rueidis

Features

Auto pipelining for non-blocking redis commands
Server-assisted client-side caching
Generic Object Mapping with client-side caching
Cache-Aside pattern with client-side caching
Distributed Locks with client-side caching
Helpers for writing tests with rueidis mock
OpenTelemetry integration
Hooks and other integrations
Go-redis like API adapter by @418Coffee
Pub/Sub, Sharded Pub/Sub, Streams
Redis Cluster, Sentinel, RedisJSON, RedisBloom, RediSearch, RedisTimeseries, etc.
Probabilistic Data Structures without Redis Stack

Getting Started

package main

import (
	"context"
	"github.com/redis/rueidis"
)

func main() {
	client, err := rueidis.NewClient(rueidis.ClientOption{InitAddress: []string{"127.0.0.1:6379"}})
	if err != nil {
		panic(err)
	}
	defer client.Close()

	ctx := context.Background()
	// SET key val NX
	err = client.Do(ctx, client.B().Set().Key("key").Value("val").Nx().Build()).Error()
	// HGETALL hm
	hm, err := client.Do(ctx, client.B().Hgetall().Key("hm").Build()).AsStrMap()
}

Check out more examples: Command Response Cheatsheet

Developer Friendly Command Builder

client.B() is the builder entry point to construct a redis command:

Developer friendly command builder
_{Recorded by @FZambia Improving Centrifugo Redis Engine throughput and allocation efficiency with Rueidis Go library}

Once a command is built, use either client.Do() or client.DoMulti() to send it to redis.

You ❗️SHOULD NOT❗️ reuse the command to another client.Do() or client.DoMulti() call because it has been recycled to the underlying sync.Pool by default.

To reuse a command, use Pin() after Build() and it will prevent the command from being recycled.

Pipelining

Auto Pipelining

All concurrent non-blocking redis commands (such as GET, SET) are automatically pipelined, which reduces the overall round trips and system calls and gets higher throughput. You can easily get the benefit of pipelining technique by just calling client.Do() from multiple goroutines concurrently. For example:

func BenchmarkPipelining(b *testing.B, client rueidis.Client) {
	// the below client.Do() operations will be issued from
	// multiple goroutines and thus will be pipelined automatically.
	b.RunParallel(func(pb *testing.PB) {
		for pb.Next() {
			client.Do(context.Background(), client.B().Get().Key("k").Build()).ToString()
		}
	})
}

Benchmark comparison with go-redis v9

Compared to go-redis, Rueidis has higher throughput across 1, 8, and 64 parallelism settings.

It is even able to achieve ~14x throughput over go-redis in a local benchmark of Macbook Pro 16” M1 Pro 2021. (see parallelism(64)-key(16)-value(64)-10)

client_test_set

Benchmark source code: https://github.com/rueian/rueidis-benchmark

A benchmark result performed on two GCP n2-highcpu-2 machines also shows that rueidis can achieve higher throughput with lower latencies: https://github.com/redis/rueidis/pull/93

Manual Pipelining

Besides auto pipelining, you can also pipeline commands manually with DoMulti():

cmds := make(rueidis.Commands, 0, 10)
for i := 0; i < 10; i++ {
    cmds = append(cmds, client.B().Set().Key("key").Value("value").Build())
}
for _, resp := range client.DoMulti(ctx, cmds...) {
    if err := resp.Error(); err != nil {
        panic(err)
    }
}

Server-Assisted Client-Side Caching

The opt-in mode of server-assisted client-side caching is enabled by default and can be used by calling DoCache() or DoMultiCache() with client-side TTLs specified.

client.DoCache(ctx, client.B().Hmget().Key("mk").Field("1", "2").Cache(), time.Minute).ToArray()
client.DoMultiCache(ctx,
    rueidis.CT(client.B().Get().Key("k1").Cache(), 1*time.Minute),
    rueidis.CT(client.B().Get().Key("k2").Cache(), 2*time.Minute))

Cached responses, including Redis Nils, will be invalidated either when being notified by redis servers or when their client-side TTLs are reached. See https://github.com/redis/rueidis/issues/534 for more details.

