Merge master

.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: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -362,7 +360,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinGCC:
     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 +633,6 @@ jobs:
 ##########################################################################################
   BuilderDebSplitted:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -682,7 +678,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinTidy:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -728,7 +723,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinDarwin:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -774,7 +768,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinAarch64:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -820,7 +813,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinFreeBSD:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -866,7 +858,6 @@ jobs:
           sudo rm -fr "$TEMP_PATH" "$CACHES_PATH"
   BuilderBinDarwinAarch64:
     needs: [DockerHubPush]
-    if: ${{ !contains(github.event.pull_request.labels.*.name, 'pr-documentation') && !contains(github.event.pull_request.labels.*.name, 'pr-doc-fix') }}
     runs-on: [self-hosted, builder]
     steps:
       - name: Set envs
@@ -912,7 +903,6 @@ jobs:
           sudo rm -fr "$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

contrib/llvm-cmake/CMakeLists.txt

@@ -1,9 +1,12 @@
-if (APPLE OR NOT ARCH_AMD64 OR SANITIZE STREQUAL "undefined")
+# During cross-compilation in our CI we have to use llvm-tblgen and other building tools
-   set (ENABLE_EMBEDDED_COMPILER_DEFAULT OFF)
+# tools to be build for host architecture and everything else for target architecture (e.g. AArch64)
+# Possible workaround is to use llvm-tblgen from some package...
+# But lets just enable LLVM for native builds
+if (CMAKE_CROSSCOMPILING OR SANITIZE STREQUAL "undefined")
+    set (ENABLE_EMBEDDED_COMPILER_DEFAULT OFF)
 else()
-   set (ENABLE_EMBEDDED_COMPILER_DEFAULT ON)
+    set (ENABLE_EMBEDDED_COMPILER_DEFAULT ON)
 endif()
-
 option (ENABLE_EMBEDDED_COMPILER "Enable support for 'compile_expressions' option for query execution" ${ENABLE_EMBEDDED_COMPILER_DEFAULT})
 
 if (NOT ENABLE_EMBEDDED_COMPILER)

docker/test/performance-comparison/compare.sh

@@ -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

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。
 

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");
@@ -563,7 +568,6 @@ 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);
@@ -571,9 +575,6 @@ 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

@@ -829,6 +829,36 @@ 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/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()

src/Common/Exception.h

@@ -35,10 +35,10 @@ public:
     {}
 
     // Format message with fmt::format, like the logging functions.
-    template <typename ...Args>
+    template <typename... Args>
-    Exception(int code, const std::string & fmt, Args&&... args)
+    Exception(int code, fmt::format_string<Args...> fmt, Args &&... args) : Exception(fmt::format(fmt, std::forward<Args>(args)...), code)
-        : Exception(fmt::format(fmt::runtime(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>
+    template <typename... Args>
-    void addMessage(const std::string& format, Args&&... 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>
+    template <typename... Args>
-    ParsingException(int code, const std::string & fmt, Args&&... args)
+    ParsingException(int code, fmt::format_string<Args...> fmt, Args &&... args) : Exception(code, fmt, std::forward<Args>(args)...)
-        : Exception(fmt::format(fmt::runtime(fmt), std::forward<Args>(args)...), code)
+    {
-    {}
+    }
 
 
     std::string displayText() const

src/Common/ProfileEvents.cpp

@@ -285,6 +285,13 @@
     \
     M(MainConfigLoads, "Number of times the main configuration was reloaded.") \
     \
+    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/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/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/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 {}: {}",
+        throw Exception(
-                        task->execution_status.code, task->execution_status.message);
+            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: {}",
+                        "Failed to predownload remaining {} bytes. Current file segment: {}, current download offset: {}, expected: {}, "
-                        file_segment->range().toString(), file_segment->getDownloadOffset(), file_offset_of_buffer_end, implementation_buffer->eof());
+                        "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/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,
+            throw Exception(
-                            "Path in the 'format_schema' setting shouldn't go outside the 'format_schema_path' directory: {} ({} not in {})",
+                ErrorCodes::BAD_ARGUMENTS,
-                            path.string());
+                "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())
+            if (isObject(type))
-                return nullptr;
+            {
+                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)
+        if (is_object)
-            return nullptr;
+            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/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
@@ -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())
+            return createTupleToObjectWrapper(*from_tuple, to_type->hasNullableSubcolumns());
-                 throw Exception(ErrorCodes::TYPE_MISMATCH,
+        }
-                    "Cast to Object can be performed only from flatten Named Tuple. Got: {}", from_type->getName());
+        else if (const auto * from_map = checkAndGetDataType<DataTypeMap>(from_type.get()))
-
+        {
-            PathsInData paths;
+            return createMapToObjectWrapper(*from_map, to_type->hasNullableSubcolumns());
-            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;
-            };
         }
         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,
+            throw Exception(
-                            "Function {} with Null argument and default implementation for Nulls "
+                ErrorCodes::LOGICAL_ERROR,
-                            "is expected to return Nullable result, got {}", result_type->getName());
+                "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())
+                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH,
-                    + ", should be at least 2.",
+                "Number of arguments for function {} doesn't match: passed {}, should be at least 2",
-                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+                getName(),
-
+                arguments.size());
-        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);
 
         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 "
+                throw Exception(
-                                    + 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>();
@@ -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())
+                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH,
-                    + ", should be at least 1",
+                "Number of arguments for function {} doesn't match: passed {}, should be at least 2",
-                ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+                getName(),
-
+                arguments.size());
-        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);
 
         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
-        /// 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.
+        /// array will be filled with [1, 3, 1, 0] but without constant string indices.
-        UInt64 * index_positions = small_index_positions_buffer;
+        Format::IndexPositions index_positions;
-
-        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();
-        }
 
         /// 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

@@ -237,7 +237,7 @@ void ParallelReadBuffer::readerThreadFunction(ReadWorkerPtr read_worker)
         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

@@ -82,8 +82,8 @@ public:
         std::unique_ptr<ReadBufferFactory> reader_factory_,
         ThreadPool * pool,
         size_t max_working_readers,
-        WorkerSetup worker_setup = {},
+        WorkerSetup worker_setup = [](ThreadStatus &){},
-        WorkerCleanup worker_cleanup = {});
+        WorkerCleanup worker_cleanup = [](ThreadStatus &){});
 
     ~ParallelReadBuffer() override { finishAndWait(); }
 

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;
+    auto object_size = S3::getObjectSize(client_ptr, bucket, key, false);
-    request.SetBucket(bucket);
-    request.SetKey(key);
 
-    auto outcome = client_ptr->HeadObject(request);
+    if (!object_size)
-    auto head_result = outcome.GetResultWithOwnership();
+    {
-    file_size = head_result.GetContentLength();
+        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 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/ReadBuffer.h>
-#include <IO/ReadHelpers.h>
 #include <IO/ReadBufferFromString.h>
+#include <IO/ReadHelpers.h>
+#include <base/types.h>
 
-
+int readAndPrint(DB::ReadBuffer & in)
-int main(int, char **)
 {
     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/AsynchronousMetrics.cpp

@@ -12,6 +12,7 @@
 #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 +608,15 @@ 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())
@@ -617,6 +627,7 @@ void AsynchronousMetrics::update(std::chrono::system_clock::time_point update_ti
     }
 #endif
 
+
     new_values["Uptime"] = getContext()->getUptimeSeconds();
 
     /// Process process memory usage according to OS

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)};

src/Interpreters/executeDDLQueryOnCluster.cpp

@@ -320,12 +320,13 @@ Chunk DDLQueryStatusSource::generate()
             if (throw_on_timeout)
             {
                 if (!first_exception)
-                    first_exception = std::make_unique<Exception>(ErrorCodes::TIMEOUT_EXCEEDED, msg_format,
+                    first_exception = std::make_unique<Exception>(
-                        node_path, timeout_seconds, num_unfinished_hosts, num_active_hosts);
+                        fmt::format(msg_format, node_path, timeout_seconds, num_unfinished_hosts, num_active_hosts),
+                        ErrorCodes::TIMEOUT_EXCEEDED);
                 return {};
             }
 
-            LOG_INFO(log, fmt::runtime(msg_format), node_path, timeout_seconds, num_unfinished_hosts, num_active_hosts);
+            LOG_INFO(log, msg_format, node_path, timeout_seconds, num_unfinished_hosts, num_active_hosts);
 
             NameSet unfinished_hosts = waiting_hosts;
             for (const auto & host_id : finished_hosts)
@@ -358,9 +359,12 @@ Chunk DDLQueryStatusSource::generate()
             /// Paradoxically, this exception will be throw even in case of "never_throw" mode.
 
             if (!first_exception)
-                first_exception = std::make_unique<Exception>(ErrorCodes::UNFINISHED,
+                first_exception = std::make_unique<Exception>(
-                    "Cannot provide query execution status. The query's node {} has been deleted by the cleaner"
+                    fmt::format(
-                    " since it was finished (or its lifetime is expired)", node_path);
+                        "Cannot provide query execution status. The query's node {} has been deleted by the cleaner"
+                        " since it was finished (or its lifetime is expired)",
+                        node_path),
+                    ErrorCodes::UNFINISHED);
             return {};
         }
 
@@ -386,7 +390,8 @@ Chunk DDLQueryStatusSource::generate()
             if (status.code != 0 && !first_exception
                 && context->getSettingsRef().distributed_ddl_output_mode != DistributedDDLOutputMode::NEVER_THROW)
             {
-                first_exception = std::make_unique<Exception>(status.code, "There was an error on [{}:{}]: {}", host, port, status.message);
+                first_exception = std::make_unique<Exception>(
+                    fmt::format("There was an error on [{}:{}]: {}", host, port, status.message), status.code);
             }
 
             ++num_hosts_finished;

src/Interpreters/executeQuery.cpp

@@ -657,6 +657,14 @@ static std::tuple<ASTPtr, BlockIO> executeQueryImpl(
                 limits.size_limits = SizeLimits(settings.max_result_rows, settings.max_result_bytes, settings.result_overflow_mode);
             }
 
+            if (const auto * insert_interpreter = typeid_cast<const InterpreterInsertQuery *>(&*interpreter))
+            {
+                /// Save insertion table (not table function). TODO: support remote() table function.
+                auto table_id = insert_interpreter->getDatabaseTable();
+                if (!table_id.empty())
+                    context->setInsertionTable(std::move(table_id));
+            }
+
             {
                 std::unique_ptr<OpenTelemetrySpanHolder> span;
                 if (context->query_trace_context.trace_id != UUID())
@@ -667,14 +675,6 @@ static std::tuple<ASTPtr, BlockIO> executeQueryImpl(
                 }
                 res = interpreter->execute();
             }
-
-            if (const auto * insert_interpreter = typeid_cast<const InterpreterInsertQuery *>(&*interpreter))
-            {
-                /// Save insertion table (not table function). TODO: support remote() table function.
-                auto table_id = insert_interpreter->getDatabaseTable();
-                if (!table_id.empty())
-                    context->setInsertionTable(std::move(table_id));
-            }
         }
 
         if (process_list_entry)

src/Processors/Formats/ISchemaReader.h

@@ -18,6 +18,10 @@ public:
 
     virtual NamesAndTypesList readSchema() = 0;
 
+    /// True if order of columns is important in format.
+    /// Exceptions: JSON, TSKV.
+    virtual bool hasStrictOrderOfColumns() const { return true; }
+
     virtual ~ISchemaReader() = default;
 
 protected:
@@ -60,6 +64,7 @@ class IRowWithNamesSchemaReader : public ISchemaReader
 public:
     IRowWithNamesSchemaReader(ReadBuffer & in_, size_t max_rows_to_read_, DataTypePtr default_type_ = nullptr);
     NamesAndTypesList readSchema() override;
+    bool hasStrictOrderOfColumns() const override { return false; }
 
 protected:
     /// Read one row and determine types of columns in it.

src/Processors/Formats/Impl/MsgPackRowInputFormat.cpp

@@ -359,7 +359,7 @@ bool MsgPackVisitor::visit_ext(const char * value, uint32_t size)
         return true;
     }
 
-    throw Exception(ErrorCodes::BAD_ARGUMENTS, "Unsupported MsgPack extension type: {%x}", type);
+    throw Exception(ErrorCodes::BAD_ARGUMENTS, "Unsupported MsgPack extension type: {:x}", type);
 }
 
 void MsgPackVisitor::parse_error(size_t, size_t) // NOLINT
@@ -498,7 +498,7 @@ DataTypePtr MsgPackSchemaReader::getDataType(const msgpack::object & object)
             msgpack::object_ext object_ext = object.via.ext;
             if (object_ext.type() == int8_t(MsgPackExtensionTypes::UUIDType))
                 return std::make_shared<DataTypeUUID>();
-            throw Exception(ErrorCodes::BAD_ARGUMENTS, "Msgpack extension type {%x} is not supported", object_ext.type());
+            throw Exception(ErrorCodes::BAD_ARGUMENTS, "Msgpack extension type {:x} is not supported", object_ext.type());
         }
     }
     __builtin_unreachable();

src/Processors/QueryPlan/ReadFromMergeTree.cpp

@@ -45,7 +45,8 @@ namespace ErrorCodes
     extern const int LOGICAL_ERROR;
 }
 
-static MergeTreeReaderSettings getMergeTreeReaderSettings(const ContextPtr & context)
+static MergeTreeReaderSettings getMergeTreeReaderSettings(
+    const ContextPtr & context, const SelectQueryInfo & query_info)
 {
     const auto & settings = context->getSettingsRef();
     return
@@ -53,6 +54,7 @@ static MergeTreeReaderSettings getMergeTreeReaderSettings(const ContextPtr & con
         .read_settings = context->getReadSettings(),
         .save_marks_in_cache = true,
         .checksum_on_read = settings.checksum_on_read,
+        .read_in_order = query_info.input_order_info != nullptr,
     };
 }
 
