's/^available:[[:space:]]\+/
 EOF
 
     # Also insert some data about the check into the CI checks table.
-    "${client[@]}" --query "INSERT INTO "'"'"gh-data"'"'".checks FORMAT TSVWithNamesAndTypes" \
+    "${client[@]}" --query "INSERT INTO "'"'"default"'"'".checks FORMAT TSVWithNamesAndTypes" \
         < ci-checks.tsv
 
     set -x

docker/test/performance-comparison/download.sh

@@ -16,26 +16,17 @@ right_sha=$4
 datasets=${CHPC_DATASETS-"hits1 hits10 hits100 values"}
 
 declare -A dataset_paths
-if [[ $S3_URL == *"s3.amazonaws.com"* ]]; then
-    dataset_paths["hits10"]="https://clickhouse-private-datasets.s3.amazonaws.com/hits_10m_single/partitions/hits_10m_single.tar"
-    dataset_paths["hits100"]="https://clickhouse-private-datasets.s3.amazonaws.com/hits_100m_single/partitions/hits_100m_single.tar"
-    dataset_paths["hits1"]="https://clickhouse-datasets.s3.amazonaws.com/hits/partitions/hits_v1.tar"
-    dataset_paths["values"]="https://clickhouse-datasets.s3.amazonaws.com/values_with_expressions/partitions/test_values.tar"
-else
-    dataset_paths["hits10"]="https://s3.mds.yandex.net/clickhouse-private-datasets/hits_10m_single/partitions/hits_10m_single.tar"
-    dataset_paths["hits100"]="https://s3.mds.yandex.net/clickhouse-private-datasets/hits_100m_single/partitions/hits_100m_single.tar"
-    dataset_paths["hits1"]="https://clickhouse-datasets.s3.yandex.net/hits/partitions/hits_v1.tar"
-    dataset_paths["values"]="https://clickhouse-datasets.s3.yandex.net/values_with_expressions/partitions/test_values.tar"
-fi
+dataset_paths["hits10"]="https://clickhouse-private-datasets.s3.amazonaws.com/hits_10m_single/partitions/hits_10m_single.tar"
+dataset_paths["hits100"]="https://clickhouse-private-datasets.s3.amazonaws.com/hits_100m_single/partitions/hits_100m_single.tar"
+dataset_paths["hits1"]="https://clickhouse-datasets.s3.amazonaws.com/hits/partitions/hits_v1.tar"
+dataset_paths["values"]="https://clickhouse-datasets.s3.amazonaws.com/values_with_expressions/partitions/test_values.tar"
 
 
 function download
 {
     # Historically there were various paths for the performance test package.
     # Test all of them.
-    declare -a urls_to_try=("https://s3.amazonaws.com/clickhouse-builds/$left_pr/$left_sha/performance/performance.tgz"
-                            "https://clickhouse-builds.s3.yandex.net/$left_pr/$left_sha/clickhouse_build_check/performance/performance.tgz"
-                           )
+    declare -a urls_to_try=("https://s3.amazonaws.com/clickhouse-builds/$left_pr/$left_sha/performance/performance.tgz")
 
     for path in "${urls_to_try[@]}"
     do

docker/test/performance-comparison/entrypoint.sh

@@ -4,7 +4,7 @@ set -ex
 CHPC_CHECK_START_TIMESTAMP="$(date +%s)"
 export CHPC_CHECK_START_TIMESTAMP
 
-S3_URL=${S3_URL:="https://clickhouse-builds.s3.yandex.net"}
+S3_URL=${S3_URL:="https://clickhouse-builds.s3.amazonaws.com"}
 
 COMMON_BUILD_PREFIX="/clickhouse_build_check"
 if [[ $S3_URL == *"s3.amazonaws.com"* ]]; then
@@ -64,9 +64,7 @@ function find_reference_sha
         # Historically there were various path for the performance test package,
         # test all of them.
         unset found
-        declare -a urls_to_try=("https://s3.amazonaws.com/clickhouse-builds/0/$REF_SHA/performance/performance.tgz"
-                                "https://clickhouse-builds.s3.yandex.net/0/$REF_SHA/clickhouse_build_check/performance/performance.tgz"
-                               )
+        declare -a urls_to_try=("https://s3.amazonaws.com/clickhouse-builds/0/$REF_SHA/performance/performance.tgz")
         for path in "${urls_to_try[@]}"
         do
             if curl_with_retry "$path"

docker/test/stateful/Dockerfile

@@ -11,7 +11,7 @@ RUN apt-get update -y \
 
 COPY s3downloader /s3downloader
 
-ENV S3_URL="https://clickhouse-datasets.s3.yandex.net"
+ENV S3_URL="https://clickhouse-datasets.s3.amazonaws.com"
 ENV DATASETS="hits visits"
 ENV EXPORT_S3_STORAGE_POLICIES=1
 

docker/test/stateful/run.sh

@@ -115,7 +115,7 @@ function run_tests()
     fi
 
     set +e
-    clickhouse-test --testname --shard --zookeeper --check-zookeeper-session --no-stateless --hung-check --print-time \
+    clickhouse-test -j 2 --testname --shard --zookeeper --check-zookeeper-session --no-stateless --hung-check --print-time \
         --skip 00168_parallel_processing_on_replicas "${ADDITIONAL_OPTIONS[@]}" \
         "$SKIP_TESTS_OPTION" 2>&1 | ts '%Y-%m-%d %H:%M:%S' | tee test_output/test_result.txt
 

docker/test/stateful/s3downloader

@@ -10,7 +10,7 @@ import requests
 import tempfile
 
 
-DEFAULT_URL = 'https://clickhouse-datasets.s3.yandex.net'
+DEFAULT_URL = 'https://clickhouse-datasets.s3.amazonaws.com'
 
 AVAILABLE_DATASETS = {
     'hits': 'hits_v1.tar',

docker/test/stateless/Dockerfile

@@ -32,7 +32,9 @@ RUN apt-get update -y \
             mysql-client=8.0* \
             postgresql-client \
             sqlite3 \
-            awscli
+            awscli \
+            openjdk-11-jre-headless
+
 
 RUN pip3 install numpy scipy pandas Jinja2
 
@@ -58,10 +60,16 @@ RUN arch=${TARGETARCH:-amd64} \
     && wget "https://dl.min.io/client/mc/release/linux-${arch}/mc" \
     && chmod +x ./mc
 
+
+RUN wget 'https://dlcdn.apache.org/hadoop/common/hadoop-3.2.2/hadoop-3.2.2.tar.gz' \
+    && tar -xvf hadoop-3.2.2.tar.gz \
+    && rm -rf hadoop-3.2.2.tar.gz
+
 ENV MINIO_ROOT_USER="clickhouse"
 ENV MINIO_ROOT_PASSWORD="clickhouse"
 ENV EXPORT_S3_STORAGE_POLICIES=1
 
 COPY run.sh /
 COPY setup_minio.sh /
+COPY setup_hdfs_minicluster.sh /
 CMD ["/bin/bash", "/run.sh"]

docker/test/stateless/run.sh

@@ -19,6 +19,7 @@ ln -s /usr/share/clickhouse-test/clickhouse-test /usr/bin/clickhouse-test
 /usr/share/clickhouse-test/config/install.sh
 
 ./setup_minio.sh
+./setup_hdfs_minicluster.sh
 
 # For flaky check we also enable thread fuzzer
 if [ "$NUM_TRIES" -gt "1" ]; then
@@ -131,8 +132,23 @@ clickhouse-client -q "system flush logs" ||:
 
 grep -Fa "Fatal" /var/log/clickhouse-server/clickhouse-server.log ||:
 pigz < /var/log/clickhouse-server/clickhouse-server.log > /test_output/clickhouse-server.log.gz &
-clickhouse-client -q "select * from system.query_log format TSVWithNamesAndTypes" | pigz > /test_output/query-log.tsv.gz &
-clickhouse-client -q "select * from system.query_thread_log format TSVWithNamesAndTypes" | pigz > /test_output/query-thread-log.tsv.gz &
+
+# Compress tables.
+#
+# NOTE:
+# - that due to tests with s3 storage we cannot use /var/lib/clickhouse/data
+#   directly
+# - even though ci auto-compress some files (but not *.tsv) it does this only
+#   for files >64MB, we want this files to be compressed explicitly
+for table in query_log zookeeper_log trace_log
+do
+    clickhouse-client -q "select * from system.$table format TSVWithNamesAndTypes" | pigz > /test_output/$table.tsv.gz &
+    if [[ -n "$USE_DATABASE_REPLICATED" ]] && [[ "$USE_DATABASE_REPLICATED" -eq 1 ]]; then
+        clickhouse-client --port 19000 -q "select * from system.$table format TSVWithNamesAndTypes" | pigz > /test_output/$table.1.tsv.gz &
+        clickhouse-client --port 29000 -q "select * from system.$table format TSVWithNamesAndTypes" | pigz > /test_output/$table.2.tsv.gz &
+    fi
+done
+wait ||:
 
 # Also export trace log in flamegraph-friendly format.
 for trace_type in CPU Memory Real
@@ -161,14 +177,6 @@ fi
 
 tar -chf /test_output/coordination.tar /var/lib/clickhouse/coordination ||:
 
-# Replace the engine with Ordinary to avoid extra symlinks stuff in artifacts.
-# (so that clickhouse-local --path can read it w/o extra care).
-sed -i -e "s/ATTACH DATABASE _ UUID '[^']*'/ATTACH DATABASE system/" -e "s/Atomic/Ordinary/" /var/lib/clickhouse/metadata/system.sql
-for table in text_log query_log zookeeper_log trace_log; do
-    sed -i "s/ATTACH TABLE _ UUID '[^']*'/ATTACH TABLE $table/" /var/lib/clickhouse/metadata/system/${table}.sql
-    tar -chf /test_output/${table}_dump.tar /var/lib/clickhouse/metadata/system.sql /var/lib/clickhouse/metadata/system/${table}.sql /var/lib/clickhouse/data/system/${table} ||:
-done
-
 if [[ -n "$USE_DATABASE_REPLICATED" ]] && [[ "$USE_DATABASE_REPLICATED" -eq 1 ]]; then
     grep -Fa "Fatal" /var/log/clickhouse-server/clickhouse-server1.log ||:
     grep -Fa "Fatal" /var/log/clickhouse-server/clickhouse-server2.log ||:
@@ -179,8 +187,6 @@ if [[ -n "$USE_DATABASE_REPLICATED" ]] && [[ "$USE_DATABASE_REPLICATED" -eq 1 ]]
     rm /var/log/clickhouse-server/clickhouse-server2.log
     mv /var/log/clickhouse-server/stderr1.log /test_output/ ||:
     mv /var/log/clickhouse-server/stderr2.log /test_output/ ||:
-    tar -chf /test_output/zookeeper_log_dump1.tar /var/lib/clickhouse1/data/system/zookeeper_log ||:
-    tar -chf /test_output/zookeeper_log_dump2.tar /var/lib/clickhouse2/data/system/zookeeper_log ||:
     tar -chf /test_output/coordination1.tar /var/lib/clickhouse1/coordination ||:
     tar -chf /test_output/coordination2.tar /var/lib/clickhouse2/coordination ||:
 fi

docker/test/stateless/setup_hdfs_minicluster.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+
+set -e -x -a -u
+
+ls -lha
+
+cd hadoop-3.2.2
+
+export JAVA_HOME=/usr
+mkdir -p target/test/data
+chown clickhouse ./target/test/data
+sudo -E -u clickhouse bin/mapred minicluster -format -nomr -nnport 12222 &
+
+while ! nc -z localhost 12222; do
+  sleep 1
+done
+
+lsof -i :12222
+
+sleep 5

docker/test/stateless/setup_minio.sh

@@ -41,6 +41,7 @@ sleep 5
 ./mc admin user add clickminio test testtest
 ./mc admin policy set clickminio readwrite user=test
 ./mc mb clickminio/test
+./mc policy set public clickminio/test
 
 
 # Upload data to Minio. By default after unpacking all tests will in

docker/test/stress/Dockerfile

@@ -29,7 +29,7 @@ COPY ./download_previous_release /download_previous_release
 COPY run.sh /
 
 ENV DATASETS="hits visits"
-ENV S3_URL="https://clickhouse-datasets.s3.yandex.net"
+ENV S3_URL="https://clickhouse-datasets.s3.amazonaws.com"
 ENV EXPORT_S3_STORAGE_POLICIES=1
 
 CMD ["/bin/bash", "/run.sh"]

docs/en/sql-reference/aggregate-functions/reference/grouparraysorted.md

@@ -0,0 +1,48 @@
+---
+toc_priority: 108
+---
+
+# groupArraySorted {#groupArraySorted}
+
+Returns an array with the first N items in ascending order.
+
+``` sql
+groupArraySorted(N)(column)
+```
+
+**Arguments**
+
+-   `N` – The number of elements to return.
+
+If the parameter is omitted, default value 10 is used.
+
+**Arguments**
+
+-   `column` – The value.
+-   `expr` — Optional. The field or expresion to sort by. If not set values are sorted by themselves.
+
+**Example**
+
+Gets the first 10 numbers:
+
+``` sql
+SELECT groupArraySorted(10)(number) FROM numbers(100)
+```
+
+``` text
+┌─groupArraySorted(10)(number)─┐
+│ [0,1,2,3,4,5,6,7,8,9]        │
+└──────────────────────────────┘
+```
+
+Or the last 10:
+
+``` sql
+SELECT groupArraySorted(10)(number, -number) FROM numbers(100)
+```
+
+``` text
+┌─groupArraySorted(10)(number, negate(number))─┐
+│ [99,98,97,96,95,94,93,92,91,90]              │
+└──────────────────────────────────────────────┘
+```

docs/en/sql-reference/aggregate-functions/reference/index.md

@@ -35,6 +35,7 @@ ClickHouse-specific aggregate functions:
 -   [groupArrayInsertAt](../../../sql-reference/aggregate-functions/reference/grouparrayinsertat.md)
 -   [groupArrayMovingAvg](../../../sql-reference/aggregate-functions/reference/grouparraymovingavg.md)
 -   [groupArrayMovingSum](../../../sql-reference/aggregate-functions/reference/grouparraymovingsum.md)
+-   [groupArraySorted](../../../sql-reference/aggregate-functions/reference/grouparraysorted.md)
 -   [groupBitAnd](../../../sql-reference/aggregate-functions/reference/groupbitand.md)
 -   [groupBitOr](../../../sql-reference/aggregate-functions/reference/groupbitor.md)
 -   [groupBitXor](../../../sql-reference/aggregate-functions/reference/groupbitxor.md)

docs/zh/engines/table-engines/mergetree-family/versionedcollapsingmergetree.md

@@ -8,7 +8,7 @@ toc_title: "版本折叠MergeTree"
 这个引擎:
 
 -   允许快速写入不断变化的对象状态。
--   删除后台中的旧对象状态。 这显着降低了存储体积。
+-   删除后台中的旧对象状态。 这显著降低了存储体积。
 
 请参阅部分 [崩溃](#table_engines_versionedcollapsingmergetree) 有关详细信息。
 

packages/clickhouse-keeper-dbg.yaml

@@ -0,0 +1,28 @@
+# package sources should be placed in ${PWD}/root
+# nfpm should run from the same directory with a config
+name: "clickhouse-keeper-dbg"
+arch: "${DEB_ARCH}" # amd64, arm64
+platform: "linux"
+version: "${CLICKHOUSE_VERSION_STRING}"
+vendor: "ClickHouse Inc."
+homepage: "https://clickhouse.com"
+license: "Apache"
+section: "database"
+priority: "optional"
+maintainer: "ClickHouse Dev Team <packages+linux@clickhouse.com>"
+description: |
+  debugging symbols for clickhouse-keeper
+    This package contains the debugging symbols for clickhouse-keeper.
+
+contents:
+- src: root/usr/lib/debug/usr/bin/clickhouse-keeper.debug
+  dst: /usr/lib/debug/usr/bin/clickhouse-keeper.debug
+# docs
+- src: ../AUTHORS
+  dst: /usr/share/doc/clickhouse-keeper-dbg/AUTHORS
+- src: ../CHANGELOG.md
+  dst: /usr/share/doc/clickhouse-keeper-dbg/CHANGELOG.md
+- src: ../LICENSE
+  dst: /usr/share/doc/clickhouse-keeper-dbg/LICENSE
+- src: ../README.md
+  dst: /usr/share/doc/clickhouse-keeper-dbg/README.md

packages/clickhouse-keeper.yaml

@@ -0,0 +1,40 @@
+# package sources should be placed in ${PWD}/root
+# nfpm should run from the same directory with a config
+name: "clickhouse-keeper"
+arch: "${DEB_ARCH}" # amd64, arm64
+platform: "linux"
+version: "${CLICKHOUSE_VERSION_STRING}"
+vendor: "ClickHouse Inc."
+homepage: "https://clickhouse.com"
+license: "Apache"
+section: "database"
+priority: "optional"
+
+conflicts:
+- clickhouse-server
+depends:
+- adduser
+suggests:
+- clickhouse-keeper-dbg
+
+maintainer: "ClickHouse Dev Team <packages+linux@clickhouse.com>"
+description: |
+  Static clickhouse-keeper binary
+    A stand-alone clickhouse-keeper package
+
+
+contents:
+- src: root/etc/clickhouse-keeper
+  dst: /etc/clickhouse-keeper
+  type: config
+- src: root/usr/bin/clickhouse-keeper
+  dst: /usr/bin/clickhouse-keeper
+# docs
+- src: ../AUTHORS
+  dst: /usr/share/doc/clickhouse-keeper/AUTHORS
+- src: ../CHANGELOG.md
+  dst: /usr/share/doc/clickhouse-keeper/CHANGELOG.md
+- src: ../LICENSE
+  dst: /usr/share/doc/clickhouse-keeper/LICENSE
+- src: ../README.md
+  dst: /usr/share/doc/clickhouse-keeper/README.md

programs/keeper/CMakeLists.txt

@@ -71,17 +71,11 @@ if (BUILD_STANDALONE_KEEPER)
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressedReadBuffer.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressedReadBufferFromFile.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressedWriteBuffer.cpp
-        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecDelta.cpp
-        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecDoubleDelta.cpp
-        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecEncrypted.cpp
-        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecGorilla.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecLZ4.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecMultiple.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecNone.cpp
-        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecT64.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecZSTD.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionFactory.cpp
-        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/getCompressionCodecForFile.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/ICompressionCodec.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/LZ4_decompress_faster.cpp
 

programs/local/LocalServer.cpp

@@ -184,6 +184,11 @@ void LocalServer::tryInitPath()
     if (path.back() != '/')
         path += '/';
 
+    fs::create_directories(fs::path(path) / "user_defined/");
+    fs::create_directories(fs::path(path) / "data/");
+    fs::create_directories(fs::path(path) / "metadata/");
+    fs::create_directories(fs::path(path) / "metadata_dropped/");
+
     global_context->setPath(path);
 
     global_context->setTemporaryStorage(path + "tmp");
@@ -565,7 +570,6 @@ void LocalServer::processConfig()
         /// Lock path directory before read
         status.emplace(fs::path(path) / "status", StatusFile::write_full_info);
 
-        fs::create_directories(fs::path(path) / "user_defined/");
         LOG_DEBUG(log, "Loading user defined objects from {}", path);
         Poco::File(path + "user_defined/").createDirectories();
         UserDefinedSQLObjectsLoader::instance().loadObjects(global_context);
@@ -573,9 +577,6 @@ void LocalServer::processConfig()
         LOG_DEBUG(log, "Loaded user defined objects.");
 
         LOG_DEBUG(log, "Loading metadata from {}", path);
-        fs::create_directories(fs::path(path) / "data/");
-        fs::create_directories(fs::path(path) / "metadata/");
-
         loadMetadataSystem(global_context);
         attachSystemTablesLocal(global_context, *createMemoryDatabaseIfNotExists(global_context, DatabaseCatalog::SYSTEM_DATABASE));
         attachInformationSchema(global_context, *createMemoryDatabaseIfNotExists(global_context, DatabaseCatalog::INFORMATION_SCHEMA));

programs/server/Server.cpp

@@ -20,6 +20,7 @@
 #include <base/phdr_cache.h>
 #include <base/ErrorHandlers.h>
 #include <base/getMemoryAmount.h>
+#include <base/getAvailableMemoryAmount.h>
 #include <base/errnoToString.h>
 #include <base/coverage.h>
 #include <base/getFQDNOrHostName.h>
@@ -45,6 +46,7 @@
 #include <Core/ServerUUID.h>
 #include <IO/HTTPCommon.h>
 #include <IO/ReadHelpers.h>
+#include <IO/ReadBufferFromFile.h>
 #include <IO/IOThreadPool.h>
 #include <IO/UseSSL.h>
 #include <Interpreters/AsynchronousMetrics.h>
@@ -80,6 +82,7 @@
 #include <Common/SensitiveDataMasker.h>
 #include <Common/ThreadFuzzer.h>
 #include <Common/getHashOfLoadedBinary.h>
+#include <Common/filesystemHelpers.h>
 #include <Common/Elf.h>
 #include <Server/MySQLHandlerFactory.h>
 #include <Server/PostgreSQLHandlerFactory.h>
@@ -505,6 +508,101 @@ void checkForUsersNotInMainConfig(
     }
 }
 
