Update to ClickHouse v24.8.4.13-lts

Author: auxten

base/base/BorrowedObjectPool.h

@@ -86,7 +86,7 @@ class BorrowedObjectPool final
     }
 
     /// Return object into pool. Client must return same object that was borrowed.
-    inline void returnObject(T && object_to_return)
+    void returnObject(T && object_to_return)
     {
         {
             std::lock_guard lock(objects_mutex);
@@ -99,28 +99,28 @@ class BorrowedObjectPool final
     }
 
     /// Max pool size
-    inline size_t maxSize() const
+    size_t maxSize() const
     {
         return max_size;
     }
 
     /// Allocated objects size by the pool. If allocatedObjectsSize == maxSize then pool is full.
-    inline size_t allocatedObjectsSize() const
+    size_t allocatedObjectsSize() const
     {
         std::lock_guard lock(objects_mutex);
         return allocated_objects_size;
     }
 
     /// Returns allocatedObjectsSize == maxSize
-    inline bool isFull() const
+    bool isFull() const
     {
         std::lock_guard lock(objects_mutex);
         return allocated_objects_size == max_size;
     }
 
     /// Borrowed objects size. If borrowedObjectsSize == allocatedObjectsSize and pool is full.
     /// Then client will wait during borrowObject function call.
-    inline size_t borrowedObjectsSize() const
+    size_t borrowedObjectsSize() const
     {
         std::lock_guard lock(objects_mutex);
         return borrowed_objects_size;
@@ -129,15 +129,15 @@ class BorrowedObjectPool final
 private:
 
     template <typename FactoryFunc>
-    inline T allocateObjectForBorrowing(const std::unique_lock<std::mutex> &, FactoryFunc && func)
+    T allocateObjectForBorrowing(const std::unique_lock<std::mutex> &, FactoryFunc && func)
     {
         ++allocated_objects_size;
         ++borrowed_objects_size;
 
         return std::forward<FactoryFunc>(func)();
     }
 
-    inline T borrowFromObjects(const std::unique_lock<std::mutex> &)
+    T borrowFromObjects(const std::unique_lock<std::mutex> &)
     {
         T dst;
         detail::moveOrCopyIfThrow(std::move(objects.back()), dst);

base/base/CMakeLists.txt

@@ -1,4 +1,4 @@
-add_compile_options($<$<OR:$<COMPILE_LANGUAGE:C>,$<COMPILE_LANGUAGE:CXX>>:${COVERAGE_FLAGS}>)
+add_compile_options("$<$<OR:$<COMPILE_LANGUAGE:C>,$<COMPILE_LANGUAGE:CXX>>:${COVERAGE_FLAGS}>")
 
 if (USE_CLANG_TIDY)
     set (CMAKE_CXX_CLANG_TIDY "${CLANG_TIDY_PATH}")
@@ -32,17 +32,9 @@ set (SRCS
     StringRef.cpp
     safeExit.cpp
     throwError.cpp
+    Numa.cpp
 )
 
-if (USE_DEBUG_HELPERS)
-    get_target_property(MAGIC_ENUM_INCLUDE_DIR ch_contrib::magic_enum INTERFACE_INCLUDE_DIRECTORIES)
-    # CMake generator expression will do insane quoting when it encounters special character like quotes, spaces, etc.
-    # Prefixing "SHELL:" will force it to use the original text.
-    set (INCLUDE_DEBUG_HELPERS "SHELL:-I\"${MAGIC_ENUM_INCLUDE_DIR}\" -include \"${ClickHouse_SOURCE_DIR}/base/base/iostream_debug_helpers.h\"")
-    # Use generator expression as we don't want to pollute CMAKE_CXX_FLAGS, which will interfere with CMake check system.
-    add_compile_options($<$<COMPILE_LANGUAGE:CXX>:${INCLUDE_DEBUG_HELPERS}>)
-endif ()
-
 add_library (common ${SRCS})
 
 if (WITH_COVERAGE)
@@ -55,6 +47,10 @@ if (TARGET ch_contrib::crc32_s390x)
     target_link_libraries(common PUBLIC ch_contrib::crc32_s390x)
 endif()
 
