How to solve the product super phenomenon when using redis flash sale?

Recently I was doing a flash sale event, and for the sake of performance and response speed, I used redis. When I was writing, I paid special attention to preventing supernormal phenomena. I used cas (check and set) optimistic locking based on redis theory. I thought that this problem should be able to eliminate it, but it still occurred. I was very confused and asked for help. The specific code is roughly as follows:

<code><?php  
header("content-type:text/html;charset=utf-8");  
$redis = new redis();  
$result = $redis->connect('10.10.10.119', 6379);  
$mywatchkey = $redis->get("mywatchkey");  
$rob_total = 100;   //抢购数量  
if($mywatchkey<$rob_total){  
    $redis->watch("mywatchkey");  
    $redis->multi();  
      
    //设置延迟，方便测试效果。  
    sleep(5);  
    //插入抢购数据  
    $redis->hSet("mywatchlist","user_id_".mt_rand(1, 9999),time());  
    $redis->set("mywatchkey",$mywatchkey+1);  
    $rob_result = $redis->exec();  
    if($rob_result){  
        $mywatchlist = $redis->hGetAll("mywatchlist");  
        echo "抢购成功！<br/>";  
        echo "剩余数量：".($rob_total-$mywatchkey-1)."<br/>";  
        echo "用户列表：<pre class="brush:php;toolbar:false">";  
        var_dump($mywatchlist);  
    }else{  
        echo "手气不好，再抢购！";exit;  
    }  
}  
?>  
</code>

Reply content:

Recently I was doing a flash sale event, and for the sake of performance and response speed, I used redis. When I was writing, I paid special attention to preventing supernormal phenomena. I used cas (check and set) optimistic locking based on redis theory. I thought that this problem should be able to eliminate it, but it still occurred. I was very confused and asked for help. The specific code is roughly as follows:

<code><?php  
header("content-type:text/html;charset=utf-8");  
$redis = new redis();  
$result = $redis->connect('10.10.10.119', 6379);  
$mywatchkey = $redis->get("mywatchkey");  
$rob_total = 100;   //抢购数量  
if($mywatchkey<$rob_total){  
    $redis->watch("mywatchkey");  
    $redis->multi();  
      
    //设置延迟，方便测试效果。  
    sleep(5);  
    //插入抢购数据  
    $redis->hSet("mywatchlist","user_id_".mt_rand(1, 9999),time());  
    $redis->set("mywatchkey",$mywatchkey+1);  
    $rob_result = $redis->exec();  
    if($rob_result){  
        $mywatchlist = $redis->hGetAll("mywatchlist");  
        echo "抢购成功！<br/>";  
        echo "剩余数量：".($rob_total-$mywatchkey-1)."<br/>";  
        echo "用户列表：<pre class="brush:php;toolbar:false">";  
        var_dump($mywatchlist);  
    }else{  
        echo "手气不好，再抢购！";exit;  
    }  
}  
?>  
</code>

I think this code will still be oversold under high concurrency conditions. Suppose: When there is only 1 prize left, three people execute $redis->watch("mywatchkey") at the same time and the data they get is 99, then there is an oversold phenomenon.

Since redis is a single-threaded read, then use the simplest queue implementation.

1. Before drawing the lottery, write the number of prizes into the redisqueueaward:100 // A list with a length of 100, the value is only used as whether you win the prize

2. Concurrent Lottery

<code>$award = $redis->lpop('award:100'); // 由于队列只有100个值，可以确保只有100个人中奖
if(!$award){
    echo "手气不好，再抢购！";exit;  
}

// 剩下就是中奖操作的事情了</code>

sleep(5);
Does your company still accept people?

Think about this situation.
mywatchkey=99
User A requests mywatchkey and gets 99.
User B requests mywatchkey and gets 99.
When A and B requests are completed, what should mywatchkey be? . 101, or 100?

This is a classic Check-then-Act error

First of all, let me ask you if the "oversold" result of this code is $mywatchkey > 100 (I think it is impossible)
Or when $mywatchkey = 100, two pages appear at the same time "Successful purchase!"
I have no idea about this The code is tested, I just guess whether it will be the following situation:

1.Session A: Perform mywatchkey check, at this time mywatchkey = 99

2.Session B: After processing mywatchkey, mywatchkey=100 at this time, but it does not affect the subsequent operations of Session A

Add two more connections, take a look, you might gain something

transactions-and-watch-statement-in-redis
redis-watch-multi-exec-by-one-client

If you want to fundamentally solve this matter, please study first:

1. Lock.

2. Traditional database transactions.

Then learn the difference between Redis and traditional databases.