+/// Unused in other builds
+#if defined(OS_LINUX)
+static String readString(const String & path)
+{
+    ReadBufferFromFile in(path);
+    String contents;
+    readStringUntilEOF(contents, in);
+    return contents;
+}
+
+static int readNumber(const String & path)
+{
+    ReadBufferFromFile in(path);
+    int result;
+    readText(result, in);
+    return result;
+}
+
+#endif
+
+static void sanityChecks(Server * server)
+{
+    std::string data_path = getCanonicalPath(server->config().getString("path", DBMS_DEFAULT_PATH));
+    std::string logs_path = server->config().getString("logger.log", "");
+
+#if defined(OS_LINUX)
+    try
+    {
+        if (readString("/sys/devices/system/clocksource/clocksource0/current_clocksource").find("tsc") == std::string::npos)
+            server->context()->addWarningMessage("Linux is not using fast TSC clock source. Performance can be degraded.");
+    }
+    catch (...)
+    {
+    }
+
+    try
+    {
+        if (readNumber("/proc/sys/vm/overcommit_memory") == 2)
+            server->context()->addWarningMessage("Linux memory overcommit is disabled.");
+    }
+    catch (...)
+    {
+    }
+
+    try
+    {
+        if (readString("/sys/kernel/mm/transparent_hugepage/enabled").find("[always]") != std::string::npos)
+            server->context()->addWarningMessage("Linux transparent hugepage are set to \"always\".");
+    }
+    catch (...)
+    {
+    }
+
+    try
+    {
+        if (readNumber("/proc/sys/kernel/pid_max") < 30000)
+            server->context()->addWarningMessage("Linux max PID is too low.");
+    }
+    catch (...)
+    {
+    }
+
+    try
+    {
+        if (readNumber("/proc/sys/kernel/threads-max") < 30000)
+            server->context()->addWarningMessage("Linux threads max count is too low.");
+    }
+    catch (...)
+    {
+    }
+
+    std::string dev_id = getBlockDeviceId(data_path);
+    if (getBlockDeviceType(dev_id) == BlockDeviceType::ROT && getBlockDeviceReadAheadBytes(dev_id) == 0)
+        server->context()->addWarningMessage("Rotational disk with disabled readahead is in use. Performance can be degraded.");
+#endif
+
+    try
+    {
+        if (getAvailableMemoryAmount() < (2l << 30))
+            server->context()->addWarningMessage("Available memory at server startup is too low (2GiB).");
+
+        if (!enoughSpaceInDirectory(data_path, 1ull << 30))
+            server->context()->addWarningMessage("Available disk space at server startup is too low (1GiB).");
+
+        if (!logs_path.empty())
+        {
+            if (!enoughSpaceInDirectory(fs::path(logs_path).parent_path(), 1ull << 30))
+                server->context()->addWarningMessage("Available disk space at server startup is too low (1GiB).");
+        }
+    }
+    catch (...)
+    {
+    }
+}
+
 int Server::main(const std::vector<std::string> & /*args*/)
 {
     Poco::Logger * log = &logger();
@@ -538,13 +636,14 @@ int Server::main(const std::vector<std::string> & /*args*/)
     global_context->addWarningMessage("Server was built in debug mode. It will work slowly.");
 #endif
 
-if (ThreadFuzzer::instance().isEffective())
-    global_context->addWarningMessage("ThreadFuzzer is enabled. Application will run slowly and unstable.");
+    if (ThreadFuzzer::instance().isEffective())
+        global_context->addWarningMessage("ThreadFuzzer is enabled. Application will run slowly and unstable.");
 
 #if defined(SANITIZER)
     global_context->addWarningMessage("Server was built with sanitizer. It will work slowly.");
 #endif
 
+    sanityChecks(this);
 
     // Initialize global thread pool. Do it before we fetch configs from zookeeper
     // nodes (`from_zk`), because ZooKeeper interface uses the pool. We will
@@ -766,6 +865,38 @@ if (ThreadFuzzer::instance().isEffective())
         }
     }
 
+    /// Try to increase limit on number of threads.
+    {
+        rlimit rlim;
+        if (getrlimit(RLIMIT_NPROC, &rlim))
+            throw Poco::Exception("Cannot getrlimit");
+
+        if (rlim.rlim_cur == rlim.rlim_max)
+        {
+            LOG_DEBUG(log, "rlimit on number of threads is {}", rlim.rlim_cur);
+        }
+        else
+        {
+            rlim_t old = rlim.rlim_cur;
+            rlim.rlim_cur = rlim.rlim_max;
+            int rc = setrlimit(RLIMIT_NPROC, &rlim);
+            if (rc != 0)
+            {
+                LOG_WARNING(log, "Cannot set max number of threads to {}. error: {}", rlim.rlim_cur, strerror(errno));
+                rlim.rlim_cur = old;
+            }
+            else
+            {
+                LOG_DEBUG(log, "Set max number of threads to {} (was {}).", rlim.rlim_cur, old);
+            }
+        }
+
+        if (rlim.rlim_cur < 30000)
+        {
+            global_context->addWarningMessage("Maximum number of threads is lower than 30000. There could be problems with handling a lot of simultaneous queries.");
+        }
+    }
+
     static ServerErrorHandler error_handler;
     Poco::ErrorHandler::set(&error_handler);
 
@@ -829,6 +960,36 @@ if (ThreadFuzzer::instance().isEffective())
         fs::create_directories(path / "metadata_dropped/");
     }
 
+#if USE_ROCKSDB
+    /// Initialize merge tree metadata cache
+    if (config().has("merge_tree_metadata_cache"))
+    {
+        fs::create_directories(path / "rocksdb/");
+        size_t size = config().getUInt64("merge_tree_metadata_cache.lru_cache_size", 256 << 20);
+        bool continue_if_corrupted = config().getBool("merge_tree_metadata_cache.continue_if_corrupted", false);
+        try
+        {
+            LOG_DEBUG(
+                log, "Initiailizing merge tree metadata cache lru_cache_size:{} continue_if_corrupted:{}", size, continue_if_corrupted);
+            global_context->initializeMergeTreeMetadataCache(path_str + "/" + "rocksdb", size);
+        }
+        catch (...)
+        {
+            if (continue_if_corrupted)
+            {
+                /// Rename rocksdb directory and reinitialize merge tree metadata cache
+                time_t now = time(nullptr);
+                fs::rename(path / "rocksdb", path / ("rocksdb.old." + std::to_string(now)));
+                global_context->initializeMergeTreeMetadataCache(path_str + "/" + "rocksdb", size);
+            }
+            else
+            {
+                throw;
+            }
+        }
+    }
+#endif
+
     if (config().has("interserver_http_port") && config().has("interserver_https_port"))
         throw Exception("Both http and https interserver ports are specified", ErrorCodes::EXCESSIVE_ELEMENT_IN_CONFIG);
 

programs/server/config.xml

@@ -1294,4 +1294,10 @@
         </tables>
     </rocksdb>
     -->
+
+    <!-- Uncomment if enable merge tree metadata cache -->
+    <merge_tree_metadata_cache>
+        <lru_cache_size>268435456</lru_cache_size>
+        <continue_if_corrupted>true</continue_if_corrupted>
+    </merge_tree_metadata_cache>
 </clickhouse>

src/Access/Common/QuotaDefs.h

@@ -13,7 +13,7 @@ enum class QuotaType
 {
     QUERIES,        /// Number of queries.
     QUERY_SELECTS,  /// Number of select queries.
-    QUERY_INSERTS,  /// Number of inserts queries.
+    QUERY_INSERTS,  /// Number of insert queries.
     ERRORS,         /// Number of queries with exceptions.
     RESULT_ROWS,    /// Number of rows returned as result.
     RESULT_BYTES,   /// Number of bytes returned as result.

src/AggregateFunctions/AggregateFunctionGroupArraySorted.cpp

@@ -0,0 +1,147 @@
+#include <AggregateFunctions/AggregateFunctionFactory.h>
+#include <AggregateFunctions/AggregateFunctionGroupArraySorted.h>
+#include <AggregateFunctions/FactoryHelpers.h>
+#include <AggregateFunctions/Helpers.h>
+#include <DataTypes/DataTypeDate.h>
+#include <DataTypes/DataTypeDateTime.h>
+#include <DataTypes/DataTypeString.h>
+#include <Common/FieldVisitorConvertToNumber.h>
+
+
+static inline constexpr UInt64 GROUP_SORTED_ARRAY_MAX_SIZE = 0xFFFFFF;
+static inline constexpr UInt64 GROUP_SORTED_ARRAY_DEFAULT_THRESHOLD = 10;
+
+
+namespace DB
+{
+struct Settings;
+
+namespace ErrorCodes
+{
+    extern const int ARGUMENT_OUT_OF_BOUND;
+    extern const int ILLEGAL_TYPE_OF_ARGUMENT;
+    extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
+}
+
+
+namespace
+{
+    template <typename T, bool expr_sorted, typename TColumnB, bool is_plain_b>
+    class AggregateFunctionGroupArraySortedNumeric : public AggregateFunctionGroupArraySorted<T, false, expr_sorted, TColumnB, is_plain_b>
+    {
+        using AggregateFunctionGroupArraySorted<T, false, expr_sorted, TColumnB, is_plain_b>::AggregateFunctionGroupArraySorted;
+    };
+
+    template <typename T, bool expr_sorted, typename TColumnB, bool is_plain_b>
+    class AggregateFunctionGroupArraySortedFieldType
+        : public AggregateFunctionGroupArraySorted<typename T::FieldType, false, expr_sorted, TColumnB, is_plain_b>
+    {
+        using AggregateFunctionGroupArraySorted<typename T::FieldType, false, expr_sorted, TColumnB, is_plain_b>::
+            AggregateFunctionGroupArraySorted;
+        DataTypePtr getReturnType() const override { return std::make_shared<DataTypeArray>(std::make_shared<T>()); }
+    };
+
+    template <template <typename, bool, typename, bool> class AggregateFunctionTemplate, typename TColumnA, bool expr_sorted, typename TColumnB, bool is_plain_b, typename... TArgs>
+    AggregateFunctionPtr
+    createAggregateFunctionGroupArraySortedTypedFinal(TArgs && ... args)
+    {
+        return AggregateFunctionPtr(new AggregateFunctionTemplate<TColumnA, expr_sorted, TColumnB, is_plain_b>(std::forward<TArgs>(args)...));
+    }
+
+    template <bool expr_sorted = false, typename TColumnB = UInt64, bool is_plain_b = false>
+    AggregateFunctionPtr
+    createAggregateFunctionGroupArraySortedTyped(const DataTypes & argument_types, const Array & params, UInt64 threshold)
+    {
+#define DISPATCH(A, C, B) \
+    if (which.idx == TypeIndex::A) \
+        return createAggregateFunctionGroupArraySortedTypedFinal<C, B, expr_sorted, TColumnB, is_plain_b>(threshold, argument_types, params);
+#define DISPATCH_NUMERIC(A) DISPATCH(A, AggregateFunctionGroupArraySortedNumeric, A)
+        WhichDataType which(argument_types[0]);
+        FOR_NUMERIC_TYPES(DISPATCH_NUMERIC)
+        DISPATCH(Enum8, AggregateFunctionGroupArraySortedNumeric, Int8)
+        DISPATCH(Enum16, AggregateFunctionGroupArraySortedNumeric, Int16)
+        DISPATCH(Date, AggregateFunctionGroupArraySortedFieldType, DataTypeDate)
+        DISPATCH(DateTime, AggregateFunctionGroupArraySortedFieldType, DataTypeDateTime)
+#undef DISPATCH
+#undef DISPATCH_NUMERIC
+
+        if (argument_types[0]->isValueUnambiguouslyRepresentedInContiguousMemoryRegion())
+        {
+            return AggregateFunctionPtr(new AggregateFunctionGroupArraySorted<StringRef, true, expr_sorted, TColumnB, is_plain_b>(
+                threshold, argument_types, params));
+        }
+        else
+        {
+            return AggregateFunctionPtr(new AggregateFunctionGroupArraySorted<StringRef, false, expr_sorted, TColumnB, is_plain_b>(
+                threshold, argument_types, params));
+        }
+    }
+
+
+    AggregateFunctionPtr createAggregateFunctionGroupArraySorted(
+        const std::string & name, const DataTypes & argument_types, const Array & params, const Settings *)
+    {
+        UInt64 threshold = GROUP_SORTED_ARRAY_DEFAULT_THRESHOLD;
+
+        if (params.size() == 1)
+        {
+            UInt64 k = applyVisitor(FieldVisitorConvertToNumber<UInt64>(), params[0]);
+
+            if (k > GROUP_SORTED_ARRAY_MAX_SIZE)
+                throw Exception(
+                    "Too large parameter(s) for aggregate function " + name + ". Maximum: " + toString(GROUP_SORTED_ARRAY_MAX_SIZE),
+                    ErrorCodes::ARGUMENT_OUT_OF_BOUND);
+
+            if (k == 0)
+                throw Exception("Parameter 0 is illegal for aggregate function " + name, ErrorCodes::ARGUMENT_OUT_OF_BOUND);
+
+            threshold = k;
+        }
+        else if (!params.empty())
+        {
+            throw Exception("Aggregate function " + name + " only supports 1 parameter.", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+        }
+
+        if (argument_types.size() == 2)
+        {
+            if (isNumber(argument_types[1]))
+            {
+#define DISPATCH2(A, B) \
+    if (which.idx == TypeIndex::A) \
+        return createAggregateFunctionGroupArraySortedTyped<true, B>(argument_types, params, threshold);
+#define DISPATCH(A) DISPATCH2(A, A)
+                WhichDataType which(argument_types[1]);
+                FOR_NUMERIC_TYPES(DISPATCH)
+                DISPATCH2(Enum8, Int8)
+                DISPATCH2(Enum16, Int16)
+#undef DISPATCH
+#undef DISPATCH2
+                throw Exception("Invalid parameter type.", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+            }
+            else if (argument_types[1]->isValueUnambiguouslyRepresentedInContiguousMemoryRegion())
+            {
+                return createAggregateFunctionGroupArraySortedTyped<true, StringRef, true>(argument_types, params, threshold);
+            }
+            else
+            {
+                return createAggregateFunctionGroupArraySortedTyped<true, StringRef, false>(argument_types, params, threshold);
+            }
+        }
+        else if (argument_types.size() == 1)
+        {
+            return createAggregateFunctionGroupArraySortedTyped<>(argument_types, params, threshold);
+        }
+        else
+        {
+            throw Exception(
+                "Aggregate function " + name + " requires one or two parameters.", ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+        }
+    }
+}
+
+void registerAggregateFunctionGroupArraySorted(AggregateFunctionFactory & factory)
+{
+    AggregateFunctionProperties properties = {.returns_default_when_only_null = false, .is_order_dependent = true};
+    factory.registerFunction("groupArraySorted", {createAggregateFunctionGroupArraySorted, properties});
+}
+}

src/AggregateFunctions/AggregateFunctionGroupArraySorted.h

@@ -0,0 +1,310 @@
+#pragma once
+
+#include <Columns/ColumnArray.h>
+#include <DataTypes/DataTypeArray.h>
+
+#include <AggregateFunctions/AggregateFunctionGroupArraySortedData.h>
+#include <AggregateFunctions/IAggregateFunction.h>
+
+namespace DB
+{
+template <typename TColumn, bool is_plain>
+inline TColumn readItem(const IColumn * column, Arena * arena, size_t row)
+{
+    if constexpr (std::is_same_v<TColumn, StringRef>)
+    {
+        if constexpr (is_plain)
+        {
+            StringRef str = column->getDataAt(row);
+            auto ptr = arena->alloc(str.size);
+            std::copy(str.data, str.data + str.size, ptr);
+            return StringRef(ptr, str.size);
+        }
+        else
+        {
+            const char * begin = nullptr;
+            return column->serializeValueIntoArena(row, *arena, begin);
+        }
+    }
+    else
+    {
+        if constexpr (std::is_same_v<TColumn, UInt64>)
+            return column->getUInt(row);
+        else
+            return column->getInt(row);
+    }
+}
+
+template <typename TColumn, typename TFilter = void>
+size_t
+getFirstNElements_low_threshold(const TColumn * data, int num_elements, int threshold, size_t * results, const TFilter * filter = nullptr)
+{
+    for (int i = 0; i < threshold; i++)
+    {
+        results[i] = 0;
+    }
+
+    threshold = std::min(num_elements, threshold);
+    int current_max = 0;
+    int cur;
+    int z;
+    for (int i = 0; i < num_elements; i++)
+    {
+        if constexpr (!std::is_same_v<TFilter, void>)
+        {
+            if (filter[i] == 0)
+                continue;
+        }
+
+        //Starting from the highest values and we look for the immediately lower than the given one
+        for (cur = current_max; cur > 0; cur--)
+        {
+            if (data[i] > data[results[cur - 1]])
+                break;
+        }
+
+        if (cur < threshold)
+        {
+            //Move all the higher values 1 position to the right
+            for (z = std::min(threshold - 1, current_max); z > cur; z--)
+                results[z] = results[z - 1];
+
+            if (current_max < threshold)
+                ++current_max;
+
+            //insert element into the given position
+            results[cur] = i;
+        }
+    }
+
+    return current_max;
+}
+
+template <typename T>
+struct SortableItem
+{
+    T a;
+    size_t b;
+    bool operator<(const SortableItem & other) const { return (this->a < other.a); }
+};
+
+template <typename TColumn, typename TFilter = void>
+size_t getFirstNElements_high_threshold(
+    const TColumn * data, size_t num_elements, size_t threshold, size_t * results, const TFilter * filter = nullptr)
+{
+    std::vector<SortableItem<TColumn>> dataIndexed(num_elements);
+    size_t num_elements_filtered = 0;
+
+    for (size_t i = 0; i < num_elements; i++)
+    {
+        if constexpr (!std::is_same_v<TFilter, void>)
+        {
+            if (filter[i] == 0)
+                continue;
+        }
+
+        dataIndexed.data()[num_elements_filtered].a = data[i];
+        dataIndexed.data()[num_elements_filtered].b = i;
+        num_elements_filtered++;
+    }
+
+    threshold = std::min(num_elements_filtered, threshold);
+
+    std::nth_element(dataIndexed.data(), dataIndexed.data() + threshold, dataIndexed.data() + num_elements_filtered);
+    std::sort(dataIndexed.data(), dataIndexed.data() + threshold);
+
+    for (size_t i = 0; i < threshold; i++)
+    {
+        results[i] = dataIndexed[i].b;
+    }
+
+    return threshold;
+}
+
+static const size_t THRESHOLD_MAX_CUSTOM_FUNCTION = 1000;
+
+template <typename TColumn>
+size_t getFirstNElements(const TColumn * data, size_t num_elements, size_t threshold, size_t * results, const UInt8 * filter = nullptr)
+{
+    if (threshold < THRESHOLD_MAX_CUSTOM_FUNCTION)
+    {
+        if (filter != nullptr)
+            return getFirstNElements_low_threshold(data, num_elements, threshold, results, filter);
+        else
+            return getFirstNElements_low_threshold(data, num_elements, threshold, results);
+    }
+    else
+    {
+        if (filter != nullptr)
+            return getFirstNElements_high_threshold(data, num_elements, threshold, results, filter);
+        else
+            return getFirstNElements_high_threshold(data, num_elements, threshold, results);
+    }
+}
+
+template <typename TColumnA, bool is_plain_a, bool use_column_b, typename TColumnB, bool is_plain_b>
+class AggregateFunctionGroupArraySorted : public IAggregateFunctionDataHelper<
+                                              AggregateFunctionGroupArraySortedData<TColumnA, use_column_b, TColumnB>,
+                                              AggregateFunctionGroupArraySorted<TColumnA, is_plain_a, use_column_b, TColumnB, is_plain_b>>
+{
+protected:
+    using State = AggregateFunctionGroupArraySortedData<TColumnA, use_column_b, TColumnB>;
+    using Base = IAggregateFunctionDataHelper<
+        AggregateFunctionGroupArraySortedData<TColumnA, use_column_b, TColumnB>,
+        AggregateFunctionGroupArraySorted>;
+
+    UInt64 threshold;
+    DataTypePtr & input_data_type;
+    mutable std::mutex mutex;
+
+    static void deserializeAndInsert(StringRef str, IColumn & data_to);
+
+public:
+    AggregateFunctionGroupArraySorted(UInt64 threshold_, const DataTypes & argument_types_, const Array & params)
+        : IAggregateFunctionDataHelper<
+            AggregateFunctionGroupArraySortedData<TColumnA, use_column_b, TColumnB>,
+            AggregateFunctionGroupArraySorted>(argument_types_, params)
+        , threshold(threshold_)
+        , input_data_type(this->argument_types[0])
+    {
+    }
+
+    void create(AggregateDataPtr place) const override
+    {
+        Base::create(place);
+        this->data(place).threshold = threshold;
+    }
+
+    String getName() const override { return "groupArraySorted"; }
+
+    DataTypePtr getReturnType() const override { return std::make_shared<DataTypeArray>(input_data_type); }
+
+    bool allocatesMemoryInArena() const override
+    {
+        if constexpr (std::is_same_v<TColumnA, StringRef>)
+            return true;
+        else
+            return false;
+    }
+
+    void add(AggregateDataPtr __restrict place, const IColumn ** columns, size_t row_num, Arena * arena) const override
+    {
+        State & data = this->data(place);
+        if constexpr (use_column_b)
+        {
+            data.add(
+                readItem<TColumnA, is_plain_a>(columns[0], arena, row_num), readItem<TColumnB, is_plain_b>(columns[1], arena, row_num));
+        }
+        else
+        {
+            data.add(readItem<TColumnA, is_plain_a>(columns[0], arena, row_num));
+        }
+    }
+
+    template <typename TColumn, bool is_plain, typename TFunc>
+    void
+    forFirstRows(size_t batch_size, const IColumn ** columns, size_t data_column, Arena * arena, ssize_t if_argument_pos, TFunc func) const
+    {
+        const TColumn * values = nullptr;
+        std::unique_ptr<std::vector<TColumn>> values_vector;
+        std::vector<size_t> best_rows(threshold);
+
+        if constexpr (std::is_same_v<TColumn, StringRef>)
+        {
+            values_vector.reset(new std::vector<TColumn>(batch_size));
+            for (size_t i = 0; i < batch_size; i++)
+                (*values_vector)[i] = readItem<TColumn, is_plain>(columns[data_column], arena, i);
+            values = (*values_vector).data();
+        }
+        else
+        {
+            const auto & column = assert_cast<const ColumnVector<TColumn> &>(*columns[data_column]);
+            values = column.getData().data();
+        }
+
+        const UInt8 * filter = nullptr;
+        StringRef refFilter;
+
+        if (if_argument_pos >= 0)
+        {
+            refFilter = columns[if_argument_pos]->getRawData();
+            filter = reinterpret_cast<const UInt8 *>(refFilter.data);
+        }
+
+        size_t num_elements = getFirstNElements(values, batch_size, threshold, best_rows.data(), filter);
+        for (size_t i = 0; i < num_elements; i++)
+        {
+            func(best_rows[i], values);
+        }
+    }
+
+    void addBatchSinglePlace(
+        size_t batch_size, AggregateDataPtr place, const IColumn ** columns, Arena * arena, ssize_t if_argument_pos) const override
+    {
+        State & data = this->data(place);
+
+        if constexpr (use_column_b)
+        {
+            forFirstRows<TColumnB, is_plain_b>(
+                batch_size, columns, 1, arena, if_argument_pos, [columns, &arena, &data](size_t row, const TColumnB * values)
+                {
+                    data.add(readItem<TColumnA, is_plain_a>(columns[0], arena, row), values[row]);
+                });
+        }
+        else
+        {
+            forFirstRows<TColumnA, is_plain_a>(
+                batch_size, columns, 0, arena, if_argument_pos, [&data](size_t row, const TColumnA * values)
+                {
+                    data.add(values[row]);
+                });
+        }
+    }
+
+    void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, Arena *) const override
+    {
+        this->data(place).merge(this->data(rhs));
+    }
+
+    void serialize(ConstAggregateDataPtr __restrict place, WriteBuffer & buf, std::optional<size_t> /* version */) const override
+    {
+        this->data(place).serialize(buf);
+    }
+
+    void
+    deserialize(AggregateDataPtr __restrict place, ReadBuffer & buf, std::optional<size_t> /* version  */, Arena * arena) const override
+    {
+        this->data(place).deserialize(buf, arena);
+    }
+
+    void insertResultInto(AggregateDataPtr __restrict place, IColumn & to, Arena * /*arena*/) const override
+    {
+        ColumnArray & arr_to = assert_cast<ColumnArray &>(to);
+        ColumnArray::Offsets & offsets_to = arr_to.getOffsets();
+
+        auto & values = this->data(place).values;
+        offsets_to.push_back(offsets_to.back() + values.size());
+
+        IColumn & data_to = arr_to.getData();
+        for (auto value : values)
+        {
+            if constexpr (std::is_same_v<TColumnA, StringRef>)
+            {
+                auto str = State::itemValue(value);
+                if constexpr (is_plain_a)
+                {
+                    data_to.insertData(str.data, str.size);
+                }
+                else
+                {
+                    data_to.deserializeAndInsertFromArena(str.data);
+                }
+            }
+            else
+            {
+                data_to.insert(State::itemValue(value));
+            }
+        }
+    }
+};
+}