@@ -82,7 +84,7 @@ ReadFromMergeTree::ReadFromMergeTree(
         getPrewhereInfo(query_info_),
         data_.getPartitionValueType(),
         virt_column_names_)})
-    , reader_settings(getMergeTreeReaderSettings(context_))
+    , reader_settings(getMergeTreeReaderSettings(context_, query_info_))
     , prepared_parts(std::move(parts_))
     , real_column_names(std::move(real_column_names_))
     , virt_column_names(std::move(virt_column_names_))
@@ -206,6 +208,7 @@ ProcessorPtr ReadFromMergeTree::createSource(
             .colums_to_read = required_columns
         };
     }
+
     return std::make_shared<TSource>(
             data, storage_snapshot, part.data_part, max_block_size, preferred_block_size_bytes,
             preferred_max_column_in_block_size_bytes, required_columns, part.ranges, use_uncompressed_cache, prewhere_info,
@@ -921,7 +924,7 @@ MergeTreeDataSelectAnalysisResultPtr ReadFromMergeTree::selectRangesToRead(
             total_marks_pk += part->index_granularity.getMarksCountWithoutFinal();
         parts_before_pk = parts.size();
 
-        auto reader_settings = getMergeTreeReaderSettings(context);
+        auto reader_settings = getMergeTreeReaderSettings(context, query_info);
 
         bool use_skip_indexes = settings.use_skip_indexes;
         if (select.final() && !settings.use_skip_indexes_if_final)

src/Storages/Kafka/ReadBufferFromKafkaConsumer.cpp

@@ -1,10 +1,13 @@
+// Needs to go first because its partial specialization of fmt::formatter
+// should be defined before any instantiation
+#include <fmt/ostream.h>
+
 #include <Storages/Kafka/ReadBufferFromKafkaConsumer.h>
 
 #include <base/logger_useful.h>
 
 #include <cppkafka/cppkafka.h>
 #include <boost/algorithm/string/join.hpp>
-#include <fmt/ostream.h>
 #include <algorithm>
 
 namespace DB

src/Storages/MergeTree/IMergeTreeDataPart.cpp

@@ -1,9 +1,11 @@
 #include "IMergeTreeDataPart.h"
 
 #include <optional>
+#include <boost/algorithm/string/join.hpp>
 #include <string_view>
 #include <Core/Defines.h>
 #include <IO/HashingWriteBuffer.h>
+#include <IO/HashingReadBuffer.h>
 #include <IO/ReadBufferFromString.h>
 #include <IO/ReadHelpers.h>
 #include <IO/WriteHelpers.h>
@@ -11,6 +13,8 @@
 #include <Storages/MergeTree/localBackup.h>
 #include <Storages/MergeTree/checkDataPart.h>
 #include <Storages/StorageReplicatedMergeTree.h>
+#include <Storages/MergeTree/PartMetadataManagerOrdinary.h>
+#include <Storages/MergeTree/PartMetadataManagerWithCache.h>
 #include <Common/StringUtils/StringUtils.h>
 #include <Common/escapeForFileName.h>
 #include <Common/ZooKeeper/ZooKeeper.h>
@@ -61,13 +65,7 @@ namespace ErrorCodes
     extern const int NOT_IMPLEMENTED;
 }
 
-static std::unique_ptr<ReadBufferFromFileBase> openForReading(const DiskPtr & disk, const String & path)
+void IMergeTreeDataPart::MinMaxIndex::load(const MergeTreeData & data, const PartMetadataManagerPtr & manager)
-{
-    size_t file_size = disk->getFileSize(path);
-    return disk->readFile(path, ReadSettings().adjustBufferSize(file_size), file_size);
-}
-
-void IMergeTreeDataPart::MinMaxIndex::load(const MergeTreeData & data, const DiskPtr & disk_, const String & part_path)
 {
     auto metadata_snapshot = data.getInMemoryMetadataPtr();
     const auto & partition_key = metadata_snapshot->getPartitionKey();
@@ -79,8 +77,8 @@ void IMergeTreeDataPart::MinMaxIndex::load(const MergeTreeData & data, const Dis
     hyperrectangle.reserve(minmax_idx_size);
     for (size_t i = 0; i < minmax_idx_size; ++i)
     {
-        String file_name = fs::path(part_path) / ("minmax_" + escapeForFileName(minmax_column_names[i]) + ".idx");
+        String file_name = "minmax_" + escapeForFileName(minmax_column_names[i]) + ".idx";
-        auto file = openForReading(disk_, file_name);
+        auto file = manager->read(file_name);
         auto serialization = minmax_column_types[i]->getDefaultSerialization();
 
         Field min_val;
@@ -192,6 +190,19 @@ void IMergeTreeDataPart::MinMaxIndex::merge(const MinMaxIndex & other)
     }
 }
 
+void IMergeTreeDataPart::MinMaxIndex::appendFiles(const MergeTreeData & data, Strings & files)
+{
+    auto metadata_snapshot = data.getInMemoryMetadataPtr();
+    const auto & partition_key = metadata_snapshot->getPartitionKey();
+    auto minmax_column_names = data.getMinMaxColumnsNames(partition_key);
+    size_t minmax_idx_size = minmax_column_names.size();
+    for (size_t i = 0; i < minmax_idx_size; ++i)
+    {
+        String file_name = "minmax_" + escapeForFileName(minmax_column_names[i]) + ".idx";
+        files.push_back(file_name);
+    }
+}
+
 
 static void incrementStateMetric(IMergeTreeDataPart::State state)
 {
@@ -299,6 +310,7 @@ IMergeTreeDataPart::IMergeTreeDataPart(
     , index_granularity_info(storage_, part_type_)
     , part_type(part_type_)
     , parent_part(parent_part_)
+    , use_metadata_cache(storage.use_metadata_cache)
 {
     if (parent_part)
         state = State::Active;
@@ -306,6 +318,8 @@ IMergeTreeDataPart::IMergeTreeDataPart(
     incrementTypeMetric(part_type);
 
     minmax_idx = std::make_shared<MinMaxIndex>();
+
+    initializePartMetadataManager();
 }
 
 IMergeTreeDataPart::IMergeTreeDataPart(
@@ -324,6 +338,7 @@ IMergeTreeDataPart::IMergeTreeDataPart(
     , index_granularity_info(storage_, part_type_)
     , part_type(part_type_)
     , parent_part(parent_part_)
+    , use_metadata_cache(storage.use_metadata_cache)
 {
     if (parent_part)
         state = State::Active;
@@ -331,6 +346,8 @@ IMergeTreeDataPart::IMergeTreeDataPart(
     incrementTypeMetric(part_type);
 
     minmax_idx = std::make_shared<MinMaxIndex>();
+
+    initializePartMetadataManager();
 }
 
 IMergeTreeDataPart::~IMergeTreeDataPart()
@@ -558,9 +575,10 @@ size_t IMergeTreeDataPart::getFileSizeOrZero(const String & file_name) const
     return checksum->second.file_size;
 }
 
-String IMergeTreeDataPart::getColumnNameWithMinimumCompressedSize(const StorageMetadataPtr & metadata_snapshot) const
+String IMergeTreeDataPart::getColumnNameWithMinimumCompressedSize(const StorageSnapshotPtr & storage_snapshot) const
 {
-    const auto & storage_columns = metadata_snapshot->getColumns().getAllPhysical();
+    auto options = GetColumnsOptions(GetColumnsOptions::AllPhysical).withExtendedObjects().withSubcolumns();
+    auto storage_columns = storage_snapshot->getColumns(options);
     MergeTreeData::AlterConversions alter_conversions;
     if (!parent_part)
         alter_conversions = storage.getAlterConversionsForPart(shared_from_this());
@@ -637,6 +655,33 @@ void IMergeTreeDataPart::loadColumnsChecksumsIndexes(bool require_columns_checks
     loadDefaultCompressionCodec();
 }
 
+void IMergeTreeDataPart::appendFilesOfColumnsChecksumsIndexes(Strings & files, bool include_projection) const
+{
+    if (isStoredOnDisk())
+    {
+        appendFilesOfUUID(files);
+        appendFilesOfColumns(files);
+        appendFilesOfChecksums(files);
+        appendFilesOfIndexGranularity(files);
+        appendFilesOfIndex(files);
+        appendFilesOfRowsCount(files);
+        appendFilesOfPartitionAndMinMaxIndex(files);
+        appendFilesOfTTLInfos(files);
+        appendFilesOfDefaultCompressionCodec(files);
+    }
+
+    if (!parent_part && include_projection)
+    {
+        for (const auto & [projection_name, projection_part] : projection_parts)
+        {
+            Strings projection_files;
+            projection_part->appendFilesOfColumnsChecksumsIndexes(projection_files, true);
+            for (const auto & projection_file : projection_files)
+                files.push_back(fs::path(projection_part->relative_path) / projection_file);
+        }
+    }
+}
+
 void IMergeTreeDataPart::loadProjections(bool require_columns_checksums, bool check_consistency)
 {
     auto metadata_snapshot = storage.getInMemoryMetadataPtr();
@@ -657,6 +702,11 @@ void IMergeTreeDataPart::loadIndexGranularity()
     throw Exception("Method 'loadIndexGranularity' is not implemented for part with type " + getType().toString(), ErrorCodes::NOT_IMPLEMENTED);
 }
 
+/// Currently we don't cache mark files of part, because cache other meta files is enough to speed up loading.
+void IMergeTreeDataPart::appendFilesOfIndexGranularity(Strings & /* files */) const
+{
+}
+
 void IMergeTreeDataPart::loadIndex()
 {
     /// It can be empty in case of mutations
@@ -680,9 +730,9 @@ void IMergeTreeDataPart::loadIndex()
             loaded_index[i]->reserve(index_granularity.getMarksCount());
         }
 
-        String index_path = fs::path(getFullRelativePath()) / "primary.idx";
+        String index_name = "primary.idx";
-        auto index_file = openForReading(volume->getDisk(), index_path);
+        String index_path = fs::path(getFullRelativePath()) / index_name;
-
+        auto index_file = metadata_manager->read(index_name);
         size_t marks_count = index_granularity.getMarksCount();
 
         Serializations key_serializations(key_size);
@@ -709,6 +759,19 @@ void IMergeTreeDataPart::loadIndex()
     }
 }
 
+void IMergeTreeDataPart::appendFilesOfIndex(Strings & files) const
+{
+    auto metadata_snapshot = storage.getInMemoryMetadataPtr();
+    if (parent_part)
+        metadata_snapshot = metadata_snapshot->projections.has(name) ? metadata_snapshot->projections.get(name).metadata : nullptr;
+
+    if (!metadata_snapshot)
+        return;
+
+    if (metadata_snapshot->hasPrimaryKey())
+        files.push_back("primary.idx");
+}
+
 NameSet IMergeTreeDataPart::getFileNamesWithoutChecksums() const
 {
     if (!isStoredOnDisk())
@@ -733,14 +796,14 @@ void IMergeTreeDataPart::loadDefaultCompressionCodec()
     }
 
     String path = fs::path(getFullRelativePath()) / DEFAULT_COMPRESSION_CODEC_FILE_NAME;
-    if (!volume->getDisk()->exists(path))
+    bool exists = metadata_manager->exists(DEFAULT_COMPRESSION_CODEC_FILE_NAME);
+    if (!exists)
     {
         default_codec = detectDefaultCompressionCodec();
     }
     else
     {
-
+        auto file_buf = metadata_manager->read(DEFAULT_COMPRESSION_CODEC_FILE_NAME);
-        auto file_buf = openForReading(volume->getDisk(), path);
         String codec_line;
         readEscapedStringUntilEOL(codec_line, *file_buf);
 
@@ -748,7 +811,13 @@ void IMergeTreeDataPart::loadDefaultCompressionCodec()
 
         if (!checkString("CODEC", buf))
         {
-            LOG_WARNING(storage.log, "Cannot parse default codec for part {} from file {}, content '{}'. Default compression codec will be deduced automatically, from data on disk", name, path, codec_line);
+            LOG_WARNING(
+                storage.log,
+                "Cannot parse default codec for part {} from file {}, content '{}'. Default compression codec will be deduced "
+                "automatically, from data on disk",
+                name,
+                path,
+                codec_line);
             default_codec = detectDefaultCompressionCodec();
         }
 
@@ -766,6 +835,11 @@ void IMergeTreeDataPart::loadDefaultCompressionCodec()
     }
 }
 