+if (TARGET ch_contrib::numactl)
+    target_link_libraries(common PUBLIC ch_contrib::numactl)
+endif()
+
 target_include_directories(common PUBLIC .. "${CMAKE_CURRENT_BINARY_DIR}/..")
 
 target_link_libraries (common

base/base/Decimal_fwd.h

@@ -44,6 +44,10 @@ concept is_over_big_int =
     || std::is_same_v<T, UInt256>
     || std::is_same_v<T, Decimal128>
     || std::is_same_v<T, Decimal256>;
+
+template <class T>
+concept is_over_big_decimal = is_decimal<T> && is_over_big_int<typename T::NativeType>;
+
 }
 
 template <> struct is_signed<DB::Decimal32> { static constexpr bool value = true; };

base/base/EnumReflection.h

@@ -32,7 +32,7 @@ constexpr void static_for(F && f)
 template <is_enum T>
 struct fmt::formatter<T> : fmt::formatter<std::string_view>
 {
-    constexpr auto format(T value, auto& format_context)
+    constexpr auto format(T value, auto& format_context) const
     {
         return formatter<string_view>::format(magic_enum::enum_name(value), format_context);
     }

base/base/Numa.cpp

@@ -0,0 +1,37 @@
+#include <base/Numa.h>
+
+#include "config.h"
+
+#if USE_NUMACTL
+#    include <numa.h>
+#endif
+
+namespace DB
+{
+
+std::optional<size_t> getNumaNodesTotalMemory()
+{
+    std::optional<size_t> total_memory;
+#if USE_NUMACTL
+    if (numa_available() != -1)
+    {
+        auto * membind = numa_get_membind();
+        if (!numa_bitmask_equal(membind, numa_all_nodes_ptr))
+        {
+            total_memory.emplace(0);
+            auto max_node = numa_max_node();
+            for (int i = 0; i <= max_node; ++i)
+            {
+                if (numa_bitmask_isbitset(membind, i))
+                    *total_memory += numa_node_size(i, nullptr);
+            }
+        }
+
+        numa_bitmask_free(membind);
+    }
+
+#endif
+    return total_memory;
+}
+
+}

base/base/Numa.h

@@ -0,0 +1,12 @@
+#pragma once
+
+#include <optional>
+
+namespace DB
+{
+
+/// return total memory of NUMA nodes the process is bound to
+/// if NUMA is not supported or process can use all nodes, std::nullopt is returned
+std::optional<size_t> getNumaNodesTotalMemory();
+
+}

base/base/StringRef.h

@@ -12,6 +12,8 @@
 #include <base/types.h>
 #include <base/unaligned.h>
 #include <base/simd.h>
+#include <fmt/core.h>
+#include <fmt/ostream.h>
 
 #include <city.h>
 
@@ -376,3 +378,5 @@ namespace PackedZeroTraits
 
 
 std::ostream & operator<<(std::ostream & os, const StringRef & str);
+
+template<> struct fmt::formatter<StringRef> : fmt::ostream_formatter {};

base/base/cgroupsv2.cpp

@@ -3,8 +3,9 @@
 #include <base/defines.h>
 
 #include <fstream>
-#include <sstream>
+#include <string>
 
+namespace fs = std::filesystem;
 
 bool cgroupsV2Enabled()
 {
@@ -13,11 +14,11 @@ bool cgroupsV2Enabled()
     {
         /// This file exists iff the host has cgroups v2 enabled.
         auto controllers_file = default_cgroups_mount / "cgroup.controllers";
-        if (!std::filesystem::exists(controllers_file))
+        if (!fs::exists(controllers_file))
             return false;
         return true;
     }
-    catch (const std::filesystem::filesystem_error &) /// all "underlying OS API errors", typically: permission denied
+    catch (const fs::filesystem_error &) /// all "underlying OS API errors", typically: permission denied
     {
         return false; /// not logging the exception as most callers fall back to cgroups v1
     }
@@ -33,8 +34,9 @@ bool cgroupsV2MemoryControllerEnabled()
     /// According to https://docs.kernel.org/admin-guide/cgroup-v2.html, file "cgroup.controllers" defines which controllers are available
     /// for the current + child cgroups. The set of available controllers can be restricted from level to level using file
     /// "cgroups.subtree_control". It is therefore sufficient to check the bottom-most nested "cgroup.controllers" file.
-    std::string cgroup = cgroupV2OfProcess();
-    auto cgroup_dir = cgroup.empty() ? default_cgroups_mount : (default_cgroups_mount / cgroup);
+    fs::path cgroup_dir = cgroupV2PathOfProcess();
+    if (cgroup_dir.empty())
+        return false;
     std::ifstream controllers_file(cgroup_dir / "cgroup.controllers");
     if (!controllers_file.is_open())
         return false;
@@ -46,25 +48,26 @@ bool cgroupsV2MemoryControllerEnabled()
 #endif
 }
 
