如何同步慢速计算并进行缓存?

huangapple go评论86阅读模式
英文:

How to syncronize slow calculation and cache it?

问题

在golang后端中,我想要为多个客户端提供一个值,我们称之为score。score会随时间变化,它的计算速度很慢。计算不依赖于先前的结果。当没有客户端时,我不想计算它。因此,计算应该只在请求时进行。但还有另一个事实-在5秒的时间段内,score不能改变。所以我尝试了不同的方法,每种方法都有其缺点:

1)在没有客户端的情况下进行昂贵的计算:

var score interface{}

// 在单独的goroutine中运行
func calculateScorePeriodically() {
	for{
		select{
		case <-time.After(5*time.Second):
			score = calculateScoreExpensiveAndSlow()
		}
	}
}

func serveScore(w http.ResponseWriter, r* http.Request) {
    b, _ := json.Marshal(score)
	w.Write(b)
}

2)阻塞所有客户端进行长时间的计算(实际上可能只向它们提供旧数据)。而且你不能将if移到互斥锁之外,因为这样多个客户端可能同时进入计算块,并且会按顺序而不是在5秒间隔内进行计算:


var (
	score interface{}
	mutex sync.Mutex
	updatedAt time.Time
)

func getCachedScore() float64 {
	mutex.Lock()
	defer mutex.Unlock()
	currentTime := time.Now()
	if currentTime.Sub(updatedAt) < 5*time.Second {
		return score
	}
	updatedAt = currentTime
	score = calculateScoreExpensiveAndSlow()
	return score
}

func serveScore(w http.ResponseWriter, r* http.Request) {
    b, _ := json.Marshal(getCachedScore())
	w.Write(b)
}

如何解决上述两个缺点?

PS. 我认为这是一个通用的问题,并且有一个模式-它有一个特殊的名称吗?

英文:

In a golang backend I want to serve a value to a multiple clients, lets call it score. Score is changing with time, and its calculation is slow. Calculation does not depend on previous results. When there are no clients I dont want to calculate it at all. So calculation should happen only on request. But also there is another fact - score cannot change within 5 seconds period. So i tried different aproaches and everything has its drawbacks:

  1. Does expensive calculation in absense of clients:
var score interface{}

// run in a separate goroutine
func calculateScorePeriodically() {
	for{
		select{
		case &lt;-time.After(5*time.Second):
			score = calculateScoreExpensiveAndSlow()
		}
	}
}

func serveScore(w http.ResponseWriter, r* http.Request) {
    b, _ := json.Marshal(score)
	w.Write(b)
}
  1. Blocks all clients for a long calculation period (but actually may just serve old data to them). And you cannot move if outside a mutex, because then multiple client may enter calculation block simultaneously and would do calculation not within 5 seconds interval but sequentially:

var (
	score interface{}
	mutex sync.Mutex
	updatedAt time.Time
)

func getCachedScore() float64 {
	mutex.Lock()
	defer mutex.Unlock()
	currentTime := time.Now()
	if currentTime.Sub(updatedAt) &lt; 5*time.Second {
		return score
	}
	updatedAt = currentTime
	score = calculateScoreExpensiveAndSlow()
	return score
}

func serveScore(w http.ResponseWriter, r* http.Request) {
    b, _ := json.Marshal(getCachedScore())
	w.Write(b)
}

How to solve both of above drawbacks?

PS. i think this is a generic problem, and a pattern - does it have a special name?

答案1

得分: 3

可能有多个解决方案。一个简单的解决方案是为计算指定一个专用的goroutine,你可以通过在通道上发送一个值来发出重新计算的信号。发送可以是非阻塞的,所以如果计算正在进行中,什么都不会发生。

这是一个可重用的缓存实现:

type cache struct {
    mu         sync.RWMutex
    value      interface{}
    updated    time.Time
    calcCh     chan struct{}
    expiration time.Duration
}

func NewCache(calc func() interface{}, expiration time.Duration) *cache {
    c := &cache{
        value:      calc(),
        updated:    time.Now(),
        calcCh:     make(chan struct{}),
        expiration: expiration,
    }

    go func() {
        for range c.calcCh {
            v := calc()

            c.mu.Lock()
            c.value, c.updated = v, time.Now()
            c.mu.Unlock()
        }
    }()

    return c
}

func (c *cache) Get() (value interface{}, updated time.Time) {
    c.mu.RLock()
    value, updated = c.value, c.updated
    c.mu.RUnlock()

    if time.Since(updated) > c.expiration {
        // 触发新的计算(将在另一个goroutine中进行)。
        // 如果计算正在进行中,进行非阻塞发送将不会产生任何效果
        select {
        case c.calcCh <- struct{}{}:
        default:
        }
    }

    return
}

func (c *cache) Stop() {
    close(c.calcCh)
}

注意:Cache.Stop()用于停止后台goroutine。在调用Cache.Stop()之后,不得调用Cache.Get()

在你的情况下使用它:

func getCachedScore() interface{} {
    // ...
}

var scoreCache = NewCache(getCachedScore, 5*time.Second)

func serveScore(w http.ResponseWriter, r *http.Request) {
    score, _ := scoreCache.Get()
    b, _ := json.Marshal(score)
    w.Write(b)
}
英文:

There may be multiple solutions. A simple solution is to have a designated goroutine for calculation, to which you can signal a need for recalculation by sending a value on a channel. The send may be non-blocking, so if a calculation is in progress, nothing will happen.

Here's a reusable cache implementation:

type cache struct {
	mu      sync.RWMutex
	value   interface{}
	updated time.Time

	calcCh     chan struct{}
	expiration time.Duration
}

func NewCache(calc func() interface{}, expiration time.Duration) *cache {
	c := &amp;cache{
		value:   calc(),
		updated: time.Now(),
		calcCh:  make(chan struct{}),
	}

	go func() {
		for range c.calcCh {
			v := calc()

			c.mu.Lock()
			c.value, c.updated = v, time.Now()
			c.mu.Unlock()
		}
	}()

	return c
}

func (c *cache) Get() (value interface{}, updated time.Time) {
	c.mu.RLock()
	value, updated = c.value, c.updated
	c.mu.RUnlock()

	if time.Since(updated) &gt; c.expiration {
		// Trigger a new calculation (will happen in another goroutine).
		// Do non-blocking send, if a calculation is in progress,
        // this will have no effect
		select {
		case c.calcCh &lt;- struct{}{}:
		default:
		}
	}

	return
}

func (c *cache) Stop() {
	close(c.calcCh)
}

Note: Cache.Stop() is to stop the background goroutine. After calling Cache.Stop(), Cache.Get() must not be called.

Using it for your case:

func getCachedScore() interface{} {
    // ...
}

var scoreCache = NewCache(getCachedScore, 5*time.Second)

func serveScore(w http.ResponseWriter, r* http.Request) {
    score, _ := scoreCache.Get()
    b, _ := json.Marshal(score)
    w.Write(b)
}

答案2

得分: 0

这是我实现的代码,与icza的答案相关,但有一些更多的功能:

package common

import (
	"context"
	"sync/atomic"
	"time"
)

type (
	CachedUpdater func() interface{}
	ChanStruct    chan struct{}
)

type Cached struct {
	value        atomic.Value  // 保存缓存值的interface{}
	updatedAt    atomic.Value  // 保存最后一次更新开始的时间time.Time
	updatePeriod time.Duration // 控制更新之间的最小时间间隔
	needUpdate   ChanStruct
}

//cachedUpdater是一个用户提供的长时间昂贵计算的函数,用于获取当前状态
func MakeCached(ctx context.Context, updatePeriod time.Duration, cachedUpdater CachedUpdater) *Cached {
	v := &Cached{
		updatePeriod: updatePeriod,
		needUpdate:   make(ChanStruct),
	}
	//v.updatedAt.Store(time.Time{}) // "从未更新过",但时间不应该是nil接口
	v.doUpdate(time.Now(), cachedUpdater)
	go v.updaterController(ctx, cachedUpdater)
	return v
}

//客户端将立即获取缓存值,并在需要时触发更新,如果值已过期
func (v *Cached) Get() interface{} {
	if v.IsExpired(time.Now()) {
		v.RequestUpdate()
	}
	return v.value.Load()
}

//updateController goroutine可以通过取消提供给maker的上下文或关闭chan来终止
func (v *Cached) Stop() {
	close(v.needUpdate)
}

//如果值已过期且更新函数可能尚未被调用,则返回true
func (v *Cached) IsExpired(currentTime time.Time) bool {
	updatedAt := v.updatedAt.Load().(time.Time)
	return currentTime.Sub(updatedAt) > v.updatePeriod
}

//请求updaterController执行更新,使用非阻塞发送到无缓冲chan。在最近更新了值的情况下,控制器可以决定不进行更新
func (v *Cached) RequestUpdate() bool {
	select {
	case v.needUpdate <- struct{}{}:
		return true
	default:
		return false
	}
}

func (v *Cached) updaterController(ctx context.Context, cachedUpdater CachedUpdater) {
	for {
		select {
		case <-ctx.Done():
			return
		case _, ok := <-v.needUpdate:
			if !ok {
				return
			}
			currentTime := time.Now()
			if !v.IsExpired(currentTime) {
				continue
			}
			v.doUpdate(currentTime, cachedUpdater)
		}
	}
}

func (v *Cached) doUpdate(currentTime time.Time, cachedUpdater CachedUpdater) {
	v.updatedAt.Store(currentTime)
	v.value.Store(cachedUpdater())
}

英文:

This is what i've implemented, correlates with icza's answer, but has some more features:

package common
import (
&quot;context&quot;
&quot;sync/atomic&quot;
&quot;time&quot;
)
type (
CachedUpdater func() interface{}
ChanStruct    chan struct{}
)
type Cached struct {
value        atomic.Value  // holds the cached value&#39;s interface{}
updatedAt    atomic.Value  // holds time.Time, time when last update sequence was started at
updatePeriod time.Duration // controls minimal anount of time between updates
needUpdate   ChanStruct
}
//cachedUpdater is a user-provided function with long expensive calculation, that gets current state
func MakeCached(ctx context.Context, updatePeriod time.Duration, cachedUpdater CachedUpdater) *Cached {
v := &amp;Cached{
updatePeriod: updatePeriod,
needUpdate:   make(ChanStruct),
}
//v.updatedAt.Store(time.Time{}) // &quot;was never updated&quot;, but time should never be nil interface
v.doUpdate(time.Now(), cachedUpdater)
go v.updaterController(ctx, cachedUpdater)
return v
}
//client will get cached value immediately, and optionally may trigger an update, if value is outdated
func (v *Cached) Get() interface{} {
if v.IsExpired(time.Now()) {
v.RequestUpdate()
}
return v.value.Load()
}
//updateController goroutine can be terminated both by cancelling context, provided to maker, or by closing chan
func (v *Cached) Stop() {
close(v.needUpdate)
}
//returns true if value is outdated and updater function was likely not called yet
func (v *Cached) IsExpired(currentTime time.Time) bool {
updatedAt := v.updatedAt.Load().(time.Time)
return currentTime.Sub(updatedAt) &gt; v.updatePeriod
}
//requests updaterController to perform update, using non-blocking send to unbuffered chan. controller can decide not to update in case if it has recently updated value
func (v *Cached) RequestUpdate() bool {
select {
case v.needUpdate &lt;- struct{}{}:
return true
default:
return false
}
}
func (v *Cached) updaterController(ctx context.Context, cachedUpdater CachedUpdater) {
for {
select {
case &lt;-ctx.Done():
return
case _, ok := &lt;-v.needUpdate:
if !ok {
return
}
currentTime := time.Now()
if !v.IsExpired(currentTime) {
continue
}
v.doUpdate(currentTime, cachedUpdater)
}
}
}
func (v *Cached) doUpdate(currentTime time.Time, cachedUpdater CachedUpdater) {
v.updatedAt.Store(currentTime)
v.value.Store(cachedUpdater())
}

huangapple
  • 本文由 发表于 2021年10月13日 18:41:39
  • 转载请务必保留本文链接:https://go.coder-hub.com/69554037.html
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定