+void IMergeTreeDataPart::appendFilesOfDefaultCompressionCodec(Strings & files)
+{
+    files.push_back(DEFAULT_COMPRESSION_CODEC_FILE_NAME);
+}
+
 CompressionCodecPtr IMergeTreeDataPart::detectDefaultCompressionCodec() const
 {
     /// In memory parts doesn't have any compression
@@ -828,7 +902,7 @@ void IMergeTreeDataPart::loadPartitionAndMinMaxIndex()
     {
         String path = getFullRelativePath();
         if (!parent_part)
-            partition.load(storage, volume->getDisk(), path);
+            partition.load(storage, metadata_manager);
 
         if (!isEmpty())
         {
@@ -836,7 +910,7 @@ void IMergeTreeDataPart::loadPartitionAndMinMaxIndex()
                 // projection parts don't have minmax_idx, and it's always initialized
                 minmax_idx->initialized = true;
             else
-                minmax_idx->load(storage, volume->getDisk(), path);
+                minmax_idx->load(storage, metadata_manager);
         }
         if (parent_part)
             return;
@@ -851,13 +925,26 @@ void IMergeTreeDataPart::loadPartitionAndMinMaxIndex()
             ErrorCodes::CORRUPTED_DATA);
 }
 
+void IMergeTreeDataPart::appendFilesOfPartitionAndMinMaxIndex(Strings & files) const
+{
+    if (storage.format_version < MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING && !parent_part)
+        return;
+
+    if (!parent_part)
+        partition.appendFiles(storage, files);
+
+    if (!isEmpty())
+        if (!parent_part)
+            minmax_idx->appendFiles(storage, files);
+}
+
 void IMergeTreeDataPart::loadChecksums(bool require)
 {
     const String path = fs::path(getFullRelativePath()) / "checksums.txt";
-
+    bool exists = metadata_manager->exists("checksums.txt");
-    if (volume->getDisk()->exists(path))
+    if (exists)
     {
-        auto buf = openForReading(volume->getDisk(), path);
+        auto buf = metadata_manager->read("checksums.txt");
         if (checksums.read(*buf))
         {
             assertEOF(*buf);
@@ -888,13 +975,18 @@ void IMergeTreeDataPart::loadChecksums(bool require)
     }
 }
 
+void IMergeTreeDataPart::appendFilesOfChecksums(Strings & files)
+{
+    files.push_back("checksums.txt");
+}
+
 void IMergeTreeDataPart::loadRowsCount()
 {
     String path = fs::path(getFullRelativePath()) / "count.txt";
 
     auto read_rows_count = [&]()
     {
-        auto buf = openForReading(volume->getDisk(), path);
+        auto buf = metadata_manager->read("count.txt");
         readIntText(rows_count, *buf);
         assertEOF(*buf);
     };
@@ -905,7 +997,8 @@ void IMergeTreeDataPart::loadRowsCount()
     }
     else if (storage.format_version >= MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING || part_type == Type::COMPACT || parent_part)
     {
-        if (!volume->getDisk()->exists(path))
+        bool exists = metadata_manager->exists("count.txt");
+        if (!exists)
             throw Exception("No count.txt in part " + name, ErrorCodes::NO_FILE_IN_DATA_PART);
 
         read_rows_count();
@@ -1003,12 +1096,17 @@ void IMergeTreeDataPart::loadRowsCount()
     }
 }
 
+void IMergeTreeDataPart::appendFilesOfRowsCount(Strings & files)
+{
+    files.push_back("count.txt");
+}
+
 void IMergeTreeDataPart::loadTTLInfos()
 {
-    String path = fs::path(getFullRelativePath()) / "ttl.txt";
+    bool exists = metadata_manager->exists("ttl.txt");
-    if (volume->getDisk()->exists(path))
+    if (exists)
     {
-        auto in = openForReading(volume->getDisk(), path);
+        auto in = metadata_manager->read("ttl.txt");
         assertString("ttl format version: ", *in);
         size_t format_version;
         readText(format_version, *in);
@@ -1030,19 +1128,29 @@ void IMergeTreeDataPart::loadTTLInfos()
     }
 }
 
+
+void IMergeTreeDataPart::appendFilesOfTTLInfos(Strings & files)
+{
+    files.push_back("ttl.txt");
+}
+
 void IMergeTreeDataPart::loadUUID()
 {
-    String path = fs::path(getFullRelativePath()) / UUID_FILE_NAME;
+    bool exists = metadata_manager->exists(UUID_FILE_NAME);
-
+    if (exists)
-    if (volume->getDisk()->exists(path))
     {
-        auto in = openForReading(volume->getDisk(), path);
+        auto in = metadata_manager->read(UUID_FILE_NAME);
         readText(uuid, *in);
         if (uuid == UUIDHelpers::Nil)
             throw Exception("Unexpected empty " + String(UUID_FILE_NAME) + " in part: " + name, ErrorCodes::LOGICAL_ERROR);
     }
 }
 
+void IMergeTreeDataPart::appendFilesOfUUID(Strings & files)
+{
+    files.push_back(UUID_FILE_NAME);
+}
+
 void IMergeTreeDataPart::loadColumns(bool require)
 {
     String path = fs::path(getFullRelativePath()) / "columns.txt";
@@ -1051,7 +1159,8 @@ void IMergeTreeDataPart::loadColumns(bool require)
         metadata_snapshot = metadata_snapshot->projections.get(name).metadata;
     NamesAndTypesList loaded_columns;
 
-    if (!volume->getDisk()->exists(path))
+    bool exists = metadata_manager->exists("columns.txt");
+    if (!exists)
     {
         /// We can get list of columns only from columns.txt in compact parts.
         if (require || part_type == Type::COMPACT)
@@ -1074,7 +1183,8 @@ void IMergeTreeDataPart::loadColumns(bool require)
     }
     else
     {
-        loaded_columns.readText(*volume->getDisk()->readFile(path));
+        auto in = metadata_manager->read("columns.txt");
+        loaded_columns.readText(*in);
 
         for (const auto & column : loaded_columns)
         {
@@ -1091,14 +1201,23 @@ void IMergeTreeDataPart::loadColumns(bool require)
     };
 
     SerializationInfoByName infos(loaded_columns, settings);
-    path = getFullRelativePath() + SERIALIZATION_FILE_NAME;
+    exists =  metadata_manager->exists(SERIALIZATION_FILE_NAME);
-    if (volume->getDisk()->exists(path))
+    if (exists)
-        infos.readJSON(*volume->getDisk()->readFile(path));
+    {
+        auto in = metadata_manager->read(SERIALIZATION_FILE_NAME);
+        infos.readJSON(*in);
+    }
 
     setColumns(loaded_columns);
     setSerializationInfos(infos);
 }
 
+void IMergeTreeDataPart::appendFilesOfColumns(Strings & files)
+{
+    files.push_back("columns.txt");
+    files.push_back(SERIALIZATION_FILE_NAME);
+}
+
 bool IMergeTreeDataPart::shallParticipateInMerges(const StoragePolicyPtr & storage_policy) const
 {
     /// `IMergeTreeDataPart::volume` describes space where current part belongs, and holds
@@ -1150,9 +1269,12 @@ try
         }
     }
 
+    metadata_manager->deleteAll(true);
+    metadata_manager->assertAllDeleted(true);
     volume->getDisk()->setLastModified(from, Poco::Timestamp::fromEpochTime(time(nullptr)));
     volume->getDisk()->moveDirectory(from, to);
     relative_path = new_relative_path;
+    metadata_manager->updateAll(true);
 
     SyncGuardPtr sync_guard;
     if (storage.getSettings()->fsync_part_directory)
@@ -1190,6 +1312,18 @@ std::optional<bool> IMergeTreeDataPart::keepSharedDataInDecoupledStorage() const
     return !storage.unlockSharedData(*this);
 }
 
+void IMergeTreeDataPart::initializePartMetadataManager()
+{
+#if USE_ROCKSDB
+    if (use_metadata_cache)
+        metadata_manager = std::make_shared<PartMetadataManagerWithCache>(this, storage.getContext()->getMergeTreeMetadataCache());
+    else
+        metadata_manager = std::make_shared<PartMetadataManagerOrdinary>(this);
+#else
+        metadata_manager = std::make_shared<PartMetadataManagerOrdinary>(this);
+#endif
+}
+
 void IMergeTreeDataPart::remove() const
 {
     std::optional<bool> keep_shared_data = keepSharedDataInDecoupledStorage();
@@ -1209,6 +1343,9 @@ void IMergeTreeDataPart::remove() const
         return;
     }
 
+    metadata_manager->deleteAll(false);
+    metadata_manager->assertAllDeleted(false);
+
     /** Atomic directory removal:
       * - rename directory to temporary name;
       * - remove it recursive.
@@ -1314,6 +1451,9 @@ void IMergeTreeDataPart::remove() const
 
 void IMergeTreeDataPart::projectionRemove(const String & parent_to, bool keep_shared_data) const
 {
+    metadata_manager->deleteAll(false);
+    metadata_manager->assertAllDeleted(false);
+
     String to = fs::path(parent_to) / relative_path;
     auto disk = volume->getDisk();
     if (checksums.empty())
@@ -1661,6 +1801,35 @@ String IMergeTreeDataPart::getZeroLevelPartBlockID(std::string_view token) const
     return info.partition_id + "_" + toString(hash_value.words[0]) + "_" + toString(hash_value.words[1]);
 }
 
+IMergeTreeDataPart::uint128 IMergeTreeDataPart::getActualChecksumByFile(const String & file_path) const
+{
+    assert(use_metadata_cache);
+
+    String file_name = std::filesystem::path(file_path).filename();
+    const auto filenames_without_checksums = getFileNamesWithoutChecksums();
+    auto it = checksums.files.find(file_name);
+    if (filenames_without_checksums.count(file_name) == 0 && it != checksums.files.end())
+    {
+        return it->second.file_hash;
+    }
+
+    if (!volume->getDisk()->exists(file_path))
+    {
+        return {};
+    }
+    std::unique_ptr<ReadBufferFromFileBase> in_file = volume->getDisk()->readFile(file_path);
+    HashingReadBuffer in_hash(*in_file);
+
+    String value;
+    readStringUntilEOF(value, in_hash);
+    return in_hash.getHash();
+}
+
+std::unordered_map<String, IMergeTreeDataPart::uint128> IMergeTreeDataPart::checkMetadata() const
+{
+    return metadata_manager->check();
+}
+
 bool isCompactPart(const MergeTreeDataPartPtr & data_part)
 {
     return (data_part && data_part->getType() == MergeTreeDataPartType::COMPACT);

src/Storages/MergeTree/IMergeTreeDataPart.h

@@ -14,6 +14,7 @@
 #include <Storages/MergeTree/MergeTreeIOSettings.h>
 #include <Storages/MergeTree/KeyCondition.h>
 #include <DataTypes/Serializations/SerializationInfo.h>
+#include <Storages/MergeTree/IPartMetadataManager.h>
 
 #include <shared_mutex>
 
@@ -60,6 +61,8 @@ public:
 
     using Type = MergeTreeDataPartType;
 
+    using uint128 = IPartMetadataManager::uint128;
+
 
     IMergeTreeDataPart(
         const MergeTreeData & storage_,
@@ -148,6 +151,7 @@ public:
     /// Initialize columns (from columns.txt if exists, or create from column files if not).
     /// Load checksums from checksums.txt if exists. Load index if required.
     void loadColumnsChecksumsIndexes(bool require_columns_checksums, bool check_consistency);
+    void appendFilesOfColumnsChecksumsIndexes(Strings & files, bool include_projection = false) const;
 
     String getMarksFileExtension() const { return index_granularity_info.marks_file_extension; }
 
@@ -164,7 +168,7 @@ public:
 
     /// Returns the name of a column with minimum compressed size (as returned by getColumnSize()).
     /// If no checksums are present returns the name of the first physically existing column.
-    String getColumnNameWithMinimumCompressedSize(const StorageMetadataPtr & metadata_snapshot) const;
+    String getColumnNameWithMinimumCompressedSize(const StorageSnapshotPtr & storage_snapshot) const;
 
     bool contains(const IMergeTreeDataPart & other) const { return info.contains(other.info); }
 
@@ -243,7 +247,7 @@ public:
     using TTLInfo = MergeTreeDataPartTTLInfo;
     using TTLInfos = MergeTreeDataPartTTLInfos;
 
-    TTLInfos ttl_infos;
+    mutable TTLInfos ttl_infos;
 
     /// Current state of the part. If the part is in working set already, it should be accessed via data_parts mutex
     void setState(State new_state) const;
@@ -300,14 +304,16 @@ public:
         {
         }
 
+        void load(const MergeTreeData & data, const PartMetadataManagerPtr & manager);
+
         using WrittenFiles = std::vector<std::unique_ptr<WriteBufferFromFileBase>>;
 
-        void load(const MergeTreeData & data, const DiskPtr & disk_, const String & part_path);
         [[nodiscard]] WrittenFiles store(const MergeTreeData & data, const DiskPtr & disk_, const String & part_path, Checksums & checksums) const;
         [[nodiscard]] WrittenFiles store(const Names & column_names, const DataTypes & data_types, const DiskPtr & disk_, const String & part_path, Checksums & checksums) const;
 
         void update(const Block & block, const Names & column_names);
         void merge(const MinMaxIndex & other);
+        static void appendFiles(const MergeTreeData & data, Strings & files);
     };
 
     using MinMaxIndexPtr = std::shared_ptr<MinMaxIndex>;
@@ -429,6 +435,12 @@ public:
     /// Required for distinguish different copies of the same part on remote FS.
     String getUniqueId() const;
 
+    /// Get checksums of metadata file in part directory
+    IMergeTreeDataPart::uint128 getActualChecksumByFile(const String & file_path) const;
+
+    /// Check metadata in cache is consistent with actual metadata on disk(if use_metadata_cache is true)
+    std::unordered_map<String, uint128> checkMetadata() const;
+
 protected:
 
