前言
最近一直在看 Kotlin 协程,因为以前也写过 Golang,所以试着去对比了一下,发现了很多好玩的事情。
一个 Kotlin 的小例子
试问下面一段代码,执行结果是什么呢?
fun main() = runBlocking {
//创建自定义线程池
val coroutineDispatcher = Executors.newFixedThreadPool(1).asCoroutineDispatcher()
val name = Thread.currentThread().name
println("main start thread-id = $name")
for (i in 0..2) {
launch(coroutineDispatcher) {
while (true) {
val j = i + 1
val filename = "/Users/xxxx/Desktop/mapping$j.txt"
val file = File(filename)
val contents = file.readText()
val name = Thread.currentThread().name
println("thread-id = $name, do work $i")
}
}
}
println("main end thread-id = $name")
}
这段代码不说 runBlocking 的情况下,启动了3个协程,且执行在一个单线程的线程池中且都是 死循环
,在死循环中读取 mapping.txt(一个大概 30M 的文件)。
我的机器是 Mac
kotlin 的版本号如下:
kotlinx-coroutines-core: 1.4.2
kotlin-stdlib: 1.4.0
结果如下:
main start thread-id = main
main end thread-id = main
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
thread-id = pool-1-thread-1, do work 0
// 无限多···
可以看到,协程的载体是 线程
,协程的代码逻辑可以理解为一个 任务
,多线程多任务且没有特殊的阻塞任务(死循环等)的情况下,任务早晚是可以被执行的。
但是这段逻辑是死循环,且是单线程,以致于 i = 0 时,创建的协程任务占用住了当前线程,第二个协程任务 无法被执行
。
所以看得出,对于 Kotlin 的协程任务因为其是 非抢占式
的,是存在不被执行的情况的(协程被 饿死了
)。
一个 Golang 的小例子
试问下面一段代码,执行结果是什么呢?
package main
import (
"fmt"
"io/ioutil"
"os"
"runtime"
"syscall"
"time"
)
func main() {
runtime.GOMAXPROCS(1)
for i := 0; i < 3; i++ {
go func(j int) {
for {
file, err := os.Open(fmt.Sprint("/Users/xxxx/Desktop/mapping", (j + 1), ".txt"))
if err != nil {
panic(err)
}
defer file.Close()
ioutil.ReadAll(file)
tid := gettid()
fmt.Printf("thread-id = %d, do work %d\n", tid, j)
}
}(i)
}
time.Sleep(1 * time.Second)
fmt.Println("finish work")
}
func gettid() (n uint64) {
r0, _, _ := syscall.RawSyscall(syscall.SYS_THREAD_SELFID, 0, 0, 0)
n = uint64(r0)
return n
}
这段代码也很简单,runtime.GOMAXPROCS(1)
设置了 P 只有一个,Golang 中的 go
关键字会启动一个协程
,其余逻辑和 Kotlin 代码的逻辑基本一致,都是启动三个协程读3个文件,那结果跟是不是像 Kotlin 一样,阻塞在第一个任务
呢?
结果如下:
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 0
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 1
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 2
thread-id = 2873233, do work 2
thread-id = 2873233, do work 2
thread-id = 2873233, do work 0
thread-id = 2873233, do work 0
thread-id = 2873233, do work 1
thread-id = 2873233, do work 1
finish work
是不是很神奇!结果 不一致
了!并没有因为死循环卡死在第一个任务上,且在1s后主线程结束、程序退出(因为代码里,主线程只是 sleep 了1s)。
这里就很神奇了,通过 gettid()
打印的线程 id 显示,这3个协程运行在了同一个线程之上,居然可以打破死循环···
gettid 的小插曲
因为电脑是 MacOSX,一开始使用 Golang 自己提供的 syscall.Gettid()
方法来获取 tid,结果发现报错:undefined Gettid()
,似乎是 MacOSX 并没有实现这个方法,所以问了大哥得到了另一种方法
func gettid() (n uint64) {
r0, _, _ := syscall.RawSyscall(syscall.SYS_THREAD_SELFID, 0, 0, 0)
n = uint64(r0)
return n
}
疑问和总结
总结
首先,确认了 Kotlin 的协程似乎并没有任务调度上的优化,只是在线程池中执行任务,存在协程饿死的情况。
其次,Golang 的协程存在一些神奇的优化,即使是死循环,任务也可以交替执行。
疑问
一开始以为是我的 go 版本比较高,因为 go1.14
加入了抢占式调度,我以为是这个的原因。后来发现上面的例子在 go1.13
乃至 go1.10
表现都一致,所以这里的真实原因是什么呢?估计是 netpoller
的作用吧。