Golang系列(36)-协程池设计与实现

2022-12-01

字数统计: 1.4k字 | 阅读时长≈ 7分

golang里可以通过关键字go很简单的开启一个协程，但是如果开启协程的速度超过回收的速度，就会导致协程泄露。本文就介绍如何通过设计一个安全的协程池来避免出现协程泄露的问题。

协程池

一个包含多个工作协程，且能同时支持并行执行任务的数据结构

实现

协程池分两种类型，一种是worker数量固定的协程池，一种是worker数量可自动伸缩的协程池

类型	协程数	特点	适用场景
FixedGoroutinePool	固定	不需要重复创建和回收协程，性能好。如果协程数分配过多，会导致资源浪费	适用于并发较低的应用场景，如定时任务
ScalableGoroutinePool	不固定	保持最低协程数量，协程数会随着并发增加而自动伸缩，在性能和资源之间有所平衡	适用于并发较高的场景，如HTTP异步请求

接口定义：

type GoroutinePool interface {
	// Schedule 调度worker执行任务
	Schedule(task func())

	// Stop 停止工作
	Stop()
}

以下是具体实现：

固定协程池

实现一个固定数量的协程池


// FixedGoroutinePool 固定数量的协程池
type FixedGoroutinePool struct {
	task  chan f
	block bool // 控制当任务已满时是否阻塞

	once sync.Once
	done chan struct{}
}

func (pool *FixedGoroutinePool) Schedule(task func()) {
	if pool.block {
		// 阻塞等待任务被执行
		pool.task <- task
		return
	}

	// 非阻塞模式
	select {
	case pool.task <- task:
	default:
	}

}

func (pool *FixedGoroutinePool) Stop() {
	pool.once.Do(func() {
		close(pool.done)
	})
}

func (pool *FixedGoroutinePool) run() {
	for {
		select {
		case <-pool.done:
			return
		case fn := <-pool.task:
			fn()
		}
	}
}

type FixedOptions struct {
	Num   int  // 协程数
	Block bool // 任务已满时是否需要阻塞
}
type fixedOption func(options *FixedOptions)

func NewFixedGoroutinePool(options ...fixedOption) *FixedGoroutinePool {
	defaultOptions := FixedOptions{Num: 100, Block: true}
	for _, setter := range options {
		setter(&defaultOptions)
	}
	pool := &FixedGoroutinePool{
		task:  make(chan f, defaultOptions.Num),
		block: defaultOptions.Block,
	}

	// 直接启动固定数量的协程
	for i := 0; i < defaultOptions.Num; i++ {
		go pool.run()
	}

	return pool
}

浮动协程池

支持自动伸缩的协程池


type ScalableGroutinePool struct {
	options Options
	// capacity of the pool.
	capacity int32

	// running is the number of the currently running goroutines.
	running int32

	// freeSignal is used to notice pool there are available
	// workers which can be sent to work.
	freeSignal chan struct{}

	// workers is a slice that store the available workers.
	workers []*Worker

	// stopped is used to check pool running status
	stopped int32

	// lock for synchronous operation
	lock sync.Mutex

	once sync.Once
	done chan struct{}
}

type Options struct {
	Max     int32 // maximum number of workers
	Idle    int   // idle number of workers
	Block   bool
	Timeout time.Duration // quit time for worker
}

type Option func(options *Options)

func NewGoroutinePool(opts ...Option) *ScalableGroutinePool {
	defaultOptions := Options{Max: 100, Idle: 10, Timeout: time.Second * 10, Block: true}
	for _, setter := range opts {
		setter(&defaultOptions)
	}
	pool := &ScalableGroutinePool{
		options:    defaultOptions,
		capacity:   defaultOptions.Max,
		running:    0,
		freeSignal: make(chan struct{}, defaultOptions.Max),
		workers:    make([]*Worker, 0, 1024),
		done:       make(chan struct{}),
	}

	// 提前准备空闲协程池
	pool.grow(defaultOptions.Idle)
	go pool.monitor()

	return pool
}

func (p *ScalableGroutinePool) monitor() {
	for {
		select {
		case <-p.done:
			return
		default:
			// 定时缩容
			p.scaleDown()
		}
	}
}