     /// Total size of all columns, calculated once in calcuateColumnSizesOnDisk
@@ -455,6 +467,11 @@ protected:
 
     std::map<String, std::shared_ptr<IMergeTreeDataPart>> projection_parts;
 
+    /// Disabled when USE_ROCKSDB is OFF or use_metadata_cache is set to false in merge tree settings
+    bool use_metadata_cache = false;
+
+    mutable PartMetadataManagerPtr metadata_manager;
+
     void removeIfNeeded();
 
     virtual void checkConsistency(bool require_part_metadata) const;
@@ -468,6 +485,9 @@ protected:
 
     std::optional<bool> keepSharedDataInDecoupledStorage() const;
 
+    void initializePartMetadataManager();
+
+
 private:
     /// In compact parts order of columns is necessary
     NameToNumber column_name_to_position;
@@ -478,36 +498,54 @@ private:
     /// Reads part unique identifier (if exists) from uuid.txt
     void loadUUID();
 
+    static void appendFilesOfUUID(Strings & files);
+
     /// Reads columns names and types from columns.txt
     void loadColumns(bool require);
 
+    static void appendFilesOfColumns(Strings & files);
+
     /// If checksums.txt exists, reads file's checksums (and sizes) from it
     void loadChecksums(bool require);
 
+    static void appendFilesOfChecksums(Strings & files);
+
     /// Loads marks index granularity into memory
     virtual void loadIndexGranularity();
 
+    virtual void appendFilesOfIndexGranularity(Strings & files) const;
+
     /// Loads index file.
     void loadIndex();
 
+    void appendFilesOfIndex(Strings & files) const;
+
     /// Load rows count for this part from disk (for the newer storage format version).
     /// For the older format version calculates rows count from the size of a column with a fixed size.
     void loadRowsCount();
 
+    static void appendFilesOfRowsCount(Strings & files);
+
     /// Loads ttl infos in json format from file ttl.txt. If file doesn't exists assigns ttl infos with all zeros
     void loadTTLInfos();
 
+    static void appendFilesOfTTLInfos(Strings & files);
+
     void loadPartitionAndMinMaxIndex();
 
     void calculateColumnsSizesOnDisk();
 
     void calculateSecondaryIndicesSizesOnDisk();
 
+    void appendFilesOfPartitionAndMinMaxIndex(Strings & files) const;
+
     /// Load default compression codec from file default_compression_codec.txt
     /// if it not exists tries to deduce codec from compressed column without
     /// any specifial compression.
     void loadDefaultCompressionCodec();
 
+    static void appendFilesOfDefaultCompressionCodec(Strings & files);
+
     /// Found column without specific compression and return codec
     /// for this column with default parameters.
     CompressionCodecPtr detectDefaultCompressionCodec() const;

src/Storages/MergeTree/IPartMetadataManager.cpp

@@ -0,0 +1,11 @@
+#include "IPartMetadataManager.h"
+
+#include <Disks/IVolume.h>
+#include <Storages/MergeTree/IMergeTreeDataPart.h>
+
+namespace DB
+{
+IPartMetadataManager::IPartMetadataManager(const IMergeTreeDataPart * part_) : part(part_), disk(part->volume->getDisk())
+{
+}
+}

src/Storages/MergeTree/IPartMetadataManager.h

@@ -0,0 +1,59 @@
+#pragma once
+
+#include <unordered_map>
+#include <city.h>
+#include <base/types.h>
+
+namespace DB
+{
+
+class IMergeTreeDataPart;
+
+class SeekableReadBuffer;
+
+class IDisk;
+using DiskPtr = std::shared_ptr<IDisk>;
+
+/// Interface for managing metadata of merge tree part.
+/// IPartMetadataManager has two implementations:
+/// - PartMetadataManagerOrdinary: manage metadata from disk directly. deleteAll/assertAllDeleted/updateAll/check
+///   are all empty implementations because they are not needed for PartMetadataManagerOrdinary(those operations
+///   are done implicitly when removing or renaming part directory).
+/// - PartMetadataManagerWithCache: manage metadata from RocksDB cache and disk.
+class IPartMetadataManager
+{
+public:
+    using uint128 = CityHash_v1_0_2::uint128;
+
+    explicit IPartMetadataManager(const IMergeTreeDataPart * part_);
+
+    virtual ~IPartMetadataManager() = default;
+
+    /// Read metadata content and return SeekableReadBuffer object.
+    virtual std::unique_ptr<SeekableReadBuffer> read(const String & file_name) const = 0;
+
+    /// Return true if metadata exists in part.
+    virtual bool exists(const String & file_name) const = 0;
+
+    /// Delete all metadatas in part.
+    /// If include_projection is true, also delete metadatas in projection parts.
+    virtual void deleteAll(bool include_projection) = 0;
+
+    /// Assert that all metadatas in part are deleted.
+    /// If include_projection is true, also assert that all metadatas in projection parts are deleted.
+    virtual void assertAllDeleted(bool include_projection) const = 0;
+
+    /// Update all metadatas in part.
+    /// If include_projection is true, also update metadatas in projection parts.
+    virtual void updateAll(bool include_projection) = 0;
+
+    /// Check all metadatas in part.
+    virtual std::unordered_map<String, uint128> check() const = 0;
+
+protected:
+    const IMergeTreeDataPart * part;
+    const DiskPtr disk;
+};
+
+using PartMetadataManagerPtr = std::shared_ptr<IPartMetadataManager>;
+}

src/Storages/MergeTree/KeyCondition.cpp

@@ -53,7 +53,7 @@ String Range::toString() const
 
 
 /// Example: for `Hello\_World% ...` string it returns `Hello_World`, and for `%test%` returns an empty string.
-static String extractFixedPrefixFromLikePattern(const String & like_pattern)
+String extractFixedPrefixFromLikePattern(const String & like_pattern)
 {
     String fixed_prefix;
 

src/Storages/MergeTree/KeyCondition.h

@@ -442,4 +442,6 @@ private:
     bool strict;
 };
 
+String extractFixedPrefixFromLikePattern(const String & like_pattern);
+
 }

src/Storages/MergeTree/MergeTreeBlockReadUtils.cpp

@@ -24,7 +24,7 @@ namespace
 /// least one existing (physical) column in part.
 bool injectRequiredColumnsRecursively(
     const String & column_name,
-    const ColumnsDescription & storage_columns,
+    const StorageSnapshotPtr & storage_snapshot,
     const MergeTreeData::AlterConversions & alter_conversions,
     const MergeTreeData::DataPartPtr & part,
     Names & columns,
@@ -36,7 +36,8 @@ bool injectRequiredColumnsRecursively(
     /// stages.
     checkStackSize();
 
-    auto column_in_storage = storage_columns.tryGetColumnOrSubcolumn(GetColumnsOptions::AllPhysical, column_name);
+    auto options = GetColumnsOptions(GetColumnsOptions::AllPhysical).withSubcolumns().withExtendedObjects();
+    auto column_in_storage = storage_snapshot->tryGetColumn(options, column_name);
     if (column_in_storage)
     {
         auto column_name_in_part = column_in_storage->getNameInStorage();
@@ -63,7 +64,8 @@ bool injectRequiredColumnsRecursively(
 
     /// Column doesn't have default value and don't exist in part
     /// don't need to add to required set.
-    const auto column_default = storage_columns.getDefault(column_name);
+    auto metadata_snapshot = storage_snapshot->getMetadataForQuery();
+    const auto column_default = metadata_snapshot->getColumns().getDefault(column_name);
     if (!column_default)
         return false;
 
@@ -73,39 +75,36 @@ bool injectRequiredColumnsRecursively(
 
     bool result = false;
     for (const auto & identifier : identifiers)
-        result |= injectRequiredColumnsRecursively(identifier, storage_columns, alter_conversions, part, columns, required_columns, injected_columns);
+        result |= injectRequiredColumnsRecursively(identifier, storage_snapshot, alter_conversions, part, columns, required_columns, injected_columns);
 
     return result;
 }
 
 }
 
-NameSet injectRequiredColumns(const MergeTreeData & storage, const StorageMetadataPtr & metadata_snapshot, const MergeTreeData::DataPartPtr & part, Names & columns)
+NameSet injectRequiredColumns(
+    const MergeTreeData & storage,
+    const StorageSnapshotPtr & storage_snapshot,
+    const MergeTreeData::DataPartPtr & part,
+    Names & columns)
 {
     NameSet required_columns{std::begin(columns), std::end(columns)};
     NameSet injected_columns;
 
     bool have_at_least_one_physical_column = false;
-
-    const auto & storage_columns = metadata_snapshot->getColumns();
     MergeTreeData::AlterConversions alter_conversions;
     if (!part->isProjectionPart())
         alter_conversions = storage.getAlterConversionsForPart(part);
+
     for (size_t i = 0; i < columns.size(); ++i)
     {
-        auto name_in_storage = Nested::extractTableName(columns[i]);
-        if (storage_columns.has(name_in_storage) && isObject(storage_columns.get(name_in_storage).type))
-        {
-            have_at_least_one_physical_column = true;
-            continue;
-        }
-
         /// We are going to fetch only physical columns
-        if (!storage_columns.hasColumnOrSubcolumn(GetColumnsOptions::AllPhysical, columns[i]))
+        auto options = GetColumnsOptions(GetColumnsOptions::AllPhysical).withSubcolumns().withExtendedObjects();
-            throw Exception("There is no physical column or subcolumn " + columns[i] + " in table.", ErrorCodes::NO_SUCH_COLUMN_IN_TABLE);
+        if (!storage_snapshot->tryGetColumn(options, columns[i]))
+            throw Exception(ErrorCodes::NO_SUCH_COLUMN_IN_TABLE, "There is no physical column or subcolumn {} in table", columns[i]);
 
         have_at_least_one_physical_column |= injectRequiredColumnsRecursively(
-            columns[i], storage_columns, alter_conversions,
+            columns[i], storage_snapshot, alter_conversions,
             part, columns, required_columns, injected_columns);
     }
 
@@ -115,7 +114,7 @@ NameSet injectRequiredColumns(const MergeTreeData & storage, const StorageMetada
         */
     if (!have_at_least_one_physical_column)
     {
-        const auto minimum_size_column_name = part->getColumnNameWithMinimumCompressedSize(metadata_snapshot);
+        const auto minimum_size_column_name = part->getColumnNameWithMinimumCompressedSize(storage_snapshot);
         columns.push_back(minimum_size_column_name);
         /// correctly report added column
         injected_columns.insert(columns.back());
@@ -271,7 +270,7 @@ MergeTreeReadTaskColumns getReadTaskColumns(
     Names pre_column_names;
 
     /// inject columns required for defaults evaluation
-    bool should_reorder = !injectRequiredColumns(storage, storage_snapshot->getMetadataForQuery(), data_part, column_names).empty();
+    bool should_reorder = !injectRequiredColumns(storage, storage_snapshot, data_part, column_names).empty();
 
     if (prewhere_info)
     {
@@ -296,7 +295,7 @@ MergeTreeReadTaskColumns getReadTaskColumns(
         if (pre_column_names.empty())
             pre_column_names.push_back(column_names[0]);
 
-        const auto injected_pre_columns = injectRequiredColumns(storage, storage_snapshot->getMetadataForQuery(), data_part, pre_column_names);
+        const auto injected_pre_columns = injectRequiredColumns(storage, storage_snapshot, data_part, pre_column_names);
         if (!injected_pre_columns.empty())
             should_reorder = true;
 

src/Storages/MergeTree/MergeTreeBlockReadUtils.h

@@ -22,7 +22,7 @@ using MergeTreeBlockSizePredictorPtr = std::shared_ptr<MergeTreeBlockSizePredict
   * so that you can calculate the DEFAULT expression for these columns.
   * Adds them to the `columns`.
   */
-NameSet injectRequiredColumns(const MergeTreeData & storage, const StorageMetadataPtr & metadata_snapshot, const MergeTreeData::DataPartPtr & part, Names & columns);
+NameSet injectRequiredColumns(const MergeTreeData & storage, const StorageSnapshotPtr & storage_snapshot, const MergeTreeData::DataPartPtr & part, Names & columns);
 
 
 /// A batch of work for MergeTreeThreadSelectBlockInputStream

src/Storages/MergeTree/MergeTreeData.cpp

@@ -214,6 +214,7 @@ MergeTreeData::MergeTreeData(
     , parts_mover(this)
     , background_operations_assignee(*this, BackgroundJobsAssignee::Type::DataProcessing, getContext())
     , background_moves_assignee(*this, BackgroundJobsAssignee::Type::Moving, getContext())
+    , use_metadata_cache(getSettings()->use_metadata_cache)
 {
     context_->getGlobalContext()->initializeBackgroundExecutorsIfNeeded();
 
@@ -333,6 +334,11 @@ MergeTreeData::MergeTreeData(
         LOG_WARNING(log, "{} Settings 'min_rows_for_wide_part', 'min_bytes_for_wide_part', "
             "'min_rows_for_compact_part' and 'min_bytes_for_compact_part' will be ignored.", reason);
 
+#if !USE_ROCKSDB
+    if (use_metadata_cache)
+        throw Exception(ErrorCodes::LOGICAL_ERROR, "Can't use merge tree metadata cache if clickhouse was compiled without rocksdb");
+#endif
+
     common_assignee_trigger = [this] (bool delay) noexcept
     {
         if (delay)
@@ -1372,7 +1378,6 @@ void MergeTreeData::loadDataParts(bool skip_sanity_checks)
     LOG_DEBUG(log, "Loaded data parts ({} items)", data_parts_indexes.size());
 }
 
