High-concurrency instance sharing: Swoole efficiently implements business through aggregation of requests

This article will share with you an example of Swoole's high-concurrency aggregation request, and introduce how to make full use of the batch processing of the database to implement business functions more efficiently through aggregation of requests in high-concurrency scenarios. This example is only used as a starting point, hoping to stimulate more in-depth thinking.

Recommended related video courses: " Tens of millions of data concurrency solutions (theoretical and practical) "

Share some high-concurrency interview questions: 15 PHP interview questions about high concurrency (summary)

The background selected for this example is the concurrent order business. Under normal circumstances, creating an order on the backend is a insert operation one by one. When concurrency is low, the database's insert operation can indeed maintain good efficiency, but when the number of requests increases, the database frequently performs a single insert This will make the overall efficiency of the order business lower (this article simply assumes that 1 order = 1insert).

Through the above description, it is actually easy to think of the areas that need optimization. Analogy to the scene of taking an elevator in real life: an elevator is filled and then goes up, which can relieve the pressure of people flow as quickly as possible.

Let’s simply implement our idea with code:

<?php

Swoole\Runtime::enableCoroutine($flags = SWOOLE_HOOK_ALL);

// 最大等待次数
const MAX_TIMES = 10;
// 按批处理时, 每一批的最大请求暂留数量
const MAX_REQUEST = 3;
// 服务端最大超时时间, 避免客户端一直等待
const MAX_TIMEOUT = 5;

Co\run(function () {
    // 请求传输的channel, 原因是不要在swoole的协程环境中, 使用多个协程修改同一个全局变量
    // 如果是golang, 当然是可以不定义这里的$rqChannel
    // 只需要简单的将下面的$rqQueue和$times定义为全局变量即可达到一样的效果
    // 但是最好的方式任然是是通过channel共享内存
    $rqChannel = new Swoole\Coroutine\Channel(MAX_REQUEST);

    // 模拟创建订单
    $createOrder = function () use ($rqChannel) {
        // 使用数组模拟请求暂留队列
        $rqQueue = [];
        // 使用等待次数模拟tick效果
        $times = MAX_TIMES;
        while (true) {
            $times--;
            // 必须带上timeout参数, 否则channel是阻塞的
            $rq = $rqChannel->pop(1);
            // 保存1个正常的请求数据
            if (!empty($rq)) {
                $rqQueue[] = $rq;
            }
            // 请求数量未达上限或者还有等待次数时, 提前进入下一次循环
            if ($times > 0 && count($rqQueue) < MAX_REQUEST) {
                continue;
            }
            // 重置等待次数
            $times = MAX_TIMES;
            // 初始化SQL
            $sql = "INSERT INTO orders VALUES ";
            $inserts = [];
            // 模拟数据验证
            $validator = function ($input): bool {
                // 为了缩减代码, 没有真的做数据验证的处理
                array_filter($input);
                return true;
            };
            // $rqQueue在协程上下文是并发安全的, 所以遍历时不用担心
            foreach ($rqQueue as $index => $rq) {
                list($data, $chan) = $rq;
                // 这里可以考虑后置执行, 原因是后面可以有一些补救逻辑
                unset($rqQueue[$index]);
                // 判断$chan是否关闭å
                if ($chan->errCode === SWOOLE_CHANNEL_CLOSED) {
                    $data = null;
                    continue;
                }
                $bool = $validator($data);
                if ($bool) {
                    $inserts[] = "({$data[&#39;user_name&#39;]}, {$data[&#39;amount&#39;]}, {$data[&#39;mobile&#39;]})";
                    $chan->push([&#39;state&#39; => 1]);
                } else {
                    $chan->push([&#39;state&#39; => 0]);
                }
                // unset($rqQueue[$index]);
            }
            $sql .= (implode(&#39;,&#39;, $inserts) . &#39;;&#39;);
            // 模拟创建订单落库的逻辑
            echo $sql;
        }
    };

    // 新手要注意这一句代码的位置, 原因是 $server->start() 之后的代码不会执行
    go($createOrder);

    // 路由处理器
    $orderHandler = function ($rq, $res) use ($rqChannel) {
        $chan = new Swoole\Coroutine\Channel(1);
        // 使用timeout参数模拟超时
        $bool = $rqChannel->push([$rq->post, $chan], MAX_TIMEOUT);
        if (!$bool) {
            // 关闭$chan
            $chan->close();
            $res->end(&#39;timeout&#39;);
        }
        if (!empty($data = $chan->pop())) {
            // 关闭$chan
            $chan->close();
            // 区分成功或失败状态再输出响应
            if ($data[&#39;state&#39;] === 1) {
                $res->end(microtime());
            } else {
                $res->end(&#39;error&#39;);
            }
        }
    };

    $server = new Co\Http\Server("0.0.0.0", 9502, false);

    $server->handle(&#39;/order/create&#39;, $orderHandler);
    // 当前协程容器的终点
    $server->start();
});

The code is still very easy to understand as a whole, Variables$rqQueue is the analogy of an elevator. The number of times a hold request has to wait for a certain period of time $times is analogous to the need for an elevator to wait for the flow of people to enter one after another. Of course, the most important thing I hope readers will pay attention to is: In a coroutine environment, do not use shared memory for communication, but use communication to share memory.

Recommended learning: swoole tutorial

The above is the detailed content of High-concurrency instance sharing: Swoole efficiently implements business through aggregation of requests. For more information, please follow other related articles on the PHP Chinese website!