ioredis
A delightful, performance-focused Redis client for Node and io.js
Support Redis >= 2.6.12 and (Node.js >= 0.11.13 or io.js).
Feature
ioredis is a robust, full-featured Redis client
used in the world's biggest online commerce company Alibaba.
- Full-featured. It supports Cluster, Sentinel, Pipelining and of course Lua scripting & Pub/Sub(with the support of binary messages).
- High performance.
- Delightful API. Supports both Node-style callbacks and promises.
- Supports command arguments and replies transform.
- Abstraction for Lua scripting, allowing you to define custom commands.
- Support for binary data.
- Support for both TCP/IP and UNIX domain sockets.
- Supports offline queue and ready checking.
- Supports ES6 types such as
Map
and Set
. - Sophisticated error handling strategy.
Links
API Documentation
Migrating from node_redis
Quick Start
Install
$ npm install ioredis
Basic Usage
var Redis = require('ioredis');
var redis = new Redis();
redis.set('foo', 'bar');
redis.get('foo', function (err, result) {
console.log(result);
});
redis.get('foo').then(function (result) {
console.log(result);
});
redis.sadd('set', 1, 3, 5, 7);
redis.sadd('set', [1, 3, 5, 7]);
Connect to Redis
When a new Redis
instance is created,
a connection to Redis will be created at the same time.
You can specify which Redis to connect to by:
new Redis()
new Redis(6380)
new Redis(6379, '192.168.1.1')
new Redis('redis://:authpassword@127.0.0.1:6380/4')
new Redis('/tmp/redis.sock')
new Redis({
port: 6379,
host: '127.0.0.1',
family: 4,
password: 'auth'
db: 0
})
Pub/Sub
Here is a simple example of the API for publish / subscribe.
This program opens two client connections. It subscribes to a channel with one connection,
and publishes to that channel with the other:
var Redis = require('ioredis');
var redis = new Redis();
var pub = new Redis();
redis.subscribe('news', 'music', function (err, count) {
pub.publish('news', 'Hello world!');
pub.publish('music', 'Hello again!');
});
redis.on('message', function (channel, message) {
console.log('Receive message %s from channel %s', message, channel);
});
redis.on('messageBuffer', function (channel, message) {
});
When a client issues a SUBSCRIBE or PSUBSCRIBE, that connection is put into a "subscriber" mode.
At that point, only commands that modify the subscription set are valid.
When the subscription set is empty, the connection is put back into regular mode.
If you need to send regular commands to Redis while in subscriber mode, just open another connection.
Handle Binary Data
Arguments can be buffers:
redis.set('foo', new Buffer('bar'));
And every command has a method that returns a Buffer (by adding a suffix of "Buffer" to the command name).
To get a buffer instead of a utf8 string:
redis.getBuffer('foo', function (err, result) {
});
Pipelining
If you want to send a batch of commands(e.g. > 100), you can use pipelining to queue
the commands in the memory, then send them to Redis all at once. This way the performance improves by 50%~300%.
redis.pipeline()
creates a Pipeline
instance. You can call any Redis
commands on it just like the Redis
instance. The commands are queued in the memory
and flushed to Redis by calling exec
method:
var pipeline = redis.pipeline();
pipeline.set('foo', 'bar');
pipeline.del('cc');
pipeline.exec(function (err, results) {
});
redis.pipeline().set('foo', 'bar').del('cc').exec(function (err, results) {
});
var promise = redis.pipeline().set('foo', 'bar').get('foo').exec();
promise.then(function (result) {
});
Each chained command can also have a callback, which will be invoked when the command
get a reply:
redis.pipeline().set('foo', 'bar').get('foo', function (err, result) {
}).exec(function (err, result) {
});
Transaction
Most of the time the transaction commands multi
& exec
are used together with pipeline.
