English | 中文

• Statically Typed - 100% statically typed using code generation.
• Scale Efficiently - thundering herd protection via pub/sub.
• Cluster Support - same API for redis & redis cluster.
• Memoize - dynamic key params based on code generation.
• Versioning - cache versioning for better management.
• Pipeline - reduce io cost by redis pipeline.

Read this first: Caches, Promises and Locks. This is how caching part works in cacheme.

go get github.com/Yiling-J/cacheme-go/cmd

After installing cacheme-go codegen, go to the root directory of your project, and run:

go run github.com/Yiling-J/cacheme-go/cmd init

The command above will generate cacheme directory under root directory:

└── cacheme
    ├── fetcher
    │   └── fetcher.go
    └── schema
        └── schema.go

Edit schema.go and add some schemas:

package schema

import (
	"time"
	cacheme "github.com/Yiling-J/cacheme-go"
)

var (
	// default prefix for redis keys
	Prefix = "cacheme"
	// store schemas
	Stores = []*cacheme.StoreSchema{
		{
			Name:         "Simple",
			Key:          "simple:{{.ID}}",
			To:           "",
			Version:      1,
			TTL:          5 * time.Minute,
			Singleflight: false,
		},
	}
)

More details here

Run code generation from the root directory of the project as follows:

go run github.com/Yiling-J/cacheme-go/cmd generate

This produces the following files:

└── cacheme
    ├── fetcher
    │   └── fetcher.go
    ├── schema
    │   └── schema.go
    └── store.go

store.go is generated based on schemas in schema.go. Adding more schemas and run generate again.

Each cache store should provide a fetch function in fetcher.go:

func Setup() {
	cacheme.SimpleCacheStore.Fetch = func(ctx context.Context, ID string) (string, error) {
		return ID, nil
	}
}

import (
	"your_project/cacheme"
	"your_project/cacheme/fetcher"
)

func main() {
	// setup fetcher
	fetcher.Setup()
	// create client
	client := cacheme.New(
		redis.NewClient(&redis.Options{
			Addr:     "localhost:6379",
			Password: "",
			DB:       0,
		}),
	)
	// or cluster client
	client := cacheme.NewCluster(
		redis.NewClusterClient(&redis.ClusterOptions{
			Addrs: []string{
				":7000",
				":7001",
				":7002"},
		}),
	)
}

Get cached result. If not in cache, call fetch function and store data to Redis.

result, err := client.SimpleCacheStore.Get(ctx, "foo")

Get multiple keys from multiple stores using pipeline. For each key, if not in cache, call fetch function and store data to Redis.

• single store

import cachemego "github.com/Yiling-J/cacheme-go"

pipeline := cachemego.NewPipeline(client.Redis())
ids := []string{"1", "2", "3", "4"}
var ps []*cacheme.SimplePromise
for _, i := range ids {
	promise, err := client.SimpleCacheStore.GetP(ctx, pipeline, i)
	ps = append(ps, promise)
}
err = pipeline.Execute(ctx)
fmt.Println(err)

for _, promise := range ps {
	r, err := promise.Result()
	fmt.Println(r, err)
}

Consider using GetM API for single store, see GetM example below.

• multiple stores

import cachemego "github.com/Yiling-J/cacheme-go"

// same pipeline for different stores
pipeline := cachemego.NewPipeline(client.Redis())

ids := []string{"1", "2", "3", "4"}
var ps []*cacheme.SimplePromise // cache string
var psf []*cacheme.FooPromise // cache model.Foo struct
for _, i := range ids {
	promise, err := client.SimpleCacheStore.GetP(ctx, pipeline, i)
	ps = append(ps, promise)
}
for _, i := range ids {
	promise, err := client.FooCacheStore.GetP(ctx, pipeline, i)
	psf = append(psf, promise)
}
// execute only once
err = pipeline.Execute(ctx)
// simple store results
for _, promise := range ps {
	r, err := promise.Result()
	fmt.Println(r, err)
}
// foo store results
for _, promise := range psf {
	r, err := promise.Result()
	fmt.Println(r, err)
}

Get multiple keys from same store, also using Redis pipeline. For each key, if not in cache, call fetch function and store data to Redis.