-std::string cgroupV2OfProcess()
+fs::path cgroupV2PathOfProcess()
 {
 #if defined(OS_LINUX)
     chassert(cgroupsV2Enabled());
     /// All PIDs assigned to a cgroup are in /sys/fs/cgroups/{cgroup_name}/cgroup.procs
     /// A simpler way to get the membership is:
     std::ifstream cgroup_name_file("/proc/self/cgroup");
     if (!cgroup_name_file.is_open())
-        return "";
+        return {};
     /// With cgroups v2, there will be a *single* line with prefix "0::/"
     /// (see https://docs.kernel.org/admin-guide/cgroup-v2.html)
     std::string cgroup;
     std::getline(cgroup_name_file, cgroup);
     static const std::string v2_prefix = "0::/";
     if (!cgroup.starts_with(v2_prefix))
-        return "";
+        return {};
     cgroup = cgroup.substr(v2_prefix.length());
-    return cgroup;
+    /// Note: The 'root' cgroup can have an empty cgroup name, this is valid
+    return default_cgroups_mount / cgroup;
 #else
-    return "";
+    return {};
 #endif
 }

base/base/cgroupsv2.h

@@ -1,7 +1,6 @@
 #pragma once
 
 #include <filesystem>
-#include <string>
 
 #if defined(OS_LINUX)
 /// I think it is possible to mount the cgroups hierarchy somewhere else (e.g. when in containers).
@@ -16,7 +15,7 @@ bool cgroupsV2Enabled();
 /// Assumes that cgroupsV2Enabled() is enabled.
 bool cgroupsV2MemoryControllerEnabled();
 
-/// Which cgroup does the process belong to?
-/// Returns an empty string if the cgroup cannot be determined.
+/// Detects which cgroup v2 the process belongs to and returns the filesystem path to the cgroup.
+/// Returns an empty path the cgroup cannot be determined.
 /// Assumes that cgroupsV2Enabled() is enabled.
-std::string cgroupV2OfProcess();
+std::filesystem::path cgroupV2PathOfProcess();

base/base/defines.h

@@ -87,10 +87,13 @@
 #   define ASAN_POISON_MEMORY_REGION(a, b)
 #endif
 
-#if !defined(ABORT_ON_LOGICAL_ERROR)
-    #if !defined(NDEBUG) || defined(ADDRESS_SANITIZER) || defined(THREAD_SANITIZER) || defined(MEMORY_SANITIZER) || defined(UNDEFINED_BEHAVIOR_SANITIZER)
-        #define ABORT_ON_LOGICAL_ERROR
-    #endif
+/// We used to have only ABORT_ON_LOGICAL_ERROR macro, but most of its uses were actually in places where we didn't care about logical errors
+/// but wanted to check exactly if the current build type is debug or with sanitizer. This new macro is introduced to fix those places.
+#if !defined(DEBUG_OR_SANITIZER_BUILD)
+#    if !defined(NDEBUG) || defined(ADDRESS_SANITIZER) || defined(THREAD_SANITIZER) || defined(MEMORY_SANITIZER) \
+        || defined(UNDEFINED_BEHAVIOR_SANITIZER)
+#        define DEBUG_OR_SANITIZER_BUILD
+#    endif
 #endif
 