Therefore by default when multi
is called, a Pipeline
instance is created automatically,
so that you can use multi
just like pipeline
:
redis.multi().set('foo', 'bar').get('foo').exec(function (err, results) {
});
If there's a syntax error in the transaction's command chain (e.g. wrong number of arguments, wrong command name, etc),
then none of the commands would be executed, and an error is returned:
redis.multi().set('foo').set('foo', 'new value').exec(function (err, results) {
});
In terms of the interface, multi
differs from pipeline
in that when specifying a callback
to each chained command, the queueing state is passed to the callback instead of the result of the command:
redis.multi().set('foo', 'bar', function (err, result) {
}).exec();
If you want to use transaction without pipeline, pass { pipeline: false } to multi
,
and every command would be sent to Redis immediately without waiting for an exec
invokation:
redis.multi({ pipeline: false });
redis.set('foo', 'bar');
redis.get('foo');
redis.exec(function (err, result) {
});
Inline transaction is supported by pipeline, that means you can group a subset commands
in the pipeline into a transaction:
redis.pipeline().get('foo').mulit().set('foo', 'bar').get('foo').exec().get('foo').exec();
Arguments & Replies Transform
Most Redis commands take one or more Strings as arguments,
and replies are sent back as a single String or an Array of Strings. However sometimes
you may want something different: For instance it would be more convenient if HGETALL
command returns a hash (e.g. {key: val1, key2: v2}
) rather than an array of key values (e.g. [key1,val1,key2,val2]
).
ioredis has a flexible system for transforming arguments and replies. There are two types
of transformers, argument transform and reply transformer:
var Redis = require('ioredis');
Redis.Command.setArgumentTransformer('hmset', function (args) {
if (args.length === 2) {
if (typeof Map !== 'undefined' && args[1] instanceof Map) {
return [args[0]].concat(utils.convertMapToArray(args[1]));
}
if ( typeof args[1] === 'object' && args[1] !== null) {
return [args[0]].concat(utils.convertObjectToArray(args[1]));
}
}
return args;
});
Redis.Command.setReplyTransformer('hgetall', function (result) {
if (Array.isArray(result)) {
var obj = {};
for (var i = 0; i < result.length; i += 2) {
obj[result[i]] = result[i + 1];
}
return obj;
}
return result;
});
There are three built-in transformers, two argument transformers for hmset
& mset
and
a reply transformer for hgetall
. Transformers for hmset
and hgetall
has been mentioned
above, and the transformer for mset
is similar to the one for hmset
:
redis.mset({ k1: 'v1', k2: 'v2' });
redis.get('k1', function (err, result) {
});
redis.mset(new Map([['k3', 'v3'], ['k4', 'v4']]));
redis.get('k3', function (err, result) {
});
Lua Scripting
ioredis supports all of the scripting commands such as EVAL
, EVALSHA
and SCRIPT
.
However it's tedious to use in real world scenarios since developers have to take
care of script caching and to detect when to use EVAL
and when to use EVALSHA
.
ioredis expose a defineCommand
method to make scripting much easier to use:
var redis = new Redis();
redis.defineCommand('echo', {
numberOfKeys: 2,
lua: 'return {KEYS[1],KEYS[2],ARGV[1],ARGV[2]}'
});
redis.echo('k1', 'k2', 'a1', 'a2', function (err, result) {
});
redis.echoBuffer('k1', 'k2', 'a1', 'a2', function (err, result) {
});
redis.pipeline().set('foo', 'bar').echo('k1', 'k2', 'a1', 'a2').exec();
If the number of keys can't be determined when defining a command, you can
omit the numberOfKeys
property, and pass the number of keys as the first argument
when you call the command:
redis.defineCommand('echoDynamicKeyNumber', {
lua: 'return {KEYS[1],KEYS[2],ARGV[1],ARGV[2]}'
});
redis.echoDynamicKeyNumber(2, 'k1', 'k2', 'a1', 'a2', function (err, result) {
});
Monitor
Redis supports the MONITOR command,
which lets you see all commands received by the Redis server across all client connections,
including from other client libraries and other computers.
The monitor
method returns a monitor instance.
After you send the MONITOR command, no other commands are valid on that connection. ioredis would emit a monitor event for every new monitor message that comes across.
The callback for the monitor event takes a timestamp from the Redis server and an array of command arguments.
Here is a simple example:
redis.monitor(function (err, monitor) {
monitor.on('monitor', function (time, args) {
});
});
Auto-reconnect
By default, ioredis will try to reconnect when the connection to Redis is lost
except when the connection is closed manually by redis.disconnect()
or redis.quit()
.