-
 /// Is the part directory old.
 /// True if its modification time and the modification time of all files inside it is less then threshold.
 /// (Only files on the first level of nesting are considered).

src/Storages/MergeTree/MergeTreeData.h

@@ -946,6 +946,7 @@ protected:
     friend class StorageReplicatedMergeTree;
     friend class MergeTreeDataWriter;
     friend class MergeTask;
+    friend class IPartMetadataManager;
 
     bool require_part_metadata;
 
@@ -1028,6 +1029,7 @@ protected:
     /// And for ReplicatedMergeTree we don't have LogEntry type for this operation.
     BackgroundJobsAssignee background_operations_assignee;
     BackgroundJobsAssignee background_moves_assignee;
+    bool use_metadata_cache;
 
     /// Strongly connected with two fields above.
     /// Every task that is finished will ask to assign a new one into an executor.

src/Storages/MergeTree/MergeTreeDataPartWide.h

@@ -69,6 +69,7 @@ private:
     ColumnSize getColumnSizeImpl(const NameAndTypePair & column, std::unordered_set<String> * processed_substreams) const;
 
     void calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size) const override;
+
 };
 
 }

src/Storages/MergeTree/MergeTreeDataSelectExecutor.cpp

@@ -877,12 +877,22 @@ RangesInDataParts MergeTreeDataSelectExecutor::filterPartsByPrimaryKeyAndSkipInd
     {
         std::atomic<size_t> total_rows{0};
 
+        /// Do not check number of read rows if we have reading
+        /// in order of sorting key with limit.
+        /// In general case, when there exists WHERE clause
+        /// it's impossible to estimate number of rows precisely,
+        /// because we can stop reading at any time.
+
         SizeLimits limits;
-        if (settings.read_overflow_mode == OverflowMode::THROW && settings.max_rows_to_read)
+        if (settings.read_overflow_mode == OverflowMode::THROW
+            && settings.max_rows_to_read
+            && !query_info.input_order_info)
             limits = SizeLimits(settings.max_rows_to_read, 0, settings.read_overflow_mode);
 
         SizeLimits leaf_limits;
-        if (settings.read_overflow_mode_leaf == OverflowMode::THROW && settings.max_rows_to_read_leaf)
+        if (settings.read_overflow_mode_leaf == OverflowMode::THROW
+            && settings.max_rows_to_read_leaf
+            && !query_info.input_order_info)
             leaf_limits = SizeLimits(settings.max_rows_to_read_leaf, 0, settings.read_overflow_mode_leaf);
 
         auto mark_cache = context->getIndexMarkCache();

src/Storages/MergeTree/MergeTreeIOSettings.h

@@ -20,6 +20,8 @@ struct MergeTreeReaderSettings
     bool save_marks_in_cache = false;
     /// Validate checksums on reading (should be always enabled in production).
     bool checksum_on_read = true;
+    /// True if we read in order of sorting key.
+    bool read_in_order = false;
 };
 
 struct MergeTreeWriterSettings

src/Storages/MergeTree/MergeTreeMetadataCache.cpp

@@ -0,0 +1,107 @@
+#include "MergeTreeMetadataCache.h"
+
+#if USE_ROCKSDB
+#include <Common/ProfileEvents.h>
+#include <base/logger_useful.h>
+
+namespace ProfileEvents
+{
+    extern const Event MergeTreeMetadataCachePut;
+    extern const Event MergeTreeMetadataCacheGet;
+    extern const Event MergeTreeMetadataCacheDelete;
+    extern const Event MergeTreeMetadataCacheSeek;
+}
+
+namespace DB
+{
+namespace ErrorCodes
+{
+    extern const int SYSTEM_ERROR;
+}
+
+
+std::unique_ptr<MergeTreeMetadataCache> MergeTreeMetadataCache::create(const String & dir, size_t size)
+{
+    assert(size != 0);
+    rocksdb::Options options;
+    rocksdb::BlockBasedTableOptions table_options;
+    rocksdb::DB * db;
+
+    options.create_if_missing = true;
+    auto cache = rocksdb::NewLRUCache(size);
+    table_options.block_cache = cache;
+    options.table_factory.reset(rocksdb::NewBlockBasedTableFactory(table_options));
+    rocksdb::Status status = rocksdb::DB::Open(options, dir, &db);
+    if (status != rocksdb::Status::OK())
+        throw Exception(
+            ErrorCodes::SYSTEM_ERROR,
+            "Fail to open rocksdb path at: {} status:{}. You can try to remove the cache (this will not affect any table data).",
+            dir,
+            status.ToString());
+    return std::make_unique<MergeTreeMetadataCache>(db);
+}
+
+MergeTreeMetadataCache::Status MergeTreeMetadataCache::put(const String & key, const String & value)
+{
+    auto options = rocksdb::WriteOptions();
+    options.sync = true;
+    options.disableWAL = false;
+    auto status = rocksdb->Put(options, key, value);
+    ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCachePut);
+    return status;
+}
+
+MergeTreeMetadataCache::Status MergeTreeMetadataCache::del(const String & key)
+{
+    auto options = rocksdb::WriteOptions();
+    options.sync = true;
+    options.disableWAL = false;
+    auto status = rocksdb->Delete(options, key);
+    ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCacheDelete);
+    LOG_TRACE(log, "Delete key:{} from MergeTreeMetadataCache status:{}", key, status.ToString());
+    return status;
+}
+
+MergeTreeMetadataCache::Status MergeTreeMetadataCache::get(const String & key, String & value)
+{
+    auto status = rocksdb->Get(rocksdb::ReadOptions(), key, &value);
+    ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCacheGet);
+    LOG_TRACE(log, "Get key:{} from MergeTreeMetadataCache status:{}", key, status.ToString());
+    return status;
+}
+
+void MergeTreeMetadataCache::getByPrefix(const String & prefix, Strings & keys, Strings & values)
+{
+    auto * it = rocksdb->NewIterator(rocksdb::ReadOptions());
+    rocksdb::Slice target(prefix);
+    for (it->Seek(target); it->Valid(); it->Next())
+    {
+        const auto key = it->key();
+        if (!key.starts_with(target))
+            break;
+
+        const auto value = it->value();
+        keys.emplace_back(key.data(), key.size());
+        values.emplace_back(value.data(), value.size());
+    }
+    LOG_TRACE(log, "Seek with prefix:{} from MergeTreeMetadataCache items:{}", prefix, keys.size());
+    ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCacheSeek);
+    delete it;
+}
+
+uint64_t MergeTreeMetadataCache::getEstimateNumKeys() const
+{
+    uint64_t keys = 0;
+    rocksdb->GetAggregatedIntProperty("rocksdb.estimate-num-keys", &keys);
+    return keys;
+}
+
+void MergeTreeMetadataCache::shutdown()
+{
+    rocksdb->Close();
+    rocksdb.reset();
+}
+
+}
+
+#endif

src/Storages/MergeTree/MergeTreeMetadataCache.h

@@ -0,0 +1,45 @@
+#pragma once
+
+#include "config_core.h"
+
+#if USE_ROCKSDB
+#include <base/types.h>
+#include <Core/Types.h>
+#include <Poco/Logger.h>
+#include <rocksdb/table.h>
+#include <rocksdb/db.h>
+
+namespace DB
+{
+class MergeTreeMetadataCache
+{
+public:
+    using Status = rocksdb::Status;
+
+    static std::unique_ptr<MergeTreeMetadataCache> create(const String & dir, size_t size);
+
+    explicit MergeTreeMetadataCache(rocksdb::DB * rocksdb_) : rocksdb{rocksdb_}
+    {
+        assert(rocksdb);
+    }
+
+    MergeTreeMetadataCache(const MergeTreeMetadataCache &) = delete;
+
+    MergeTreeMetadataCache & operator=(const MergeTreeMetadataCache &) = delete;
+
+    Status put(const String & key, const String & value);
+    Status del(const String & key);
+    Status get(const String & key, String & value);
+    void getByPrefix(const String & prefix, Strings & keys, Strings & values);
+    uint64_t getEstimateNumKeys() const;
+
+    void shutdown();
+private:
+    std::unique_ptr<rocksdb::DB> rocksdb;
+    Poco::Logger * log = &Poco::Logger::get("MergeTreeMetadataCache");
+};
+
+using MergeTreeMetadataCachePtr = std::shared_ptr<MergeTreeMetadataCache>;
+}
+
+#endif

src/Storages/MergeTree/MergeTreePartition.cpp

@@ -178,12 +178,6 @@ namespace
     };
 }
 
-static std::unique_ptr<ReadBufferFromFileBase> openForReading(const DiskPtr & disk, const String & path)
-{
-    size_t file_size = disk->getFileSize(path);
-    return disk->readFile(path, ReadSettings().adjustBufferSize(file_size), file_size);
-}
-
 String MergeTreePartition::getID(const MergeTreeData & storage) const
 {
     return getID(storage.getInMemoryMetadataPtr()->getPartitionKey().sample_block);
@@ -373,15 +367,15 @@ void MergeTreePartition::serializeText(const MergeTreeData & storage, WriteBuffe
     }
 }
 
-void MergeTreePartition::load(const MergeTreeData & storage, const DiskPtr & disk, const String & part_path)
+void MergeTreePartition::load(const MergeTreeData & storage, const PartMetadataManagerPtr & manager)
 {
     auto metadata_snapshot = storage.getInMemoryMetadataPtr();
     if (!metadata_snapshot->hasPartitionKey())
         return;
 
     const auto & partition_key_sample = adjustPartitionKey(metadata_snapshot, storage.getContext()).sample_block;
-    auto partition_file_path = part_path + "partition.dat";
+
-    auto file = openForReading(disk, partition_file_path);
+    auto file = manager->read("partition.dat");
     value.resize(partition_key_sample.columns());
     for (size_t i = 0; i < partition_key_sample.columns(); ++i)
         partition_key_sample.getByPosition(i).type->getDefaultSerialization()->deserializeBinary(value[i], *file);
@@ -402,7 +396,9 @@ std::unique_ptr<WriteBufferFromFileBase> MergeTreePartition::store(const Block &
     auto out = disk->writeFile(part_path + "partition.dat");
     HashingWriteBuffer out_hashing(*out);
     for (size_t i = 0; i < value.size(); ++i)
+    {
         partition_key_sample.getByPosition(i).type->getDefaultSerialization()->serializeBinary(value[i], out_hashing);
+    }
 
     out_hashing.next();
     checksums.files["partition.dat"].file_size = out_hashing.count();
@@ -462,4 +458,14 @@ KeyDescription MergeTreePartition::adjustPartitionKey(const StorageMetadataPtr &
     return partition_key;
 }
 
+
+void MergeTreePartition::appendFiles(const MergeTreeData & storage, Strings& files)
+{
+    auto metadata_snapshot = storage.getInMemoryMetadataPtr();
+    if (!metadata_snapshot->hasPartitionKey())
+        return;
+
+    files.push_back("partition.dat");
+}
+
 }

src/Storages/MergeTree/MergeTreePartition.h

@@ -4,6 +4,7 @@
 #include <Disks/IDisk.h>
 #include <IO/WriteBuffer.h>
 #include <Storages/KeyDescription.h>
+#include <Storages/MergeTree/IPartMetadataManager.h>
 #include <Core/Field.h>
 
 namespace DB
@@ -37,7 +38,8 @@ public:
 
     void serializeText(const MergeTreeData & storage, WriteBuffer & out, const FormatSettings & format_settings) const;
 
-    void load(const MergeTreeData & storage, const DiskPtr & disk, const String & part_path);
+    void load(const MergeTreeData & storage, const PartMetadataManagerPtr & manager);
+
     /// Store functions return write buffer with written but not finalized data.
     /// User must call finish() for returned object.
     [[nodiscard]] std::unique_ptr<WriteBufferFromFileBase> store(const MergeTreeData & storage, const DiskPtr & disk, const String & part_path, MergeTreeDataPartChecksums & checksums) const;
@@ -47,6 +49,8 @@ public:
 
     void create(const StorageMetadataPtr & metadata_snapshot, Block block, size_t row, ContextPtr context);
 
+    static void appendFiles(const MergeTreeData & storage, Strings & files);
+
     /// Adjust partition key and execute its expression on block. Return sample block according to used expression.
     static NamesAndTypesList executePartitionByExpression(const StorageMetadataPtr & metadata_snapshot, Block & block, ContextPtr context);
 

src/Storages/MergeTree/MergeTreeSelectProcessor.cpp

@@ -39,9 +39,12 @@ MergeTreeSelectProcessor::MergeTreeSelectProcessor(
 {
     /// Actually it means that parallel reading from replicas enabled
     /// and we have to collaborate with initiator.
-    /// In this case we won't set approximate rows, because it will be accounted multiple times
+    /// In this case we won't set approximate rows, because it will be accounted multiple times.
-    if (!extension_.has_value())
+    /// Also do not count amount of read rows if we read in order of sorting key,
+    /// because we don't know actual amount of read rows in case when limit is set.
+    if (!extension_.has_value() && !reader_settings.read_in_order)
         addTotalRowsApprox(total_rows);
+
     ordered_names = header_without_virtual_columns.getNames();
 }
 

src/Storages/MergeTree/MergeTreeSequentialSource.cpp

@@ -41,7 +41,7 @@ MergeTreeSequentialSource::MergeTreeSequentialSource(
     addTotalRowsApprox(data_part->rows_count);
 