 /// chassert(x) is similar to assert(x), but:
@@ -101,7 +104,7 @@
 /// Also it makes sense to call abort() instead of __builtin_unreachable() in debug builds,
 /// because SIGABRT is easier to debug than SIGTRAP (the second one makes gdb crazy)
 #if !defined(chassert)
-    #if defined(ABORT_ON_LOGICAL_ERROR)
+#    if defined(DEBUG_OR_SANITIZER_BUILD)
         // clang-format off
         #include <base/types.h>
         namespace DB

base/base/demangle.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <cstdlib>
 #include <memory>
 #include <string>
 

base/base/extended_types.h

@@ -108,6 +108,14 @@ struct make_unsigned // NOLINT(readability-identifier-naming)
     using type = std::make_unsigned_t<T>;
 };
 
+template <> struct make_unsigned<Int8> { using type = UInt8; };
+template <> struct make_unsigned<UInt8> { using type = UInt8; };
+template <> struct make_unsigned<Int16> { using type = UInt16; };
+template <> struct make_unsigned<UInt16> { using type = UInt16; };
+template <> struct make_unsigned<Int32> { using type = UInt32; };
+template <> struct make_unsigned<UInt32> { using type = UInt32; };
+template <> struct make_unsigned<Int64> { using type = UInt64; };
+template <> struct make_unsigned<UInt64> { using type = UInt64; };
 template <> struct make_unsigned<Int128> { using type = UInt128; };
 template <> struct make_unsigned<UInt128> { using type = UInt128; };
 template <> struct make_unsigned<Int256>  { using type = UInt256; };
@@ -121,6 +129,14 @@ struct make_signed // NOLINT(readability-identifier-naming)
     using type = std::make_signed_t<T>;
 };
 
+template <> struct make_signed<Int8> { using type = Int8; };
+template <> struct make_signed<UInt8> { using type = Int8; };
+template <> struct make_signed<Int16> { using type = Int16; };
+template <> struct make_signed<UInt16> { using type = Int16; };
+template <> struct make_signed<Int32> { using type = Int32; };
+template <> struct make_signed<UInt32> { using type = Int32; };
+template <> struct make_signed<Int64> { using type = Int64; };
+template <> struct make_signed<UInt64> { using type = Int64; };
 template <> struct make_signed<Int128>  { using type = Int128; };
 template <> struct make_signed<UInt128> { using type = Int128; };
 template <> struct make_signed<Int256>  { using type = Int256; };

base/base/getFQDNOrHostName.cpp

@@ -6,6 +6,9 @@ namespace
 {
     std::string getFQDNOrHostNameImpl()
     {
+#if defined(OS_DARWIN)
+        return Poco::Net::DNS::hostName();
+#else
         try
         {
             return Poco::Net::DNS::thisHost().name();
@@ -14,6 +17,7 @@ namespace
         {
             return Poco::Net::DNS::hostName();
         }
+#endif
     }
 }
 

base/base/getMemoryAmount.cpp

@@ -2,15 +2,14 @@
 
 #include <base/cgroupsv2.h>
 #include <base/getPageSize.h>
+#include <base/Numa.h>
 
 #include <fstream>
-#include <stdexcept>
 
 #include <unistd.h>
 #include <sys/types.h>
 #include <sys/param.h>
 
-
 namespace
 {
 
@@ -23,8 +22,9 @@ std::optional<uint64_t> getCgroupsV2MemoryLimit()
     if (!cgroupsV2MemoryControllerEnabled())
         return {};
 
-    std::string cgroup = cgroupV2OfProcess();
-    auto current_cgroup = cgroup.empty() ? default_cgroups_mount : (default_cgroups_mount / cgroup);
+    std::filesystem::path current_cgroup = cgroupV2PathOfProcess();
+    if (current_cgroup.empty())
+        return {};
 
     /// Open the bottom-most nested memory limit setting file. If there is no such file at the current
     /// level, try again at the parent level as memory settings are inherited.
@@ -62,6 +62,9 @@ uint64_t getMemoryAmountOrZero()
 
     uint64_t memory_amount = num_pages * page_size;
 
+    if (auto total_numa_memory = DB::getNumaNodesTotalMemory(); total_numa_memory.has_value())
+        memory_amount = *total_numa_memory;
+
     /// Respect the memory limit set by cgroups v2.
     auto limit_v2 = getCgroupsV2MemoryLimit();
     if (limit_v2.has_value() && *limit_v2 < memory_amount)