redis 压力测试与qps监控

1 背景

由于股票撮合中,我们使用zset构建到价成交,故这里对rangebyscore命令进行原位压力测试

撮合、止盈止损

使用redis zset触发股票成交

行情丢失导致无法撮合成交【重点】

redis线程池如何定,为什么开10个disruptor消费线程(redis连接):

1)io密集型4核2(n+1);

2)从第2点本地压测结果看,10线程已80%满足最高qps;

3)disruptor太多线程不好

 

2 首先压一波本地

压测设备:mac 2016 12'

 2.1 docker

 redis-benchmark -h 127.0.0.1 -p 63790 -c 100 -n 10000 script load "redis.call('zrangebyscore','sh111111','3','9')"

   java benchmark  java 代码  redis命令行
1  807  729  866
10 3115 3115  3187
50 4467 4235  4640
100 4375 4417  5238
500      5747

 

 *java benchmark与java代码都存在从池拿连接的操作

2.2 native

 redis-benchmark -h 127.0.0.1 -p 6379 -c 1 -n 10000 script load "redis.call('zrangebyscore','sh111111','3','9')"

   java benchmark  java 代码  redis命令行
1 11729  6050 10131
10 28597 18653 21000
50 31943 29056 23584
100 29476 28438 24875
500     24937

 

那么我们看到局域网及docker的测试,可能经过网卡后,10线程qps为3k,这个值与官方宣称的10w相去甚远,所以我看下往上其它人的压测结果

 

3 其它参考:

3.1 openresty-redis在不同网络环境下QPS对比讲解

http://blog.sina.cn/dpool/blog/s/blog_6145ed810102vefe.html?from=groupmessage&isappinstalled=0

redis相对openresty网络环境redis(requests per second)openresty(requests per second)

本地52631.58

局域网3105.59  与我docker测试水平相当

公网(纽约节点)169.95

 

3.2 memcache、redis、tair性能对比测试报告

http://blog.sina.cn/dpool/blog/s/blog_6145ed810102vefe.html?from=groupmessage&isappinstalled=0

以单线程通过各缓存客户端get调用向服务端获取数据,比较10000操作所消耗的时间

redis 1k对象 1260qps

并发1000个线程通过缓存软件的客户get调用向服务端获取数据,每个线程完成10000次的操作

redis 1k对象 11430qps  这个数据比我测试的要大三倍

 

4 阿里云redis qps 10线程,4.7万qps

https://zhuanlan.zhihu.com/p/78034665?utm_source=wechat_session&utm_medium=social&utm_oi=1003056052560101376&from=singlemessage&isappinstalled=0&wechatShare=1&s_s_i=Nxnfuuur16PoKq5S8w%2Bv7CqmqZ5fwF2fxQZXH9O4%2FPM%3D&s_r=1

阿里云社区版

社区 标准版双副本 1g主从 redis5.0 号称8w qps(集群256分片2560w qps),企业版24w(集群6144w):https://help.aliyun.com/document_detail/26350.html

施压机 :4 vCPU 8 GiB (I/O优化)ecs.c6.xlarge   10Mbps (峰值)

 

 

5 后话,为什么redis 多线程客户端获得更大qps,大到什么程度

以一个例子说明,假设:

一次命令时间(borrow|return resource + Jedis执行命令(含网络) )的平均耗时约为1ms,一个连接的QPS大约是1000 业务期望的QPS是50000 那么理论上需要的资源池大小是50000 / 1000 = 50个。但事实上这是个理论值,还要考虑到要比理论值预留一些资源,通常来讲maxTotal可以比理论值大一些。

但这个值不是越大越好,一方面连接太多占用客户端和服务端资源,另一方面对于Redis这种高QPS的服务器,一个大命令的阻塞即使设置再大资源池仍然会无济于事。  

https://cloud.tencent.com/developer/article/1425158

注意,redis多线程qps并不像理论的那样,多个线程qps=单个线程*线程数(有点像负载均衡),因为线程之间相互切换吞吐量相互制约,成非线性关系

 

 

 

6 性能监控:

参考1 https://www.cnblogs.com/cheyunhua/p/9068029.html 设置redis最大内存,类似于java内存的xmx

参考2 https://blog.csdn.net/z644041867/article/details/77965521 性能监控指标

redis-cli info | grep -w "connected_clients" |awk -F':' '{print $2}'

redis-cli info | grep -w "used_memory_rss_human" |awk -F':' '{print $2}' 类似于java内存jmx监控的commited和used

redis-cli info | grep -w "used_memory_peak_human" |awk -F':' '{print $2}'

redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'

 

redis-benchmark -h 127.0.0.1 -p 6379 -c 1 -n 1000000 script load "redis.call('zrangebyscore','sh111111','3','9')"
^Cript load redis.call('zrangebyscore','sh111111','3','9'): 10026.05

JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
0
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
9666
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
9473
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10249
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10590
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10486
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10421
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10450
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10673
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10707
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10655
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10530
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10570
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
10396
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
9595
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
9010
JoycedeMacBook:redis-5.0.5 joyce$ redis-cli info | grep -w "instantaneous_ops_per_sec" |awk -F':' '{print $2}'
9414

 

 