src/AggregateFunctions/AggregateFunctionGroupArraySortedData.h

@@ -0,0 +1,162 @@
+#pragma once
+
+#include <IO/ReadBuffer.h>
+#include <IO/ReadHelpers.h>
+#include <IO/VarInt.h>
+#include <IO/WriteBuffer.h>
+#include <IO/WriteHelpers.h>
+
+
+static inline constexpr UInt64 GROUP_SORTED_DEFAULT_THRESHOLD = 0xFFFFFF;
+
+namespace DB
+{
+template <typename T>
+static void writeOneItem(WriteBuffer & buf, T item)
+{
+    if constexpr (std::numeric_limits<T>::is_signed)
+    {
+        writeVarInt(item, buf);
+    }
+    else
+    {
+        writeVarUInt(item, buf);
+    }
+}
+
+static void writeOneItem(WriteBuffer & buf, const StringRef & item)
+{
+    writeBinary(item, buf);
+}
+
+template <typename T>
+static void readOneItem(ReadBuffer & buf, Arena * /*arena*/, T & item)
+{
+    if constexpr (std::numeric_limits<T>::is_signed)
+    {
+        DB::Int64 val;
+        readVarT(val, buf);
+        item = val;
+    }
+    else
+    {
+        DB::UInt64 val;
+        readVarT(val, buf);
+        item = val;
+    }
+}
+
+static void readOneItem(ReadBuffer & buf, Arena * arena, StringRef & item)
+{
+    item = readStringBinaryInto(*arena, buf);
+}
+
+template <typename Storage>
+struct AggregateFunctionGroupArraySortedDataBase
+{
+    typedef typename Storage::value_type ValueType;
+    AggregateFunctionGroupArraySortedDataBase(UInt64 threshold_ = GROUP_SORTED_DEFAULT_THRESHOLD) : threshold(threshold_) { }
+
+    virtual ~AggregateFunctionGroupArraySortedDataBase() { }
+    inline void narrowDown()
+    {
+        while (values.size() > threshold)
+            values.erase(--values.end());
+    }
+
+    void merge(const AggregateFunctionGroupArraySortedDataBase & other)
+    {
+        values.merge(Storage(other.values));
+        narrowDown();
+    }
+
+    void serialize(WriteBuffer & buf) const
+    {
+        writeOneItem(buf, UInt64(values.size()));
+        for (auto value : values)
+        {
+            serializeItem(buf, value);
+        }
+    }
+
+    virtual void serializeItem(WriteBuffer & buf, ValueType & val) const = 0;
+    virtual ValueType deserializeItem(ReadBuffer & buf, Arena * arena) const = 0;
+
+    void deserialize(ReadBuffer & buf, Arena * arena)
+    {
+        values.clear();
+        UInt64 length;
+        readOneItem(buf, nullptr, length);
+
+        while (length--)
+        {
+            values.insert(deserializeItem(buf, arena));
+        }
+
+        narrowDown();
+    }
+
+    UInt64 threshold;
+    Storage values;
+};
+
+template <typename T, bool expr_sorted, typename TIndex>
+struct AggregateFunctionGroupArraySortedData
+{
+};
+
+template <typename T, typename TIndex>
+struct AggregateFunctionGroupArraySortedData<T, true, TIndex> : public AggregateFunctionGroupArraySortedDataBase<std::multimap<TIndex, T>>
+{
+    using Base = AggregateFunctionGroupArraySortedDataBase<std::multimap<TIndex, T>>;
+    using Base::Base;
+
+    void add(T item, TIndex weight)
+    {
+        Base::values.insert({weight, item});
+        Base::narrowDown();
+    }
+
+    void serializeItem(WriteBuffer & buf, typename Base::ValueType & value) const override
+    {
+        writeOneItem(buf, value.first);
+        writeOneItem(buf, value.second);
+    }
+
+    virtual typename Base::ValueType deserializeItem(ReadBuffer & buf, Arena * arena) const override
+    {
+        TIndex first;
+        T second;
+        readOneItem(buf, arena, first);
+        readOneItem(buf, arena, second);
+
+        return {first, second};
+    }
+
+    static T itemValue(typename Base::ValueType & value) { return value.second; }
+};
+
+template <typename T, typename TIndex>
+struct AggregateFunctionGroupArraySortedData<T, false, TIndex> : public AggregateFunctionGroupArraySortedDataBase<std::multiset<T>>
+{
+    using Base = AggregateFunctionGroupArraySortedDataBase<std::multiset<T>>;
+    using Base::Base;
+
+    void add(T item)
+    {
+        Base::values.insert(item);
+        Base::narrowDown();
+    }
+
+    void serializeItem(WriteBuffer & buf, typename Base::ValueType & value) const override { writeOneItem(buf, value); }
+
+    typename Base::ValueType deserializeItem(ReadBuffer & buf, Arena * arena) const override
+    {
+        T value;
+        readOneItem(buf, arena, value);
+        return value;
+    }
+
+    static T itemValue(typename Base::ValueType & value) { return value; }
+};
+}

src/AggregateFunctions/AggregateFunctionSumMap.cpp

@@ -67,7 +67,7 @@ auto parseArguments(const std::string & name, const DataTypes & arguments)
         values_types.push_back(array_type->getNestedType());
     }
 
-    return std::tuple{std::move(keys_type), std::move(values_types), tuple_argument};
+    return std::tuple<DataTypePtr, DataTypes, bool>{std::move(keys_type), std::move(values_types), tuple_argument};
 }
 
 // This function instantiates a particular overload of the sumMap family of

src/AggregateFunctions/registerAggregateFunctions.cpp

@@ -59,6 +59,7 @@ void registerAggregateFunctionNothing(AggregateFunctionFactory &);
 void registerAggregateFunctionExponentialMovingAverage(AggregateFunctionFactory &);
 void registerAggregateFunctionSparkbar(AggregateFunctionFactory &);
 void registerAggregateFunctionIntervalLengthSum(AggregateFunctionFactory &);
+void registerAggregateFunctionGroupArraySorted(AggregateFunctionFactory & factory);
 
 class AggregateFunctionCombinatorFactory;
 void registerAggregateFunctionCombinatorIf(AggregateFunctionCombinatorFactory &);
@@ -130,6 +131,7 @@ void registerAggregateFunctions()
         registerAggregateFunctionIntervalLengthSum(factory);
         registerAggregateFunctionExponentialMovingAverage(factory);
         registerAggregateFunctionSparkbar(factory);
+        registerAggregateFunctionGroupArraySorted(factory);
 
         registerWindowFunctions(factory);
     }

src/CMakeLists.txt

@@ -494,6 +494,11 @@ endif()
 
 target_link_libraries (clickhouse_common_io PUBLIC ch_contrib::fast_float)
 
+if (USE_ORC)
+    dbms_target_link_libraries(PUBLIC ${ORC_LIBRARIES})
+    dbms_target_include_directories(SYSTEM BEFORE PUBLIC ${ORC_INCLUDE_DIR} "${CMAKE_BINARY_DIR}/contrib/orc/c++/include")
+endif ()
+
 if (TARGET ch_contrib::rocksdb)
     dbms_target_link_libraries(PUBLIC ch_contrib::rocksdb)
 endif()
@@ -573,10 +578,6 @@ if (ENABLE_TESTS)
         target_link_libraries(unit_tests_dbms PRIVATE ch_contrib::simdjson)
     endif()
 
-    if(TARGET ch_contrib::rapidjson)
-        target_include_directories(unit_tests_dbms PRIVATE ch_contrib::rapidjson)
-    endif()
-
     if (TARGET ch_contrib::yaml_cpp)
         target_link_libraries(unit_tests_dbms PRIVATE ch_contrib::yaml_cpp)
     endif()

src/Columns/ColumnObject.cpp

@@ -521,7 +521,7 @@ ColumnObject::ColumnObject(bool is_nullable_)
 {
 }
 
-ColumnObject::ColumnObject(SubcolumnsTree && subcolumns_, bool is_nullable_)
+ColumnObject::ColumnObject(Subcolumns && subcolumns_, bool is_nullable_)
     : is_nullable(is_nullable_)
     , subcolumns(std::move(subcolumns_))
     , num_rows(subcolumns.empty() ? 0 : (*subcolumns.begin())->data.size())
@@ -696,7 +696,7 @@ const ColumnObject::Subcolumn & ColumnObject::getSubcolumn(const PathInData & ke
 ColumnObject::Subcolumn & ColumnObject::getSubcolumn(const PathInData & key)
 {
     if (const auto * node = subcolumns.findLeaf(key))
-        return const_cast<SubcolumnsTree::Node *>(node)->data;
+        return const_cast<Subcolumns::Node *>(node)->data;
 
     throw Exception(ErrorCodes::ILLEGAL_COLUMN, "There is no subcolumn {} in ColumnObject", key.getPath());
 }
@@ -794,7 +794,7 @@ bool ColumnObject::isFinalized() const
 void ColumnObject::finalize()
 {
     size_t old_size = size();
-    SubcolumnsTree new_subcolumns;
+    Subcolumns new_subcolumns;
     for (auto && entry : subcolumns)
     {
         const auto & least_common_type = entry->data.getLeastCommonType();

src/Columns/ColumnObject.h

@@ -138,20 +138,20 @@ public:
         size_t num_of_defaults_in_prefix = 0;
     };
 
-    using SubcolumnsTree = SubcolumnsTree<Subcolumn>;
+    using Subcolumns = SubcolumnsTree<Subcolumn>;
 
 private:
     /// If true then all subcolumns are nullable.
     const bool is_nullable;
 
-    SubcolumnsTree subcolumns;
+    Subcolumns subcolumns;
     size_t num_rows;
 
 public:
     static constexpr auto COLUMN_NAME_DUMMY = "_dummy";
 
     explicit ColumnObject(bool is_nullable_);
-    ColumnObject(SubcolumnsTree && subcolumns_, bool is_nullable_);
+    ColumnObject(Subcolumns && subcolumns_, bool is_nullable_);
 
     /// Checks that all subcolumns have consistent sizes.
     void checkConsistency() const;
@@ -173,8 +173,8 @@ public:
     /// It cares about consistency of sizes of Nested arrays.
     void addNestedSubcolumn(const PathInData & key, const FieldInfo & field_info, size_t new_size);
 
-    const SubcolumnsTree & getSubcolumns() const { return subcolumns; }
-    SubcolumnsTree & getSubcolumns() { return subcolumns; }
+    const Subcolumns & getSubcolumns() const { return subcolumns; }
+    Subcolumns & getSubcolumns() { return subcolumns; }
     PathsInData getKeys() const;
 
     /// Finalizes all subcolumns.

src/Common/Exception.h

@@ -35,10 +35,10 @@ public:
     {}
 
     // Format message with fmt::format, like the logging functions.
-    template <typename ...Args>
-    Exception(int code, const std::string & fmt, Args&&... args)
-        : Exception(fmt::format(fmt::runtime(fmt), std::forward<Args>(args)...), code)
-    {}
+    template <typename... Args>
+    Exception(int code, fmt::format_string<Args...> fmt, Args &&... args) : Exception(fmt::format(fmt, std::forward<Args>(args)...), code)
+    {
+    }
 
     struct CreateFromPocoTag {};
     struct CreateFromSTDTag {};
@@ -52,10 +52,10 @@ public:
     const char * what() const throw() override { return message().data(); }
 
     /// Add something to the existing message.
-    template <typename ...Args>
-    void addMessage(const std::string& format, Args&&... args)
+    template <typename... Args>
+    void addMessage(fmt::format_string<Args...> format, Args &&... args)
     {
-        extendedMessage(fmt::format(fmt::runtime(format), std::forward<Args>(args)...));
+        extendedMessage(fmt::format(format, std::forward<Args>(args)...));
     }
 
     void addMessage(const std::string& message)
@@ -117,10 +117,10 @@ public:
     ParsingException(int code, const std::string & message);
 
     // Format message with fmt::format, like the logging functions.
-    template <typename ...Args>
-    ParsingException(int code, const std::string & fmt, Args&&... args)
-        : Exception(fmt::format(fmt::runtime(fmt), std::forward<Args>(args)...), code)
-    {}
+    template <typename... Args>
+    ParsingException(int code, fmt::format_string<Args...> fmt, Args &&... args) : Exception(code, fmt, std::forward<Args>(args)...)
+    {
+    }
 
 
     std::string displayText() const

src/Common/FileSegment.cpp

@@ -437,6 +437,7 @@ String FileSegment::stateToString(FileSegment::State state)
         case FileSegment::State::SKIP_CACHE:
             return "SKIP_CACHE";
     }
+    __builtin_unreachable();
 }
 
 String FileSegmentsHolder::toString()

src/Common/HashTable/FixedHashTable.h

@@ -67,6 +67,9 @@ struct FixedHashTableCalculatedSize
 {
     size_t getSize(const Cell * buf, const typename Cell::State & state, size_t num_cells) const
     {
+        if (!buf)
+            return 0;
+
         size_t res = 0;
         for (const Cell * end = buf + num_cells; buf != end; ++buf)
             if (!buf->isZero(state))
@@ -76,6 +79,9 @@ struct FixedHashTableCalculatedSize
 
     bool isEmpty(const Cell * buf, const typename Cell::State & state, size_t num_cells) const
     {
+        if (!buf)
+            return true;
+
         for (const Cell * end = buf + num_cells; buf != end; ++buf)
             if (!buf->isZero(state))
                 return false;

src/Common/HashTable/TwoLevelHashTable.h

@@ -94,6 +94,12 @@ public:
 
     TwoLevelHashTable() = default;
 
+    explicit TwoLevelHashTable(size_t size_hint)
+    {
+        for (auto & impl : impls)
+            impl.reserve(size_hint / NUM_BUCKETS);
+    }
+
     /// Copy the data from another (normal) hash table. It should have the same hash function.
     template <typename Source>
     explicit TwoLevelHashTable(const Source & src)

src/Common/ProfileEvents.cpp

@@ -9,6 +9,7 @@
     M(SelectQuery, "Same as Query, but only for SELECT queries.") \
     M(InsertQuery, "Same as Query, but only for INSERT queries.") \
     M(AsyncInsertQuery, "Same as InsertQuery, but only for asynchronous INSERT queries.") \
+    M(AsyncInsertBytes, "Data size in bytes of asynchronous INSERT queries.") \
     M(FailedQuery, "Number of failed queries.") \
     M(FailedSelectQuery, "Same as FailedQuery, but only for SELECT queries.") \
     M(FailedInsertQuery, "Same as FailedQuery, but only for INSERT queries.") \
@@ -284,6 +285,16 @@
     \
     M(MainConfigLoads, "Number of times the main configuration was reloaded.") \
     \
+    M(AggregationPreallocatedElementsInHashTables, "How many elements were preallocated in hash tables for aggregation.") \
+    M(AggregationHashTablesInitializedAsTwoLevel, "How many hash tables were inited as two-level for aggregation.") \
+    \
+    M(MergeTreeMetadataCacheGet, "Number of rocksdb reads(used for merge tree metadata cache)") \
+    M(MergeTreeMetadataCachePut, "Number of rocksdb puts(used for merge tree metadata cache)") \
+    M(MergeTreeMetadataCacheDelete, "Number of rocksdb deletes(used for merge tree metadata cache)") \
+    M(MergeTreeMetadataCacheSeek, "Number of rocksdb seeks(used for merge tree metadata cache)") \
+    M(MergeTreeMetadataCacheHit, "Number of times the read of meta file was done from MergeTree metadata cache") \
+    M(MergeTreeMetadataCacheMiss, "Number of times the read of meta file was not done from MergeTree metadata cache") \
+    \
     M(ScalarSubqueriesGlobalCacheHit, "Number of times a read from a scalar subquery was done using the global cache") \
     M(ScalarSubqueriesLocalCacheHit, "Number of times a read from a scalar subquery was done using the local cache") \
     M(ScalarSubqueriesCacheMiss, "Number of times a read from a scalar subquery was not cached and had to be calculated completely")

src/Common/RangeGenerator.h

@@ -0,0 +1,46 @@
+#pragma once
+
+#include <optional>
+#include <cmath>
+
+namespace DB
+{
+
+class RangeGenerator
+{
+public:
+    explicit RangeGenerator(size_t total_size_, size_t range_step_, size_t range_start = 0)
+        : from(range_start), range_step(range_step_), total_size(total_size_)
+    {
+    }
+
+    size_t totalRanges() const { return static_cast<size_t>(round(static_cast<float>(total_size - from) / range_step)); }
+
+    using Range = std::pair<size_t, size_t>;
+
+    // return upper exclusive range of values, i.e. [from_range, to_range>
+    std::optional<Range> nextRange()
+    {
+        if (from >= total_size)
+        {
+            return std::nullopt;
+        }
+
+        auto to = from + range_step;
+        if (to >= total_size)
+        {
+            to = total_size;
+        }
+
+        Range range{from, to};
+        from = to;
+        return range;
+    }
+
+private:
+    size_t from;
+    size_t range_step;
+    size_t total_size;
+};
+
+}

src/Common/filesystemHelpers.cpp

@@ -4,6 +4,8 @@
 #if defined(__linux__)
 #    include <cstdio>
 #    include <mntent.h>
+#    include <sys/stat.h>
+#    include <sys/sysmacros.h>
 #endif
 #include <cerrno>
 #include <Poco/Version.h>
@@ -13,6 +15,9 @@
 #include <unistd.h>
 #include <sys/types.h>
 #include <utime.h>
+#include <IO/ReadBufferFromFile.h>
+#include <IO/Operators.h>
+#include <IO/WriteBufferFromString.h>
 
 namespace fs = std::filesystem;
 
@@ -24,6 +29,7 @@ namespace ErrorCodes
     extern const int LOGICAL_ERROR;
     extern const int SYSTEM_ERROR;
     extern const int NOT_IMPLEMENTED;
+    extern const int CANNOT_STAT;
     extern const int CANNOT_STATVFS;
     extern const int PATH_ACCESS_DENIED;
     extern const int CANNOT_CREATE_FILE;
@@ -57,6 +63,68 @@ std::unique_ptr<TemporaryFile> createTemporaryFile(const std::string & path)
     return std::make_unique<TemporaryFile>(path);
 }
 