Benchmark

Server-assisted client-side caching can dramatically boost latencies and throughput just like having a redis replica right inside your application. For example:

client_test_get

Benchmark source code: https://github.com/rueian/rueidis-benchmark

Client-Side Caching Helpers

Use CacheTTL() to check the remaining client-side TTL in seconds:

client.DoCache(ctx, client.B().Get().Key("k1").Cache(), time.Minute).CacheTTL() == 60

Use IsCacheHit() to verify if the response came from the client-side memory:

client.DoCache(ctx, client.B().Get().Key("k1").Cache(), time.Minute).IsCacheHit() == true

If the OpenTelemetry is enabled by the rueidisotel.NewClient(option), then there are also two metrics instrumented:

rueidis_do_cache_miss
rueidis_do_cache_hits

MGET/JSON.MGET Client-Side Caching Helpers

rueidis.MGetCache and rueidis.JsonMGetCache are handy helpers fetching multiple keys across different slots through the client-side caching. They will first group keys by slot to build MGET or JSON.MGET commands respectively and then send requests with only cache missed keys to redis nodes.

Broadcast Mode Client-Side Caching

Although the default is opt-in mode, you can use broadcast mode by specifying your prefixes in ClientOption.ClientTrackingOptions:

client, err := rueidis.NewClient(rueidis.ClientOption{
	InitAddress:           []string{"127.0.0.1:6379"},
	ClientTrackingOptions: []string{"PREFIX", "prefix1:", "PREFIX", "prefix2:", "BCAST"},
})
if err != nil {
	panic(err)
}
client.DoCache(ctx, client.B().Get().Key("prefix1:1").Cache(), time.Minute).IsCacheHit() == false
client.DoCache(ctx, client.B().Get().Key("prefix1:1").Cache(), time.Minute).IsCacheHit() == true

Please make sure that commands passed to DoCache() and DoMultiCache() are covered by your prefixes. Otherwise, their client-side cache will not be invalidated by redis.

Client-Side Caching with Cache Aside Pattern

Cache-Aside is a widely used caching strategy. rueidisaside can help you cache data into your client-side cache backed by Redis. For example:

client, err := rueidisaside.NewClient(rueidisaside.ClientOption{
    ClientOption: rueidis.ClientOption{InitAddress: []string{"127.0.0.1:6379"}},
})
if err != nil {
    panic(err)
}
val, err := client.Get(context.Background(), time.Minute, "mykey", func(ctx context.Context, key string) (val string, err error) {
    if err = db.QueryRowContext(ctx, "SELECT val FROM mytab WHERE id = ?", key).Scan(&val); err == sql.ErrNoRows {
        val = "_nil_" // cache nil to avoid penetration.
        err = nil     // clear err in case of sql.ErrNoRows.
    }
    return
})
// ...

Please refer to the full example at rueidisaside.

Context Cancellation

client.Do(), client.DoMulti(), client.DoCache(), and client.DoMultiCache() can return early if the context is canceled or the deadline is reached.

ctx, cancel := context.WithTimeout(context.Background(), time.Second)
defer cancel()
client.Do(ctx, client.B().Set().Key("key").Value("val").Nx().Build()).Error() == context.DeadlineExceeded

Please note that though operations can return early, the command is likely sent already.

Alternative PubSub Hooks

The client.Receive() requires users to provide a subscription command in advance. There is an alternative Dedicatedclient.SetPubSubHooks() that allows users to subscribe/unsubscribe channels later.

c, cancel := client.Dedicate()
defer cancel()

wait := c.SetPubSubHooks(rueidis.PubSubHooks{
	OnMessage: func(m rueidis.PubSubMessage) {
		// Handle message. This callback will be called sequentially but in another goroutine.
	}
})
c.Do(ctx, c.B().Subscribe().Channel("ch").Build())
err := <-wait // disconnected with err

If the hooks are not nil, the above wait channel is guaranteed to be closed when the hooks will not be called anymore, and produce at most one error describing the reason. Users can use this channel to detect disconnection.

CAS Transaction

To do a CAS Transaction (WATCH + MULTI + EXEC), a dedicated connection should be used because there should be no unintentional write commands between WATCH and EXEC. Otherwise, the EXEC may not fail as expected.

client.Dedicated(func(c rueidis.DedicatedClient) error {
    // watch keys first
    c.Do(ctx, c.B().Watch().Key("k1", "k2").Build())
    // perform read here
    c.Do(ctx, c.B().Mget().Key("k1", "k2").Build())
    // perform write with MULTI EXEC
    c.DoMulti(
        ctx,
        c.B().Multi().Build(),
        c.B().Set().Key("k1").Value("1").Build(),
        c.B().Set().Key("k2").Value("2").Build(),
        c.B().Exec().Build(),
    )
    return nil
})

Or use Dedicate() and invoke cancel() when finished to put the connection back to the pool.

c, cancel := client.Dedicate()
defer cancel()

c.Do(ctx, c.B().Watch().Key("k1", "k2").Build())
// do the rest CAS operations with the `client` who occupies a connection

However, occupying a connection is not good in terms of throughput. It is better to use Lua script to perform optimistic locking instead.