// grow 自动扩容worker数量
func (p *ScalableGroutinePool) grow(n int) {
	for i := 0; i < n; i++ {
		// create instance of Worker
		w := NewWorker(10, func() {
			atomic.AddInt32(&p.running, -1)
		}, p.releaseWorker)

		// push into the slice
		p.workers = append(p.workers, w)
		p.running++
		p.freeSignal <- struct{}{}
	}
}

type f func()

// Stop 停止协程池
func (p *ScalableGroutinePool) Stop() {
	p.once.Do(func() {
		atomic.StoreInt32(&p.stopped, 1)
		close(p.done)
		p.lock.Lock()
		for _, worker := range p.workers {
			worker.Stop()
		}
		p.lock.Unlock()
	})
}

// Schedule 执行任务
func (p *ScalableGroutinePool) Schedule(task func()) {
	// 判断pool是否已关闭
	if atomic.LoadInt32(&p.stopped) == 1 {
		return
	}

	// 调度一个worker来执行任务
	p.acquireWorker().Submit(task, p.options.Block)
}

// acquireWorker 获取一个worker实例
func (p *ScalableGroutinePool) acquireWorker() (w *Worker) {
	p.lock.Lock()
	defer p.lock.Unlock()
	// 查看当前可用worker
	available := len(p.workers)
	if p.running < p.capacity {
		if available == 0 {
			p.grow(p.options.Idle)
		}

		<-p.freeSignal
		w = p.popLastWorker()
		return
	}

	// 当可用worker数为0且协程数达到上限时，
	// 因为此时已被lock住，且无法通过releaseWorker释放，所以会导致死锁
	// 所以这种情况下必须先释放锁
	p.lock.Unlock()
	<-p.freeSignal
	p.lock.Lock()
	w = p.popLastWorker()
	return w
}

// popLastWorker 获取空闲队列里最尾部的一个worker
func (p *ScalableGroutinePool) popLastWorker() (w *Worker) {
	// 取数组最后一个worker
	n := len(p.workers) - 1
	w = p.workers[n]
	p.workers[n] = nil
	p.workers = p.workers[:n]
	return
}

// releaseWorker puts a worker back into free pool, recycling the goroutines.
func (p *ScalableGroutinePool) releaseWorker(worker *Worker) {
	p.lock.Lock()
	p.workers = append(p.workers, worker)
	p.lock.Unlock()
	p.freeSignal <- struct{}{}
}

// scaleDown 缩容
func (p *ScalableGroutinePool) scaleDown() {
	// 根据时间控制每隔一段时间按策略缩容
	time.Sleep(p.options.Timeout)
	p.lock.Lock()
	defer p.lock.Unlock()

	// 低于空闲数量则不缩容
	available := len(p.workers)
	if available <= p.options.Idle {
		return
	}

	num := (available - p.options.Idle) / 4
	for i := 0; i < num; i++ {
		p.workers[i].Stop()
	}
	p.workers = p.workers[num:]
}


type Worker struct {
	task chan f
	once sync.Once
	done chan struct{}
}

func NewWorker(cap int, beforeCloseCallback func(), afterCallback func(w *Worker)) *Worker {
	w := &Worker{
		task: make(chan f, cap),
		done: make(chan struct{}),
	}
	go w.run(beforeCloseCallback, afterCallback)
	return w
}

func (w *Worker) run(beforeCloseCallback func(), afterCallback func(w *Worker)) {
	var (
		fn     f
		opened bool
	)
	for {
		select {
		case <-w.done:
			beforeCloseCallback()
			return
		case fn, opened = <-w.task:
			if !opened {
				return
			}
			fn()
			//回收复用
			afterCallback(w)
		}
	}
}

// stop this worker.
func (w *Worker) Stop() {
	w.once.Do(func() {
		close(w.done)
		close(w.task)
	})
}

// Submit sends a task to this worker.
func (w *Worker) Submit(task f, block bool) {
	if block {
		select {
		case <-w.done:
		case w.task <- task:
		}
		return
	}

	select {
	case w.task <- task:
	default:
	}
}

本文作者： Hongker
本文链接： https://hongker.github.io/2022/12/01/golang-worker-pool/
版权声明： 本博客所有文章除特别声明外，均采用 MIT 许可协议。转载请注明出处！