错误使用map引发的血案
前言
最近业务中,同事使用map
来接收返回的结果,使用waitGroup
来并发的处理执行返回的结果,结果上线之后,直接崩了。
日志大量的数据库缓存池连接失败
{"ecode":-500,"message":"timed out while checking out a connection from connection pool"}
{"ecode":-500,"message":"connection(xxxxxxxxx:xxxxx) failed to write: context deadline exceeded"}
场景复原
先来看来伪代码
一个全局的map
,然后WaitGroup
开启一组协程并发的读写数据,写入内容到map
中。
package main
import (
"fmt"
"sync"
"time"
)
var count = 300
func main() {
var data = make(map[int]string, count)
var wg sync.WaitGroup
for i := 0; i < count; i++ {
wg.Add(1)
go func(i int) {
defer wg.Done()
time.Sleep(time.Second * 1)
mockSqlPool()
data[i] = "test"
}(i)
}
fmt.Println("-----------WaitGroup执行结束了-----------")
wg.Wait()
}
// 模拟数据库的连接和释放
func mockSqlPool() {
defer fmt.Println("关闭pool")
fmt.Println("我是pool")
}
运行的输出
...
我是pool
关闭pool
我是pool
fatal error: 关闭pool
concurrent map writes
我是pool
goroutine 56 [running]:
runtime.throw(0x10d3923, 0x15)
/usr/local/go/src/runtime/panic.go:774 +0x72 fp=0xc00023cf20 sp=0xc00023cef0 pc=0x10298d2
runtime.mapassign_fast64(0x10b29e0, 0xc000066180, 0x16, 0x0)
/usr/local/go/src/runtime/map_fast64.go:101 +0x350 fp=0xc00023cf60 sp=0xc00023cf20 pc=0x100f620
main.main.func1(0xc00008c004, 0xc000066180, 0x16)
/Users/yj/Go/src/Go-POINT/map/main.go:23 +0x87 fp=0xc00023cfc8 sp=0xc00023cf60 pc=0x109a297
runtime.goexit()
/usr/local/go/src/runtime/asm_amd64.s:1357 +0x1 fp=0xc00023cfd0 sp=0xc00023cfc8 pc=0x1053a51
created by main.main
/Users/yj/Go/src/Go-POINT/map/main.go:18 +0xbb
goroutine 1 [semacquire]:
sync.runtime_Semacquire(0xc00008c004)
/usr/local/go/src/runtime/sema.go:56 +0x42
sync.(*WaitGroup).Wait(0xc00008c004)
/usr/local/go/src/sync/waitgroup.go:130 +0x64
main.main()
/Users/yj/Go/src/Go-POINT/map/main.go:27 +0x138
goroutine 22 [semacquire]:
internal/poll.runtime_Semacquire(0xc00008606c)
/usr/local/go/src/runtime/sema.go:61 +0x42
internal/poll.(*fdMutex).rwlock(0xc000086060, 0xc000030500, 0x1097137)
/usr/local/go/src/internal/poll/fd_mutex.go:154 +0xad
internal/poll.(*FD).writeLock(...)
/usr/local/go/src/internal/poll/fd_mutex.go:239
internal/poll.(*FD).Write(0xc000086060, 0xc000226030, 0xb, 0x10, 0x0, 0x0, 0x0)
/usr/local/go/src/internal/poll/fd_unix.go:255 +0x5e
os.(*File).write(...)
/usr/local/go/src/os/file_unix.go:276
os.(*File).Write(0xc000084008, 0xc000226030, 0xb, 0x10, 0xc0000306b0, 0x103d37e, 0xc00000c060)
/usr/local/go/src/os/file.go:153 +0x77
fmt.Fprintln(0x10ec4e0, 0xc000084008, 0xc000030730, 0x1, 0x1, 0x10459b6, 0xc00000c060, 0x3)
/usr/local/go/src/fmt/print.go:265 +0x8b
fmt.Println(...)
/usr/local/go/src/fmt/print.go:274
main.mockSqlPool()
/Users/yj/Go/src/Go-POINT/map/main.go:35 +0x104
main.main.func1(0xc00008c004, 0xc000066180, 0x4)
/Users/yj/Go/src/Go-POINT/map/main.go:21 +0x63
created by main.main
/Users/yj/Go/src/Go-POINT/map/main.go:18 +0xbb
goroutine 192 [semacquire]:
internal/poll.runtime_Semacquire(0xc00009e06c)
/usr/local/go/src/runtime/sema.go:61 +0x42
internal/poll.(*fdMutex).rwlock(0xc00009e060, 0x10fae00, 0xc00023ad00)
/usr/local/go/src/internal/poll/fd_mutex.go:154 +0xe9
internal/poll.(*FD).writeLock(...)
/usr/local/go/src/internal/poll/fd_mutex.go:239
internal/poll.(*FD).Write(0xc00009e060, 0xc000246100, 0xb, 0x10, 0x0, 0x0, 0x0)
/usr/local/go/src/internal/poll/fd_unix.go:255 +0x6f
os.(*File).write(...)
