单机模拟高并发进行压力测试

版权声明:本文为博主原创文章,遵循 CC 4.0 BY-SA 版权协议,转载请附上原文出处链接和本声明。
本文链接:https://blog.csdn.net/ligen2012/article/details/78723383

本文主要介绍如何使用netty实现C100K级别的并发长连接

Socket五元组

理论上,客户端IP与端口,服务端IP与端口以及通讯协议(TCP|UDP)构成了一个连接的五元组。改变其中任何一个元素都可以新建一个连接。如果想要单台机器或者两台机器模拟数十万的连接数,仅仅靠系统分配的端口数目是不够的,我们必须要改变其中的IP元素。本文采取的方式是利用linux的secondary ip机制,为服务端分配多个IP,理论上最多可以分配(2^32-1)个IP,连接数目可达2^16 * IP数目,远远超出我们的目标连接数。

调整系统参数限制

端口范围检查

lg@lenove:~$ sudo sysctl --all | grep port_range
net.ipv4.ip_local_port_range = 10000    60000 //有5w可用

tcp设置检查

lg@lenove:~$ sudo sysctl --all | grep tw_reuse

net.ipv4.tcp_tw_reuse = 1 //允许将TIME-WAIT sockets重新用于新的TCP连接,优化客户端处理能力

检查系统文件句柄限制

lg@lenove:~$ ulimit -Hn

1000000 //100w

左右互博:单机模拟100k并发

配置secondary ip, 这里配了20个,理论上可以支持100W个TCP连接(20个IP*5w个端口)

lg@lenove:~$ seq 20 | xargs -i sudo ip addr add 127.0.1.{}/8 dev lo 

lg@lenove:~$ ip addr show
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.1/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.2/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.3/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.4/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.5/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.6/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.7/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.8/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.9/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.10/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.11/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.12/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.13/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.14/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.15/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.16/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.17/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.18/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.19/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet 127.0.1.20/8 scope host secondary lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host 
       valid_lft forever preferred_lft forever

红蓝对抗:双机直连模拟100k并发

双机模拟和单机模拟原理类似,不过在路由设置上麻烦一些。
首先,两台主机需要用双绞线直连,以规避路由器的连接数目限制,同时获得带宽保证。

服务器端 IP设置

lg@lenove:~$ sudo ip addr add 192.168.255.1/30 brd + label enp4s0:server dev enp4s0

lg@lenove:~$ seq 1 20 | xargs -i sudo ip addr add 192.168.0.{}/24 dev enp4s0

lg@lenove:~$ ip addr show
2: enp4s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP group default qlen 1000
    link/ether c8:5b:76:b2:50:33 brd ff:ff:ff:ff:ff:ff
    inet 192.168.16.102/24 brd 192.168.16.255 scope global enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.255.1/30 brd 192.168.255.3 scope global enp4s0:server
       valid_lft forever preferred_lft forever
    inet 192.168.0.1/24 scope global enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.2/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.3/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.4/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.5/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.6/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.7/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.8/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.9/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.10/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.11/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.12/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.13/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.14/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.15/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.16/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.17/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.18/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.19/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet 192.168.0.20/24 scope global secondary enp4s0
       valid_lft forever preferred_lft forever
    inet6 fe80::b684:f2df:7ecc:a317/64 scope link 
       valid_lft forever preferred_lft forever

客户端IP设置

lg@lenove:~$ sudo ip addr add 192.168.255.2/30 brd + label enp2s0:client dev enp2s0
lg@lenove:~$ sudo ip route add 192.168.0.0/24 via 192.168.255.1

lg@lenove:~$ ip route show
192.168.0.0/24 via 192.168.255.1 dev enp2s0

注意,服务器地址为192.168.255.1/30,客户端地址为192.168.255.2/30,两者处于同一个网段,并且,客户端将192.168.0.0/24网段的网关指向了192.168.255.1。

带宽测速,将近一千兆:

lg@lenove:~$ iperf -s
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 4.00 KByte (default)
------------------------------------------------------------
[  4] local 192.168.255.1 port 5001 connected with 192.168.255.2 port 38656
[ ID] Interval       Transfer     Bandwidth
[  4]  0.0-10.0 sec  1.10 GBytes   940 Mbits/sec

客户端代码实现

因项目涉及到公司业务,不便上传源码。现摘取关键实现部分:

构造虚拟主机列表

private final static int V_HOST_SIZE = 20;
private static Stream<String> hostStream() {
        return IntStream.rangeClosed(1, V_HOST_SIZE)
                .mapToObj(GatewayServerTest::ip);
    }
private static String ip(int index) {
        return String.format("127.0.1.%d", index);
//        return String.format("192.168.0.%d", index);
    }

阻塞,等待连接完成

private static void sync(ChannelFuture channelFuture) {
        handleException(channelFuture::sync);
    }

private static <T> T handleException(SupplierWithException<T> supplier) {
        try {
            return supplier.get();
        } catch (Exception e) {
            throw new RuntimeException(e.getMessage());
        }
    }