+#if !defined(__linux__)
+[[noreturn]]
+#endif
+String getBlockDeviceId([[maybe_unused]] const String & path)
+{
+#if defined(__linux__)
+    struct stat sb;
+    if (lstat(path.c_str(), &sb))
+        throwFromErrnoWithPath("Cannot lstat " + path, path, ErrorCodes::CANNOT_STAT);
+    WriteBufferFromOwnString ss;
+    ss << major(sb.st_dev) << ":" << minor(sb.st_dev);
+    return ss.str();
+#else
+    throw DB::Exception("The function getDeviceId is supported on Linux only", ErrorCodes::NOT_IMPLEMENTED);
+#endif
+}
+
+#if !defined(__linux__)
+[[noreturn]]
+#endif
+BlockDeviceType getBlockDeviceType([[maybe_unused]] const String & device_id)
+{
+#if defined(__linux__)
+    try
+    {
+        ReadBufferFromFile in("/sys/dev/block/" + device_id + "/queue/rotational");
+        int rotational;
+        readText(rotational, in);
+        return rotational ? BlockDeviceType::ROT : BlockDeviceType::NONROT;
+    }
+    catch (...)
+    {
+        return BlockDeviceType::UNKNOWN;
+    }
+#else
+    throw DB::Exception("The function getDeviceType is supported on Linux only", ErrorCodes::NOT_IMPLEMENTED);
+#endif
+}
+
+#if !defined(__linux__)
+[[noreturn]]
+#endif
+UInt64 getBlockDeviceReadAheadBytes([[maybe_unused]] const String & device_id)
+{
+#if defined(__linux__)
+    try
+    {
+        ReadBufferFromFile in("/sys/dev/block/" + device_id + "/queue/read_ahead_kb");
+        int read_ahead_kb;
+        readText(read_ahead_kb, in);
+        return read_ahead_kb * 1024;
+    }
+    catch (...)
+    {
+        return static_cast<UInt64>(-1);
+    }
+#else
+    throw DB::Exception("The function getDeviceType is supported on Linux only", ErrorCodes::NOT_IMPLEMENTED);
+#endif
+}
+
+/// Returns name of filesystem mounted to mount_point
 std::filesystem::path getMountPoint(std::filesystem::path absolute_path)
 {
     if (absolute_path.is_relative())

src/Common/filesystemHelpers.h

@@ -18,6 +18,31 @@ using TemporaryFile = Poco::TemporaryFile;
 bool enoughSpaceInDirectory(const std::string & path, size_t data_size);
 std::unique_ptr<TemporaryFile> createTemporaryFile(const std::string & path);
 
+// Determine what block device is responsible for specified path
+#if !defined(__linux__)
+[[noreturn]]
+#endif
+String getBlockDeviceId([[maybe_unused]] const String & path);
+
+enum class BlockDeviceType
+{
+    UNKNOWN = 0, // we were unable to determine device type
+    NONROT = 1, // not a rotational device (SSD, NVME, etc)
+    ROT = 2 // rotational device (HDD)
+};
+
+// Try to determine block device type
+#if !defined(__linux__)
+[[noreturn]]
+#endif
+BlockDeviceType getBlockDeviceType([[maybe_unused]] const String & device_id);
+
+// Get size of read-ahead in bytes for specified block device
+#if !defined(__linux__)
+[[noreturn]]
+#endif
+UInt64 getBlockDeviceReadAheadBytes([[maybe_unused]] const String & device_id);
+
 /// Returns mount point of filesystem where absolute_path (must exist) is located
 std::filesystem::path getMountPoint(std::filesystem::path absolute_path);
 

src/Common/format.h

@@ -0,0 +1,178 @@
+#pragma once
+
+#include <base/types.h>
+#include <Common/Exception.h>
+#include <Common/PODArray.h>
+#include <Common/StringUtils/StringUtils.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int BAD_ARGUMENTS;
+}
+
+namespace Format
+{
+    using IndexPositions = PODArrayWithStackMemory<UInt64, 64>;
+
+    static inline void parseNumber(const String & description, UInt64 l, UInt64 r, UInt64 & res, UInt64 argument_number)
+    {
+        res = 0;
+        for (UInt64 pos = l; pos < r; ++pos)
+        {
+            if (!isNumericASCII(description[pos]))
+                throw Exception("Not a number in curly braces at position " + std::to_string(pos), ErrorCodes::BAD_ARGUMENTS);
+            res = res * 10 + description[pos] - '0';
+            if (res >= argument_number)
+                throw Exception(
+                    "Too big number for arguments, must be at most " + std::to_string(argument_number - 1), ErrorCodes::BAD_ARGUMENTS);
+        }
+    }
+
+    static inline void init(
+        const String & pattern,
+        size_t argument_number,
+        const std::vector<std::optional<String>> & constant_strings,
+        IndexPositions & index_positions,
+        std::vector<String> & substrings)
+    {
+        /// Is current position after open curly brace.
+        bool is_open_curly = false;
+        /// The position of last open token.
+        size_t last_open = -1;
+
+        /// Is formatting in a plain {} token.
+        std::optional<bool> is_plain_numbering;
+        UInt64 index_if_plain = 0;
+
+        /// Left position of adding substrings, just to the closed brace position or the start of the string.
+        /// Invariant --- the start of substring is in this position.
+        size_t start_pos = 0;
+
+        /// A flag to decide whether we should glue the constant strings.
+        bool glue_to_next = false;
+
+        /// Handling double braces (escaping).
+        auto double_brace_removal = [](String & str)
+        {
+            size_t i = 0;
+            bool should_delete = true;
+            str.erase(
+                std::remove_if(
+                    str.begin(),
+                    str.end(),
+                    [&i, &should_delete, &str](char)
+                    {
+                        bool is_double_brace = (str[i] == '{' && str[i + 1] == '{') || (str[i] == '}' && str[i + 1] == '}');
+                        ++i;
+                        if (is_double_brace && should_delete)
+                        {
+                            should_delete = false;
+                            return true;
+                        }
+                        should_delete = true;
+                        return false;
+                    }),
+                str.end());
+        };
+
+        index_positions.emplace_back();
+
+        for (size_t i = 0; i < pattern.size(); ++i)
+        {
+            if (pattern[i] == '{')
+            {
+                /// Escaping handling
+                /// It is safe to access because of null termination
+                if (pattern[i + 1] == '{')
+                {
+                    ++i;
+                    continue;
+                }
+
+                if (is_open_curly)
+                    throw Exception("Two open curly braces without close one at position " + std::to_string(i), ErrorCodes::BAD_ARGUMENTS);
+
+                String to_add = String(pattern.data() + start_pos, i - start_pos);
+                double_brace_removal(to_add);
+                if (!glue_to_next)
+                    substrings.emplace_back(to_add);
+                else
+                    substrings.back() += to_add;
+
+                glue_to_next = false;
+
+                is_open_curly = true;
+                last_open = i + 1;
+            }
+            else if (pattern[i] == '}')
+            {
+                if (pattern[i + 1] == '}')
+                {
+                    ++i;
+                    continue;
+                }
+
+                if (!is_open_curly)
+                    throw Exception("Closed curly brace without open one at position " + std::to_string(i), ErrorCodes::BAD_ARGUMENTS);
+
+                is_open_curly = false;
+
+                if (last_open == i)
+                {
+                    if (is_plain_numbering && !*is_plain_numbering)
+                        throw Exception(
+                            "Cannot switch from automatic field numbering to manual field specification", ErrorCodes::BAD_ARGUMENTS);
+                    is_plain_numbering = true;
+                    if (index_if_plain >= argument_number)
+                        throw Exception("Argument is too big for formatting", ErrorCodes::BAD_ARGUMENTS);
+                    index_positions.back() = index_if_plain++;
+                }
+                else
+                {
+                    if (is_plain_numbering && *is_plain_numbering)
+                        throw Exception(
+                            "Cannot switch from automatic field numbering to manual field specification", ErrorCodes::BAD_ARGUMENTS);
+                    is_plain_numbering = false;
+
+                    UInt64 arg;
+                    parseNumber(pattern, last_open, i, arg, argument_number);
+
+                    if (arg >= argument_number)
+                        throw Exception(
+                            "Argument is too big for formatting. Note that indexing starts from zero", ErrorCodes::BAD_ARGUMENTS);
+
+                    index_positions.back() = arg;
+                }
+
+                if (!constant_strings.empty() && constant_strings[index_positions.back()])
+                {
+                    /// The next string should be glued to last `A {} C`.format('B') -> `A B C`.
+                    glue_to_next = true;
+                    substrings.back() += *constant_strings[index_positions.back()];
+                }
+                else
+                    index_positions.emplace_back(); /// Otherwise we commit arg number and proceed.
+
+                start_pos = i + 1;
+            }
+        }
+
+        if (is_open_curly)
+            throw Exception("Last open curly brace is not closed", ErrorCodes::BAD_ARGUMENTS);
+
+        String to_add = String(pattern.data() + start_pos, pattern.size() - start_pos);
+        double_brace_removal(to_add);
+
+        if (!glue_to_next)
+            substrings.emplace_back(to_add);
+        else
+            substrings.back() += to_add;
+
+        index_positions.pop_back();
+    }
+}
+
+}

src/Common/tests/gtest_global_context.cpp

@@ -1,7 +1,18 @@
 #include "gtest_global_context.h"
 
 const ContextHolder & getContext()
+{
+    return getMutableContext();
+}
+
+ContextHolder & getMutableContext()
 {
     static ContextHolder holder;
     return holder;
 }
+
+void destroyContext()
+{
+    auto & holder = getMutableContext();
+    return holder.destroy();
+}

src/Common/tests/gtest_global_context.h

@@ -16,6 +16,17 @@ struct ContextHolder
     }
 
     ContextHolder(ContextHolder &&) = default;
+
+    void destroy()
+    {
+        context->shutdown();
+        context.reset();
+        shared_context.reset();
+    }
 };
 
 const ContextHolder & getContext();
+
+ContextHolder & getMutableContext();
+
+void destroyContext();

src/Compression/CompressionFactory.cpp

@@ -165,25 +165,36 @@ void registerCodecNone(CompressionCodecFactory & factory);
 void registerCodecLZ4(CompressionCodecFactory & factory);
 void registerCodecLZ4HC(CompressionCodecFactory & factory);
 void registerCodecZSTD(CompressionCodecFactory & factory);
+void registerCodecMultiple(CompressionCodecFactory & factory);
+
+
+/// Keeper use only general-purpose codecs, so we don't need these special codecs
+/// in standalone build
+#ifndef KEEPER_STANDALONE_BUILD
+
 void registerCodecDelta(CompressionCodecFactory & factory);
 void registerCodecT64(CompressionCodecFactory & factory);
 void registerCodecDoubleDelta(CompressionCodecFactory & factory);
 void registerCodecGorilla(CompressionCodecFactory & factory);
 void registerCodecEncrypted(CompressionCodecFactory & factory);
-void registerCodecMultiple(CompressionCodecFactory & factory);
+
+#endif
 
 CompressionCodecFactory::CompressionCodecFactory()
 {
-    registerCodecLZ4(*this);
     registerCodecNone(*this);
+    registerCodecLZ4(*this);
     registerCodecZSTD(*this);
     registerCodecLZ4HC(*this);
+    registerCodecMultiple(*this);
+
+#ifndef KEEPER_STANDALONE_BUILD
     registerCodecDelta(*this);
     registerCodecT64(*this);
     registerCodecDoubleDelta(*this);
     registerCodecGorilla(*this);
     registerCodecEncrypted(*this);
-    registerCodecMultiple(*this);
+#endif
 
     default_codec = get("LZ4", {});
 }

src/Core/Settings.h

@@ -500,6 +500,10 @@ class IColumn;
     M(Bool, optimize_rewrite_sum_if_to_count_if, true, "Rewrite sumIf() and sum(if()) function countIf() function when logically equivalent", 0) \
     M(UInt64, insert_shard_id, 0, "If non zero, when insert into a distributed table, the data will be inserted into the shard `insert_shard_id` synchronously. Possible values range from 1 to `shards_number` of corresponding distributed table", 0) \
     \
+    M(Bool, collect_hash_table_stats_during_aggregation, true, "Enable collecting hash table statistics to optimize memory allocation", 0) \
+    M(UInt64, max_entries_for_hash_table_stats, 10'000, "How many entries hash table statistics collected during aggregation is allowed to have", 0) \
+    M(UInt64, max_size_to_preallocate_for_aggregation, 10'000'000, "For how many elements it is allowed to preallocate space in all hash tables in total before aggregation", 0) \
+    \
     /** Experimental feature for moving data between shards. */ \
     \
     M(Bool, allow_experimental_query_deduplication, false, "Experimental data deduplication for SELECT queries based on part UUIDs", 0) \

src/Core/SettingsEnums.h

@@ -187,5 +187,4 @@ DECLARE_SETTING_ENUM_WITH_RENAME(EnumComparingMode, FormatSettings::EnumComparin
 DECLARE_SETTING_ENUM_WITH_RENAME(EscapingRule, FormatSettings::EscapingRule)
 
 DECLARE_SETTING_ENUM_WITH_RENAME(MsgPackUUIDRepresentation, FormatSettings::MsgPackUUIDRepresentation)
-
 }

src/DataTypes/Serializations/SerializationObject.cpp

@@ -63,12 +63,12 @@ private:
     size_t num_dimensions_to_keep;
 };
 
-using Node = typename ColumnObject::SubcolumnsTree::Node;
+using Node = typename ColumnObject::Subcolumns::Node;
 
 /// Finds a subcolumn from the same Nested type as @entry and inserts
 /// an array with default values with consistent sizes as in Nested type.
 bool tryInsertDefaultFromNested(
-    const std::shared_ptr<Node> & entry, const ColumnObject::SubcolumnsTree & subcolumns)
+    const std::shared_ptr<Node> & entry, const ColumnObject::Subcolumns & subcolumns)
 {
     if (!entry->path.hasNested())
         return false;
@@ -198,7 +198,7 @@ void SerializationObject<Parser>::deserializeWholeText(IColumn & column, ReadBuf
 template <typename Parser>
 void SerializationObject<Parser>::deserializeTextEscaped(IColumn & column, ReadBuffer & istr, const FormatSettings &) const
 {
-    deserializeTextImpl(column, [&](String & s) { readEscapedStringInto(s, istr); });
+    deserializeTextImpl(column, [&](String & s) { readEscapedString(s, istr); });
 }
 
 template <typename Parser>

src/Databases/DatabaseOnDisk.cpp

@@ -6,6 +6,7 @@
 #include <IO/WriteHelpers.h>
 #include <Interpreters/Context.h>
 #include <Interpreters/InterpreterCreateQuery.h>
+#include <Interpreters/ApplyWithSubqueryVisitor.h>
 #include <Parsers/ASTCreateQuery.h>
 #include <Parsers/ASTFunction.h>
 #include <Parsers/ParserCreateQuery.h>
@@ -55,6 +56,9 @@ std::pair<String, StoragePtr> createTableFromAST(
     ast_create_query.attach = true;
     ast_create_query.setDatabase(database_name);
 
+    if (ast_create_query.select && ast_create_query.isView())
+        ApplyWithSubqueryVisitor().visit(*ast_create_query.select);
+
     if (ast_create_query.as_table_function)
     {
         const auto & factory = TableFunctionFactory::instance();

src/Databases/DatabaseReplicatedWorker.cpp

@@ -179,8 +179,12 @@ String DatabaseReplicatedDDLWorker::tryEnqueueAndExecuteEntry(DDLLogEntry & entr
 
     if (!task->was_executed)
     {
-        throw Exception(ErrorCodes::LOGICAL_ERROR, "Entry {} was executed, but was not committed: code {}: {}",
-                        task->execution_status.code, task->execution_status.message);
+        throw Exception(
+            ErrorCodes::LOGICAL_ERROR,
+            "Entry {} was executed, but was not committed: code {}: {}",
+            task->entry_name,
+            task->execution_status.code,
+            task->execution_status.message);
     }
 
     try_node->setAlreadyRemoved();

src/Dictionaries/XDBCDictionarySource.cpp

@@ -50,7 +50,7 @@ namespace
         {
             if (!qualified_name.database.empty())
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
-                    "Dictionary source of type {} specifies a schema but schema is not supported by {}-driver",
+                    "Dictionary source specifies a schema but schema is not supported by {}-driver",
                     bridge_.getName());
         }
 

src/Disks/IO/CachedReadBufferFromRemoteFS.cpp

@@ -392,8 +392,13 @@ void CachedReadBufferFromRemoteFS::predownload(FileSegmentPtr & file_segment)
                 if (bytes_to_predownload)
                     throw Exception(
                         ErrorCodes::LOGICAL_ERROR,
-                        "Failed to predownload remaining {} bytes. Current file segment: {}, current download offset: {}, expected: {}, eof: {}",
-                        file_segment->range().toString(), file_segment->getDownloadOffset(), file_offset_of_buffer_end, implementation_buffer->eof());
+                        "Failed to predownload remaining {} bytes. Current file segment: {}, current download offset: {}, expected: {}, "
+                        "eof: {}",
+                        bytes_to_predownload,
+                        file_segment->range().toString(),
+                        file_segment->getDownloadOffset(),
+                        file_offset_of_buffer_end,
+                        implementation_buffer->eof());
 
                 auto result = implementation_buffer->hasPendingData();
 

src/Disks/IO/CachedReadBufferFromRemoteFS.h

@@ -96,6 +96,7 @@ private:
             case ReadType::REMOTE_FS_READ_AND_PUT_IN_CACHE:
                 return "REMOTE_FS_READ_AND_PUT_IN_CACHE";
         }
+        __builtin_unreachable();
     }
     size_t first_offset = 0;
 };

src/Disks/IO/ReadBufferFromRemoteFSGather.cpp

@@ -44,7 +44,7 @@ SeekableReadBufferPtr ReadBufferFromS3Gather::createImplementationBuffer(const S
     {
         return std::make_unique<ReadBufferFromS3>(
             client_ptr, bucket, fs::path(metadata.remote_fs_root_path) / path, max_single_read_retries,
-            settings, /* use_external_buffer */true, read_until_position, /* restricted_seek */true);
+            settings, /* use_external_buffer */true, /* offset */ 0, read_until_position, /* restricted_seek */true);
     };
 
     if (with_cache)

src/Formats/FormatSchemaInfo.cpp

@@ -85,9 +85,12 @@ FormatSchemaInfo::FormatSchemaInfo(const String & format_schema, const String &
     else if (path.has_parent_path() && !fs::weakly_canonical(default_schema_directory_path / path).string().starts_with(fs::weakly_canonical(default_schema_directory_path).string()))
     {
         if (is_server)
-            throw Exception(ErrorCodes::BAD_ARGUMENTS,
-                            "Path in the 'format_schema' setting shouldn't go outside the 'format_schema_path' directory: {} ({} not in {})",
-                            path.string());
+            throw Exception(
+                ErrorCodes::BAD_ARGUMENTS,
+                "Path in the 'format_schema' setting shouldn't go outside the 'format_schema_path' directory: {} ({} not in {})",
+                default_schema_directory(),
+                path.string(),
+                default_schema_directory());
         path = default_schema_directory_path / path;
         schema_path = path.filename();
         schema_directory = path.parent_path() / "";

src/Formats/JSONEachRowUtils.cpp

@@ -9,6 +9,7 @@
 #include <DataTypes/DataTypeArray.h>
 #include <DataTypes/DataTypeTuple.h>
 #include <DataTypes/DataTypeMap.h>
+#include <DataTypes/DataTypeObject.h>
 #include <Common/JSONParsers/SimdJSONParser.h>
 #include <Common/JSONParsers/RapidJSONParser.h>
 #include <Common/JSONParsers/DummyJSONParser.h>
@@ -158,22 +159,37 @@ DataTypePtr getDataTypeFromJSONFieldImpl(const Element & field)
     {
         auto object = field.getObject();
         DataTypePtr value_type;
+        bool is_object = false;
         for (const auto key_value_pair : object)
         {
             auto type = getDataTypeFromJSONFieldImpl(key_value_pair.second);
             if (!type)
-                return nullptr;
+                continue;
 
-            if (value_type && value_type->getName() != type->getName())
-                return nullptr;
+            if (isObject(type))
+            {
+                is_object = true;
+                break;
+            }
 
-            value_type = type;
+            if (!value_type)
+            {
+                value_type = type;
+            }
+            else if (!value_type->equals(*type))
+            {
+                is_object = true;
+                break;
+            }
         }
 
-        if (!value_type)
-            return nullptr;
+        if (is_object)
+            return std::make_shared<DataTypeObject>("json", false);
 
-        return std::make_shared<DataTypeMap>(std::make_shared<DataTypeString>(), value_type);
+        if (value_type)
+            return std::make_shared<DataTypeMap>(std::make_shared<DataTypeString>(), value_type);
+
+        return nullptr;
     }
 
     throw Exception{ErrorCodes::INCORRECT_DATA, "Unexpected JSON type"};

src/Formats/ReadSchemaUtils.cpp

@@ -7,6 +7,8 @@
 #include <Formats/ReadSchemaUtils.h>
 #include <Processors/Formats/ISchemaReader.h>
 #include <Common/assert_cast.h>
+#include <Interpreters/Context.h>
+#include <Storages/IStorage.h>
 
 namespace DB
 {
@@ -17,6 +19,28 @@ namespace ErrorCodes
     extern const int BAD_ARGUMENTS;
 }
 
+static std::optional<NamesAndTypesList> getOrderedColumnsList(
+    const NamesAndTypesList & columns_list, const Names & columns_order_hint)
+{
+    if (columns_list.size() != columns_order_hint.size())
+        return {};
+
+    std::unordered_map<String, DataTypePtr> available_columns;
+    for (const auto & [name, type] : columns_list)
+        available_columns.emplace(name, type);
+
+    NamesAndTypesList res;
+    for (const auto & name : columns_order_hint)
+    {
+        auto it = available_columns.find(name);
+        if (it == available_columns.end())
+            return {};
+
+        res.emplace_back(name, it->second);
+    }
+    return res;
+}
+
 ColumnsDescription readSchemaFromFormat(
     const String & format_name,
     const std::optional<FormatSettings> & format_settings,
@@ -52,6 +76,22 @@ ColumnsDescription readSchemaFromFormat(
         {
             throw Exception(ErrorCodes::CANNOT_EXTRACT_TABLE_STRUCTURE, "Cannot extract table structure from {} format file. Error: {}", format_name, e.message());
         }
+
+        /// If we have "INSERT SELECT" query then try to order
+        /// columns as they are ordered in table schema for formats
+        /// without strict column order (like JSON and TSKV).
+        /// It will allow to execute simple data loading with query
+        /// "INSERT INTO table SELECT * FROM ..."
+        const auto & insertion_table = context->getInsertionTable();
+        if (!schema_reader->hasStrictOrderOfColumns() && !insertion_table.empty())
+        {
+            auto storage = DatabaseCatalog::instance().getTable(insertion_table, context);
+            auto metadata = storage->getInMemoryMetadataPtr();
+            auto names_in_storage = metadata->getColumns().getNamesOfPhysical();
+            auto ordered_list = getOrderedColumnsList(names_and_types, names_in_storage);
+            if (ordered_list)
+                names_and_types = *ordered_list;
+        }
     }
     else
         throw Exception(ErrorCodes::BAD_ARGUMENTS, "{} file format doesn't support schema inference", format_name);

src/Formats/registerFormats.cpp

@@ -13,6 +13,7 @@ void registerFileSegmentationEngineCSV(FormatFactory & factory);
 void registerFileSegmentationEngineJSONEachRow(FormatFactory & factory);
 void registerFileSegmentationEngineRegexp(FormatFactory & factory);
 void registerFileSegmentationEngineJSONAsString(FormatFactory & factory);
+void registerFileSegmentationEngineJSONAsObject(FormatFactory & factory);
 void registerFileSegmentationEngineJSONCompactEachRow(FormatFactory & factory);
 