qs, err := client.SimpleCacheStore.GetM("foo").GetM("bar").GetM("xyz").Do(ctx)
// qs is a queryset struct, support two methods: GetSlice and Get
// GetSlice return ordered results slice
r, err := qs.GetSlice() // r: {foo_result, bar_result, xyz_result}
// Get return result of given param
r, err := qs.Get("foo") // r: foo_result
r, err := qs.Get("bar") // r: bar_result
r, err := qs.Get("fake") // error, because "fake" not in queryset

err := client.SimpleCacheStore.Invalid(ctx, "foo")

err := client.SimpleCacheStore.Update(ctx, "foo")

// invalid all version 1 simple cache
client.SimpleCacheStore.InvalidAll(ctx, "1")

Each schema has 5 fields:

• Name - store name, will be struct name in generated code, capital first.
• Key - key with variable using go template syntax, Variable name will be used in code generation.
• To - cached value, type of value will be used in code generation. Examples:
  • string: ""
  • int: 1
  • struct: model.Foo{}
  • struct pointer: &model.Foo{}
  • slice: []model.Foo{}
  • map: map[model.Foo]model.Bar{}
• Version - version interface, can be string, int, or callable func() string.
• TTL - redis ttl using go time.
• Singleflight - bool, if true, concurrent requests to same key on same executable will call Redis only once

• Duplicate name/key is not allowed.
• Everytime you update schema, run code generation again.
• Not all store API support Singleflight option:
  • Get: support.
  • GetM: support. singleflight key will be the combination of all keys, order by alphabetical.

   // these two will use same singleflight group key
   store.GetM("foo").GetM("bar").GetM("xyz").Do(ctx)
   Store.GetM("bar").GetM("foo").GetM("xyz").Do(ctx)

  • GetP: not support.
• Version callable can help you managing version better. Example:

   // models.go
   const FooCacheVersion = "1"
   type Foo struct {}
   const BarCacheVersion = "1"
   type Bar struct {Foo: Foo}

   // schema.go
   // version has 3 parts: foo version & bar version & global version number
   // if you change struct, update FooCacheVersion or BarCacheVersion
   // if you change fetcher function or ttl or something else, change global version number
   {
   	Name:    "Bar",
   	Key:     "bar:{{.ID}}:info",
   	To:      model.Bar{},
   	Version: func() string {return model.FooCacheVersion + model.BarCacheVersion + "1"},
   	TTL:     5 * time.Minute,
   },

• If set Singleflight to true, Cacheme Get command will be wrapped in a singleflight, so concurrent requests to same key will call Redis only once. Let's use some example to explain this:
  • you have some products to sell, and thousands people will view the detail at same time, so the product key product:1:info may be hit 100000 times per second. Now you should turn on singleflight, and the actually redis hit may reduce to 5000.
  • you have cache for user shopping cart user:123:cart, only the user himself can see that. Now no need to use singleflight, becauese there should't be concurrent requests to that key.
  • you are using serverless platform, AWS Lambda or similar. So each request runs in isolated environment, can't talk to each other through channels. Then singleflight make no sense.
• Full redis key has 3 parts: prefix + schema key + version. Schema Keycategory:{{.categoryID}}:book:{{.bookID}} with prefix cacheme, version 1 will generate key:

   cacheme:category:1:book:3:v1
  

  Also you will see categoryID and bookID in generated code, as fetch func params.

You can use custom logger with cacheme, your logger should implement cacheme logger interface:

type Logger interface {
	Log(store string, key string, op string)
}

Here store is the store tag, key is cache key without prefix, op is operation type. Default logger is NOPLogger, just return and do nothing.

logger := &YourCustomLogger{}
client.SetLogger(logger)

• HIT: cache hit to redis, if you enable singleflight, grouped requests only log once.
• MISS: cache miss
• FETCH: fetch data from fetcher

Parallel benchmarks of Cacheme alongside go-redis/cache:

• params: 10000/1000000 hits, 10 keys loop, TTL 10s, SetParallelism(100), singleflight on
• go-redis/cache without local cache

cpu: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
BenchmarkCachemeGetParallel-12    	   10000	    198082 ns/op
BenchmarkCacheGetParallel-12    	   10000	    189766 ns/op
BenchmarkCachemeGetParallel-12    	 1000000	      9501 ns/op
BenchmarkCacheGetParallel-12    	 1000000	      4323 ns/op

At 10000 hits, result almost same. At 1000000 hits, go-redis/cache is about 2 times faster than Cacheme. but keep in mind, go-redis/cache is based on singleflight only, not truly distributed. This bench case is single executable, not the real load case.