创建连接,非阻塞模式

private static Stream<ChannelFuture> connect(Bootstrap bootstrap, String host, int n) {
    return Stream.generate(() -> connect(bootstrap, host)).limit(n);
    }

private static ChannelFuture connect(Bootstrap bootstrap, String host) {
    return handleException(() -> bootstrap.connect(host, port));
    }

main方法,主要目的是,每次发起DELTA个连接,平均分配给V_HOST_SIZE个虚拟主机,直到连接数目达到CONN_SIZE


private final static int CONN_SIZE = 100000;
private final static int DELTA = 100;

public static void main(String[] args) {
    EventLoopGroup ioGroup = new NioEventLoopGroup(4);

    final Bootstrap bootstrap = new Bootstrap();
    bootstrap.group(ioGroup)
            .channel(NioSocketChannel.class)
            .option(ChannelOption.ALLOCATOR, PooledByteBufAllocator.DEFAULT)
            .handler(new ChannelInitializer<SocketChannel>() {
                @Override
                public void initChannel(final SocketChannel ch) throws Exception {
                    ch.pipeline().addLast(
                            new CustomHandler()
                    );
                }
            });

    final AtomicInteger started = new AtomicInteger(0);

    while (started.get() < CONN_SIZE) {

        final List<ChannelFuture> futures = hostStream()
                .flatMap(host -> connect(bootstrap, host, DELTA / V_HOST_SIZE))
                .collect(Collectors.toList());

        futures.stream()
                .peek(future -> started.incrementAndGet())
                .forEach(GatewayServerTest::sync);

    try {
        Thread.sleep(1000 * 60 * 60 * 24);
    } catch (InterruptedException e) {
        e.printStackTrace();
    } finally {
        ioGroup.shutdownGracefully();
    }
}

CustomHandler的作用主要是定时向服务器发送GPS数据,一秒一次

public class CustomHandler extends ChannelInboundHandlerAdapter {


    @Override
    public void channelActive(ChannelHandlerContext ctx) throws Exception {

        ctx.executor().execute(new GpsTask(ctx));

    }


    private final static class GpsTask implements Runnable {
        private ChannelHandlerContext context;

        public GpsTask(ChannelHandlerContext context) {
            this.context = context;
        }


        public void run() {
           context.writeAndFlush("116.40, 39.91");
           context.executor().schedule(this, 1, TimeUnit.SECONDS);
        }
    }
}

在笔者的Thinkpad E470上面,左右互博模式的并发连接数可以达到20W,30w时出现了机器卡死现象;红蓝对抗模式下,可以达到30w以上的并发连接数,但会出现丢失连接的现象。

附录–相关异常

  • NoRouteToHostException: 可能是端口耗尽所致
展开阅读全文

Jmeter高并发压力测试

07-03

<p>n <span style="font-size:18px;">本课程是一门具有很强实践性质的“项目实战”课程,俗称“</span><span style="font-size:18px;">Java</span><span style="font-size:18px;">商城秒杀系统”,主要介绍了当前一些电商平台或者商城在举办一些商品秒杀、抢购活动时所涉及的相关业务流程,其中,主要包括“秒杀活动列表的展示”、“秒杀活动详情的获取”、“用户登录认证服务”、“高并发秒杀</span><span style="font-size:18px;">-</span><span style="font-size:18px;">抢购业务”、“秒杀成功订单的生成”以及“邮件通知服务”等等,如下图所示:</span>n</p>n<p>n <span style="font-size:18px;"><img src="https://img-bss.csdn.net/201907030743388775.png" alt="" /><br /></span>n</p>n<p>n <span style="font-size:18px;"><br /></span>n</p>n<p>n <span style="font-size:18px;">n </span></p><p>n 其中,该“秒杀系统”所对应的整体业务流程图如下图所示:<span></span>n </p>n<img src="https://img-bss.csdn.net/201907030744225540.png" alt="" /><br />nn<p>n <span style="font-size:18px;"><br /></span>n</p>n<p>n <span style="font-size:18px;"><br /></span>n</p>n<p>n <span style="font-size:18px;">n </span></p><p>n 从该整体业务流程图中可以看出,该秒杀系统业务的梳理<span>debug</span>自认为还是比较清楚的。而值得一提的是,本“秒杀系统”涉及的相关技术栈还是比较新颖、主流的,特别是在如今微服务、分布式系统架构横行的时代,本系统涉及的相关技术栈还是很值得各位小伙伴学习的!如下图所示:<span></span>n </p>nnn<p>n <span style="font-size:18px;"><img src="https://img-bss.csdn.net/201907030745139190.png" alt="" /><br /></span>n</p>n<p>n <span style="font-size:18px;"><br /></span>n</p>n<p>n <span style="font-size:18px;">n </span></p><p>n 除此之外,由于本课程属于“项目实战”性质的课程,故而学完本课程后,其整体的收益相对于“单一技术栈”的学习收益将会更多!如下图所示:<span></span>n </p>n<img src="https://img-bss.csdn.net/201907030745435683.png" alt="" /><br />nn<p>n <br /></p>

没有更多推荐了,返回首页