/usr/local/go/src/os/file_unix.go:276
os.(*File).Write(0xc00009c008, 0xc000246100, 0xb, 0x10, 0xc000124580, 0x40, 0x0)
/usr/local/go/src/os/file.go:153 +0xa7
fmt.Fprintln(0x1158520, 0xc00009c008, 0xc00014d728, 0x1, 0x1, 0x107e3e6, 0xc0000d8100, 0x16)
/usr/local/go/src/fmt/print.go:265 +0xb3
fmt.Println(...)
/usr/local/go/src/fmt/print.go:274
main.mockSqlPool()
/Users/yj/Go/src/Go-POINT/map/main.go:35 +0x129
main.main.func1(0xc0000a0004, 0xc000088180, 0x8f)
/Users/yj/Go/src/Go-POINT/map/main.go:21 +0x75
created by main.main
/Users/yj/Go/src/Go-POINT/map/main.go:18 +0x102
goroutine 193 [semacquire]:
internal/poll.runtime_Semacquire(0xc00009e06c)
/usr/local/go/src/runtime/sema.go:61 +0x42
internal/poll.(*fdMutex).rwlock(0xc00009e060, 0x10fae00, 0xc000286410)
/usr/local/go/src/internal/poll/fd_mutex.go:154 +0xe9
internal/poll.(*FD).writeLock(...)
/usr/local/go/src/internal/poll/fd_mutex.go:239
internal/poll.(*FD).Write(0xc00009e060, 0xc0000a01a0, 0xb, 0x10, 0x0, 0x0, 0x0)
/usr/local/go/src/internal/poll/fd_unix.go:255 +0x6f
os.(*File).write(...)
/usr/local/go/src/os/file_unix.go:276
os.(*File).Write(0xc00009c008, 0xc0000a01a0, 0xb, 0x10, 0xc0001245c0, 0x40, 0x0)
/usr/local/go/src/os/file.go:153 +0xa7
fmt.Fprintln(0x1158520, 0xc00009c008, 0xc00014df28, 0x1, 0x1, 0x107e3e6, 0xc0000d8100, 0x17)
/usr/local/go/src/fmt/print.go:265 +0xb3
fmt.Println(...)
/usr/local/go/src/fmt/print.go:274
main.mockSqlPool()
/Users/yj/Go/src/Go-POINT/map/main.go:35 +0x129
main.main.func1(0xc0000a0004, 0xc000088180, 0x90)
/Users/yj/Go/src/Go-POINT/map/main.go:21 +0x75
created by main.main
/Users/yj/Go/src/Go-POINT/map/main.go:18 +0x102
goroutine 194 [semacquire]:
internal/poll.runtime_Semacquire(0xc00009e06c)
/usr/local/go/src/runtime/sema.go:61 +0x42
internal/poll.(*fdMutex).rwlock(0xc00009e060, 0x10fae00, 0xc00023add0)
/usr/local/go/src/internal/poll/fd_mutex.go:154 +0xe9
internal/poll.(*FD).writeLock(...)
/usr/local/go/src/internal/poll/fd_mutex.go:239
internal/poll.(*FD).Write(0xc00009e060, 0xc000246110, 0xb, 0x10, 0x0, 0x0, 0x0)
/usr/local/go/src/internal/poll/fd_unix.go:255 +0x6f
os.(*File).write(...)
/usr/local/go/src/os/file_unix.go:276
os.(*File).Write(0xc00009c008, 0xc000246110, 0xb, 0x10, 0xc000124600, 0x40, 0x0)
/usr/local/go/src/os/file.go:153 +0xa7
fmt.Fprintln(0x1158520, 0xc00009c008, 0xc000146728, 0x1, 0x1, 0x107e3e6, 0xc0000d8100, 0x18)
/usr/local/go/src/fmt/print.go:265 +0xb3
fmt.Println(...)
/usr/local/go/src/fmt/print.go:274
main.mockSqlPool()
/Users/yj/Go/src/Go-POINT/map/main.go:35 +0x129
main.main.func1(0xc0000a0004, 0xc000088180, 0x91)
/Users/yj/Go/src/Go-POINT/map/main.go:21 +0x75
created by main.main
/Users/yj/Go/src/Go-POINT/map/main.go:18 +0x102
会发现很多goroutine
处于semacquire
状态,说明这些goroutine
正在等待被信号量唤醒。但是这时候waitGroup
已经因为panic
退出了,这些goroutine
不会在通过waitGroup.Done()
退出,造成这些goroutine
一直阻塞到这,最后的结果就是这些goroutine
占用的数据库连接不能被释放。
关于waitGroup
的信号量
整个Wait()
会被runtime_Semacquire
阻塞,直到等到全部退出的信号量;
Done()
会在最后一次的时候通过runtime_Semrelease
发出取消阻塞的信号,然后被runtime_Semacquire
阻塞的Wait()
就可以退出了;
上面涉及到的几种状态
- semacquire 状态,这个状态表示等待调用
- Waiting 等待状态。线程在等待某件事的发生。例如等待网络数据、硬盘;调用操作系统 API;等待内存同步访问条件 ready,如 atomic, mutexes
- Runnable 就绪状态。只要给 CPU 资源我就能运行
原因
上面的错误原因有两个:
-
1、
map
不是并发安全,并发写的时候会触发panic
; -
2、避免在循环中连接数据库;
参考
【map 并发崩溃一例】https://xargin.com/map-concurrent-throw/