     /// Add columns because we don't want to read empty blocks
-    injectRequiredColumns(storage, storage_snapshot->metadata, data_part, columns_to_read);
+    injectRequiredColumns(storage, storage_snapshot, data_part, columns_to_read);
     NamesAndTypesList columns_for_reader;
     if (take_column_types_from_storage)
     {

src/Storages/MergeTree/MergeTreeSettings.h

@@ -139,6 +139,7 @@ struct Settings;
     /** Experimental/work in progress feature. Unsafe for production. */ \
     M(UInt64, part_moves_between_shards_enable, 0, "Experimental/Incomplete feature to move parts between shards. Does not take into account sharding expressions.", 0) \
     M(UInt64, part_moves_between_shards_delay_seconds, 30, "Time to wait before/after moving parts between shards.", 0) \
+    M(Bool, use_metadata_cache, false, "Experimental feature to speed up parts loading process by using MergeTree metadata cache", 0) \
     \
     /** Obsolete settings. Kept for backward compatibility only. */ \
     M(UInt64, min_relative_delay_to_yield_leadership, 120, "Obsolete setting, does nothing.", 0) \

src/Storages/MergeTree/PartMetadataManagerOrdinary.cpp

@@ -0,0 +1,33 @@
+#include "PartMetadataManagerOrdinary.h"
+
+#include <IO/ReadBufferFromFileBase.h>
+#include <Disks/IDisk.h>
+#include <Storages/MergeTree/IMergeTreeDataPart.h>
+
+namespace DB
+{
+
+static std::unique_ptr<ReadBufferFromFileBase> openForReading(const DiskPtr & disk, const String & path)
+{
+    size_t file_size = disk->getFileSize(path);
+    return disk->readFile(path, ReadSettings().adjustBufferSize(file_size), file_size);
+}
+
+PartMetadataManagerOrdinary::PartMetadataManagerOrdinary(const IMergeTreeDataPart * part_) : IPartMetadataManager(part_)
+{
+}
+
+
+std::unique_ptr<SeekableReadBuffer> PartMetadataManagerOrdinary::read(const String & file_name) const
+{
+    String file_path = fs::path(part->getFullRelativePath()) / file_name;
+    return openForReading(disk, file_path);
+}
+
+bool PartMetadataManagerOrdinary::exists(const String & file_name) const
+{
+    return disk->exists(fs::path(part->getFullRelativePath()) / file_name);
+}
+
+
+}

src/Storages/MergeTree/PartMetadataManagerOrdinary.h

@@ -0,0 +1,29 @@
+#pragma once
+
+#include <Storages/MergeTree/IPartMetadataManager.h>
+
+namespace DB
+{
+
+class PartMetadataManagerOrdinary : public IPartMetadataManager
+{
+public:
+    explicit PartMetadataManagerOrdinary(const IMergeTreeDataPart * part_);
+
+    ~PartMetadataManagerOrdinary() override = default;
+
+    std::unique_ptr<SeekableReadBuffer> read(const String & file_name) const override;
+
+    bool exists(const String & file_name) const override;
+
+    void deleteAll(bool /*include_projection*/) override {}
+
+    void assertAllDeleted(bool /*include_projection*/) const override {}
+
+    void updateAll(bool /*include_projection*/) override {}
+
+    std::unordered_map<String, uint128> check() const override { return {}; }
+};
+
+
+}

src/Storages/MergeTree/PartMetadataManagerWithCache.cpp

@@ -0,0 +1,273 @@
+#include "PartMetadataManagerWithCache.h"
+
+#if USE_ROCKSDB
+#include <Common/hex.h>
+#include <Common/ErrorCodes.h>
+#include <IO/HashingReadBuffer.h>
+#include <IO/ReadBufferFromString.h>
+#include <Storages/MergeTree/IMergeTreeDataPart.h>
+
+namespace ProfileEvents
+{
+    extern const Event MergeTreeMetadataCacheHit;
+    extern const Event MergeTreeMetadataCacheMiss;
+}
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int LOGICAL_ERROR;
+    extern const int CORRUPTED_DATA;
+    extern const int NO_SUCH_PROJECTION_IN_TABLE;
+}
+
+PartMetadataManagerWithCache::PartMetadataManagerWithCache(const IMergeTreeDataPart * part_, const MergeTreeMetadataCachePtr & cache_)
+    : IPartMetadataManager(part_), cache(cache_)
+{
+}
+
+String PartMetadataManagerWithCache::getKeyFromFilePath(const String & file_path) const
+{
+    return disk->getName() + ":" + file_path;
+}
+
+String PartMetadataManagerWithCache::getFilePathFromKey(const String & key) const
+{
+    return key.substr(disk->getName().size() + 1);
+}
+
+std::unique_ptr<SeekableReadBuffer> PartMetadataManagerWithCache::read(const String & file_name) const
+{
+    String file_path = fs::path(part->getFullRelativePath()) / file_name;
+    String key = getKeyFromFilePath(file_path);
+    String value;
+    auto status = cache->get(key, value);
+    if (!status.ok())
+    {
+        ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCacheMiss);
+        auto in = disk->readFile(file_path);
+        readStringUntilEOF(value, *in);
+        cache->put(key, value);
+    }
+    else
+    {
+        ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCacheHit);
+    }
+    return std::make_unique<ReadBufferFromOwnString>(value);
+}
+
+bool PartMetadataManagerWithCache::exists(const String & file_name) const
+{
+    String file_path = fs::path(part->getFullRelativePath()) / file_name;
+    String key = getKeyFromFilePath(file_path);
+    String value;
+    auto status = cache->get(key, value);
+    if (status.ok())
+    {
+        ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCacheHit);
+        return true;
+    }
+    else
+    {
+        ProfileEvents::increment(ProfileEvents::MergeTreeMetadataCacheMiss);
+        return disk->exists(fs::path(part->getFullRelativePath()) / file_name);
+    }
+}
+
+void PartMetadataManagerWithCache::deleteAll(bool include_projection)
+{
+    Strings file_names;
+    part->appendFilesOfColumnsChecksumsIndexes(file_names, include_projection);
+
+    String value;
+    for (const auto & file_name : file_names)
+    {
+        String file_path = fs::path(part->getFullRelativePath()) / file_name;
+        String key = getKeyFromFilePath(file_path);
+        auto status = cache->del(key);
+        if (!status.ok())
+        {
+            status = cache->get(key, value);
+            if (status.IsNotFound())
+                continue;
+
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR,
+                "deleteAll failed include_projection:{} status:{}, file_path:{}",
+                include_projection,
+                status.ToString(),
+                file_path);
+        }
+    }
+}
+
+void PartMetadataManagerWithCache::updateAll(bool include_projection)
+{
+    Strings file_names;
+    part->appendFilesOfColumnsChecksumsIndexes(file_names, include_projection);
+
+    String value;
+    String read_value;
+    for (const auto & file_name : file_names)
+    {
+        String file_path = fs::path(part->getFullRelativePath()) / file_name;
+        if (!disk->exists(file_path))
+            continue;
+        auto in = disk->readFile(file_path);
+        readStringUntilEOF(value, *in);
+
+        String key = getKeyFromFilePath(file_path);
+        auto status = cache->put(key, value);
+        if (!status.ok())
+        {
+            status = cache->get(key, read_value);
+            if (status.IsNotFound() || read_value == value)
+                continue;
+
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR,
+                "updateAll failed include_projection:{} status:{}, file_path:{}",
+                include_projection,
+                status.ToString(),
+                file_path);
+        }
+    }
+}
+
+void PartMetadataManagerWithCache::assertAllDeleted(bool include_projection) const
+{
+    Strings keys;
+    std::vector<uint128> _;
+    getKeysAndCheckSums(keys, _);
+    if (keys.empty())
+        return;
+
+    String file_path;
+    String file_name;
+    for (const auto & key : keys)
+    {
+        file_path = getFilePathFromKey(key);
+        file_name = fs::path(file_path).filename();
+
+        /// Metadata file belongs to current part
+        if (fs::path(part->getFullRelativePath()) / file_name == file_path)
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR,
+                "Data part {} with type {} with meta file {} still in cache",
+                part->name,
+                part->getType().toString(),
+                file_path);
+
+        /// File belongs to projection part of current part
+        if (!part->isProjectionPart() && include_projection)
+        {
+            const auto & projection_parts = part->getProjectionParts();
+            for (const auto & [projection_name, projection_part] : projection_parts)
+            {
+                if (fs::path(projection_part->getFullRelativePath()) / file_name == file_path)
+                {
+                    throw Exception(
+                        ErrorCodes::LOGICAL_ERROR,
+                        "Data part {} with type {} with meta file {} with projection name still in cache",
+                        part->name,
+                        part->getType().toString(),
+                        file_path,
+                        projection_name);
+                }
+            }
+        }
+    }
+}
+
+void PartMetadataManagerWithCache::getKeysAndCheckSums(Strings & keys, std::vector<uint128> & checksums) const
+{
+    String prefix = getKeyFromFilePath(fs::path(part->getFullRelativePath()) / "");
+    Strings values;
+    cache->getByPrefix(prefix, keys, values);
+    size_t size = keys.size();
+    for (size_t i = 0; i < size; ++i)
+    {
+        ReadBufferFromString rbuf(values[i]);
+        HashingReadBuffer hbuf(rbuf);
+        checksums.push_back(hbuf.getHash());
+    }
+}
+
+std::unordered_map<String, IPartMetadataManager::uint128> PartMetadataManagerWithCache::check() const
+{
+    /// Only applies for normal part stored on disk
+    if (part->isProjectionPart() || !part->isStoredOnDisk())
+        return {};
+
+    /// The directory of projection part is under the directory of its parent part
+    const auto filenames_without_checksums = part->getFileNamesWithoutChecksums();
+
+    std::unordered_map<String, uint128> results;
+    Strings keys;
+    std::vector<uint128> cache_checksums;
+    std::vector<uint128> disk_checksums;
+    getKeysAndCheckSums(keys, cache_checksums);
+    for (size_t i = 0; i < keys.size(); ++i)
+    {
+        const auto & key = keys[i];
+        String file_path = getFilePathFromKey(key);
+        String file_name = fs::path(file_path).filename();
+        results.emplace(file_name, cache_checksums[i]);
+
+        /// File belongs to normal part
+        if (fs::path(part->getFullRelativePath()) / file_name == file_path)
+        {
+            auto disk_checksum = part->getActualChecksumByFile(file_path);
+            if (disk_checksum != cache_checksums[i])
+                throw Exception(
+                    ErrorCodes::CORRUPTED_DATA,
+                    "Checksums doesn't match in part {}. Expected: {}. Found {}.",
+                    part->name,
+                    getHexUIntUppercase(disk_checksum.first) + getHexUIntUppercase(disk_checksum.second),
+                    getHexUIntUppercase(cache_checksums[i].first) + getHexUIntUppercase(cache_checksums[i].second));
+
+            disk_checksums.push_back(disk_checksum);
+            continue;
+        }
+
+        /// File belongs to projection part
+        String proj_dir_name = fs::path(file_path).parent_path().filename();
+        auto pos = proj_dir_name.find_last_of('.');
+        if (pos == String::npos)
+        {
+            throw Exception(
+                ErrorCodes::NO_SUCH_PROJECTION_IN_TABLE,
+                "There is no projection in part: {} contains file: {} with directory name: {}",
+                part->name,
+                file_path,
+                proj_dir_name);
+        }
+
+        String proj_name = proj_dir_name.substr(0, pos);
+        const auto & projection_parts = part->getProjectionParts();
+        auto it = projection_parts.find(proj_name);
+        if (it == projection_parts.end())
+        {
+            throw Exception(
+                ErrorCodes::NO_SUCH_PROJECTION_IN_TABLE,
+                "There is no projection {} in part: {} contains file: {}",
+                proj_name, part->name, file_path);
+        }
+
+        auto disk_checksum = it->second->getActualChecksumByFile(file_path);
+        if (disk_checksum != cache_checksums[i])
+            throw Exception(
+                ErrorCodes::CORRUPTED_DATA,
+                "Checksums doesn't match in projection part {} {}. Expected: {}. Found {}.",
+                part->name, proj_name,
+                getHexUIntUppercase(disk_checksum.first) + getHexUIntUppercase(disk_checksum.second),
+                getHexUIntUppercase(cache_checksums[i].first) + getHexUIntUppercase(cache_checksums[i].second));
+        disk_checksums.push_back(disk_checksum);
+    }
+    return results;
+}
+
+}
+#endif

src/Storages/MergeTree/PartMetadataManagerWithCache.h

@@ -0,0 +1,56 @@
+#pragma once
+
+#include "config_core.h"
+
+#if USE_ROCKSDB
+#include <Storages/MergeTree/IPartMetadataManager.h>
+#include <Storages/MergeTree/MergeTreeMetadataCache.h>
+
+namespace DB
+{
+
+/// PartMetadataManagerWithCache stores metadatas of part in RocksDB as cache layer to speed up
+/// loading process of merge tree table.
+class PartMetadataManagerWithCache : public IPartMetadataManager
+{
+public:
+    PartMetadataManagerWithCache(const IMergeTreeDataPart * part_, const MergeTreeMetadataCachePtr & cache_);
+
+    ~PartMetadataManagerWithCache() override = default;
+
+    /// First read the metadata from RocksDB cache, then from disk.
+    std::unique_ptr<SeekableReadBuffer> read(const String & file_name) const override;
+
+    /// First judge existence of the metadata in RocksDB cache, then in disk.
+    bool exists(const String & file_name) const override;
+
+    /// Delete all metadatas in part from RocksDB cache.
+    void deleteAll(bool include_projection) override;
+
+    /// Assert all metadatas in part from RocksDB cache are deleted.
+    void assertAllDeleted(bool include_projection) const override;
+
+    /// Update all metadatas in part from RocksDB cache.
+    /// Need to be called after part directory is renamed.
+    void updateAll(bool include_projection) override;
+
+    /// Check if all metadatas in part from RocksDB cache are up to date.
+    std::unordered_map<String, uint128> check() const override;
+
+private:
+    /// Get cache key from path of metadata file.
+    /// Format: <disk_name>:relative/full/path/of/metadata/file
+    String getKeyFromFilePath(const String & file_path) const;
+
+    /// Get metadata file path from cache key.
+    String getFilePathFromKey(const String & key) const;
+
+    /// Get cache keys and checksums of corresponding metadata in a part(including projection parts)
+    void getKeysAndCheckSums(Strings & keys, std::vector<uint128> & checksums) const;
+
+
+    MergeTreeMetadataCachePtr cache;
+};
+
+}
+#endif

src/Storages/MergeTree/ReplicatedMergeTreePartCheckThread.cpp

@@ -399,6 +399,7 @@ CheckResult ReplicatedMergeTreePartCheckThread::checkPart(const String & part_na
         LOG_WARNING(log, "We have part {} covering part {}", part->name, part_name);
     }
 