 /// Formats for both input/output.
@@ -103,6 +104,7 @@ void registerProtobufSchemaReader(FormatFactory & factory);
 void registerProtobufListSchemaReader(FormatFactory & factory);
 void registerLineAsStringSchemaReader(FormatFactory & factory);
 void registerJSONAsStringSchemaReader(FormatFactory & factory);
+void registerJSONAsObjectSchemaReader(FormatFactory & factory);
 void registerRawBLOBSchemaReader(FormatFactory & factory);
 void registerMsgPackSchemaReader(FormatFactory & factory);
 void registerCapnProtoSchemaReader(FormatFactory & factory);
@@ -123,6 +125,7 @@ void registerFormats()
     registerFileSegmentationEngineJSONEachRow(factory);
     registerFileSegmentationEngineRegexp(factory);
     registerFileSegmentationEngineJSONAsString(factory);
+    registerFileSegmentationEngineJSONAsObject(factory);
     registerFileSegmentationEngineJSONCompactEachRow(factory);
 
     registerInputFormatNative(factory);
@@ -207,6 +210,7 @@ void registerFormats()
     registerProtobufListSchemaReader(factory);
     registerLineAsStringSchemaReader(factory);
     registerJSONAsStringSchemaReader(factory);
+    registerJSONAsObjectSchemaReader(factory);
     registerRawBLOBSchemaReader(factory);
     registerMsgPackSchemaReader(factory);
     registerCapnProtoSchemaReader(factory);

src/Functions/FunctionsConversion.h

@@ -53,6 +53,7 @@
 #include <DataTypes/DataTypeLowCardinality.h>
 #include <Columns/ColumnLowCardinality.h>
 #include <Interpreters/Context.h>
+#include <Common/HashTable/HashMap.h>
 
 
 namespace DB
@@ -886,7 +887,7 @@ struct ConvertImplGenericToString
         const IColumn & col_from = *col_with_type_and_name.column;
 
         size_t size = col_from.size();
-        auto col_to = result_type->createColumn();
+        auto col_to = removeNullable(result_type)->createColumn();
 
         {
             ColumnStringHelpers::WriteHelper write_helper(
@@ -3140,52 +3141,138 @@ private:
         }
     }
 
+    WrapperType createTupleToObjectWrapper(const DataTypeTuple & from_tuple, bool has_nullable_subcolumns) const
+    {
+        if (!from_tuple.haveExplicitNames())
+            throw Exception(ErrorCodes::TYPE_MISMATCH,
+            "Cast to Object can be performed only from flatten Named Tuple. Got: {}", from_tuple.getName());
+
+        PathsInData paths;
+        DataTypes from_types;
+
+        std::tie(paths, from_types) = flattenTuple(from_tuple.getPtr());
+        auto to_types = from_types;
+
+        for (auto & type : to_types)
+        {
+            if (isTuple(type) || isNested(type))
+                throw Exception(ErrorCodes::TYPE_MISMATCH,
+                    "Cast to Object can be performed only from flatten Named Tuple. Got: {}",
+                    from_tuple.getName());
+
+            type = recursiveRemoveLowCardinality(type);
+        }
+
+        return [element_wrappers = getElementWrappers(from_types, to_types),
+            has_nullable_subcolumns, from_types, to_types, paths]
+            (ColumnsWithTypeAndName & arguments, const DataTypePtr &, const ColumnNullable * nullable_source, size_t input_rows_count)
+        {
+            size_t tuple_size = to_types.size();
+            auto flattened_column = flattenTuple(arguments.front().column);
+            const auto & column_tuple = assert_cast<const ColumnTuple &>(*flattened_column);
+
+            if (tuple_size != column_tuple.getColumns().size())
+                throw Exception(ErrorCodes::TYPE_MISMATCH,
+                    "Expected tuple with {} subcolumn, but got {} subcolumns",
+                    tuple_size, column_tuple.getColumns().size());
+
+            auto res = ColumnObject::create(has_nullable_subcolumns);
+            for (size_t i = 0; i < tuple_size; ++i)
+            {
+                ColumnsWithTypeAndName element = {{column_tuple.getColumns()[i], from_types[i], "" }};
+                auto converted_column = element_wrappers[i](element, to_types[i], nullable_source, input_rows_count);
+                res->addSubcolumn(paths[i], converted_column->assumeMutable());
+            }
+
+            return res;
+        };
+    }
+
+    WrapperType createMapToObjectWrapper(const DataTypeMap & from_map, bool has_nullable_subcolumns) const
+    {
+        auto key_value_types = from_map.getKeyValueTypes();
+
+        if (!isStringOrFixedString(key_value_types[0]))
+            throw Exception(ErrorCodes::TYPE_MISMATCH,
+                "Cast to Object from Map can be performed only from Map "
+                "with String or FixedString key. Got: {}", from_map.getName());
+
+        const auto & value_type = key_value_types[1];
+        auto to_value_type = value_type;
+
+        if (!has_nullable_subcolumns && value_type->isNullable())
+            to_value_type = removeNullable(value_type);
+
+        if (has_nullable_subcolumns && !value_type->isNullable())
+            to_value_type = makeNullable(value_type);
+
+        DataTypes to_key_value_types{std::make_shared<DataTypeString>(), std::move(to_value_type)};
+        auto element_wrappers = getElementWrappers(key_value_types, to_key_value_types);
+
+        return [has_nullable_subcolumns, element_wrappers, key_value_types, to_key_value_types]
+            (ColumnsWithTypeAndName & arguments, const DataTypePtr &, const ColumnNullable * nullable_source, size_t) -> ColumnPtr
+        {
+            const auto & column_map = assert_cast<const ColumnMap &>(*arguments.front().column);
+            const auto & offsets = column_map.getNestedColumn().getOffsets();
+            auto key_value_columns = column_map.getNestedData().getColumnsCopy();
+
+            for (size_t i = 0; i < 2; ++i)
+            {
+                ColumnsWithTypeAndName element{{key_value_columns[i], key_value_types[i], ""}};
+                key_value_columns[i] = element_wrappers[i](element, to_key_value_types[i], nullable_source, key_value_columns[i]->size());
+            }
+
+            const auto & key_column_str = assert_cast<const ColumnString &>(*key_value_columns[0]);
+            const auto & value_column = *key_value_columns[1];
+
+            using SubcolumnsMap = HashMap<StringRef, MutableColumnPtr, StringRefHash>;
+            SubcolumnsMap subcolumns;
+
+            for (size_t row = 0; row < offsets.size(); ++row)
+            {
+                for (size_t i = offsets[static_cast<ssize_t>(row) - 1]; i < offsets[row]; ++i)
+                {
+                    auto ref = key_column_str.getDataAt(i);
+
+                    bool inserted;
+                    SubcolumnsMap::LookupResult it;
+                    subcolumns.emplace(ref, it, inserted);
+                    auto & subcolumn = it->getMapped();
+
+                    if (inserted)
+                        subcolumn = value_column.cloneEmpty()->cloneResized(row);
+
+                    /// Map can have duplicated keys. We insert only first one.
+                    if (subcolumn->size() == row)
+                        subcolumn->insertFrom(value_column, i);
+                }
+
+                /// Insert default values for keys missed in current row.
+                for (const auto & [_, subcolumn] : subcolumns)
+                    if (subcolumn->size() == row)
+                        subcolumn->insertDefault();
+            }
+
+            auto column_object = ColumnObject::create(has_nullable_subcolumns);
+            for (auto && [key, subcolumn] : subcolumns)
+            {
+                PathInData path(key.toView());
+                column_object->addSubcolumn(path, std::move(subcolumn));
+            }
+
+            return column_object;
+        };
+    }
+
     WrapperType createObjectWrapper(const DataTypePtr & from_type, const DataTypeObject * to_type) const
     {
         if (const auto * from_tuple = checkAndGetDataType<DataTypeTuple>(from_type.get()))
         {
-            if (!from_tuple->haveExplicitNames())
-                 throw Exception(ErrorCodes::TYPE_MISMATCH,
-                    "Cast to Object can be performed only from flatten Named Tuple. Got: {}", from_type->getName());
-
-            PathsInData paths;
-            DataTypes from_types;
-
-            std::tie(paths, from_types) = flattenTuple(from_type);
-            auto to_types = from_types;
-
-            for (auto & type : to_types)
-            {
-                if (isTuple(type) || isNested(type))
-                     throw Exception(ErrorCodes::TYPE_MISMATCH,
-                        "Cast to Object can be performed only from flatten Named Tuple. Got: {}", from_type->getName());
-
-                type = recursiveRemoveLowCardinality(type);
-            }
-
-            return [element_wrappers = getElementWrappers(from_types, to_types),
-                has_nullable_subcolumns = to_type->hasNullableSubcolumns(), from_types, to_types, paths]
-                (ColumnsWithTypeAndName & arguments, const DataTypePtr &, const ColumnNullable * nullable_source, size_t input_rows_count)
-            {
-                size_t tuple_size = to_types.size();
-                auto flattened_column = flattenTuple(arguments.front().column);
-                const auto & column_tuple = assert_cast<const ColumnTuple &>(*flattened_column);
-
-                if (tuple_size != column_tuple.getColumns().size())
-                    throw Exception(ErrorCodes::TYPE_MISMATCH,
-                        "Expected tuple with {} subcolumn, but got {} subcolumns",
-                        tuple_size, column_tuple.getColumns().size());
-
-                auto res = ColumnObject::create(has_nullable_subcolumns);
-                for (size_t i = 0; i < tuple_size; ++i)
-                {
-                    ColumnsWithTypeAndName element = {{column_tuple.getColumns()[i], from_types[i], "" }};
-                    auto converted_column = element_wrappers[i](element, to_types[i], nullable_source, input_rows_count);
-                    res->addSubcolumn(paths[i], converted_column->assumeMutable());
-                }
-
-                return res;
-            };
+            return createTupleToObjectWrapper(*from_tuple, to_type->hasNullableSubcolumns());
+        }
+        else if (const auto * from_map = checkAndGetDataType<DataTypeMap>(from_type.get()))
+        {
+            return createMapToObjectWrapper(*from_map, to_type->hasNullableSubcolumns());
         }
         else if (checkAndGetDataType<DataTypeString>(from_type.get()))
         {
@@ -3199,7 +3286,7 @@ private:
         }
 
         throw Exception(ErrorCodes::TYPE_MISMATCH,
-            "Cast to Object can be performed only from flatten named tuple or string. Got: {}", from_type->getName());
+            "Cast to Object can be performed only from flatten named Tuple, Map or String. Got: {}", from_type->getName());
     }
 
     template <typename FieldType>

src/Functions/FunctionsStringArray.h

@@ -259,7 +259,7 @@ public:
             throw Exception(
                 ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH,
                 "Function '{}' needs at least 2 arguments, at most 3 arguments; passed {}.",
-                arguments.size());
+                name, arguments.size());
 
         if (!isString(arguments[0]))
             throw Exception("Illegal type " + arguments[0]->getName() + " of first argument of function " + getName() + ". Must be String.",

src/Functions/IFunction.cpp

@@ -181,9 +181,12 @@ ColumnPtr IExecutableFunction::defaultImplementationForNulls(
         // Default implementation for nulls returns null result for null arguments,
         // so the result type must be nullable.
         if (!result_type->isNullable())
-            throw Exception(ErrorCodes::LOGICAL_ERROR,
-                            "Function {} with Null argument and default implementation for Nulls "
-                            "is expected to return Nullable result, got {}", result_type->getName());
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR,
+                "Function {} with Null argument and default implementation for Nulls "
+                "is expected to return Nullable result, got {}",
+                getName(),
+                result_type->getName());
 
         return result_type->createColumnConstWithDefaultValue(input_rows_count);
     }

src/Functions/castOrDefault.cpp

@@ -231,7 +231,7 @@ private:
             {
                 throw Exception(ErrorCodes::BAD_ARGUMENTS,
                     "Function {} decimal scale should have native UInt type. Actual {}",
-                    scale_argument.type->getName());
+                    getName(), scale_argument.type->getName());
             }
 
             scale = arguments[additional_argument_index].column->getUInt(0);

src/Functions/concat.cpp

@@ -52,23 +52,21 @@ public:
     {
         if (arguments.size() < 2)
             throw Exception(
-                "Number of arguments for function " + getName() + " doesn't match: passed " + toString(arguments.size())
-                    + ", should be at least 2.",
-                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
-
-        if (arguments.size() > FormatImpl::argument_threshold)
-            throw Exception(
-                "Number of arguments for function " + getName() + " doesn't match: passed " + toString(arguments.size())
-                    + ", should be at most " + std::to_string(FormatImpl::argument_threshold),
-                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH,
+                "Number of arguments for function {} doesn't match: passed {}, should be at least 2",
+                getName(),
+                arguments.size());
 
         for (const auto arg_idx : collections::range(0, arguments.size()))
         {
             const auto * arg = arguments[arg_idx].get();
             if (!isStringOrFixedString(arg))
-                throw Exception{"Illegal type " + arg->getName() + " of argument " + std::to_string(arg_idx + 1) + " of function "
-                                    + getName(),
-                                ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT};
+                throw Exception(
+                    ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
+                    "Illegal type {} of argument {} of function {}",
+                    arg->getName(),
+                    arg_idx + 1,
+                    getName());
         }
 
         return std::make_shared<DataTypeString>();
@@ -125,7 +123,7 @@ private:
         std::vector<const ColumnString::Chars *> data(num_arguments);
         std::vector<const ColumnString::Offsets *> offsets(num_arguments);
         std::vector<size_t> fixed_string_sizes(num_arguments);
-        std::vector<String> constant_strings(num_arguments);
+        std::vector<std::optional<String>> constant_strings(num_arguments);
         bool has_column_string = false;
         bool has_column_fixed_string = false;
         for (size_t i = 0; i < num_arguments; ++i)

src/Functions/dateName.cpp

@@ -112,7 +112,7 @@ public:
             || (res = executeType<DataTypeDateTime64>(arguments, result_type))))
             throw Exception(
                 ErrorCodes::ILLEGAL_COLUMN,
-                "Illegal column {} of function {], must be Date or DateTime.",
+                "Illegal column {} of function {}, must be Date or DateTime.",
                 arguments[1].column->getName(),
                 getName());
 

src/Functions/flattenTuple.cpp

@@ -0,0 +1,68 @@
+#include <Functions/IFunction.h>
+#include <Functions/FunctionFactory.h>
+#include <Functions/FunctionHelpers.h>
+#include <DataTypes/DataTypeTuple.h>
+#include <DataTypes/ObjectUtils.h>
+#include <Columns/ColumnTuple.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int ILLEGAL_COLUMN;
+    extern const int ILLEGAL_TYPE_OF_ARGUMENT;
+}
+
+namespace
+{
+
+class FunctionFlattenTuple : public IFunction
+{
+public:
+    static constexpr auto name = "flattenTuple";
+    static FunctionPtr create(ContextPtr) { return std::make_shared<FunctionFlattenTuple>(); }
+
+    String getName() const override { return name; }
+    size_t getNumberOfArguments() const override { return 1; }
+    bool isSuitableForShortCircuitArgumentsExecution(const DataTypesWithConstInfo &) const override { return true; }
+    bool useDefaultImplementationForConstants() const override { return true; }
+
+    DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
+    {
+        const auto & type = arguments[0];
+        const auto * type_tuple = checkAndGetDataType<DataTypeTuple>(type.get());
+        if (!type_tuple || !type_tuple->haveExplicitNames())
+            throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
+                "Argument for function '{}' must be Named Tuple. Got '{}'",
+                getName(), type->getName());
+
+        auto [paths, types] = flattenTuple(type);
+        Names names;
+        names.reserve(paths.size());
+        for (const auto & path : paths)
+            names.push_back(path.getPath());
+
+        return std::make_shared<DataTypeTuple>(types, names);
+    }
+
+    ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t /*input_rows_count*/) const override
+    {
+        auto column = arguments.at(0).column;
+        if (!checkAndGetColumn<ColumnTuple>(column.get()))
+            throw Exception(ErrorCodes::ILLEGAL_COLUMN,
+                "Illegal column {} of first argument of function {}. Expected ColumnTuple",
+                column->getName(), getName());
+
+        return flattenTuple(column);
+    }
+};
+
+}
+
+void registerFunctionFlattenTuple(FunctionFactory & factory)
+{
+    factory.registerFunction<FunctionFlattenTuple>();
+}
+
+}

src/Functions/formatString.cpp

@@ -45,25 +45,23 @@ public:
 
     DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
     {
-        if (arguments.empty())
+        if (arguments.size() < 2)
             throw Exception(
-                "Number of arguments for function " + getName() + " doesn't match: passed " + toString(arguments.size())
-                    + ", should be at least 1",
-                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
-
-        if (arguments.size() > FormatImpl::argument_threshold)
-            throw Exception(
-                "Number of arguments for function " + getName() + " doesn't match: passed " + toString(arguments.size())
-                    + ", should be at most " + std::to_string(FormatImpl::argument_threshold),
-                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH,
+                "Number of arguments for function {} doesn't match: passed {}, should be at least 2",
+                getName(),
+                arguments.size());
 
         for (const auto arg_idx : collections::range(0, arguments.size()))
         {
             const auto * arg = arguments[arg_idx].get();
             if (!isStringOrFixedString(arg))
                 throw Exception(
-                    "Illegal type " + arg->getName() + " of argument " + std::to_string(arg_idx + 1) + " of function " + getName(),
-                    ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+                    ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
+                    "Illegal type {} of argument {} of function {}",
+                    arg->getName(),
+                    arg_idx + 1,
+                    getName());
         }
 
         return std::make_shared<DataTypeString>();
@@ -84,7 +82,7 @@ public:
         std::vector<const ColumnString::Chars *> data(arguments.size() - 1);
         std::vector<const ColumnString::Offsets *> offsets(arguments.size() - 1);
         std::vector<size_t> fixed_string_sizes(arguments.size() - 1);
-        std::vector<String> constant_strings(arguments.size() - 1);
+        std::vector<std::optional<String>> constant_strings(arguments.size() - 1);
 
         bool has_column_string = false;
         bool has_column_fixed_string = false;

src/Functions/formatString.h

@@ -4,8 +4,10 @@
 #include <base/types.h>
 #include <Common/Exception.h>
 #include <Common/StringUtils/StringUtils.h>
+#include <Common/format.h>
 #include <Common/memcpySmall.h>
 
+
 #include <algorithm>
 #include <optional>
 #include <string>
@@ -15,15 +17,9 @@
 
 namespace DB
 {
-namespace ErrorCodes
-{
-    extern const int BAD_ARGUMENTS;
-}
 
 struct FormatImpl
 {
-    static constexpr size_t small_argument_threshold = 1024;
-    static constexpr size_t argument_threshold = std::numeric_limits<UInt32>::max();
     static constexpr size_t right_padding = 15;
 
     template <typename... Args>
@@ -39,165 +35,10 @@ struct FormatImpl
             format<false, false>(std::forward<Args>(args)...);
     }
 