It's very flexible to control how long to wait to reconnect after disconnected
using the retryStrategy
option:
var redis = new Redis({
retryStrategy: function (times) {
var delay = Math.min(times * 2, 2000);
return delay;
}
});
retryStrategy
is a function that will be called when the connection is lost.
The argument times
represents this is the nth reconnection being made and
the return value represents how long(ms) to wait to reconnect. When the
return value isn't a number, ioredis will stop trying reconnecting and the connection
will be lost forever if user don't call redis.connect()
manually.
Offline Queue
When a command can't be processed by Redis(e.g. the connection hasn't been established or
Redis is loading data from disk), by default it's added to the offline queue and will be
executed when it can be processed. You can disable this feature by set enableOfflineQueue
option to false
:
var redis = new Redis({ enableOfflineQueue: false });
Sentinel
ioredis supports Sentinel out of the box. It works transparently as all features that work when
you connect to a single node also work when you connect to a sentinel group. Make sure to run Redis 2.8+ if you want to use this feature.
To connect using Sentinel, use:
var redis = new Redis({
sentinels: [{ host: 'localhost', port: 26379 }, { host: 'localhost', port: 26380 }],
name: 'mymaster'
});
redis.set('foo', 'bar');
The arguments passed to the constructor are different from ones you used to connect to a single node, where:
name
identifies a group of Redis instances composed of a master and one or more slaves (mymaster
in the example);sentinels
are a list of sentinels to connect to. The list does not need to enumerate all your sentinel instances, but a few so that if one is down the client will try the next one.
ioredis guarantees that the node you connected with is always a master even after a failover. When a failover happens, instead of trying to reconnect with the failed node(which will be demoted to slave when it's available again), ioredis will ask sentinels for the new master node and connect to it. All commands sent during the failover are queued and will be executed when the new connection is established so that none of the commands will be lost.
It's possible to connect to a slave instead of a master by specifying the option role
with the value of slave
, and ioredis will try to connect to a random slave of the specified master, with the guarantee that the connected node is always a slave. If the current node is promoted to master owing to a failover, ioredis will disconnect with it and ask sentinels for another slave node to connect to.
Cluster
Support for Cluster is currently experimental. It's not recommended to use it in production.
If you encounter any problems, welcome to submit an issue :-).
You can connect to a cluster like this:
var Redis = require('ioredis');
var cluster = new Redis.Cluster([{
port: 6380,
host: '127.0.0.1'
}, {
port: 6381,
host: '127.0.0.1'
}]);
cluster.set('foo', 'bar');
cluster.get('foo', function (err, res) {
});
When using Redis.Cluster
to connect to a cluster, there are some differences from using Redis
:
- The argument is a list of nodes of the cluster you want to connect.
Just like Sentinel, the list does not need to enumerate all your cluster nodes,
but a few so that if one is down the client will try the next one, and the client will discover other nodes automatically when at least one node is connnected.
- Some comands can't be used in the cluster mode, e.g.
info
and pipeline
, custom commands also don't work(currently).
hiredis
If hiredis is installed(by npm install hiredis
),
ioredis will use it by default. Otherwise, a pure JavaScript parser will be used.
Typically there's not much differences between them in terms of performance.
Benchmark
Compares with node_redis:
> npm run bench
simple set
65,438 op/s » ioredis
36,954 op/s » node_redis
simple get
71,109 op/s » ioredis
36,825 op/s » node_redis
simple get with pipeline
11,123 op/s » ioredis
3,820 op/s » node_redis
lrange 100
58,812 op/s » ioredis
46,703 op/s » node_redis
Suites: 4
Benches: 8
Elapsed: 61,715.11 ms
You can find the code at benchmark.js
.
Motivation
Originally we used the Redis client node_redis,
but over a period of time we found that it's not robust enough for us to use
in our production environment. The library has some non-trivial bugs and many unresolved
issues on the GitHub(165 so far). For instance:
var redis = require('redis');
var client = redis.createClient();
client.set('foo', 'message');
client.set('bar', 'Hello world');
client.mget('foo', 'bar');
client.subscribe('channel');
client.on('message', function (msg) {
console.log('received ', msg);
});
I submitted some pull requests but sadly none of them has been merged, so here's ioredis.
Acknowledge
The JavaScript and hiredis parsers are modified from node_redis (MIT License, Copyright (c) 2010 Matthew Ranney, http://ranney.com/).
License
MIT