+    part->checkMetadata();
     return {part_name, true, ""};
 }
 

src/Storages/MergeTree/tests/gtest_merge_tree_metadata_cache.cpp

@@ -0,0 +1,83 @@
+#include "config_core.h"
+
+#if USE_ROCKSDB
+#include <gtest/gtest.h>
+#include <rocksdb/table.h>
+#include <rocksdb/db.h>
+#include <Interpreters/Context.h>
+#include <Storages/MergeTree/MergeTreeMetadataCache.h>
+
+using namespace DB;
+
+class MergeTreeMetadataCacheTest : public ::testing::Test
+{
+public:
+    void SetUp() override
+    {
+        cache = MergeTreeMetadataCache::create("./db/", 268435456);
+    }
+
+    void TearDown() override
+    {
+        cache->shutdown();
+        cache.reset();
+    }
+
+    MergeTreeMetadataCachePtr cache;
+};
+
+TEST_F(MergeTreeMetadataCacheTest, testCommon)
+{
+    std::vector<String> files
+        = {"columns.txt", "checksums.txt", "primary.idx", "count.txt", "partition.dat", "minmax_p.idx", "default_compression_codec.txt"};
+    String prefix = "data/test_metadata_cache/check_part_metadata_cache/201806_1_1_0_4/";
+
+    for (const auto & file : files)
+    {
+        auto status = cache->put(prefix + file, prefix + file);
+        ASSERT_EQ(status.code(), rocksdb::Status::Code::kOk);
+    }
+
+    for (const auto & file : files)
+    {
+        String value;
+        auto status = cache->get(prefix + file, value);
+        ASSERT_EQ(status.code(), rocksdb::Status::Code::kOk);
+        ASSERT_EQ(value, prefix + file);
+    }
+
+    {
+        Strings keys;
+        Strings values;
+        cache->getByPrefix(prefix, keys, values);
+        ASSERT_EQ(keys.size(), files.size());
+        ASSERT_EQ(values.size(), files.size());
+        for (size_t i = 0; i < files.size(); ++i)
+        {
+            ASSERT_EQ(values[i], keys[i]);
+        }
+    }
+
+    for (const auto & file : files)
+    {
+        auto status = cache->del(prefix + file);
+        ASSERT_EQ(status.code(), rocksdb::Status::Code::kOk);
+    }
+
+    for (const auto & file : files)
+    {
+        String value;
+        auto status = cache->get(prefix + file, value);
+        ASSERT_EQ(status.code(), rocksdb::Status::Code::kNotFound);
+    }
+
+    {
+        Strings keys;
+        Strings values;
+        cache->getByPrefix(prefix, keys, values);
+        ASSERT_EQ(keys.size(), 0);
+        ASSERT_EQ(values.size(), 0);
+    }
+}
+
+#endif

src/Storages/StorageMergeTree.cpp

@@ -1616,6 +1616,8 @@ CheckResults StorageMergeTree::checkData(const ASTPtr & query, ContextPtr local_
                 auto out = disk->writeFile(tmp_checksums_path, 4096);
                 part->checksums.write(*out);
                 disk->moveFile(tmp_checksums_path, checksums_path);
+
+                part->checkMetadata();
                 results.emplace_back(part->name, true, "Checksums recounted and written to disk.");
             }
             catch (const Exception & ex)
@@ -1632,6 +1634,7 @@ CheckResults StorageMergeTree::checkData(const ASTPtr & query, ContextPtr local_
             try
             {
                 checkDataPart(part, true);
+                part->checkMetadata();
                 results.emplace_back(part->name, true, "");
             }
             catch (const Exception & ex)

src/Storages/StorageS3.cpp

@@ -1,4 +1,6 @@
 #include <Common/config.h>
+#include "IO/ParallelReadBuffer.h"
+#include "IO/IOThreadPool.h"
 #include "Parsers/ASTCreateQuery.h"
 
 #if USE_AWS_S3
@@ -238,7 +240,8 @@ StorageS3Source::StorageS3Source(
     String compression_hint_,
     const std::shared_ptr<Aws::S3::S3Client> & client_,
     const String & bucket_,
-    std::shared_ptr<IteratorWrapper> file_iterator_)
+    std::shared_ptr<IteratorWrapper> file_iterator_,
+    const size_t download_thread_num_)
     : SourceWithProgress(getHeader(sample_block_, need_path, need_file))
     , WithContext(context_)
     , name(std::move(name_))
@@ -254,6 +257,7 @@ StorageS3Source::StorageS3Source(
     , with_file_column(need_file)
     , with_path_column(need_path)
     , file_iterator(file_iterator_)
+    , download_thread_num(download_thread_num_)
 {
     initialize();
 }
@@ -275,28 +279,79 @@ bool StorageS3Source::initialize()
 
     file_path = fs::path(bucket) / current_key;
 
-    read_buf = wrapReadBufferWithCompressionMethod(
+    read_buf = wrapReadBufferWithCompressionMethod(createS3ReadBuffer(current_key), chooseCompressionMethod(current_key, compression_hint));
-        std::make_unique<ReadBufferFromS3>(client, bucket, current_key, max_single_read_retries, getContext()->getReadSettings()),
+
-        chooseCompressionMethod(current_key, compression_hint));
     auto input_format = getContext()->getInputFormat(format, *read_buf, sample_block, max_block_size, format_settings);
     QueryPipelineBuilder builder;
     builder.init(Pipe(input_format));
 
     if (columns_desc.hasDefaults())
     {
-        builder.addSimpleTransform([&](const Block & header)
+        builder.addSimpleTransform(
-        {
+            [&](const Block & header)
-            return std::make_shared<AddingDefaultsTransform>(header, columns_desc, *input_format, getContext());
+            { return std::make_shared<AddingDefaultsTransform>(header, columns_desc, *input_format, getContext()); });
-        });
     }
 
     pipeline = std::make_unique<QueryPipeline>(QueryPipelineBuilder::getPipeline(std::move(builder)));
     reader = std::make_unique<PullingPipelineExecutor>(*pipeline);
 
-    initialized = false;
     return true;
 }
 
+std::unique_ptr<ReadBuffer> StorageS3Source::createS3ReadBuffer(const String & key)
+{
+    const size_t object_size = DB::S3::getObjectSize(client, bucket, key, false);
+
+    auto download_buffer_size = getContext()->getSettings().max_download_buffer_size;
+    const bool use_parallel_download = download_buffer_size > 0 && download_thread_num > 1;
+    const bool object_too_small = object_size < download_thread_num * download_buffer_size;
+    if (!use_parallel_download || object_too_small)
+    {
+        LOG_TRACE(log, "Downloading object of size {} from S3 in single thread", object_size);
+        return std::make_unique<ReadBufferFromS3>(client, bucket, key, max_single_read_retries, getContext()->getReadSettings());
+    }
+
+    assert(object_size > 0);
+
+    if (download_buffer_size < DBMS_DEFAULT_BUFFER_SIZE)
+    {
+        LOG_WARNING(log, "Downloading buffer {} bytes too small, set at least {} bytes", download_buffer_size, DBMS_DEFAULT_BUFFER_SIZE);
+        download_buffer_size = DBMS_DEFAULT_BUFFER_SIZE;
+    }
+
+    auto factory = std::make_unique<ReadBufferS3Factory>(
+        client, bucket, key, download_buffer_size, object_size, max_single_read_retries, getContext()->getReadSettings());
+    LOG_TRACE(
+        log, "Downloading from S3 in {} threads. Object size: {}, Range size: {}.", download_thread_num, object_size, download_buffer_size);
+
+    ThreadGroupStatusPtr running_group = CurrentThread::isInitialized() && CurrentThread::get().getThreadGroup()
+        ? CurrentThread::get().getThreadGroup()
+        : MainThreadStatus::getInstance().getThreadGroup();
+
+    ContextPtr query_context = CurrentThread::isInitialized() ? CurrentThread::get().getQueryContext() : nullptr;
+
+    auto worker_cleanup = [has_running_group = running_group == nullptr](ThreadStatus & thread_status)
+    {
+        if (has_running_group)
+            thread_status.detachQuery(false);
+    };
+
+    auto worker_setup = [query_context = std::move(query_context),
+                         running_group = std::move(running_group)](ThreadStatus & thread_status)
+    {
+        /// Save query context if any, because cache implementation needs it.
+        if (query_context)
+            thread_status.attachQueryContext(query_context);
+
+        /// To be able to pass ProfileEvents.
+        if (running_group)
+            thread_status.attachQuery(running_group);
+    };
+
+    return std::make_unique<ParallelReadBuffer>(
+        std::move(factory), &IOThreadPool::get(), download_thread_num, std::move(worker_setup), std::move(worker_cleanup));
+}
+
 String StorageS3Source::getName() const
 {
     return name;
@@ -670,6 +725,7 @@ Pipe StorageS3::read(
         block_for_format = storage_snapshot->metadata->getSampleBlock();
     }
 
+    const size_t max_download_threads = local_context->getSettingsRef().max_download_threads;
     for (size_t i = 0; i < num_streams; ++i)
     {
         pipes.emplace_back(std::make_shared<StorageS3Source>(
@@ -686,7 +742,8 @@ Pipe StorageS3::read(
             compression_method,
             client_auth.client,
             client_auth.uri.bucket,
-            iterator_wrapper));
+            iterator_wrapper,
+            max_download_threads));
     }
     auto pipe = Pipe::unitePipes(std::move(pipes));
 

src/Storages/StorageS3.h

@@ -74,7 +74,8 @@ public:
         String compression_hint_,
         const std::shared_ptr<Aws::S3::S3Client> & client_,
         const String & bucket,
-        std::shared_ptr<IteratorWrapper> file_iterator_);
+        std::shared_ptr<IteratorWrapper> file_iterator_,
+        size_t download_thread_num);
 
     String getName() const override;
 
@@ -101,13 +102,17 @@ private:
     std::unique_ptr<PullingPipelineExecutor> reader;
     /// onCancel and generate can be called concurrently
     std::mutex reader_mutex;
-    bool initialized = false;
     bool with_file_column = false;
     bool with_path_column = false;
     std::shared_ptr<IteratorWrapper> file_iterator;
+    size_t download_thread_num = 1;
+
+    Poco::Logger * log = &Poco::Logger::get("StorageS3Source");
 
     /// Recreate ReadBuffer and BlockInputStream for each file.
     bool initialize();
+
+    std::unique_ptr<ReadBuffer> createS3ReadBuffer(const String & key);
 };
 
 /**

src/Storages/StorageSnapshot.cpp

@@ -51,40 +51,42 @@ NamesAndTypesList StorageSnapshot::getColumns(const GetColumnsOptions & options)
 NamesAndTypesList StorageSnapshot::getColumnsByNames(const GetColumnsOptions & options, const Names & names) const
 {
     NamesAndTypesList res;
-    const auto & columns = getMetadataForQuery()->getColumns();
     for (const auto & name : names)
+        res.push_back(getColumn(options, name));
+    return res;
+}
+
+std::optional<NameAndTypePair> StorageSnapshot::tryGetColumn(const GetColumnsOptions & options, const String & column_name) const
+{
+    const auto & columns = getMetadataForQuery()->getColumns();
+    auto column = columns.tryGetColumn(options, column_name);
+    if (column && (!isObject(column->type) || !options.with_extended_objects))
+        return column;
+
+    if (options.with_extended_objects)
     {
-        auto column = columns.tryGetColumn(options, name);
+        auto object_column = object_columns.tryGetColumn(options, column_name);
-        if (column && !isObject(column->type))
+        if (object_column)
-        {
+            return object_column;
-            res.emplace_back(std::move(*column));
-            continue;
-        }
-
-        if (options.with_extended_objects)
-        {
-            auto object_column = object_columns.tryGetColumn(options, name);
-            if (object_column)
-            {
-                res.emplace_back(std::move(*object_column));
-                continue;
-            }
-        }
-
-        if (options.with_virtuals)
-        {
-            auto it = virtual_columns.find(name);
-            if (it != virtual_columns.end())
-            {
-                res.emplace_back(name, it->second);
-                continue;
-            }
-        }
-
-        throw Exception(ErrorCodes::NO_SUCH_COLUMN_IN_TABLE, "There is no column {} in table", name);
     }
 
