Question

相比 glibc ptmalloc，Google tcmalloc 和 Facebook jemalloc 对高并发长时间运行的后台服务效率更高。

$ apt install -y libjemalloc-dev

使用方式：

• 无需修改源码，直接 -ljemalloc 链接即可。
• 已有程序，用 LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so 启动。

$ gcc -g -O0 -o test main.c -ljemalloc   # 链接参数放输入文件后面。

$ MALLOC_CONF=stats_print:true ./test    # 检查 jemalloc 是否启用。

垃圾回收

虽然有点另类，但的确有个发展了很多年的 C/C++ 垃圾回收器。

The Boehm-Demers-Weiser conservative garbage collector can be used as a garbage collecting replacement for C malloc or C++ new. It allows you to allocate memory basically as you normally would, without explicitly deallocating memory that is no longer useful. The collector automatically recycles memory when it determines that it can no longer be otherwise accessed.

$ apt install -y libgc-dev

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <gc/gc.h>

void test () 
{
    char *s = GC_MALLOC(1<<20);  // 1 MB
    // char *s = malloc(1<<20);
    *s = 0x10; 
}

int main (void)
{
    for (int i = 0; i < 100; i++) {
        printf("%d, heap = %ld\n", i, GC_get_heap_size());
        test();
        sleep(1);
    }

    // getchar();
    return 0; 
}

$ gcc -o test test.c -lgc

安全检查

利用 Valgrind 检查内存安全问题，诸如：内存泄漏（memory leak），使用未初始化变量（uninitialized）等。

$ apt install -y valgrind

$ valgrind --leak-check=full --show-reachable=yes --log-file=mem.log ./test

还可利用 cachegrind 检查 CPU 缓存命中情况。

I1 : L1 指令； D1 : L1 数据； LL : 最后一级缓存。

rd : 读； wr : 写。

$ valgrind --tool=cachegrind ./test

==3369== Cachegrind, a cache and branch-prediction profiler
==3369== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
==3369== Command: ./test
==3369== 
--3369-- warning: L3 cache found, using its data for the LL simulation.
0
==3369== 
==3369== I   refs:      179,941
==3369== I1  misses:      1,113
==3369== LLi misses:      1,091
==3369== I1  miss rate:    0.62%
==3369== LLi miss rate:    0.61%
==3369== 
==3369== D   refs:       53,145  (41,400 rd   + 11,745 wr)
==3369== D1  misses:      3,181  ( 2,509 rd   +    672 wr)
==3369== LLd misses:      2,635  ( 2,035 rd   +    600 wr)
==3369== D1  miss rate:     6.0% (   6.1%     +    5.7%  )
==3369== LLd miss rate:     5.0% (   4.9%     +    5.1%  )
==3369== 
==3369== LL refs:         4,294  ( 3,622 rd   +    672 wr)
==3369== LL misses:       3,726  ( 3,126 rd   +    600 wr)
==3369== LL miss rate:      1.6% (   1.4%     +    5.1%  )

• Facebook jemalloc: Getting Started
• A garbage collector for C and C++
• Valgrind User Manual