A map + linked-list backed LRU cache. Cached items belong to a primary and secondary cache key. This allows you to generate multiple versions of the same object yet purge all variations. For example:
// 100MB
cache := lrucache.New(lrucache.Configure())
cache.Set("video/43", ".json", video1)
cache.Set("video/43", ".xml", video2)
// will remove both entries from the cache
cache.Remove("video/43")
The cache takes some liberties. For example, a delay prevents recently promoted items from being re-promoted (thus minimizing the occurrences of a global write lock on the linked list). Furthermore, the garbage collection keeps some artifacts around. We've found these compromises suitable for caching HTTP responses.
There are now a number of respectable caching libraries for Go, such as GroupCache and Vitess.
The cache is thread-safe.
Start by creating an instance of the cache:
cache := lrucache.New(lrucache.Configure())
The cache size, itemsToPrune and callback can be specific via the configuration object:
cache := lrucache.New(
lrucache.Configure()
.Size(1 * 1024 * 1024 * 1024) //1GB
.ItemsToPrune(5000) //# of items to prune when a GC runs
.Callback(func(){ atomic.AddUint64(&gcs, 1) }) //called whenever a GC runs
)
After creating an instance, you can Get, Set or Remove. Items added to the cache must implement the lrucache.CacheItem interface, which simply defines Debug() []byte and Expiry() time.Time:
type Response struct {
body []byte
statusCode int
}
func (r *Response) Debug() []byte {
// Adds arbitrary data to the output of cache.Debug() for this line item.
// In this case, returning len(body) might be useful, or just do nothing
// if you aren't using Debug
return []byte{}
}
func (r *Response) Expires() time.Time {
return time.Now().Add(time.Hour)
}
Install using the "go get" command:
go get github.com/viki-org/lrucache