Merge pull request #16527 from kitaisreal/mmap-read-buffer-from-file-descriptor-use-getpagesize

MMapReadBufferFromFileDescriptor use getpagesize

base/common/CMakeLists.txt

@@ -6,6 +6,7 @@ set (SRCS
     demangle.cpp
     getFQDNOrHostName.cpp
     getMemoryAmount.cpp
+    getPageSize.cpp
     getThreadId.cpp
     JSON.cpp
     LineReader.cpp

base/common/getMemoryAmount.cpp

@@ -1,5 +1,6 @@
 #include <stdexcept>
 #include "common/getMemoryAmount.h"
+#include "common/getPageSize.h"
 
 #include <unistd.h>
 #include <sys/types.h>
@@ -18,7 +19,7 @@ uint64_t getMemoryAmountOrZero()
     if (num_pages <= 0)
         return 0;
 
-    int64_t page_size = sysconf(_SC_PAGESIZE);
+    int64_t page_size = getPageSize();
     if (page_size <= 0)
         return 0;
 

base/common/getPageSize.cpp

@@ -0,0 +1,8 @@
+#include "common/getPageSize.h"
+
+#include <unistd.h>
+
+Int64 getPageSize()
+{
+    return sysconf(_SC_PAGESIZE);
+}

base/common/getPageSize.h

@@ -0,0 +1,6 @@
+#pragma once
+
+#include "common/types.h"
+
+/// Get memory page size
+Int64 getPageSize();

base/common/ya.make

@@ -47,6 +47,7 @@ SRCS(
     errnoToString.cpp
     getFQDNOrHostName.cpp
     getMemoryAmount.cpp
+    getPageSize.cpp
     getResource.cpp
     getThreadId.cpp
     mremap.cpp

src/Common/Allocator.h

@@ -26,6 +26,7 @@
     #define DISABLE_MREMAP 1
 #endif
 #include <common/mremap.h>
+#include <common/getPageSize.h>
 
 #include <Common/MemoryTracker.h>
 #include <Common/Exception.h>
@@ -59,7 +60,6 @@
   */
 extern const size_t MMAP_THRESHOLD;
 
-static constexpr size_t MMAP_MIN_ALIGNMENT = 4096;
 static constexpr size_t MALLOC_MIN_ALIGNMENT = 8;
 
 namespace DB
@@ -194,10 +194,11 @@ private:
     void * allocNoTrack(size_t size, size_t alignment)
     {
         void * buf;
+        size_t mmap_min_alignment = ::getPageSize();
 
         if (size >= MMAP_THRESHOLD)
         {
-            if (alignment > MMAP_MIN_ALIGNMENT)
+            if (alignment > mmap_min_alignment)
                 throw DB::Exception(fmt::format("Too large alignment {}: more than page size when allocating {}.",
                     ReadableSize(alignment), ReadableSize(size)), DB::ErrorCodes::BAD_ARGUMENTS);
 

src/Common/Arena.h

@@ -83,10 +83,11 @@ private:
     /// Last contiguous chunk of memory.
     Chunk * head;
     size_t size_in_bytes;
+    size_t page_size;
 
-    static size_t roundUpToPageSize(size_t s)
+    static size_t roundUpToPageSize(size_t s, size_t page_size)
     {
-        return (s + 4096 - 1) / 4096 * 4096;
+        return (s + page_size - 1) / page_size * page_size;
     }
 
     /// If chunks size is less than 'linear_growth_threshold', then use exponential growth, otherwise - linear growth
@@ -113,7 +114,7 @@ private:
         }
 
         assert(size_after_grow >= min_next_size);
-        return roundUpToPageSize(size_after_grow);
+        return roundUpToPageSize(size_after_grow, page_size);
     }
 
     /// Add next contiguous chunk of memory with size not less than specified.
@@ -129,7 +130,8 @@ private:
 public:
     Arena(size_t initial_size_ = 4096, size_t growth_factor_ = 2, size_t linear_growth_threshold_ = 128 * 1024 * 1024)
         : growth_factor(growth_factor_), linear_growth_threshold(linear_growth_threshold_),
-        head(new Chunk(initial_size_, nullptr)), size_in_bytes(head->size())
+        head(new Chunk(initial_size_, nullptr)), size_in_bytes(head->size()),
+        page_size(static_cast<size_t>(::getPageSize()))
     {
     }
 

src/Common/ArrayCache.h

@@ -13,6 +13,8 @@
 #include <boost/noncopyable.hpp>
 #include <ext/scope_guard.h>
 
+#include <common/getPageSize.h>
+
 #include <Common/Exception.h>
 #include <Common/randomSeed.h>
 #include <Common/formatReadable.h>
@@ -326,8 +328,6 @@ private:
         return (x + (rounding - 1)) / rounding * rounding;
     }
 
-    static constexpr size_t page_size = 4096;
-
     /// Sizes and addresses of allocated memory will be aligned to specified boundary.
     static constexpr size_t alignment = 16;
 
@@ -505,6 +505,7 @@ private:
 
         /// If nothing was found and total size of allocated chunks plus required size is lower than maximum,
         ///  allocate a new chunk.
+        size_t page_size = static_cast<size_t>(::getPageSize());
         size_t required_chunk_size = std::max(min_chunk_size, roundUp(size, page_size));
         if (total_chunks_size + required_chunk_size <= max_total_size)
         {

src/Common/MemoryStatisticsOS.cpp

@@ -8,10 +8,11 @@
 
 #include "MemoryStatisticsOS.h"
 
+#include <common/logger_useful.h>
+#include <common/getPageSize.h>
 #include <Common/Exception.h>
 #include <IO/ReadBufferFromMemory.h>
 #include <IO/ReadHelpers.h>
-#include <common/logger_useful.h>
 
 
 namespace DB
@@ -26,7 +27,6 @@ namespace ErrorCodes
 }
 
 static constexpr auto filename = "/proc/self/statm";
-static constexpr size_t PAGE_SIZE = 4096;
 
 MemoryStatisticsOS::MemoryStatisticsOS()
 {
@@ -93,11 +93,12 @@ MemoryStatisticsOS::Data MemoryStatisticsOS::get() const
     skipWhitespaceIfAny(in);
     readIntText(data.data_and_stack, in);
 
-    data.virt *= PAGE_SIZE;
-    data.resident *= PAGE_SIZE;
-    data.shared *= PAGE_SIZE;
-    data.code *= PAGE_SIZE;
-    data.data_and_stack *= PAGE_SIZE;
+    size_t page_size = static_cast<size_t>(::getPageSize());
+    data.virt *= page_size;
+    data.resident *= page_size;
+    data.shared *= page_size;
+    data.code *= page_size;
+    data.data_and_stack *= page_size;
 
     return data;
 }

src/Common/StringSearcher.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <common/getPageSize.h>
 #include <Common/Exception.h>
 #include <Common/StringUtils/StringUtils.h>
 #include <Common/UTF8Helpers.h>
@@ -37,7 +38,7 @@ struct StringSearcherBase
 {
 #ifdef __SSE2__
     static constexpr auto n = sizeof(__m128i);
-    const int page_size = getpagesize();
+    const int page_size = ::getPageSize();
 
     bool pageSafe(const void * const ptr) const
     {

src/IO/MMapReadBufferFromFileDescriptor.cpp

@@ -6,6 +6,7 @@
 #include <Common/ProfileEvents.h>
 #include <Common/formatReadable.h>
 #include <Common/Exception.h>
+#include <common/getPageSize.h>
 #include <IO/WriteHelpers.h>
 #include <IO/MMapReadBufferFromFileDescriptor.h>
 
@@ -38,7 +39,9 @@ void MMapReadBufferFromFileDescriptor::init(int fd_, size_t offset, size_t lengt
                 ErrorCodes::CANNOT_ALLOCATE_MEMORY);
 
         BufferBase::set(static_cast<char *>(buf), length, 0);
-        ReadBuffer::padded = (length % 4096) > 0 && (length % 4096) <= (4096 - 15); /// TODO determine page size
+
+        size_t page_size = static_cast<size_t>(::getPageSize());
+        ReadBuffer::padded = (length % page_size) > 0 && (length % page_size) <= (page_size - 15);
     }
 }
 

src/IO/tests/o_direct_and_dirty_pages.cpp

@@ -3,6 +3,7 @@
 #include <iostream>
 
 #include <common/types.h>
+#include <common/getPageSize.h>
 #include <IO/WriteHelpers.h>
 #include <IO/ReadHelpers.h>
 #include <IO/WriteBufferFromFile.h>
@@ -16,6 +17,7 @@ int main(int, char **)
     {
         static const size_t N = 100000;
         static const size_t BUF_SIZE = 1048576;
+        size_t page_size = static_cast<size_t>(::getPageSize());
 
         ReadBufferFromFile rand_in("/dev/urandom");
         unsigned rand = 0;
@@ -33,7 +35,7 @@ int main(int, char **)
         }
 
         {
-            ReadBufferFromFile rb("test1", BUF_SIZE, O_RDONLY | O_DIRECT, nullptr, 4096);
+            ReadBufferFromFile rb("test1", BUF_SIZE, O_RDONLY | O_DIRECT, nullptr, page_size);
             String res;
             for (size_t i = 0; i < N; ++i)
                 readStringBinary(res, rb);
@@ -44,14 +46,14 @@ int main(int, char **)
         /// Write to file with O_DIRECT, read as usual.
 
         {
-            WriteBufferFromFile wb("test2", BUF_SIZE, O_WRONLY | O_CREAT | O_TRUNC | O_DIRECT, 0666, nullptr, 4096);
+            WriteBufferFromFile wb("test2", BUF_SIZE, O_WRONLY | O_CREAT | O_TRUNC | O_DIRECT, 0666, nullptr, page_size);
 
             for (size_t i = 0; i < N; ++i)
                 writeStringBinary(test, wb);
 
-            if (wb.offset() % 4096 != 0)
+            if (wb.offset() % page_size != 0)
             {
-                size_t pad = 4096 - wb.offset() % 4096;
+                size_t pad = page_size - wb.offset() % page_size;
                 memset(wb.position(), 0, pad);
                 wb.position() += pad;
             }

src/Interpreters/tests/internal_iotop.cpp

@@ -5,6 +5,7 @@
 #include <Common/TaskStatsInfoGetter.h>
 #include <Poco/File.h>
 #include <Common/Stopwatch.h>
+#include <common/getPageSize.h>
 #include <common/getThreadId.h>
 #include <IO/WriteBufferFromString.h>
 #include <linux/taskstats.h>
@@ -61,8 +62,9 @@ static void do_io(size_t id)
     std::string path_dst = "test_out_" + std::to_string(id);
 
     {
+        size_t page_size = static_cast<size_t>(::getPageSize());
         ReadBufferFromFile rb("/dev/urandom");
-        WriteBufferFromFile wb(path_dst, DBMS_DEFAULT_BUFFER_SIZE, O_WRONLY | O_CREAT | O_TRUNC | O_DIRECT, 0666, nullptr, 4096);
+        WriteBufferFromFile wb(path_dst, DBMS_DEFAULT_BUFFER_SIZE, O_WRONLY | O_CREAT | O_TRUNC | O_DIRECT, 0666, nullptr, page_size);
         copyData(rb, wb, copy_size);
         wb.close();
     }

src/Storages/StorageStripeLog.cpp

@@ -40,8 +40,6 @@
 namespace DB
 {
 
-#define INDEX_BUFFER_SIZE 4096
-
 namespace ErrorCodes
 {
     extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
@@ -319,7 +317,7 @@ Pipe StorageStripeLog::read(
         return Pipe(std::make_shared<NullSource>(metadata_snapshot->getSampleBlockForColumns(column_names, getVirtuals(), getStorageID())));
     }
 
-    CompressedReadBufferFromFile index_in(disk->readFile(index_file, INDEX_BUFFER_SIZE));
+    CompressedReadBufferFromFile index_in(disk->readFile(index_file, 4096));
     std::shared_ptr<const IndexForNativeFormat> index{std::make_shared<IndexForNativeFormat>(index_in, column_names_set)};
 
     size_t size = index->blocks.size();

utils/iotest/iotest.cpp

@@ -6,6 +6,7 @@
 #include <Common/Stopwatch.h>
 #include <Common/ThreadPool.h>
 #include <Common/randomSeed.h>
+#include <common/getPageSize.h>
 
 #include <cstdlib>
 #include <iomanip>
@@ -46,7 +47,7 @@ void thread(int fd, int mode, size_t min_offset, size_t max_offset, size_t block
 {
     using namespace DB;
 
-    Memory<> direct_buf(block_size, sysconf(_SC_PAGESIZE));
+    Memory<> direct_buf(block_size, ::getPageSize());
     std::vector<char> simple_buf(block_size);
 
     char * buf;

utils/iotest/iotest_aio.cpp

@@ -14,6 +14,7 @@ int main(int, char **) { return 0; }
 #include <Common/ThreadPool.h>
 #include <Common/Stopwatch.h>
 #include <Common/randomSeed.h>
+#include <common/getPageSize.h>
 #include <pcg_random.hpp>
 #include <IO/BufferWithOwnMemory.h>
 #include <IO/ReadHelpers.h>
@@ -52,7 +53,7 @@ void thread(int fd, int mode, size_t min_offset, size_t max_offset, size_t block
 
     std::vector<Memory<>> buffers(buffers_count);
     for (size_t i = 0; i < buffers_count; ++i)
-        buffers[i] = Memory<>(block_size, sysconf(_SC_PAGESIZE));
+        buffers[i] = Memory<>(block_size, ::getPageSize());
 
     pcg64_fast rng(randomSeed());