-    return res;
+    if (options.with_virtuals)
+    {
+        auto it = virtual_columns.find(column_name);
+        if (it != virtual_columns.end())
+            return NameAndTypePair(column_name, it->second);
+    }
+
+    return {};
+}
+
+NameAndTypePair StorageSnapshot::getColumn(const GetColumnsOptions & options, const String & column_name) const
+{
+    auto column = tryGetColumn(options, column_name);
+    if (!column)
+        throw Exception(ErrorCodes::NO_SUCH_COLUMN_IN_TABLE, "There is no column {} in table", column_name);
+
+    return *column;
 }
 
 Block StorageSnapshot::getSampleBlockForColumns(const Names & column_names) const

src/Storages/StorageSnapshot.h

@@ -61,6 +61,10 @@ struct StorageSnapshot
     /// Get columns with types according to options only for requested names.
     NamesAndTypesList getColumnsByNames(const GetColumnsOptions & options, const Names & names) const;
 
+    /// Get column with type according to options for requested name.
+    std::optional<NameAndTypePair> tryGetColumn(const GetColumnsOptions & options, const String & column_name) const;
+    NameAndTypePair getColumn(const GetColumnsOptions & options, const String & column_name) const;
+
     /// Block with ordinary + materialized + aliases + virtuals + subcolumns.
     Block getSampleBlockForColumns(const Names & column_names) const;
 

src/Storages/StorageURL.cpp

@@ -560,6 +560,8 @@ ColumnsDescription IStorageURLBase::getTableStructureFromData(
     const std::optional<FormatSettings> & format_settings,
     ContextPtr context)
 {
+    context->getRemoteHostFilter().checkURL(Poco::URI(uri));
+
     Poco::Net::HTTPBasicCredentials credentials;
 
     std::vector<String> urls_to_check;

src/Storages/System/StorageSystemMergeTreeMetadataCache.cpp

@@ -0,0 +1,143 @@
+#include <Storages/System/StorageSystemMergeTreeMetadataCache.h>
+
+#if USE_ROCKSDB
+#include <DataTypes/DataTypeDateTime.h>
+#include <DataTypes/DataTypeString.h>
+#include <DataTypes/DataTypesNumber.h>
+#include <Interpreters/Context.h>
+#include <Parsers/ASTExpressionList.h>
+#include <Parsers/ASTFunction.h>
+#include <Parsers/ASTIdentifier.h>
+#include <Parsers/ASTLiteral.h>
+#include <Parsers/ASTSelectQuery.h>
+#include <Storages/MergeTree/KeyCondition.h>
+#include <Storages/MergeTree/MergeTreeMetadataCache.h>
+#include <Common/typeid_cast.h>
+
+namespace DB
+{
+namespace ErrorCodes
+{
+    extern const int BAD_ARGUMENTS;
+}
+
+NamesAndTypesList StorageSystemMergeTreeMetadataCache::getNamesAndTypes()
+{
+    return {
+        {"key", std::make_shared<DataTypeString>()},
+        {"value", std::make_shared<DataTypeString>()},
+    };
+}
+
+static bool extractKeyImpl(const IAST & elem, String & res, bool & precise)
+{
+    const auto * function = elem.as<ASTFunction>();
+    if (!function)
+        return false;
+
+    if (function->name == "and")
+    {
+        for (const auto & child : function->arguments->children)
+        {
+            bool tmp_precise = false;
+            if (extractKeyImpl(*child, res, tmp_precise))
+            {
+                precise = tmp_precise;
+                return true;
+            }
+        }
+        return false;
+    }
+
+    if (function->name == "equals" || function->name == "like")
+    {
+        const auto & args = function->arguments->as<ASTExpressionList &>();
+        const IAST * value;
+
+        if (args.children.size() != 2)
+            return false;
+
+        const ASTIdentifier * ident;
+        if ((ident = args.children.at(0)->as<ASTIdentifier>()))
+            value = args.children.at(1).get();
+        else if ((ident = args.children.at(1)->as<ASTIdentifier>()))
+            value = args.children.at(0).get();
+        else
+            return false;
+
+        if (ident->name() != "key")
+            return false;
+
+        const auto * literal = value->as<ASTLiteral>();
+        if (!literal)
+            return false;
+
+        if (literal->value.getType() != Field::Types::String)
+            return false;
+
+        res = literal->value.safeGet<String>();
+        precise = function->name == "equals";
+        return true;
+    }
+    return false;
+}
+
+
+/// Retrieve from the query a condition of the form `key= 'key'`, from conjunctions in the WHERE clause.
+static String extractKey(const ASTPtr & query, bool& precise)
+{
+    const auto & select = query->as<ASTSelectQuery &>();
+    if (!select.where())
+        return "";
+
+    String res;
+    return extractKeyImpl(*select.where(), res, precise) ? res : "";
+}
+
+
+void StorageSystemMergeTreeMetadataCache::fillData(MutableColumns & res_columns, ContextPtr context, const SelectQueryInfo & query_info) const
+{
+    bool precise = false;
+    String key = extractKey(query_info.query, precise);
+    if (key.empty())
+        throw Exception(
+            "SELECT from system.merge_tree_metadata_cache table must contain condition like key = 'key' or key LIKE 'prefix%' in WHERE clause.", ErrorCodes::BAD_ARGUMENTS);
+
+    auto cache = context->getMergeTreeMetadataCache();
+    if (precise)
+    {
+        String value;
+        if (cache->get(key, value) != MergeTreeMetadataCache::Status::OK())
+            return;
+
+        size_t col_num = 0;
+        res_columns[col_num++]->insert(key);
+        res_columns[col_num++]->insert(value);
+    }
+    else
+    {
+        String target = extractFixedPrefixFromLikePattern(key);
+        if (target.empty())
+            throw Exception(
+                "SELECT from system.merge_tree_metadata_cache table must contain condition like key = 'key' or key LIKE 'prefix%' in WHERE clause.", ErrorCodes::BAD_ARGUMENTS);
+
+        Strings keys;
+        Strings values;
+        keys.reserve(4096);
+        values.reserve(4096);
+        cache->getByPrefix(target, keys, values);
+        if (keys.empty())
+            return;
+
+        assert(keys.size() == values.size());
+        for (size_t i = 0; i < keys.size(); ++i)
+        {
+            size_t col_num = 0;
+            res_columns[col_num++]->insert(keys[i]);
+            res_columns[col_num++]->insert(values[i]);
+        }
+    }
+}
+
+}
+#endif

src/Storages/System/StorageSystemMergeTreeMetadataCache.h

@@ -0,0 +1,32 @@
+#pragma once
+
+#include "config_core.h"
+
+#if USE_ROCKSDB
+#include <base/shared_ptr_helper.h>
+#include <Storages/System/IStorageSystemOneBlock.h>
+
+
+namespace DB
+{
+class Context;
+
+
+/// Implements `merge_tree_metadata_cache` system table, which allows you to view the metadata cache data in rocksdb for testing purposes.
+class StorageSystemMergeTreeMetadataCache : public shared_ptr_helper<StorageSystemMergeTreeMetadataCache>, public IStorageSystemOneBlock<StorageSystemMergeTreeMetadataCache>
+{
+    friend struct shared_ptr_helper<StorageSystemMergeTreeMetadataCache>;
+
+public:
+    std::string getName() const override { return "SystemMergeTreeMetadataCache"; }
+
+    static NamesAndTypesList getNamesAndTypes();
+
+protected:
+    using IStorageSystemOneBlock::IStorageSystemOneBlock;
+
+    void fillData(MutableColumns & res_columns, ContextPtr context, const SelectQueryInfo & query_info) const override;
+};
+
+}
+#endif

src/Storages/System/attachSystemTables.cpp

@@ -75,6 +75,7 @@
 
 #if USE_ROCKSDB
 #include <Storages/RocksDB/StorageSystemRocksDB.h>
+#include <Storages/System/StorageSystemMergeTreeMetadataCache.h>
 #endif
 
 
@@ -129,6 +130,7 @@ void attachSystemTablesLocal(ContextPtr context, IDatabase & system_database)
 #endif
 #if USE_ROCKSDB
     attach<StorageSystemRocksDB>(context, system_database, "rocksdb");
+    attach<StorageSystemMergeTreeMetadataCache>(context, system_database, "merge_tree_metadata_cache");
 #endif
 }
 

tests/ci/cherry_pick_utils/parser.py

@@ -20,8 +20,6 @@ class Description:
 
     def __init__(self, pull_request):
         self.label_name = str()
-        self.legal = False
-
         self._parse(pull_request["bodyText"])
 
     def _parse(self, text):
@@ -39,12 +37,6 @@ class Description:
                 category = stripped
                 next_category = False
 
-            if (
-                stripped
-                == "I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en"
-            ):
-                self.legal = True
-
             category_headers = (
                 "Category (leave one):",
                 "Changelog category (leave one):",

tests/ci/clickhouse_helper.py

@@ -10,13 +10,13 @@ from get_robot_token import get_parameter_from_ssm
 class ClickHouseHelper:
     def __init__(self, url=None):
         if url is None:
-            self.url = get_parameter_from_ssm("clickhouse-test-stat-url2")
+            url = get_parameter_from_ssm("clickhouse-test-stat-url")
-            self.auth = {
+
-                "X-ClickHouse-User": get_parameter_from_ssm(
+        self.url = url
-                    "clickhouse-test-stat-login2"
+        self.auth = {
-                ),
+            "X-ClickHouse-User": get_parameter_from_ssm("clickhouse-test-stat-login"),
-                "X-ClickHouse-Key": "",
+            "X-ClickHouse-Key": get_parameter_from_ssm("clickhouse-test-stat-password"),
-            }
+        }
 
     @staticmethod
     def _insert_json_str_info_impl(url, auth, db, table, json_str):
@@ -179,7 +179,7 @@ def mark_flaky_tests(clickhouse_helper, check_name, test_results):
             check_name=check_name
         )
 
-        tests_data = clickhouse_helper.select_json_each_row("gh-data", query)
+        tests_data = clickhouse_helper.select_json_each_row("default", query)
         master_failed_tests = {row["test_name"] for row in tests_data}
         logging.info("Found flaky tests: %s", ", ".join(master_failed_tests))
 

tests/ci/commit_status_helper.py

@@ -59,3 +59,17 @@ def post_commit_status_to_file(file_path, description, state, report_url):
     with open(file_path, "w", encoding="utf-8") as f:
         out = csv.writer(f, delimiter="\t")
         out.writerow([state, report_url, description])
+
+
+def remove_labels(gh, pr_info, labels_names):
+    repo = gh.get_repo(GITHUB_REPOSITORY)
+    pull_request = repo.get_pull(pr_info.number)
+    for label in labels_names:
+        pull_request.remove_from_labels(label)
+
+
+def post_labels(gh, pr_info, labels_names):
+    repo = gh.get_repo(GITHUB_REPOSITORY)
+    pull_request = repo.get_pull(pr_info.number)
+    for label in labels_names:
+        pull_request.add_to_labels(label)

tests/ci/compatibility_check.py

@@ -197,4 +197,4 @@ if __name__ == "__main__":
         report_url,
         CHECK_NAME,
     )
-    ch_helper.insert_events_into(db="gh-data", table="checks", events=prepared_events)
+    ch_helper.insert_events_into(db="default", table="checks", events=prepared_events)

tests/ci/docker_images_check.py

@@ -459,7 +459,7 @@ def main():
         NAME,
     )
     ch_helper = ClickHouseHelper()
-    ch_helper.insert_events_into(db="gh-data", table="checks", events=prepared_events)
+    ch_helper.insert_events_into(db="default", table="checks", events=prepared_events)
 
 
 if __name__ == "__main__":

tests/ci/docker_manifests_merge.py

@@ -234,7 +234,7 @@ def main():
         NAME,
     )
     ch_helper = ClickHouseHelper()
-    ch_helper.insert_events_into(db="gh-data", table="checks", events=prepared_events)
+    ch_helper.insert_events_into(db="default", table="checks", events=prepared_events)
 
 
 if __name__ == "__main__":

tests/ci/docs_check.py

@@ -114,4 +114,4 @@ if __name__ == "__main__":
         report_url,
         NAME,
     )
-    ch_helper.insert_events_into(db="gh-data", table="checks", events=prepared_events)
+    ch_helper.insert_events_into(db="default", table="checks", events=prepared_events)

tests/ci/fast_test_check.py

@@ -204,7 +204,7 @@ if __name__ == "__main__":
         report_url,
         NAME,
     )
-    ch_helper.insert_events_into(db="gh-data", table="checks", events=prepared_events)
+    ch_helper.insert_events_into(db="default", table="checks", events=prepared_events)
 
     # Refuse other checks to run if fast test failed
     if state != "success":

tests/ci/functional_test_check.py

@@ -356,7 +356,7 @@ if __name__ == "__main__":
         report_url,
         check_name_with_group,
     )
-    ch_helper.insert_events_into(db="gh-data", table="checks", events=prepared_events)
+    ch_helper.insert_events_into(db="default", table="checks", events=prepared_events)
 
     if state != "success":
         if "force-tests" in pr_info.labels: