一、简介
二、带着问题
三、源码
    3.1 代码结构
    3.2 核心代码

源码学习 https://github.com/muesli/cache2go

一、简介

cache2go是一个简单的缓存库，大概500行代码。

• 支持并发安全
• 支持过期清除
• 支持访问计数
• 配置回调函数

二、带着问题

1. 并发的使用
2. 过期清除如何实现的
3. 回掉函数的配置

三、源码

3.1 代码结构

• Cache.go 缓存库
• CacheTable.go 缓存表
• CacheItem.go 缓存项

3.2 核心代码

1. CacheItem

type CacheItem struct {
    sync.RWMutex

    key interface{}
    data interface{}
    
    //每个item都有一个访问时间和访问次数，以及存活时间。
    lifeSpan time.Duration
    createdOn time.Time
    accessedOn time.Time
    accessCount int64

    aboutToExpire []func(key interface{})   //支持回调函数
}

每次访问item时更新accessedOn和accessCount；
每次清理时，判断now-accessedOn是否已大于存活时间，来决定清理；

// KeepAlive marks an item to be kept for another expireDuration period.
func (item *CacheItem) KeepAlive() {
    item.Lock()
    defer item.Unlock()
    item.accessedOn = time.Now()
    item.accessCount++
}

2. CacheTable

// CacheTable is a table within the cache
type CacheTable struct {
    sync.RWMutex

    // The table's name.
    name string
    // All cached items.
    items map[interface{}]*CacheItem

    // Timer responsible for triggering cleanup.
    cleanupTimer *time.Timer   //以表的维度，来清理过期数据
    // Current timer duration.
    cleanupInterval time.Duration
    
    // Callback method triggered when trying to load a non-existing key.
    loadData func(key interface{}, args ...interface{}) *CacheItem
    // Callback method triggered when adding a new item to the cache.
    addedItem []func(item *CacheItem)
    // Callback method triggered before deleting an item from the cache.
    aboutToDeleteItem []func(item *CacheItem)
}

Add
回调函数的调用；
每次添加，如果存活时间<清理周期，会触发一次清理；

func (table *CacheTable) Add(key interface{}, lifeSpan time.Duration, data interface{}) *CacheItem {
    item := NewCacheItem(key, lifeSpan, data)

    // Add item to cache.
    table.Lock()
    table.items[item.key] = item

    // Cache values so we don't keep blocking the mutex.
    expDur := table.cleanupInterval
    addedItem := table.addedItem
    table.Unlock()

    // Trigger callback after adding an item to cache. 回掉函数的触发
    if addedItem != nil {
        for _, callback := range addedItem {
            callback(item)
        }
    }

    // If we haven't set up any expiration check timer or found a more imminent item.
    if item.lifeSpan > 0 && (expDur == 0 || item.lifeSpan < expDur) {
        table.expirationCheck()
    }

    return item
}

Delete

// Delete an item from the cache.
func (table *CacheTable) Delete(key interface{}) (*CacheItem, error) {
    table.Lock()
    defer table.Unlock()

    r, ok := table.items[key]
    if !ok {
        return nil, ErrKeyNotFound
    }

    // Cache value so we don't keep blocking the mutex.
    aboutToDeleteItem := table.aboutToDeleteItem
    table.Unlock()

    // Trigger callbacks before deleting an item from cache.
    if aboutToDeleteItem != nil {
        for _, callback := range aboutToDeleteItem {
            callback(r)
        }
    }

    r.RLock()
    defer r.RUnlock()
    if r.aboutToExpire != nil {
        for _, callback := range r.aboutToExpire {
            callback(key)
        }
    }

    table.Lock()
    delete(table.items, key)

    return r, nil
}

Value
如果存在，更新item的访问；
如果不存在，支持自定义加载函数；

func (table *CacheTable) Value(key interface{}, args ...interface{}) (*CacheItem, error) {
    table.RLock()
    r, ok := table.items[key]
    loadData := table.loadData
    table.RUnlock()

    if ok {
        // Update access counter and timestamp.
        r.KeepAlive()
        return r, nil
    }

    // Item doesn't exist in cache. Try and fetch it with a data-loader.
    if loadData != nil {
        item := loadData(key, args...)
        if item != nil {
            table.Add(key, item.lifeSpan, item.data)
            return item, nil
        }

        return nil, ErrKeyNotFoundOrLoadable
    }

    return nil, ErrKeyNotFound
}

定期清理
清理实际上是遍历每个item，判断它的过期时间，过期就删除；
清理的触发：一是每次添加时，二是每次清理之后创建一个计时器；
非固定周期的清理，基于item的存活时间来调整；

// Expiration check loop, triggered by a self-adjusting timer.
func (table *CacheTable) expirationCheck() {
    table.Lock()
    if table.cleanupTimer != nil {
        table.cleanupTimer.Stop()
    }
    if table.cleanupInterval > 0 {
        table.log("Expiration check triggered after", table.cleanupInterval, "for table", table.name)
    } else {
        table.log("Expiration check installed for table", table.name)
    }

    // To be more accurate with timers, we would need to update 'now' on every
    // loop iteration. Not sure it's really efficient though.
    now := time.Now()
    smallestDuration := 0 * time.Second
    for key, item := range table.items {
        // Cache values so we don't keep blocking the mutex.
        item.RLock()
        lifeSpan := item.lifeSpan
        accessedOn := item.accessedOn
        item.RUnlock()

        if lifeSpan == 0 {
            continue
        }
        if now.Sub(accessedOn) >= lifeSpan {
            // Item has excessed its lifespan.
            table.deleteInternal(key)
        } else {
            // Find the item chronologically closest to its end-of-lifespan.
            if smallestDuration == 0 || lifeSpan-now.Sub(accessedOn) < smallestDuration {
                smallestDuration = lifeSpan - now.Sub(accessedOn)
            }
        }
    }

    // Setup the interval for the next cleanup run.
    table.cleanupInterval = smallestDuration
    if smallestDuration > 0 {
        table.cleanupTimer = time.AfterFunc(smallestDuration, func() {
            go table.expirationCheck()
        })
    }
    table.Unlock()
}

四、思考

• 锁的类型：创建后一成不变的可以不加锁，只读加读锁，写入加写锁；
• 锁的范围：加锁获取数据后，后续操作缓存数据不再加锁；
• 锁的粒度：对数据表的操作，获取表锁。对于某个item的操作，获取item的锁；