性能测试疑问

[CarlZhang111]

说明

• 测试环境： arm笔记本
• 测试版本：commit: abacb8362f67c07eb1a2b5b7406bab0dbdb5abfe

准备工作

• 修改shmcache_set.c代码如下：

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include <chrono>
#include <iostream>
#include <thread>
#include "fastcommon/logger.h"
#include "fastcommon/shared_func.h"
#include "shmcache/shmcache.h"

static void usage(const char *prog)
{
    fprintf(stderr, "shmcache set key value.\n"
         "Usage: %s [config_filename] <key> <value> <ttl>\n", prog);
}

int main(int argc, char *argv[])
{
	int result;
    int index;
    char *config_filename;
    struct shmcache_context context;
    struct shmcache_key_info key;
    char *value;
    int value_len;
    int ttl;

    if (argc >= 2 && (strcmp(argv[1], "-h") == 0 ||
                strcmp(argv[1], "help") == 0 ||
                strcmp(argv[1], "--help") == 0))
    {
        usage(argv[0]);
        return 0;
    }
    if (argc < 4) {
        usage(argv[0]);
        return EINVAL;
    }

    config_filename = "/etc/libshmcache.conf";
    if (isFile(argv[1])) {
        if (argc < 5) {
            usage(argv[0]);
            return EINVAL;
        }
        config_filename = argv[1];
        index = 2;
    } else {
        index = 1;
    }

	log_init();
    if ((result=shmcache_init_from_file(&context, config_filename)) != 0) {
        return result;
    }

    key.data = argv[index++];
    key.length = strlen(key.data);
    value = argv[index++];
    value_len = strlen(value);
    ttl = atoi(argv[index++]);
    while(1) {
      auto start_time = std::chrono::high_resolution_clock::now();
      result = shmcache_set(&context, &key, value, value_len, ttl);
      //if (result == 0) {
      //  printf("set key: %s success.\n", key.data);
      //} else {
      //  fprintf(stderr, "set key: %s fail, errno: %d\n", key.data, result);
      //}

      auto end_time = std::chrono::high_resolution_clock::now();
      auto duration =
        std::chrono::duration_cast<std::chrono::microseconds>(end_time - start_time);
      if (duration.count() > 100) {
        std::cout << "Set data with time: " << duration.count() << std::endl;
      }
      std::this_thread::sleep_for(std::chrono::milliseconds(10));
    }
    return result;
}

• /etc/libshmcache.conf内容：

# shared memory type, shm or mmap
# shm for SystemV pure shared memory
# mmap for POSIX shared memory based file
# default value is mmap
# Note: when type is shm, the shm limit is too small in FreeBSD and MacOS,
#  you should increase following kernel parameters:
#    kern.sysv.shmmax
#    kern.sysv.shmall
#    kern.sysv.shmseg
#  you can refer to http://www.unidata.ucar.edu/software/mcidas/2008/users_guide/workstation.html
type = mmap

# the lock filename
filename = /tmp/libshmcache

# the memory limit
# the oldest memory will be recycled when this max memory reached
max_memory = 32M

# the min memory
# default: 0, means do NOT set min memory
min_memory = 0

# the memory segment size for incremental memory allocation
segment_size = 2M

# the key number limit
max_key_count = 200000

# the size limit for one value
max_value_size = 256K

# the hash function in libfastcommon/src/hash.h
# default: simple_hash
hash_function = simple_hash

# recycle key number once when reach max keys
# <= 0 means recycle one memory striping
# default: 0
recycle_key_once = 0

# if recycle valid entries by FIFO
# default: false
recycle_valid_entries = true

# value allocator policy
# avg. key TTL threshold for recycling memory
# <= 0 for never recycle memory until reach memory limit (max_memory)
# unit: second
# default: 0
value_policy.avg_key_ttl = 5400

# when the remain memory <= this parameter, discard it
value_policy.discard_memory_size = 128

# when a value allocator allocate fail times > this parameter,
# means it is almost full, discard it
# default: 5
value_policy.max_fail_times = 5