Memory Consumption Consideration

Each underlying connection in rueidis allocates a ring buffer for pipelining. Its size is controlled by the ClientOption.RingScaleEachConn and the default value is 10 which results into each ring of size 2^10.

If you have many rueidis connections, you may find that they occupy quite an amount of memory. In that case, you may consider reducing ClientOption.RingScaleEachConn to 8 or 9 at the cost of potential throughput degradation.

You may also consider setting the value of ClientOption.PipelineMultiplex to -1, which will let rueidis use only 1 connection for pipelining to each redis node.

Instantiating a new Redis Client

You can create a new redis client using NewClient and provide several options.

// Connect to a single redis node:
client, err := rueidis.NewClient(rueidis.ClientOption{
    InitAddress: []string{"127.0.0.1:6379"},
})

// Connect to a redis cluster
client, err := rueidis.NewClient(rueidis.ClientOption{
    InitAddress: []string{"127.0.0.1:7001", "127.0.0.1:7002", "127.0.0.1:7003"},
    ShuffleInit: true,
})

// Connect to a redis cluster and use replicas for read operations
client, err := rueidis.NewClient(rueidis.ClientOption{
    InitAddress: []string{"127.0.0.1:7001", "127.0.0.1:7002", "127.0.0.1:7003"},
    SendToReplicas: func(cmd rueidis.Completed) bool {
        return cmd.IsReadOnly()
    },
})

// Connect to sentinels
client, err := rueidis.NewClient(rueidis.ClientOption{
    InitAddress: []string{"127.0.0.1:26379", "127.0.0.1:26380", "127.0.0.1:26381"},
    Sentinel: rueidis.SentinelOption{
        MasterSet: "my_master",
    },
})

Arbitrary Command

If you want to construct commands that are absent from the command builder, you can use client.B().Arbitrary():

// This will result in [ANY CMD k1 k2 a1 a2]
client.B().Arbitrary("ANY", "CMD").Keys("k1", "k2").Args("a1", "a2").Build()

Working with JSON, Raw `[]byte`, and Vector Similarity Search

The command builder treats all the parameters as Redis strings, which are binary safe. This means that users can store []byte directly into Redis without conversion. And the rueidis.BinaryString helper can convert []byte to string without copying. For example:

client.B().Set().Key("b").Value(rueidis.BinaryString([]byte{...})).Build()

Treating all the parameters as Redis strings also means that the command builder doesn’t do any quoting, conversion automatically for users.

When working with RedisJSON, users frequently need to prepare JSON strings in Redis strings. And rueidis.JSON can help:

client.B().JsonSet().Key("j").Path("$.myStrField").Value(rueidis.JSON("str")).Build()
// equivalent to
client.B().JsonSet().Key("j").Path("$.myStrField").Value(`"str"`).Build()

When working with vector similarity search, users can use rueidis.VectorString32 and rueidis.VectorString64 to build queries:

cmd := client.B().FtSearch().Index("idx").Query("*=>[KNN 5 @vec $V]").
    Params().Nargs(2).NameValue().NameValue("V", rueidis.VectorString64([]float64{...})).
    Dialect(2).Build()
n, resp, err := client.Do(ctx, cmd).AsFtSearch()

Command Response Cheatsheet

While the command builder is developer-friendly, the response parser is a little unfriendly. Developers must know what type of Redis response will be returned from the server beforehand and which parser they should use.