-    static void parseNumber(const String & description, UInt64 l, UInt64 r, UInt64 & res)
-    {
-        res = 0;
-        for (UInt64 pos = l; pos < r; ++pos)
-        {
-            if (!isNumericASCII(description[pos]))
-                throw Exception("Not a number in curly braces at position " + std::to_string(pos), ErrorCodes::BAD_ARGUMENTS);
-            res = res * 10 + description[pos] - '0';
-            if (res >= argument_threshold)
-                throw Exception(
-                    "Too big number for arguments, must be at most " + std::to_string(argument_threshold), ErrorCodes::BAD_ARGUMENTS);
-        }
-    }
-
-    static inline void init(
-        const String & pattern,
-        const std::vector<const ColumnString::Chars *> & data,
-        size_t argument_number,
-        const std::vector<String> & constant_strings,
-        UInt64 * index_positions_ptr,
-        std::vector<String> & substrings)
-    {
-        /// Is current position after open curly brace.
-        bool is_open_curly = false;
-        /// The position of last open token.
-        size_t last_open = -1;
-
-        /// Is formatting in a plain {} token.
-        std::optional<bool> is_plain_numbering;
-        UInt64 index_if_plain = 0;
-
-        /// Left position of adding substrings, just to the closed brace position or the start of the string.
-        /// Invariant --- the start of substring is in this position.
-        size_t start_pos = 0;
-
-        /// A flag to decide whether we should glue the constant strings.
-        bool glue_to_next = false;
-
-        /// Handling double braces (escaping).
-        auto double_brace_removal = [](String & str)
-        {
-            size_t i = 0;
-            bool should_delete = true;
-            str.erase(
-                std::remove_if(
-                    str.begin(),
-                    str.end(),
-                    [&i, &should_delete, &str](char)
-                    {
-                        bool is_double_brace = (str[i] == '{' && str[i + 1] == '{') || (str[i] == '}' && str[i + 1] == '}');
-                        ++i;
-                        if (is_double_brace && should_delete)
-                        {
-                            should_delete = false;
-                            return true;
-                        }
-                        should_delete = true;
-                        return false;
-                    }),
-                str.end());
-        };
-
-        for (size_t i = 0; i < pattern.size(); ++i)
-        {
-            if (pattern[i] == '{')
-            {
-                /// Escaping handling
-                /// It is safe to access because of null termination
-                if (pattern[i + 1] == '{')
-                {
-                    ++i;
-                    continue;
-                }
-
-                if (is_open_curly)
-                    throw Exception("Two open curly braces without close one at position " + std::to_string(i), ErrorCodes::BAD_ARGUMENTS);
-
-                String to_add = String(pattern.data() + start_pos, i - start_pos);
-                double_brace_removal(to_add);
-                if (!glue_to_next)
-                    substrings.emplace_back(to_add);
-                else
-                    substrings.back() += to_add;
-
-                glue_to_next = false;
-
-                is_open_curly = true;
-                last_open = i + 1;
-            }
-            else if (pattern[i] == '}')
-            {
-                if (pattern[i + 1] == '}')
-                {
-                    ++i;
-                    continue;
-                }
-
-                if (!is_open_curly)
-                    throw Exception("Closed curly brace without open one at position " + std::to_string(i), ErrorCodes::BAD_ARGUMENTS);
-
-                is_open_curly = false;
-
-                if (last_open == i)
-                {
-                    if (is_plain_numbering && !*is_plain_numbering)
-                        throw Exception(
-                            "Cannot switch from automatic field numbering to manual field specification", ErrorCodes::BAD_ARGUMENTS);
-                    is_plain_numbering = true;
-                    if (index_if_plain >= argument_number)
-                        throw Exception("Argument is too big for formatting", ErrorCodes::BAD_ARGUMENTS);
-                    *index_positions_ptr = index_if_plain++;
-                }
-                else
-                {
-                    if (is_plain_numbering && *is_plain_numbering)
-                        throw Exception(
-                            "Cannot switch from automatic field numbering to manual field specification", ErrorCodes::BAD_ARGUMENTS);
-                    is_plain_numbering = false;
-
-                    UInt64 arg;
-                    parseNumber(pattern, last_open, i, arg);
-
-                    if (arg >= argument_number)
-                        throw Exception(
-                            "Argument is too big for formatting. Note that indexing starts from zero", ErrorCodes::BAD_ARGUMENTS);
-
-                    *index_positions_ptr = arg;
-                }
-
-                /// Constant string.
-                if (!data[*index_positions_ptr])
-                {
-                    /// The next string should be glued to last `A {} C`.format('B') -> `A B C`.
-                    glue_to_next = true;
-                    substrings.back() += constant_strings[*index_positions_ptr];
-                }
-                else
-                    ++index_positions_ptr; /// Otherwise we commit arg number and proceed.
-
-                start_pos = i + 1;
-            }
-        }
-
-        if (is_open_curly)
-            throw Exception("Last open curly brace is not closed", ErrorCodes::BAD_ARGUMENTS);
-
-        String to_add = String(pattern.data() + start_pos, pattern.size() - start_pos);
-        double_brace_removal(to_add);
-
-        if (!glue_to_next)
-            substrings.emplace_back(to_add);
-        else
-            substrings.back() += to_add;
-    }
-
     /// data for ColumnString and ColumnFixed. Nullptr means no data, it is const string.
     /// offsets for ColumnString, nullptr is an indicator that there is a fixed string rather than ColumnString.
     /// fixed_string_N for savings N to fixed strings.
-    /// constant_strings for constant strings. If data[i] is nullptr, than it is constant string.
+    /// constant_strings for constant strings. If data[i] is nullptr, it is constant string.
     /// res_data is result_data, res_offsets is offset result.
     /// input_rows_count is the number of rows processed.
     /// Precondition: data.size() == offsets.size() == fixed_string_N.size() == constant_strings.size().
@@ -207,29 +48,22 @@ struct FormatImpl
         const std::vector<const ColumnString::Chars *> & data,
         const std::vector<const ColumnString::Offsets *> & offsets,
         [[maybe_unused]] /* Because sometimes !has_column_fixed_string */ const std::vector<size_t> & fixed_string_N,
-        const std::vector<String> & constant_strings,
+        const std::vector<std::optional<String>> & constant_strings,
         ColumnString::Chars & res_data,
         ColumnString::Offsets & res_offsets,
         size_t input_rows_count)
     {
         const size_t argument_number = offsets.size();
 
-        UInt64 small_index_positions_buffer[small_argument_threshold];
-        /// The subsequent indexes of strings we should use. e.g `Hello world {1} {3} {1} {0}` this array will be filled with [1, 3, 1, 0, ... (garbage)] but without constant string indices.
-        UInt64 * index_positions = small_index_positions_buffer;
-
-        std::unique_ptr<UInt64[]> big_index_positions_buffer;
-        if (argument_number > small_argument_threshold)
-        {
-            big_index_positions_buffer.reset(new UInt64[argument_number]);
-            index_positions = big_index_positions_buffer.get();
-        }
+        /// The subsequent indexes of strings we should use. e.g `Hello world {1} {3} {1} {0}` this
+        /// array will be filled with [1, 3, 1, 0] but without constant string indices.
+        Format::IndexPositions index_positions;
 
         /// Vector of substrings of pattern that will be copied to the answer, not string view because of escaping and iterators invalidation.
         /// These are exactly what is between {} tokens, for `Hello {} world {}` we will have [`Hello `, ` world `, ``].
         std::vector<String> substrings;
 
-        init(pattern, data, argument_number, constant_strings, index_positions, substrings);
+        Format::init(pattern, argument_number, constant_strings, index_positions, substrings);
 
         UInt64 final_size = 0;
 
@@ -271,7 +105,7 @@ struct FormatImpl
                 for (size_t j = 1; j < substrings.size(); ++j)
                 {
                     UInt64 arg = index_positions[j - 1];
-                    auto offset_ptr = offsets[arg];
+                    const auto * offset_ptr = offsets[arg];
                     UInt64 arg_offset = 0;
                     UInt64 size = 0;
 

src/Functions/makeDate.cpp

@@ -0,0 +1,155 @@
+#include <Functions/IFunction.h>
+#include <Functions/FunctionFactory.h>
+#include <DataTypes/DataTypeDate.h>
+#include <DataTypes/DataTypeDate32.h>
+#include <DataTypes/DataTypesNumber.h>
+#include <Columns/ColumnsNumber.h>
+#include <Interpreters/castColumn.h>
+
+#include <Common/DateLUT.h>
+#include <Common/typeid_cast.h>
+
+#include <array>
+
+namespace DB
+{
+namespace ErrorCodes
+{
+    extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
+    extern const int ILLEGAL_TYPE_OF_ARGUMENT;
+}
+
+namespace
+{
+
+// A helper function to simplify comparisons of valid YYYY-MM-DD values for <,>,=
+inline constexpr Int64 YearMonthDayToSingleInt(Int64 year, Int64 month, Int64 day)
+{
+    return year * 512 + month * 32 + day;
+}
+
+// Common implementation for makeDate, makeDate32
+template <typename Traits>
+class FunctionMakeDate : public IFunction
+{
+private:
+    static constexpr std::array<const char*, 3> argument_names = {"year", "month", "day"};
+
+public:
+    static constexpr auto name = Traits::name;
+
+    static FunctionPtr create(ContextPtr) { return std::make_shared<FunctionMakeDate>(); }
+
+    String getName() const override { return name; }
+
+    bool isVariadic() const override { return false; }
+
+    size_t getNumberOfArguments() const override { return argument_names.size(); }
+
+    bool isInjective(const ColumnsWithTypeAndName &) const override
+    {
+        return false; // {year,month,day} that are out of supported range are converted into a default value
+    }
+
+    bool isSuitableForShortCircuitArgumentsExecution(const DataTypesWithConstInfo & /*arguments*/) const override { return true; }
+
+    bool useDefaultImplementationForNulls() const override { return true; }
+
+    bool useDefaultImplementationForConstants() const override { return true; }
+
+    DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
+    {
+        if (arguments.size() != argument_names.size())
+            throw Exception(ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH,
+                "Function {} requires 3 arguments, but {} given", getName(), arguments.size());
+
+        for (size_t i = 0; i < argument_names.size(); ++i)
+        {
+            DataTypePtr argument_type = arguments[i];
+            if (!isNumber(argument_type))
+                throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
+                    "Argument '{}' for function {} must be number", std::string(argument_names[i]), getName());
+        }
+
+        return std::make_shared<typename Traits::ReturnDataType>();
+    }
+
+    ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
+    {
+        const DataTypePtr converted_argument_type = std::make_shared<DataTypeFloat32>();
+        Columns converted_arguments;
+        converted_arguments.reserve(arguments.size());
+        for (const auto & argument : arguments)
+        {
+            ColumnPtr argument_column = castColumn(argument, converted_argument_type);
+            argument_column = argument_column->convertToFullColumnIfConst();
+            converted_arguments.push_back(argument_column);
+        }
+
+        auto res_column = Traits::ReturnColumnType::create(input_rows_count);
+        auto & result_data = res_column->getData();
+
+        const auto & year_data = typeid_cast<const ColumnFloat32 &>(*converted_arguments[0]).getData();
+        const auto & month_data = typeid_cast<const ColumnFloat32 &>(*converted_arguments[1]).getData();
+        const auto & day_data = typeid_cast<const ColumnFloat32 &>(*converted_arguments[2]).getData();
+
+        const auto & date_lut = DateLUT::instance();
+
+        for (size_t i = 0; i < input_rows_count; ++i)
+        {
+            const auto year = year_data[i];
+            const auto month = month_data[i];
+            const auto day = day_data[i];
+
+            Int32 day_num = 0;
+
+            if (year >= Traits::MIN_YEAR &&
+                year <= Traits::MAX_YEAR &&
+                month >= 1 && month <= 12 &&
+                day >= 1 && day <= 31 &&
+                YearMonthDayToSingleInt(year, month, day) <= Traits::MAX_DATE)
+            {
+                day_num = date_lut.makeDayNum(year, month, day);
+            }
+
+            result_data[i] = day_num;
+        }
+
+        return res_column;
+    }
+};
+
+// makeDate(year, month, day)
+struct MakeDateTraits
+{
+    static constexpr auto name = "makeDate";
+    using ReturnDataType = DataTypeDate;
+    using ReturnColumnType = ColumnUInt16;
+
+    static constexpr auto MIN_YEAR = 1970;
+    static constexpr auto MAX_YEAR = 2149;
+    // This date has the maximum day number that fits in 16-bit uint
+    static constexpr auto MAX_DATE = YearMonthDayToSingleInt(MAX_YEAR, 6, 6);
+};
+
+// makeDate32(year, month, day)
+struct MakeDate32Traits
+{
+    static constexpr auto name = "makeDate32";
+    using ReturnDataType = DataTypeDate32;
+    using ReturnColumnType = ColumnInt32;
+
+    static constexpr auto MIN_YEAR = 1925;
+    static constexpr auto MAX_YEAR = 2283;
+    static constexpr auto MAX_DATE = YearMonthDayToSingleInt(MAX_YEAR, 11, 11);
+};
+
+}
+
+void registerFunctionsMakeDate(FunctionFactory & factory)
+{
+    factory.registerFunction<FunctionMakeDate<MakeDateTraits>>();
+    factory.registerFunction<FunctionMakeDate<MakeDate32Traits>>();
+}
+
+}

src/Functions/registerFunctions.cpp

@@ -8,6 +8,7 @@ namespace DB
 void registerFunctionsArithmetic(FunctionFactory &);
 void registerFunctionsArray(FunctionFactory &);
 void registerFunctionsTuple(FunctionFactory &);
+void registerFunctionsMakeDate(FunctionFactory &);
 void registerFunctionsMap(FunctionFactory &);
 void registerFunctionsBitmap(FunctionFactory &);
 void registerFunctionsBinaryRepr(FunctionFactory &);
@@ -73,6 +74,7 @@ void registerFunctions()
     registerFunctionsArithmetic(factory);
     registerFunctionsArray(factory);
     registerFunctionsTuple(factory);
+    registerFunctionsMakeDate(factory);
     registerFunctionsMap(factory);
     registerFunctionsBitmap(factory);
     registerFunctionsBinaryRepr(factory);

src/Functions/registerFunctionsMiscellaneous.cpp

@@ -80,6 +80,7 @@ void registerFunctionInitialQueryID(FunctionFactory & factory);
 void registerFunctionServerUUID(FunctionFactory &);
 void registerFunctionZooKeeperSessionUptime(FunctionFactory &);
 void registerFunctionGetOSKernelVersion(FunctionFactory &);
+void registerFunctionFlattenTuple(FunctionFactory &);
 
 #if USE_ICU
 void registerFunctionConvertCharset(FunctionFactory &);
@@ -166,6 +167,7 @@ void registerFunctionsMiscellaneous(FunctionFactory & factory)
     registerFunctionServerUUID(factory);
     registerFunctionZooKeeperSessionUptime(factory);
     registerFunctionGetOSKernelVersion(factory);
+    registerFunctionFlattenTuple(factory);
 
 #if USE_ICU
     registerFunctionConvertCharset(factory);

src/IO/ParallelReadBuffer.cpp

@@ -13,18 +13,11 @@ namespace ErrorCodes
 
 }
 
-ParallelReadBuffer::ParallelReadBuffer(
-    std::unique_ptr<ReadBufferFactory> reader_factory_,
-    ThreadPool * pool_,
-    size_t max_working_readers_,
-    WorkerSetup worker_setup_,
-    WorkerCleanup worker_cleanup_)
+ParallelReadBuffer::ParallelReadBuffer(std::unique_ptr<ReadBufferFactory> reader_factory_, CallbackRunner schedule_, size_t max_working_readers_)
     : SeekableReadBufferWithSize(nullptr, 0)
-    , pool(pool_)
     , max_working_readers(max_working_readers_)
+    , schedule(std::move(schedule_))
     , reader_factory(std::move(reader_factory_))
-    , worker_setup(std::move(worker_setup_))
-    , worker_cleanup(std::move(worker_cleanup_))
 {
     std::unique_lock<std::mutex> lock{mutex};
     addReaders(lock);
@@ -40,30 +33,8 @@ bool ParallelReadBuffer::addReaderToPool(std::unique_lock<std::mutex> & /*buffer
 
     auto worker = read_workers.emplace_back(std::make_shared<ReadWorker>(std::move(reader)));
 
-    pool->scheduleOrThrow(
-        [&, this, worker = std::move(worker)]() mutable
-        {
-            ThreadStatus thread_status;
+    schedule([this, worker = std::move(worker)]() mutable { readerThreadFunction(std::move(worker)); });
 
-            {
-                std::lock_guard lock{mutex};
-                ++active_working_reader;
-            }
-
-            SCOPE_EXIT({
-                worker_cleanup(thread_status);
-
-                std::lock_guard lock{mutex};
-                --active_working_reader;
-                if (active_working_reader == 0)
-                {
-                    readers_done.notify_all();
-                }
-            });
-            worker_setup(thread_status);
-
-            readerThreadFunction(std::move(worker));
-        });
     return true;
 }
 
@@ -232,12 +203,27 @@ bool ParallelReadBuffer::nextImpl()
 
 void ParallelReadBuffer::readerThreadFunction(ReadWorkerPtr read_worker)
 {
+    {
+        std::lock_guard lock{mutex};
+        ++active_working_reader;
+    }
+
+    SCOPE_EXIT({
+        std::lock_guard lock{mutex};
+        --active_working_reader;
+        if (active_working_reader == 0)
+        {
+            readers_done.notify_all();
+        }
+    });
+
     try
     {
         while (!emergency_stop && !read_worker->cancel)
         {
             if (!read_worker->reader->next())
-                throw Exception("Failed to read all the data from the reader", ErrorCodes::LOGICAL_ERROR);
+                throw Exception(
+                    ErrorCodes::LOGICAL_ERROR, "Failed to read all the data from the reader, missing {} bytes", read_worker->bytes_left);
 
             if (emergency_stop || read_worker->cancel)
                 break;

src/IO/ParallelReadBuffer.h

@@ -3,6 +3,7 @@
 #include <IO/BufferWithOwnMemory.h>
 #include <IO/ReadBuffer.h>
 #include <IO/SeekableReadBuffer.h>
+#include <Interpreters/threadPoolCallbackRunner.h>
 #include <Common/ArenaWithFreeLists.h>
 #include <Common/ThreadPool.h>
 
@@ -76,14 +77,7 @@ public:
         virtual std::optional<size_t> getTotalSize() = 0;
     };
 
-    using WorkerSetup = std::function<void(ThreadStatus &)>;
-    using WorkerCleanup = std::function<void(ThreadStatus &)>;
-    explicit ParallelReadBuffer(
-        std::unique_ptr<ReadBufferFactory> reader_factory_,
-        ThreadPool * pool,
-        size_t max_working_readers,
-        WorkerSetup worker_setup = {},
-        WorkerCleanup worker_cleanup = {});
+    explicit ParallelReadBuffer(std::unique_ptr<ReadBufferFactory> reader_factory_, CallbackRunner schedule_, size_t max_working_readers);
 
     ~ParallelReadBuffer() override { finishAndWait(); }
 
@@ -140,16 +134,14 @@ private:
 
     Segment current_segment;
 
-    ThreadPool * pool;
     size_t max_working_readers;
     size_t active_working_reader{0};
     // Triggered when all reader workers are done
     std::condition_variable readers_done;
 
-    std::unique_ptr<ReadBufferFactory> reader_factory;
+    CallbackRunner schedule;
 
-    WorkerSetup worker_setup;
-    WorkerCleanup worker_cleanup;
+    std::unique_ptr<ReadBufferFactory> reader_factory;
 
     /**
      * FIFO queue of readers.

src/IO/ReadBufferFromFileDescriptor.h

@@ -39,6 +39,10 @@ public:
     {
     }
 
+    virtual ~ReadBufferFromFileDescriptor() override
+    {
+    }
+
     int getFD() const
     {
         return fd;
@@ -80,6 +84,9 @@ public:
     {
         use_pread = true;
     }
+    virtual ~ReadBufferFromFileDescriptorPRead() override
+    {
+    }
 };
 
 }

src/IO/ReadBufferFromS3.cpp

@@ -1,4 +1,5 @@
 #include <Common/config.h>
+#include "IO/S3Common.h"
 
 #if USE_AWS_S3
 
@@ -42,6 +43,7 @@ ReadBufferFromS3::ReadBufferFromS3(
     UInt64 max_single_read_retries_,
     const ReadSettings & settings_,
     bool use_external_buffer_,
+    size_t offset_,
     size_t read_until_position_,
     bool restricted_seek_)
     : SeekableReadBufferWithSize(nullptr, 0)
@@ -49,9 +51,10 @@ ReadBufferFromS3::ReadBufferFromS3(
     , bucket(bucket_)
     , key(key_)
     , max_single_read_retries(max_single_read_retries_)
+    , offset(offset_)
+    , read_until_position(read_until_position_)
     , read_settings(settings_)
     , use_external_buffer(use_external_buffer_)
-    , read_until_position(read_until_position_)
     , restricted_seek(restricted_seek_)
 {
 }
@@ -210,13 +213,14 @@ std::optional<size_t> ReadBufferFromS3::getTotalSize()
     if (file_size)
         return file_size;
 
-    Aws::S3::Model::HeadObjectRequest request;
-    request.SetBucket(bucket);
-    request.SetKey(key);
+    auto object_size = S3::getObjectSize(client_ptr, bucket, key, false);
 
-    auto outcome = client_ptr->HeadObject(request);
-    auto head_result = outcome.GetResultWithOwnership();
-    file_size = head_result.GetContentLength();
+    if (!object_size)
+    {
+        return std::nullopt;
+    }
+
+    file_size = object_size;
     return file_size;
 }
 
@@ -234,6 +238,11 @@ void ReadBufferFromS3::setReadUntilPosition(size_t position)
     }
 }
 
+SeekableReadBuffer::Range ReadBufferFromS3::getRemainingReadRange() const
+{
+    return Range{.left = static_cast<size_t>(offset), .right = read_until_position ? std::optional{read_until_position - 1} : std::nullopt};
+}
+
 std::unique_ptr<ReadBuffer> ReadBufferFromS3::initialize()
 {
     Aws::S3::Model::GetObjectRequest req;
@@ -272,6 +281,36 @@ std::unique_ptr<ReadBuffer> ReadBufferFromS3::initialize()
         throw Exception(outcome.GetError().GetMessage(), ErrorCodes::S3_ERROR);
 }
 
+SeekableReadBufferPtr ReadBufferS3Factory::getReader()
+{
+    const auto next_range = range_generator.nextRange();
+    if (!next_range)
+    {
+        return nullptr;
+    }
+
+    auto reader = std::make_shared<ReadBufferFromS3>(
+        client_ptr,
+        bucket,
+        key,
+        s3_max_single_read_retries,
+        read_settings,
+        false /*use_external_buffer*/,
+        next_range->first,
+        next_range->second);
+    return reader;
+}
+
+off_t ReadBufferS3Factory::seek(off_t off, [[maybe_unused]] int whence)
+{
+    range_generator = RangeGenerator{object_size, range_step, static_cast<size_t>(off)};
+    return off;
+}
+
+std::optional<size_t> ReadBufferS3Factory::getTotalSize()
+{
+    return object_size;
+}
 }
 
 #endif

src/IO/ReadBufferFromS3.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <Common/RangeGenerator.h>
 #include <Common/config.h>
 
 #if USE_AWS_S3
@@ -7,6 +8,7 @@
 #include <memory>
 
 #include <IO/HTTPCommon.h>
+#include <IO/ParallelReadBuffer.h>
 #include <IO/ReadBuffer.h>
 #include <IO/ReadSettings.h>
 #include <IO/SeekableReadBuffer.h>
@@ -30,7 +32,9 @@ private:
     String bucket;
     String key;
     UInt64 max_single_read_retries;
+
     off_t offset = 0;
+    off_t read_until_position = 0;
 