# sleep time to avoid other processes read dirty data when
# recycle more than one valid (in TTL / not expired) KV entries
# 0 for never sleep
# unit: microsecond (us)
# default: 0
value_policy.sleep_us_when_recycle_valid_entries = 1000

# try lock interval in us, must great than zero
# unit: microsecond (us)
# default value is 200 us
lock_policy.trylock_interval_us = 200

# the interval to detect deadlock caused by the crushed process
# must great than zero
# unit: millisecond (ms)
# default value is 1000 ms
lock_policy.detect_deadlock_interval_ms = 1000

# standard log level as syslog, case insensitive, value list:
## emerg for emergency
## alert
## crit for critical
## error
## warn for warning
## notice
## info
## debug
log_level = info

编译完成后运行：

./shmcache_set test value 1

期望输出

正常情况下不会输出Set data with time

实际输出

[2024-12-24 02:09:19] INFO - file: shmopt.c, line: 86, striping_count: 2, striping.count.current: 0, context->memory->vm_info.segment.size: 2097152, context->memory->vm_info.striping.size: 1048576
[2024-12-24 02:09:19] INFO - file: shmopt.c, line: 109, pid: 4085, create value segment #1, size: 2097152
[2024-12-24 02:09:19] INFO - file: shmcache.c, line: 219, pid: 4085, init share memory first time, hashtable segment size: 4409992
Set data with time: 886μs
Set data with time: 4824μs
Set data with time: 1208μs
Set data with time: 1082μs
Set data with time: 4517μs
Set data with time: 7079μs
Set data with time: 1225μs
Set data with time: 1427μs
Set data with time: 7082μs
Set data with time: 1891μs
Set data with time: 2711μs
Set data with time: 2747μs
Set data with time: 493μs
Set data with time: 2631μs
Set data with time: 1148μs
Set data with time: 1111μs
Set data with time: 1209μs
Set data with time: 1107μs
Set data with time: 1130μs
Set data with time: 2402μs
Set data with time: 4471μs
Set data with time: 5060μs
Set data with time: 1192μs
Set data with time: 824μs
Set data with time: 849μs
Set data with time: 816μs
Set data with time: 2900μs
Set data with time: 1277μs
Set data with time: 1566μs
Set data with time: 7107μs
Set data with time: 6883μs
Set data with time: 212μs
Set data with time: 1088μs
Set data with time: 513μs
Set data with time: 4049μs
Set data with time: 3732μs
Set data with time: 927μs
Set data with time: 715μs
Set data with time: 802μs
Set data with time: 1274μs
Set data with time: 698μs
Set data with time: 2899μs
Set data with time: 2292μs
Set data with time: 777μs
Set data with time: 848μs
Set data with time: 542μs
Set data with time: 6138μs
Set data with time: 908μs
Set data with time: 1144μs
Set data with time: 1642μs
Set data with time: 448μs
Set data with time: 467μs
Set data with time: 791μs
Set data with time: 580μs
Set data with time: 136μs
Set data with time: 267μs
Set data with time: 1582μs
Set data with time: 5285μs
Set data with time: 3284μs
Set data with time: 3466μs
Set data with time: 4630μs

问题

请问这种写入单值是否正常情况下会存在耗时7ms+现象，如果正常的话，此repo的读写性能大概是什么样子？或有什么要求
期待回复，感谢

[happyfish100]

从测试结果来看，感觉不太正常。
会不会是你笔记本环境的问题呢？比如笔记本负载比较高，或者大量使用到了交换内存？

[CarlZhang111]

和在docker里面跑测试程序有关系吗？因为不止我自己的笔记本有这个问题，其他笔记本测试出来也是最多2ms耗时，平时笔记本没有任何操作，只用来测试

[happyfish100]

你把 type = mmap 改为 type = shm，采用纯内存缓存方式测试一下看看效果

[CarlZhang111]

改为shm跑了十几分钟到是没出现，谢谢

[happyfish100]

那应该就是采用 mmap方式，操作系统会把写入的数据不定期刷到文件 （磁盘 ）导致的。