Error Handling: If an incorrect parser function is chosen, an errParse will be returned. Here’s an example using ToArray which demonstrates this scenario:

// Attempt to parse the response. If a parsing error occurs, check if the error is a parse error and handle it.
// Normally, you should fix the code by choosing the correct parser function.
// For instance, use ToString() if the expected response is a string, or ToArray() if the expected response is an array as follows:
if err := client.Do(ctx, client.B().Get().Key("k").Build()).ToArray(); IsParseErr(err) {
    fmt.Println("Parsing error:", err)
}

It is hard to remember what type of message will be returned and which parsing to use. So, here are some common examples:

// GET
client.Do(ctx, client.B().Get().Key("k").Build()).ToString()
client.Do(ctx, client.B().Get().Key("k").Build()).AsInt64()
// MGET
client.Do(ctx, client.B().Mget().Key("k1", "k2").Build()).ToArray()
// SET
client.Do(ctx, client.B().Set().Key("k").Value("v").Build()).Error()
// INCR
client.Do(ctx, client.B().Incr().Key("k").Build()).AsInt64()
// HGET
client.Do(ctx, client.B().Hget().Key("k").Field("f").Build()).ToString()
// HMGET
client.Do(ctx, client.B().Hmget().Key("h").Field("a", "b").Build()).ToArray()
// HGETALL
client.Do(ctx, client.B().Hgetall().Key("h").Build()).AsStrMap()
// ZRANGE
client.Do(ctx, client.B().Zrange().Key("k").Min("1").Max("2").Build()).AsStrSlice()
// ZRANK
client.Do(ctx, client.B().Zrank().Key("k").Member("m").Build()).AsInt64()
// ZSCORE
client.Do(ctx, client.B().Zscore().Key("k").Member("m").Build()).AsFloat64()
// ZRANGE
client.Do(ctx, client.B().Zrange().Key("k").Min("0").Max("-1").Build()).AsStrSlice()
client.Do(ctx, client.B().Zrange().Key("k").Min("0").Max("-1").Withscores().Build()).AsZScores()
// ZPOPMIN
client.Do(ctx, client.B().Zpopmin().Key("k").Build()).AsZScore()
client.Do(ctx, client.B().Zpopmin().Key("myzset").Count(2).Build()).AsZScores()
// SCARD
client.Do(ctx, client.B().Scard().Key("k").Build()).AsInt64()
// SMEMBERS
client.Do(ctx, client.B().Smembers().Key("k").Build()).AsStrSlice()
// LINDEX
client.Do(ctx, client.B().Lindex().Key("k").Index(0).Build()).ToString()
// LPOP
client.Do(ctx, client.B().Lpop().Key("k").Build()).ToString()
client.Do(ctx, client.B().Lpop().Key("k").Count(2).Build()).AsStrSlice()
// SCAN
client.Do(ctx, client.B().Scan().Cursor(0).Build()).AsScanEntry()
// FT.SEARCH
client.Do(ctx, client.B().FtSearch().Index("idx").Query("@f:v").Build()).AsFtSearch()
// GEOSEARCH
client.Do(ctx, client.B().Geosearch().Key("k").Fromlonlat(1, 1).Bybox(1).Height(1).Km().Build()).AsGeosearch()

Use DecodeSliceOfJSON to scan array result

DecodeSliceOfJSON is useful when you would like to scan the results of an array into a slice of a specific struct.

type User struct {
	Name string `json:"name"`
}

// Set some values
if err = client.Do(ctx, client.B().Set().Key("user1").Value(`{"name": "name1"}`).Build()).Error(); err != nil {
	return err
}
if err = client.Do(ctx, client.B().Set().Key("user2").Value(`{"name": "name2"}`).Build()).Error(); err != nil {
	return err
}

// Scan MGET results into []*User
var users []*User // or []User is also scannable
if err := rueidis.DecodeSliceOfJSON(client.Do(ctx, client.B().Mget().Key("user1", "user2").Build()), &users); err != nil {
	return err
}

for _, user := range users {
	fmt.Printf("%+v\n", user)
}
/*
&{name:name1}
&{name:name2}
*/

!!!!!! DO NOT DO THIS !!!!!!

Please make sure that all values in the result have the same JSON structures.

// Set a pure string value
if err = client.Do(ctx, client.B().Set().Key("user1").Value("userName1").Build()).Error(); err != nil {
	return err
}

// Bad
users := make([]*User, 0)
if err := rueidis.DecodeSliceOfJSON(client.Do(ctx, client.B().Mget().Key("user1").Build()), &users); err != nil {
	return err
}
// -> Error: invalid character 'u' looking for the beginning of the value
// in this case, use client.Do(ctx, client.B().Mget().Key("user1").Build()).AsStrSlice()

Generate command builders

Command builders are generated based on the definitions in ./hack/cmds by running:

go generate

Testing

Please use the ./dockertest.sh script for running test cases locally. And please try your best to have 100% test coverage on code changes.