     Aws::S3::Model::GetObjectResult read_result;
     std::unique_ptr<ReadBuffer> impl;
@@ -45,6 +49,7 @@ public:
         UInt64 max_single_read_retries_,
         const ReadSettings & settings_,
         bool use_external_buffer = false,
+        size_t offset_ = 0,
         size_t read_until_position_ = 0,
         bool restricted_seek_ = false);
 
@@ -58,7 +63,7 @@ public:
 
     void setReadUntilPosition(size_t position) override;
 
-    Range getRemainingReadRange() const override { return Range{ .left = static_cast<size_t>(offset), .right = read_until_position }; }
+    Range getRemainingReadRange() const override;
 
     size_t getFileOffsetOfBufferEnd() const override { return offset; }
 
@@ -69,13 +74,55 @@ private:
 
     bool use_external_buffer;
 
-    off_t read_until_position = 0;
-
     /// There is different seek policy for disk seek and for non-disk seek
     /// (non-disk seek is applied for seekable input formats: orc, arrow, parquet).
     bool restricted_seek;
 };
 
+/// Creates separate ReadBufferFromS3 for sequence of ranges of particular object
+class ReadBufferS3Factory : public ParallelReadBuffer::ReadBufferFactory
+{
+public:
+    explicit ReadBufferS3Factory(
+        std::shared_ptr<Aws::S3::S3Client> client_ptr_,
+        const String & bucket_,
+        const String & key_,
+        size_t range_step_,
+        size_t object_size_,
+        UInt64 s3_max_single_read_retries_,
+        const ReadSettings & read_settings_)
+        : client_ptr(client_ptr_)
+        , bucket(bucket_)
+        , key(key_)
+        , read_settings(read_settings_)
+        , range_generator(object_size_, range_step_)
+        , range_step(range_step_)
+        , object_size(object_size_)
+        , s3_max_single_read_retries(s3_max_single_read_retries_)
+    {
+        assert(range_step > 0);
+        assert(range_step < object_size);
+    }
+
+    SeekableReadBufferPtr getReader() override;
+
+    off_t seek(off_t off, [[maybe_unused]] int whence) override;
+
+    std::optional<size_t> getTotalSize() override;
+
+private:
+    std::shared_ptr<Aws::S3::S3Client> client_ptr;
+    const String bucket;
+    const String key;
+    ReadSettings read_settings;
+
+    RangeGenerator range_generator;
+    size_t range_step;
+    size_t object_size;
+
+    UInt64 s3_max_single_read_retries;
+};
+
 }
 
 #endif

src/IO/ReadBufferFromString.h

@@ -15,4 +15,11 @@ public:
 
     explicit ReadBufferFromString(std::string_view s) : ReadBufferFromMemory(s.data(), s.size()) {}
 };
+
+class ReadBufferFromOwnString : public String, public ReadBufferFromString
+{
+public:
+    explicit ReadBufferFromOwnString(const String & s_): String(s_), ReadBufferFromString(*this) {}
+};
+
 }

src/IO/ReadWriteBufferFromHTTP.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include <functional>
+#include <Common/RangeGenerator.h>
 #include <IO/ConnectionTimeouts.h>
 #include <IO/HTTPCommon.h>
 #include <IO/ParallelReadBuffer.h>
@@ -635,43 +636,6 @@ public:
     void buildNewSession(const Poco::URI & uri) override { session = makeHTTPSession(uri, timeouts); }
 };
 
-class RangeGenerator
-{
-public:
-    explicit RangeGenerator(size_t total_size_, size_t range_step_, size_t range_start = 0)
-        : from(range_start), range_step(range_step_), total_size(total_size_)
-    {
-    }
-
-    size_t totalRanges() const { return static_cast<size_t>(round(static_cast<float>(total_size - from) / range_step)); }
-
-    using Range = std::pair<size_t, size_t>;
-
-    // return upper exclusive range of values, i.e. [from_range, to_range>
-    std::optional<Range> nextRange()
-    {
-        if (from >= total_size)
-        {
-            return std::nullopt;
-        }
-
-        auto to = from + range_step;
-        if (to >= total_size)
-        {
-            to = total_size;
-        }
-
-        Range range{from, to};
-        from = to;
-        return std::move(range);
-    }
-
-private:
-    size_t from;
-    size_t range_step;
-    size_t total_size;
-};
-
 class ReadWriteBufferFromHTTP : public detail::ReadWriteBufferFromHTTPBase<std::shared_ptr<UpdatableSession>>
 {
     using Parent = detail::ReadWriteBufferFromHTTPBase<std::shared_ptr<UpdatableSession>>;

src/IO/S3Common.cpp

@@ -24,6 +24,7 @@
 #    include <aws/core/utils/UUID.h>
 #    include <aws/core/http/HttpClientFactory.h>
 #    include <aws/s3/S3Client.h>
+#    include <aws/s3/model/HeadObjectRequest.h>  // Y_IGNORE
 
 #    include <IO/S3/PocoHTTPClientFactory.h>
 #    include <IO/S3/PocoHTTPClient.h>
@@ -682,6 +683,7 @@ namespace DB
 namespace ErrorCodes
 {
     extern const int BAD_ARGUMENTS;
+    extern const int S3_ERROR;
 }
 
 namespace S3
@@ -839,6 +841,26 @@ namespace S3
             throw Exception(ErrorCodes::BAD_ARGUMENTS, "Bucket name length is out of bounds in virtual hosted style S3 URI:     {}{}",
                             quoteString(bucket), !uri.empty() ? " (" + uri.toString() + ")" : "");
     }
+
+    size_t getObjectSize(std::shared_ptr<Aws::S3::S3Client> client_ptr, const String & bucket, const String & key, bool throw_on_error)
+    {
+        Aws::S3::Model::HeadObjectRequest req;
+        req.SetBucket(bucket);
+        req.SetKey(key);
+
+        Aws::S3::Model::HeadObjectOutcome outcome = client_ptr->HeadObject(req);
+
+        if (outcome.IsSuccess())
+        {
+            auto read_result = outcome.GetResultWithOwnership();
+            return static_cast<size_t>(read_result.GetContentLength());
+        }
+        else if (throw_on_error)
+        {
+            throw DB::Exception(outcome.GetError().GetMessage(), ErrorCodes::S3_ERROR);
+        }
+        return 0;
+    }
 }
 
 }

src/IO/S3Common.h

@@ -75,6 +75,8 @@ struct URI
     static void validateBucket(const String & bucket, const Poco::URI & uri);
 };
 
+size_t getObjectSize(std::shared_ptr<Aws::S3::S3Client> client_ptr, const String & bucket, const String & key, bool throw_on_error = true);
+
 }
 
 #endif

src/IO/examples/read_buffer.cpp

@@ -2,18 +2,15 @@
 
 #include <iostream>
 
-#include <base/types.h>
-#include <IO/ReadHelpers.h>
+#include <IO/ReadBuffer.h>
 #include <IO/ReadBufferFromString.h>
+#include <IO/ReadHelpers.h>
+#include <base/types.h>
 
-
-int main(int, char **)
+int readAndPrint(DB::ReadBuffer & in)
 {
     try
     {
-        std::string s = "-123456 123.456 вася пе\\tтя\t'\\'xyz\\\\'";
-        DB::ReadBufferFromString in(s);
-
         DB::Int64 a;
         DB::Float64 b;
         DB::String c, d;
@@ -31,12 +28,32 @@ int main(int, char **)
 
         std::cout << a << ' ' << b << ' ' << c << '\t' << '\'' << d << '\'' << std::endl;
         std::cout << in.count() << std::endl;
+        return 0;
     }
     catch (const DB::Exception & e)
     {
         std::cerr << e.what() << ", " << e.displayText() << std::endl;
         return 1;
     }
+}
+
+int main(int, char **)
+{
+    {
+        std::string s = "-123456 123.456 вася пе\\tтя\t'\\'xyz\\\\'";
+        DB::ReadBufferFromString in(s);
+        if (readAndPrint(in))
+            std::cout << "readAndPrint from ReadBufferFromString failed" << std::endl;
+    }
+
+
+    std::shared_ptr<DB::ReadBufferFromOwnString> in;
+    {
+        std::string s = "-123456 123.456 вася пе\\tтя\t'\\'xyz\\\\'";
+        in = std::make_shared<DB::ReadBufferFromOwnString>(s);
+    }
+    if (readAndPrint(*in))
+        std::cout << "readAndPrint from ReadBufferFromOwnString failed" << std::endl;
 
     return 0;
 }

src/Interpreters/Aggregator.cpp

@@ -1,4 +1,6 @@
+#include <algorithm>
 #include <future>
+#include <numeric>
 #include <Poco/Util/Application.h>
 
 #include <base/sort.h>
@@ -15,6 +17,7 @@
 #include <IO/WriteBufferFromFile.h>
 #include <Compression/CompressedWriteBuffer.h>
 #include <Interpreters/Aggregator.h>
+#include <Common/LRUCache.h>
 #include <Common/MemoryTracker.h>
 #include <Common/CurrentThread.h>
 #include <Common/typeid_cast.h>
@@ -27,12 +30,236 @@
 #include <Interpreters/JIT/CompiledExpressionCache.h>
 #include <Core/ProtocolDefines.h>
 
+#include <Parsers/ASTSelectQuery.h>
 
 namespace ProfileEvents
 {
-    extern const Event ExternalAggregationWritePart;
-    extern const Event ExternalAggregationCompressedBytes;
-    extern const Event ExternalAggregationUncompressedBytes;
+extern const Event ExternalAggregationWritePart;
+extern const Event ExternalAggregationCompressedBytes;
+extern const Event ExternalAggregationUncompressedBytes;
+extern const Event AggregationPreallocatedElementsInHashTables;
+extern const Event AggregationHashTablesInitializedAsTwoLevel;
+}
+
+namespace
+{
+/** Collects observed HashMap-s sizes to avoid redundant intermediate resizes.
+  */
+class HashTablesStatistics
+{
+public:
+    struct Entry
+    {
+        size_t sum_of_sizes; // used to determine if it's better to convert aggregation to two-level from the beginning
+        size_t median_size; // roughly the size we're going to preallocate on each thread
+    };
+
+    using Cache = DB::LRUCache<UInt64, Entry>;
+    using CachePtr = std::shared_ptr<Cache>;
+    using Params = DB::Aggregator::Params::StatsCollectingParams;
+
+    /// Collection and use of the statistics should be enabled.
+    std::optional<Entry> getSizeHint(const Params & params)
+    {
+        if (!params.isCollectionAndUseEnabled())
+            throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Collection and use of the statistics should be enabled.");
+
+        std::lock_guard lock(mutex);
+        const auto cache = getHashTableStatsCache(params, lock);
+        if (const auto hint = cache->get(params.key))
+        {
+            LOG_DEBUG(
+                &Poco::Logger::get("Aggregator"),
+                "An entry for key={} found in cache: sum_of_sizes={}, median_size={}",
+                params.key,
+                hint->sum_of_sizes,
+                hint->median_size);
+            return *hint;
+        }
+        return std::nullopt;
+    }
+
+    /// Collection and use of the statistics should be enabled.
+    void update(size_t sum_of_sizes, size_t median_size, const Params & params)
+    {
+        if (!params.isCollectionAndUseEnabled())
+            throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Collection and use of the statistics should be enabled.");
+
+        std::lock_guard lock(mutex);
+        const auto cache = getHashTableStatsCache(params, lock);
+        const auto hint = cache->get(params.key);
+        // We'll maintain the maximum among all the observed values until the next prediction turns out to be too wrong.
+        if (!hint || sum_of_sizes < hint->sum_of_sizes / 2 || hint->sum_of_sizes < sum_of_sizes || median_size < hint->median_size / 2
+            || hint->median_size < median_size)
+        {
+            LOG_DEBUG(
+                &Poco::Logger::get("Aggregator"),
+                "Statistics updated for key={}: new sum_of_sizes={}, median_size={}",
+                params.key,
+                sum_of_sizes,
+                median_size);
+            cache->set(params.key, std::make_shared<Entry>(Entry{.sum_of_sizes = sum_of_sizes, .median_size = median_size}));
+        }
+    }
+
+    std::optional<DB::HashTablesCacheStatistics> getCacheStats() const
+    {
+        std::lock_guard lock(mutex);
+        if (hash_table_stats)
+        {
+            size_t hits = 0, misses = 0;
+            hash_table_stats->getStats(hits, misses);
+            return DB::HashTablesCacheStatistics{.entries = hash_table_stats->count(), .hits = hits, .misses = misses};
+        }
+        return std::nullopt;
+    }
+
+    static size_t calculateCacheKey(const DB::ASTPtr & select_query)
+    {
+        if (!select_query)
+            throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Query ptr cannot be null");
+
+        const auto & select = select_query->as<DB::ASTSelectQuery &>();
+
+        // It may happen in some corner cases like `select 1 as num group by num`.
+        if (!select.tables())
+            return 0;
+
+        SipHash hash;
+        hash.update(select.tables()->getTreeHash());
+        if (const auto where = select.where())
+            hash.update(where->getTreeHash());
+        if (const auto group_by = select.groupBy())
+            hash.update(group_by->getTreeHash());
+        return hash.get64();
+    }
+
+private:
+    CachePtr getHashTableStatsCache(const Params & params, const std::lock_guard<std::mutex> &)
+    {
+        if (!hash_table_stats || hash_table_stats->maxSize() != params.max_entries_for_hash_table_stats)
+            hash_table_stats = std::make_shared<Cache>(params.max_entries_for_hash_table_stats);
+        return hash_table_stats;
+    }
+
+    mutable std::mutex mutex;
+    CachePtr hash_table_stats;
+};
+
+HashTablesStatistics & getHashTablesStatistics()
+{
+    static HashTablesStatistics hash_tables_stats;
+    return hash_tables_stats;
+}
+
+bool worthConvertToTwoLevel(
+    size_t group_by_two_level_threshold, size_t result_size, size_t group_by_two_level_threshold_bytes, auto result_size_bytes)
+{
+    // params.group_by_two_level_threshold will be equal to 0 if we have only one thread to execute aggregation (refer to AggregatingStep::transformPipeline).
+    return (group_by_two_level_threshold && result_size >= group_by_two_level_threshold)
+        || (group_by_two_level_threshold_bytes && result_size_bytes >= static_cast<Int64>(group_by_two_level_threshold_bytes));
+}
+
+DB::AggregatedDataVariants::Type convertToTwoLevelTypeIfPossible(DB::AggregatedDataVariants::Type type)
+{
+    using Type = DB::AggregatedDataVariants::Type;
+    switch (type)
+    {
+#define M(NAME) \
+    case Type::NAME: \
+        return Type::NAME##_two_level;
+        APPLY_FOR_VARIANTS_CONVERTIBLE_TO_TWO_LEVEL(M)
+#undef M
+        default:
+            return type;
+    }
+    __builtin_unreachable();
+}
+
+void initDataVariantsWithSizeHint(
+    DB::AggregatedDataVariants & result, DB::AggregatedDataVariants::Type method_chosen, const DB::Aggregator::Params & params)
+{
+    const auto & stats_collecting_params = params.stats_collecting_params;
+    if (stats_collecting_params.isCollectionAndUseEnabled())
+    {
+        if (auto hint = getHashTablesStatistics().getSizeHint(stats_collecting_params))
+        {
+            const auto max_threads = params.group_by_two_level_threshold != 0 ? std::max(params.max_threads, 1ul) : 1;
+            const auto lower_limit = hint->sum_of_sizes / max_threads;
+            const auto upper_limit = stats_collecting_params.max_size_to_preallocate_for_aggregation / max_threads;
+            const auto adjusted = std::min(std::max(lower_limit, hint->median_size), upper_limit);
+            if (worthConvertToTwoLevel(
+                    params.group_by_two_level_threshold,
+                    hint->sum_of_sizes,
+                    /*group_by_two_level_threshold_bytes*/ 0,
+                    /*result_size_bytes*/ 0))
+                method_chosen = convertToTwoLevelTypeIfPossible(method_chosen);
+            result.init(method_chosen, adjusted);
+            ProfileEvents::increment(ProfileEvents::AggregationHashTablesInitializedAsTwoLevel, result.isTwoLevel());
+            return;
+        }
+    }
+    result.init(method_chosen);
+}
+
+/// Collection and use of the statistics should be enabled.
+void updateStatistics(const DB::ManyAggregatedDataVariants & data_variants, const DB::Aggregator::Params::StatsCollectingParams & params)
+{
+    if (!params.isCollectionAndUseEnabled())
+        throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Collection and use of the statistics should be enabled.");
+
+    std::vector<size_t> sizes(data_variants.size());
+    for (size_t i = 0; i < data_variants.size(); ++i)
+        sizes[i] = data_variants[i]->size();
+    const auto median_size = sizes.begin() + sizes.size() / 2; // not precisely though...
+    std::nth_element(sizes.begin(), median_size, sizes.end());
+    const auto sum_of_sizes = std::accumulate(sizes.begin(), sizes.end(), 0ull);
+    getHashTablesStatistics().update(sum_of_sizes, *median_size, params);
+}
+
+// The std::is_constructible trait isn't suitable here because some classes have template constructors with semantics different from providing size hints.
+// Also string hash table variants are not supported due to the fact that both local perf tests and tests in CI showed slowdowns for them.
+template <typename...>
+struct HasConstructorOfNumberOfElements : std::false_type
+{
+};
+
+template <typename... Ts>
+struct HasConstructorOfNumberOfElements<HashMapTable<Ts...>> : std::true_type
+{
+};
+
+template <typename Key, typename Cell, typename Hash, typename Grower, typename Allocator, template <typename...> typename ImplTable>
+struct HasConstructorOfNumberOfElements<TwoLevelHashMapTable<Key, Cell, Hash, Grower, Allocator, ImplTable>> : std::true_type
+{
+};
+
+template <typename... Ts>
+struct HasConstructorOfNumberOfElements<HashTable<Ts...>> : std::true_type
+{
+};
+
+template <typename... Ts>
+struct HasConstructorOfNumberOfElements<TwoLevelHashTable<Ts...>> : std::true_type
+{
+};
+
+template <template <typename> typename Method, typename Base>
+struct HasConstructorOfNumberOfElements<Method<Base>> : HasConstructorOfNumberOfElements<Base>
+{
+};
+
+template <typename Method>
+auto constructWithReserveIfPossible(size_t size_hint)
+{
+    if constexpr (HasConstructorOfNumberOfElements<typename Method::Data>::value)
+    {
+        ProfileEvents::increment(ProfileEvents::AggregationPreallocatedElementsInHashTables, size_hint);
+        return std::make_unique<Method>(size_hint);
+    }
+    else
+        return std::make_unique<Method>();
+}
 }
 
 namespace DB
@@ -64,6 +291,10 @@ AggregatedDataVariants::~AggregatedDataVariants()
     }
 }
 
+std::optional<HashTablesCacheStatistics> getHashTablesCacheStatistics()
+{
+    return getHashTablesStatistics().getCacheStats();
+}
 
 void AggregatedDataVariants::convertToTwoLevel()
 {
@@ -88,6 +319,47 @@ void AggregatedDataVariants::convertToTwoLevel()
     }
 }
 
+void AggregatedDataVariants::init(Type type_, std::optional<size_t> size_hint)
+{
+    switch (type_)
+    {
+        case Type::EMPTY:
+            break;
+        case Type::without_key:
+            break;
+
+#define M(NAME, IS_TWO_LEVEL) \
+    case Type::NAME: \
+        if (size_hint) \
+            (NAME) = constructWithReserveIfPossible<decltype(NAME)::element_type>(*size_hint); \
+        else \
+            (NAME) = std::make_unique<decltype(NAME)::element_type>(); \
+        break;
+            APPLY_FOR_AGGREGATED_VARIANTS(M)
+#undef M
+    }
+
+    type = type_;
+}
+
+Aggregator::Params::StatsCollectingParams::StatsCollectingParams() = default;
+
+Aggregator::Params::StatsCollectingParams::StatsCollectingParams(
+    const ASTPtr & select_query_,
+    bool collect_hash_table_stats_during_aggregation_,
+    size_t max_entries_for_hash_table_stats_,
+    size_t max_size_to_preallocate_for_aggregation_)
+    : key(collect_hash_table_stats_during_aggregation_ ? HashTablesStatistics::calculateCacheKey(select_query_) : 0)
+    , max_entries_for_hash_table_stats(max_entries_for_hash_table_stats_)
+    , max_size_to_preallocate_for_aggregation(max_size_to_preallocate_for_aggregation_)
+{
+}
+
+bool Aggregator::Params::StatsCollectingParams::isCollectionAndUseEnabled() const
+{
+    return key != 0;
+}
+
 Block Aggregator::getHeader(bool final) const
 {
     return params.getHeader(final);
@@ -237,8 +509,7 @@ public:
 
 #endif
 
-Aggregator::Aggregator(const Params & params_)
-    : params(params_)
+Aggregator::Aggregator(const Params & params_) : params(params_)
 {
     /// Use query-level memory tracker
     if (auto * memory_tracker_child = CurrentThread::getMemoryTracker())
@@ -292,7 +563,6 @@ Aggregator::Aggregator(const Params & params_)
 #if USE_EMBEDDED_COMPILER
     compileAggregateFunctionsIfNeeded();
 #endif
-
 }
 
 #if USE_EMBEDDED_COMPILER
@@ -958,7 +1228,7 @@ bool Aggregator::executeOnBlock(Columns columns, UInt64 num_rows, AggregatedData
     /// How to perform the aggregation?
     if (result.empty())
     {
-        result.init(method_chosen);
+        initDataVariantsWithSizeHint(result, method_chosen, params);
         result.keys_size = params.keys_size;
         result.key_sizes = key_sizes;
         LOG_TRACE(log, "Aggregation method: {}", result.getMethodName());
@@ -1038,9 +1308,8 @@ bool Aggregator::executeOnBlock(Columns columns, UInt64 num_rows, AggregatedData
     /// Here all the results in the sum are taken into account, from different threads.
     auto result_size_bytes = current_memory_usage - memory_usage_before_aggregation;
 
