0x1 背景
golang的协程是好用,但是有时候瓶颈并不在语言上,而是在后面的数据源上面,例如我们常见的mysql,redis等,当一个后端服务很多请求的时候,语言是能hold得住,但是 mysql产生错误,比如 too many connection, too many time_wait 等等这些,今天我们就分析一下怎么解决这种问题
0x2 代码范例
请查看main.go, halokid (有帮忙的话请start或者follow一下哦,谢谢)
0x3 分析
只执行ini函数, 检查mysql的进程显示为(原有的mysql是没有进程在处理的)
没执行前
mysql> show processlist;
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| 4 | event_scheduler | localhost | NULL | Daemon | 2304 | Waiting on empty queue | NULL |
| 9 | root | 10.244.1.1:64000 | test | Sleep | 1315 | | NULL |
| 10 | root | 10.244.1.1:64022 | test | Query | 0 | starting | show processlist |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
3 rows in set (0.01 sec)
复制代码
执行后
mysql> show processlist;
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| 4 | event_scheduler | localhost | NULL | Daemon | 2284 | Waiting on empty queue | NULL |
| 9 | root | 10.244.1.1:64000 | test | Sleep | 1295 | | NULL |
| 10 | root | 10.244.1.1:64022 | test | Query | 0 | starting | show processlist |
| 13 | root | 10.244.1.1:52134 | test | Sleep | 20 | | NULL |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
4 rows in set (0.00 sec)
复制代码
可见执行 db.Ping() 之后, process多了一个 Sleep 的连接,就是放了一个连接进 连接池
运行
db.SetMaxOpenConns(10)
db.SetMaxIdleConns(5)
复制代码
两句之后, 连接池并没有改变, 可见上面的逻辑是在 数据库处理逻辑真实执行的时候才生效的
真正的连接池建立是发生在执行协程查询的逻辑时
fmt.Println("mysql连接池测试开始...")
var wg sync.WaitGroup
wg.Add(10)
for i := 0; i < 10; i++ {
go func() {
defer wg.Done()
GetUser()
}()
}
wg.Wait()
复制代码
运行上面的逻辑代码之后,执行协程查询
mysql> show processlist;
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| 4 | event_scheduler | localhost | NULL | Daemon | 4397 | Waiting on empty queue | NULL |
| 9 | root | 10.244.1.1:64000 | test | Sleep | 3408 | | NULL |
| 10 | root | 10.244.1.1:64022 | test | Query | 0 | starting | show processlist |
| 19 | root | 10.244.1.1:54823 | test | Sleep | 952 | | NULL |
| 20 | root | 10.244.1.1:54824 | test | Sleep | 1104 | | NULL |
| 47 | root | 10.244.1.1:57906 | test | Sleep | 0 | | NULL |
| 48 | root | 10.244.1.1:57909 | test | Sleep | 0 | | NULL |
| 49 | root | 10.244.1.1:57912 | test | Sleep | 0 | | NULL |
| 50 | root | 10.244.1.1:57907 | test | Sleep | 0 | | NULL |
| 51 | root | 10.244.1.1:57908 | test | Sleep | 0 | | NULL |
| 52 | root | 10.244.1.1:57913 | test | Sleep | 0 | | NULL |
| 53 | root | 10.244.1.1:57911 | test | Sleep | 0 | | NULL |
| 54 | root | 10.244.1.1:57910 | test | Sleep | 0 | | NULL |
| 55 | root | 10.244.1.1:57915 | test | Sleep | 0 | | NULL |
| 56 | root | 10.244.1.1:57914 | test | Sleep | 0 | | NULL |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
15 rows in set (0.00 sec)
复制代码
执行完在等待
mysql> show processlist;
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| 4 | event_scheduler | localhost | NULL | Daemon | 3931 | Waiting on empty queue | NULL |
| 9 | root | 10.244.1.1:64000 | test | Sleep | 2942 | | NULL |
| 10 | root | 10.244.1.1:64022 | test | Query | 0 | starting | show processlist |
| 19 | root | 10.244.1.1:54823 | test | Sleep | 486 | | NULL |
| 20 | root | 10.244.1.1:54824 | test | Sleep | 638 | | NULL |
| 32 | root | 10.244.1.1:56588 | test | Sleep | 22 | | NULL |
| 33 | root | 10.244.1.1:56591 | test | Sleep | 22 | | NULL |
| 34 | root | 10.244.1.1:56589 | test | Sleep | 22 | | NULL |
| 35 | root | 10.244.1.1:56590 | test | Sleep | 22 | | NULL |
| 36 | root | 10.244.1.1:56592 | test | Sleep | 22 | | NULL |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
10 rows in set (0.00 sec)
复制代码
协程执行完之后
mysql> show processlist;
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| 4 | event_scheduler | localhost | NULL | Daemon | 3941 | Waiting on empty queue | NULL |
| 9 | root | 10.244.1.1:64000 | test | Sleep | 2952 | | NULL |
| 10 | root | 10.244.1.1:64022 | test | Query | 0 | starting | show processlist |
| 19 | root | 10.244.1.1:54823 | test | Sleep | 496 | | NULL |
| 20 | root | 10.244.1.1:54824 | test | Sleep | 648 | | NULL |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
5 rows in set (0.00 sec)
复制代码
我们发现最大连接控制在了10个, 执行完之后还有5个连接在保持着
这里有一个很重要的问题,就是连接池的过期时间
0x4 深入分析 我们把 db.SetConnMaxLifetime(15 * time.Second), 连接池连接的生命周期设置为 15秒, 我们会发现15秒之后,连接池的连接都会断掉
mysql> show processlist;
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
| 4 | event_scheduler | localhost | NULL | Daemon | 4987 | Waiting on empty queue | NULL |
| 9 | root | 10.244.1.1:64000 | test | Sleep | 3998 | | NULL |
| 10 | root | 10.244.1.1:64022 | test | Query | 0 | starting | show processlist |
| 19 | root | 10.244.1.1:54823 | test | Sleep | 1542 | | NULL |
| 20 | root | 10.244.1.1:54824 | test | Sleep | 1694 | | NULL |
+----+-----------------+------------------+------+---------+------+------------------------+------------------+
5 rows in set (0.00 sec)
复制代码
30秒之后再次查询数据库
time.Sleep(30 * time.Second)
rows, err := db.Query("select name from users")
fmt.Println("err -----", err)
defer rows.Close()
for rows.Next(){
var name string
rows.Scan(&name)
fmt.Println("name---", name)
}
复制代码
这个时候发现程序会重新发起新的db连接
总结:
mysql服务端的连接生命周期
还有一种请况就是,我们的程序的连接池生命周期设置大于mysql服务器的生命周期设置, 这个时候就会有一种请况,假如我们重复用连接池的连接,会产生 连接错误的问题,解决方法有两种:
可以在程序里面设置生命周期时间小于mysql服务端的连接生命周期时间就可以了
增加程序的重连(keepalive)机制,就是定时发送一个连接包服务端
关于第2点我们我们以后可以再发散来说,一般如果允许的话,用第一种方式即可。
mysql> show variables like 'mysqlx_wait_timeout';
+---------------------+-------+
| Variable_name | Value |
+---------------------+-------+
| mysqlx_wait_timeout | 28800 |
+---------------------+-------+
1 row in set (0.00 sec)
复制代码