Caching via collection

The go-zero microservices framework provides many out-of-the-box tools, good tools can not only improve the performance of the service but also improve the robustness of the code to avoid errors, to achieve a uniform style of code for others to read, etc. This series of articles will introduce the use of the go-zero framework tools and their implementation principles

In-process caching tool collection.Cache#

When doing server development, I believe we will encounter the use of cache, go-zero provides a simple cache package collection.Cache, simple to use as follows

// Initialize the cache, where WithLimit can specify the maximum number of caches
c, err := collection.NewCache(time.Minute, collection.WithLimit(10000))
if err ! = nil {
// Set the cache
c.Set("key", user)
// Get the cache, ok: if it exists
v, ok := c.Get("key")
// Delete the cache
// Get the cache, if the key does not exist, then func will be called to generate the cache
v, err := c.Take("key", func() (interface{}, error) {
return user, nil

The built features implemented by cache include

  • automatic cache expiration, you can specify the expiration time
  • Cache size limit, you can specify the number of caches
  • cache add, delete and change
  • Cache hit rate statistics
  • Concurrency security
  • Cache hitting

Implementation principle. Cache automatic invalidation is managed using TimingWheel

timingWheel, err := NewTimingWheel(time.Second, slots, func(k, v interface{}) {
key, ok := k.(string)
if !ok {

Cache size limit, is the use of LRU elimination policy, when the new cache will check whether the limit has been exceeded, the specific code in the keyLru implementation

func (klru *keyLru) add(key string) {
if elem, ok := klru.elements[key]; ok {
MoveToFront(elem) return
// Add new item
elem := klru.evicts.PushFront(key)
klru.items[key] = elem
// Verify size not exceeded
if klru.evicts.Len() > klru.limit {

Cache hit statistics are implemented in code as cacheStat, which automatically counts when a cache hit is lost and prints the status of hits, qps, etc. used at regular intervals.

The printout will look like this

cache(proc) - qpm: 2, hit_ratio: 50.0%, elements: 0, hit: 1, miss: 1

Cache hit containment is implemented using syncx.SingleFlight, which is a request for the same key at the same time, SingleFlight(), which will request the same key at the same time, will be added later. The specific implementation is described in:

func (c *Cache) Take(key string, fetch func() (interface{}, error)) (interface{}, error) {
val, fresh, err := c.barrier.DoEx(key, func() (interface{}, error) {
v, e := fetch()
if e ! = nil {
return nil, e
c.Set(key, v)
return v, nil
if err ! = nil {
return nil, err
if fresh {
return val, nil
} else {
// got the result from previous ongoing query
return val, nil

This article introduces the Cache tool in the go-zero framework, which is very useful in real projects. Using good tools for improving service performance and development efficiency are very helpful, I hope this article can bring you some gains.

Last updated on