-    bool worth_convert_to_two_level
-        = (params.group_by_two_level_threshold && result_size >= params.group_by_two_level_threshold)
-        || (params.group_by_two_level_threshold_bytes && result_size_bytes >= static_cast<Int64>(params.group_by_two_level_threshold_bytes));
+    bool worth_convert_to_two_level = worthConvertToTwoLevel(
+        params.group_by_two_level_threshold, result_size, params.group_by_two_level_threshold_bytes, result_size_bytes);
 
     /** Converting to a two-level data structure.
       * It allows you to make, in the subsequent, an effective merge - either economical from memory or parallel.
@@ -1327,10 +1596,7 @@ void Aggregator::convertToBlockImpl(
 
 
 template <typename Mapped>
-inline void Aggregator::insertAggregatesIntoColumns(
-    Mapped & mapped,
-    MutableColumns & final_aggregate_columns,
-    Arena * arena) const
+inline void Aggregator::insertAggregatesIntoColumns(Mapped & mapped, MutableColumns & final_aggregate_columns, Arena * arena) const
 {
     /** Final values of aggregate functions are inserted to columns.
       * Then states of aggregate functions, that are not longer needed, are destroyed.
@@ -2179,6 +2445,9 @@ ManyAggregatedDataVariants Aggregator::prepareVariantsToMerge(ManyAggregatedData
 
     LOG_TRACE(log, "Merging aggregated data");
 
+    if (params.stats_collecting_params.isCollectionAndUseEnabled())
+        updateStatistics(data_variants, params.stats_collecting_params);
+
     ManyAggregatedDataVariants non_empty_data;
     non_empty_data.reserve(data_variants.size());
     for (auto & data : data_variants)
@@ -2388,9 +2657,8 @@ bool Aggregator::mergeOnBlock(Block block, AggregatedDataVariants & result, bool
     /// Here all the results in the sum are taken into account, from different threads.
     auto result_size_bytes = current_memory_usage - memory_usage_before_aggregation;
 
-    bool worth_convert_to_two_level
-        = (params.group_by_two_level_threshold && result_size >= params.group_by_two_level_threshold)
-        || (params.group_by_two_level_threshold_bytes && result_size_bytes >= static_cast<Int64>(params.group_by_two_level_threshold_bytes));
+    bool worth_convert_to_two_level = worthConvertToTwoLevel(
+        params.group_by_two_level_threshold, result_size, params.group_by_two_level_threshold_bytes, result_size_bytes);
 
     /** Converting to a two-level data structure.
       * It allows you to make, in the subsequent, an effective merge - either economical from memory or parallel.

src/Interpreters/Aggregator.h

@@ -34,6 +34,7 @@
 #include <Columns/ColumnNullable.h>
 #include <Columns/ColumnLowCardinality.h>
 
+#include <Parsers/IAST_fwd.h>
 
 namespace DB
 {
@@ -129,6 +130,7 @@ private:
 template <typename Base>
 struct AggregationDataWithNullKeyTwoLevel : public Base
 {
+    using Base::Base;
     using Base::impls;
 
     AggregationDataWithNullKeyTwoLevel() = default;
@@ -183,6 +185,8 @@ struct AggregationMethodOneNumber
 
     AggregationMethodOneNumber() = default;
 
+    explicit AggregationMethodOneNumber(size_t size_hint) : data(size_hint) { }
+
     template <typename Other>
     explicit AggregationMethodOneNumber(const Other & other) : data(other.data)
     {
@@ -225,6 +229,8 @@ struct AggregationMethodString
     {
     }
 
+    explicit AggregationMethodString(size_t size_hint) : data(size_hint) { }
+
     using State = ColumnsHashing::HashMethodString<typename Data::value_type, Mapped>;
 
     static const bool low_cardinality_optimization = false;
@@ -250,6 +256,8 @@ struct AggregationMethodStringNoCache
 
     AggregationMethodStringNoCache() = default;
 
+    explicit AggregationMethodStringNoCache(size_t size_hint) : data(size_hint) { }
+
     template <typename Other>
     explicit AggregationMethodStringNoCache(const Other & other) : data(other.data)
     {
@@ -280,6 +288,8 @@ struct AggregationMethodFixedString
 
     AggregationMethodFixedString() = default;
 
+    explicit AggregationMethodFixedString(size_t size_hint) : data(size_hint) { }
+
     template <typename Other>
     explicit AggregationMethodFixedString(const Other & other) : data(other.data)
     {
@@ -309,6 +319,8 @@ struct AggregationMethodFixedStringNoCache
 
     AggregationMethodFixedStringNoCache() = default;
 
+    explicit AggregationMethodFixedStringNoCache(size_t size_hint) : data(size_hint) { }
+
     template <typename Other>
     explicit AggregationMethodFixedStringNoCache(const Other & other) : data(other.data)
     {
@@ -382,6 +394,8 @@ struct AggregationMethodKeysFixed
 
     AggregationMethodKeysFixed() = default;
 
+    explicit AggregationMethodKeysFixed(size_t size_hint) : data(size_hint) { }
+
     template <typename Other>
     explicit AggregationMethodKeysFixed(const Other & other) : data(other.data)
     {
@@ -473,6 +487,8 @@ struct AggregationMethodSerialized
 
     AggregationMethodSerialized() = default;
 
+    explicit AggregationMethodSerialized(size_t size_hint) : data(size_hint) { }
+
     template <typename Other>
     explicit AggregationMethodSerialized(const Other & other) : data(other.data)
     {
@@ -652,21 +668,7 @@ struct AggregatedDataVariants : private boost::noncopyable
 
     ~AggregatedDataVariants();
 
-    void init(Type type_)
-    {
-        switch (type_)
-        {
-            case Type::EMPTY:       break;
-            case Type::without_key: break;
-
-        #define M(NAME, IS_TWO_LEVEL) \
-            case Type::NAME: (NAME) = std::make_unique<decltype(NAME)::element_type>(); break;
-            APPLY_FOR_AGGREGATED_VARIANTS(M)
-        #undef M
-        }
-
-        type = type_;
-    }
+    void init(Type type_, std::optional<size_t> size_hint = std::nullopt);
 
     /// Number of rows (different keys).
     size_t size() const
@@ -929,29 +931,61 @@ public:
         bool compile_aggregate_expressions;
         size_t min_count_to_compile_aggregate_expression;
 
+        struct StatsCollectingParams
+        {
+            StatsCollectingParams();
+
+            StatsCollectingParams(
+                const ASTPtr & select_query_,
+                bool collect_hash_table_stats_during_aggregation_,
+                size_t max_entries_for_hash_table_stats_,
+                size_t max_size_to_preallocate_for_aggregation_);
+
+            bool isCollectionAndUseEnabled() const;
+
+            const UInt64 key = 0;
+            const size_t max_entries_for_hash_table_stats = 0;
+            const size_t max_size_to_preallocate_for_aggregation = 0;
+        };
+        StatsCollectingParams stats_collecting_params;
+
         Params(
             const Block & src_header_,
-            const ColumnNumbers & keys_, const AggregateDescriptions & aggregates_,
-            bool overflow_row_, size_t max_rows_to_group_by_, OverflowMode group_by_overflow_mode_,
-            size_t group_by_two_level_threshold_, size_t group_by_two_level_threshold_bytes_,
+            const ColumnNumbers & keys_,
+            const AggregateDescriptions & aggregates_,
+            bool overflow_row_,
+            size_t max_rows_to_group_by_,
+            OverflowMode group_by_overflow_mode_,
+            size_t group_by_two_level_threshold_,
+            size_t group_by_two_level_threshold_bytes_,
             size_t max_bytes_before_external_group_by_,
             bool empty_result_for_aggregation_by_empty_set_,
-            VolumePtr tmp_volume_, size_t max_threads_,
+            VolumePtr tmp_volume_,
+            size_t max_threads_,
             size_t min_free_disk_space_,
             bool compile_aggregate_expressions_,
             size_t min_count_to_compile_aggregate_expression_,
-            const Block & intermediate_header_ = {})
-            : src_header(src_header_),
-            intermediate_header(intermediate_header_),
-            keys(keys_), aggregates(aggregates_), keys_size(keys.size()), aggregates_size(aggregates.size()),
-            overflow_row(overflow_row_), max_rows_to_group_by(max_rows_to_group_by_), group_by_overflow_mode(group_by_overflow_mode_),
-            group_by_two_level_threshold(group_by_two_level_threshold_), group_by_two_level_threshold_bytes(group_by_two_level_threshold_bytes_),
-            max_bytes_before_external_group_by(max_bytes_before_external_group_by_),
-            empty_result_for_aggregation_by_empty_set(empty_result_for_aggregation_by_empty_set_),
-            tmp_volume(tmp_volume_), max_threads(max_threads_),
-            min_free_disk_space(min_free_disk_space_),
-            compile_aggregate_expressions(compile_aggregate_expressions_),
-            min_count_to_compile_aggregate_expression(min_count_to_compile_aggregate_expression_)
+            const Block & intermediate_header_ = {},
+            const StatsCollectingParams & stats_collecting_params_ = {})
+            : src_header(src_header_)
+            , intermediate_header(intermediate_header_)
+            , keys(keys_)
+            , aggregates(aggregates_)
+            , keys_size(keys.size())
+            , aggregates_size(aggregates.size())
+            , overflow_row(overflow_row_)
+            , max_rows_to_group_by(max_rows_to_group_by_)
+            , group_by_overflow_mode(group_by_overflow_mode_)
+            , group_by_two_level_threshold(group_by_two_level_threshold_)
+            , group_by_two_level_threshold_bytes(group_by_two_level_threshold_bytes_)
+            , max_bytes_before_external_group_by(max_bytes_before_external_group_by_)
+            , empty_result_for_aggregation_by_empty_set(empty_result_for_aggregation_by_empty_set_)
+            , tmp_volume(tmp_volume_)
+            , max_threads(max_threads_)
+            , min_free_disk_space(min_free_disk_space_)
+            , compile_aggregate_expressions(compile_aggregate_expressions_)
+            , min_count_to_compile_aggregate_expression(min_count_to_compile_aggregate_expression_)
+            , stats_collecting_params(stats_collecting_params_)
         {
         }
 
@@ -1350,4 +1384,13 @@ APPLY_FOR_AGGREGATED_VARIANTS(M)
 
 #undef M
 
+
+struct HashTablesCacheStatistics
+{
+    size_t entries = 0;
+    size_t hits = 0;
+    size_t misses = 0;
+};
+
+std::optional<HashTablesCacheStatistics> getHashTablesCacheStatistics();
 }

src/Interpreters/AsynchronousInsertQueue.cpp

@@ -32,6 +32,7 @@ namespace CurrentMetrics
 namespace ProfileEvents
 {
     extern const Event AsyncInsertQuery;
+    extern const Event AsyncInsertBytes;
 }
 
 namespace DB
@@ -222,7 +223,9 @@ void AsynchronousInsertQueue::pushImpl(InsertData::EntryPtr entry, QueueIterator
     if (!data)
         data = std::make_unique<InsertData>();
 
-    data->size += entry->bytes.size();
+    size_t entry_data_size = entry->bytes.size();
+
+    data->size += entry_data_size;
     data->last_update = std::chrono::steady_clock::now();
     data->entries.emplace_back(entry);
 
@@ -239,6 +242,7 @@ void AsynchronousInsertQueue::pushImpl(InsertData::EntryPtr entry, QueueIterator
 
     CurrentMetrics::add(CurrentMetrics::PendingAsyncInsert);
     ProfileEvents::increment(ProfileEvents::AsyncInsertQuery);
+    ProfileEvents::increment(ProfileEvents::AsyncInsertBytes, entry_data_size);
 }
 
 void AsynchronousInsertQueue::waitForProcessingQuery(const String & query_id, const Milliseconds & timeout)

src/Interpreters/AsynchronousMetrics.cpp

@@ -1,3 +1,4 @@
+#include <Interpreters/Aggregator.h>
 #include <Interpreters/AsynchronousMetrics.h>
 #include <Interpreters/AsynchronousMetricLog.h>
 #include <Interpreters/JIT/CompiledExpressionCache.h>
@@ -12,6 +13,7 @@
 #include <Storages/MarkCache.h>
 #include <Storages/StorageMergeTree.h>
 #include <Storages/StorageReplicatedMergeTree.h>
+#include <Storages/MergeTree/MergeTreeMetadataCache.h>
 #include <IO/UncompressedCache.h>
 #include <IO/MMappedFileCache.h>
 #include <IO/ReadHelpers.h>
@@ -607,6 +609,15 @@ void AsynchronousMetrics::update(std::chrono::system_clock::time_point update_ti
         }
     }
 
+#if USE_ROCKSDB
+    {
+        if (auto metadata_cache = getContext()->tryGetMergeTreeMetadataCache())
+        {
+            new_values["MergeTreeMetadataCacheSize"] = metadata_cache->getEstimateNumKeys();
+        }
+    }
+#endif
+
 #if USE_EMBEDDED_COMPILER
     {
         if (auto * compiled_expression_cache = CompiledExpressionCacheFactory::instance().tryGetCache())
@@ -617,8 +628,18 @@ void AsynchronousMetrics::update(std::chrono::system_clock::time_point update_ti
     }
 #endif
 
+
     new_values["Uptime"] = getContext()->getUptimeSeconds();
 
+    {
+        if (const auto stats = getHashTablesCacheStatistics())
+        {
+            new_values["HashTableStatsCacheEntries"] = stats->entries;
+            new_values["HashTableStatsCacheHits"] = stats->hits;
+            new_values["HashTableStatsCacheMisses"] = stats->misses;
+        }
+    }
+
     /// Process process memory usage according to OS
 #if defined(OS_LINUX) || defined(OS_FREEBSD)
     {

src/Interpreters/CatBoostModel.cpp

@@ -169,6 +169,7 @@ public:
         if (columns.size() != float_features_count + cat_features_count)
             throw Exception(ErrorCodes::BAD_ARGUMENTS,
                 "Number of columns is different with number of features: columns size {} float features size {} + cat features size {}",
+                columns.size(),
                 float_features_count,
                 cat_features_count);
 

src/Interpreters/Context.cpp

@@ -85,11 +85,15 @@
 #include <Storages/MergeTree/BackgroundJobsAssignee.h>
 #include <Storages/MergeTree/MergeTreeBackgroundExecutor.h>
 #include <Storages/MergeTree/MergeTreeDataPartUUID.h>
+#include <Storages/MergeTree/MergeTreeMetadataCache.h>
 #include <Interpreters/SynonymsExtensions.h>
 #include <Interpreters/Lemmatizers.h>
 #include <Interpreters/ClusterDiscovery.h>
 #include <filesystem>
 
+#if USE_ROCKSDB
+#include <rocksdb/table.h>
+#endif
 
 namespace fs = std::filesystem;
 
@@ -276,6 +280,11 @@ struct ContextSharedPart
 
     Context::ConfigReloadCallback config_reload_callback;
 
+#if USE_ROCKSDB
+    /// Global merge tree metadata cache, stored in rocksdb.
+    MergeTreeMetadataCachePtr merge_tree_metadata_cache;
+#endif
+
     ContextSharedPart()
         : access_control(std::make_unique<AccessControl>())
         , global_overcommit_tracker(&process_list)
@@ -410,6 +419,15 @@ struct ContextSharedPart
             trace_collector.reset();
             /// Stop zookeeper connection
             zookeeper.reset();
+
+#if USE_ROCKSDB
+            /// Shutdown merge tree metadata cache
+            if (merge_tree_metadata_cache)
+            {
+                merge_tree_metadata_cache->shutdown();
+                merge_tree_metadata_cache.reset();
+            }
+#endif
         }
 
         /// Can be removed w/o context lock
@@ -2048,6 +2066,23 @@ zkutil::ZooKeeperPtr Context::getAuxiliaryZooKeeper(const String & name) const
     return zookeeper->second;
 }
 
+#if USE_ROCKSDB
+MergeTreeMetadataCachePtr Context::getMergeTreeMetadataCache() const
+{
+    auto cache = tryGetMergeTreeMetadataCache();
+    if (!cache)
+        throw Exception(
+            ErrorCodes::LOGICAL_ERROR,
+            "Merge tree metadata cache is not initialized, please add config merge_tree_metadata_cache in config.xml and restart");
+    return cache;
+}
+
+MergeTreeMetadataCachePtr Context::tryGetMergeTreeMetadataCache() const
+{
+    return shared->merge_tree_metadata_cache;
+}
+#endif
+
 void Context::resetZooKeeper() const
 {
     std::lock_guard lock(shared->zookeeper_mutex);
@@ -2291,6 +2326,13 @@ void Context::initializeTraceCollector()
     shared->initializeTraceCollector(getTraceLog());
 }
 
+#if USE_ROCKSDB
+void Context::initializeMergeTreeMetadataCache(const String & dir, size_t size)
+{
+    shared->merge_tree_metadata_cache = MergeTreeMetadataCache::create(dir, size);
+}
+#endif
+
 bool Context::hasTraceCollector() const
 {
     return shared->hasTraceCollector();

src/Interpreters/Context.h

@@ -16,6 +16,7 @@
 #include <base/types.h>
 #include <Storages/MergeTree/ParallelReplicasReadingCoordinator.h>
 
+
 #include "config_core.h"
 
 #include <boost/container/flat_set.hpp>
@@ -152,6 +153,12 @@ using ReadTaskCallback = std::function<String()>;
 
 using MergeTreeReadTaskCallback = std::function<std::optional<PartitionReadResponse>(PartitionReadRequest)>;
 
+
+#if USE_ROCKSDB
+class MergeTreeMetadataCache;
+using MergeTreeMetadataCachePtr = std::shared_ptr<MergeTreeMetadataCache>;
+#endif
+
 /// An empty interface for an arbitrary object that may be attached by a shared pointer
 /// to query context, when using ClickHouse as a library.
 struct IHostContext
@@ -179,6 +186,7 @@ private:
     std::unique_ptr<ContextSharedPart> shared;
 };
 
+
 /** A set of known objects that can be used in the query.
   * Consists of a shared part (always common to all sessions and queries)
   *  and copied part (which can be its own for each session or query).
@@ -680,6 +688,11 @@ public:
 
     UInt32 getZooKeeperSessionUptime() const;
 
+#if USE_ROCKSDB
+    MergeTreeMetadataCachePtr getMergeTreeMetadataCache() const;
+    MergeTreeMetadataCachePtr tryGetMergeTreeMetadataCache() const;
+#endif
+
 #if USE_NURAFT
     std::shared_ptr<KeeperDispatcher> & getKeeperDispatcher() const;
 #endif
@@ -769,6 +782,10 @@ public:
     /// Call after initialization before using trace collector.
     void initializeTraceCollector();
 
+#if USE_ROCKSDB
+    void initializeMergeTreeMetadataCache(const String & dir, size_t size);
+#endif
+
     bool hasTraceCollector() const;
 
     /// Nullptr if the query log is not ready for this moment.

src/Interpreters/DatabaseCatalog.cpp

@@ -233,7 +233,7 @@ DatabaseAndTable DatabaseCatalog::getTableImpl(
         {
             assert(!db_and_table.first && !db_and_table.second);
             if (exception)
-                exception->emplace(ErrorCodes::UNKNOWN_TABLE, "Table {} doesn't exist", table_id.getNameForLogs());
+                exception->emplace(fmt::format("Table {} doesn't exist", table_id.getNameForLogs()), ErrorCodes::UNKNOWN_TABLE);
             return {};
         }
 
@@ -263,7 +263,7 @@ DatabaseAndTable DatabaseCatalog::getTableImpl(
         /// If table_id has no UUID, then the name of database was specified by user and table_id was not resolved through context.
         /// Do not allow access to TEMPORARY_DATABASE because it contains all temporary tables of all contexts and users.
         if (exception)
-            exception->emplace(ErrorCodes::DATABASE_ACCESS_DENIED, "Direct access to `{}` database is not allowed", String(TEMPORARY_DATABASE));
+            exception->emplace(fmt::format("Direct access to `{}` database is not allowed", TEMPORARY_DATABASE), ErrorCodes::DATABASE_ACCESS_DENIED);
         return {};
     }
 
@@ -274,7 +274,7 @@ DatabaseAndTable DatabaseCatalog::getTableImpl(
         if (databases.end() == it)
         {
             if (exception)
-                exception->emplace(ErrorCodes::UNKNOWN_DATABASE, "Database {} doesn't exist", backQuoteIfNeed(table_id.getDatabaseName()));
+                exception->emplace(fmt::format("Database {} doesn't exist", backQuoteIfNeed(table_id.getDatabaseName())), ErrorCodes::UNKNOWN_DATABASE);
             return {};
         }
         database = it->second;
@@ -282,7 +282,7 @@ DatabaseAndTable DatabaseCatalog::getTableImpl(
 
     auto table = database->tryGetTable(table_id.table_name, context_);
     if (!table && exception)
-            exception->emplace(ErrorCodes::UNKNOWN_TABLE, "Table {} doesn't exist", table_id.getNameForLogs());
+            exception->emplace(fmt::format("Table {} doesn't exist", table_id.getNameForLogs()), ErrorCodes::UNKNOWN_TABLE);
     if (!table)
         database = nullptr;
 

src/Interpreters/InterpreterInsertQuery.cpp

@@ -358,6 +358,7 @@ BlockIO InterpreterInsertQuery::execute()
 
                 auto new_context = Context::createCopy(context);
                 new_context->setSettings(new_settings);
+                new_context->setInsertionTable(getContext()->getInsertionTable());
 
                 InterpreterSelectWithUnionQuery interpreter_select{
                     query.select, new_context, SelectQueryOptions(QueryProcessingStage::Complete, 1)};