7 测试代码:

package redis;

import com.alibaba.fastjson.JSON;
import com.alibaba.fastjson.JSONObject;
import com.google.protobuf.InvalidProtocolBufferException;
import ip.IpPool;
import org.apache.commons.pool2.impl.GenericObjectPool;
import org.apache.commons.pool2.impl.GenericObjectPoolConfig;
import org.openjdk.jmh.annotations.*;
import org.openjdk.jmh.runner.Runner;
import org.openjdk.jmh.runner.RunnerException;
import org.openjdk.jmh.runner.options.Options;
import org.openjdk.jmh.runner.options.OptionsBuilder;
import org.redisson.Redisson;
import org.redisson.api.RBucket;
import org.redisson.api.RedissonClient;
import org.redisson.config.Config;
import redis.clients.jedis.Jedis;
import redis.clients.jedis.JedisPool;
import redis.clients.jedis.JedisPoolConfig;
import serial.MyBaseBean;
import serial.MyBaseProto;

import java.util.Set;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.TimeUnit;

/**
 * Created by joyce on 2019/10/24.
 */
@BenchmarkMode(Mode.Throughput)//基准测试类型
@OutputTimeUnit(TimeUnit.SECONDS)//基准测试结果的时间类型
@Threads(10)//测试线程数量(IO密集)
@State(Scope.Thread)//该状态为每个线程独享
public class YaliRedis {

    private static JedisPool jedisPool;
    private static final int threadCount = 10;

    @Setup
    public static void init() {
        JedisPoolConfig config = new JedisPoolConfig();
        config.setMaxTotal(800);
        config.setMaxIdle(800);
        jedisPool = new JedisPool(config,"localhost",63790,2000,"test");
//        Jedis jedis = jedisPool.getResource();
//        String test = jedis.get("testkey");
//        System.out.println(test);
//        Set<String> set = jedis.zrangeByScore("sh111111", 3,9);
//        System.out.println(set.size());
//        jedis.close();

    }
    @TearDown
    public static void destroy() {
        jedisPool.close();
    }

    public static void main(String[] args) throws Exception {

    //    initData();

        if(false) {
            Options opt = new OptionsBuilder().include(YaliRedis.class.getSimpleName()).forks(1).warmupIterations(1)
                    .measurementIterations(3).build();

            new Runner(opt).run();
        } else {
            init();
            final int perThread = 10000;
            CountDownLatch countDownLatchMain = new CountDownLatch(threadCount);
            CountDownLatch countDownLatchSub = new CountDownLatch(1);

            for(int i=0; i<threadCount; ++i) {
                new Thread(new Runnable() {
                    @Override
                    public void run() {
                        try {
                            countDownLatchSub.await();
                            Set<String> set = null;
                            for(int j=0; j<perThread; ++j)
                                set = testZSet();
                            System.out.println(set.size());

                        } catch (Exception e) {
                            e.printStackTrace();
                        } finally {
                            countDownLatchMain.countDown();
                        }

                    }
                }).start();
            }
            long st = (System.currentTimeMillis());
            countDownLatchSub.countDown();
            countDownLatchMain.await();
            System.out.println(System.currentTimeMillis() - st);
            System.out.println(threadCount * perThread * 1000 / (System.currentTimeMillis() - st));
        }

    }

    @Benchmark
    public static Set<String> testZSet() {
        Jedis jedis = null;
        jedis = jedisPool.getResource();
        Set<String> set = jedis.zrangeByScore("sh111111", 3,9);
        jedis.close();
        return set;
    }

//    @Benchmark
    public static void test() {
        Jedis jedis = null;
        jedis = jedisPool.getResource();
        jedis.get("testkey");
        jedis.close();
    }

//    @Benchmark
    public static void testJson() {

        Jedis jedis = null;
        jedis = jedisPool.getResource();
        String xx = jedis.get("testjson");
        JSONObject userJson = JSONObject.parseObject(xx);
        MyBaseBean user = JSON.toJavaObject(userJson,MyBaseBean.class);
        jedis.close();
    }

//    @Benchmark
    public static void testPb() {

        Jedis jedis = null;
        jedis = jedisPool.getResource();
        byte [] bytes = jedis.get("testpb".getBytes());
        try {
            MyBaseProto.BaseProto baseProto = MyBaseProto.BaseProto.parseFrom(bytes);
        } catch (InvalidProtocolBufferException e) {
            e.printStackTrace();
        }
        jedis.close();
    }

    public static void initData() {
        Jedis jedis = new Jedis("localhost", 63790);
        jedis.auth("test");
        for(int i=1; i<=9; ++i) {
            jedis.zadd("sh111111", i, String.valueOf(i*100));
        }
    }

}

 

8 备用:

1 redis-benchmark.     +     ( java bench jedis )

1)redis 本机

redis-benchmark -h 127.0.0.1 -p 6379 -c 1000 -n 10000 script load "redis.call('zrangebyscore','sh111111','3','9)"

1 th

10000 (11500)

50 th

24000 (33000)

100 th

25000 (30000)

500 th

26000 (20000)

1000 th

24000

 

2)docker

redis-benchmark -h 127.0.0.1 -p 63790 -c 100 -n 10000 script load "redis.call('zrangebyscore','sh111111','3','9)"

1 th

640 (700)

50 th

3900 (3300)

100 th

4400 (3800)

500 th

6000 (4500). 约80%

1000 th

5300

 

 

posted on 2019-10-25 13:59  silyvin  阅读(5511)  评论(0编辑  收藏  举报