rate-limiter-flexible

node-rate-limiter-flexible

Flexible rate limiter and anti-DDoS protector works in process Memory, Cluster, MongoDB or Redis allows to control requests rate in single process or distributed environment.

It uses fixed window as it is much faster than rolling window. See comparative benchmarks with other libraries here

Advantages:

• block strategy against really powerful DDoS attacks (like 100k requests per sec) Read about it and benchmarking here
• backed on native Promises
• works in Cluster without additional software See RateLimiterCluster benchmark and detailed description here
• actions can be done evenly over duration window to cut off picks
• no race conditions
• covered by tests
• no prod dependencies
• Redis and Mongo errors don't result to broken app if insuranceLimiter set up
• useful penalty and reward methods to change limits on some results of an action

Links

• RateLimiterRedis
• RateLimiterMongo
• RateLimiterCluster
• RateLimiterMemory
• RateLimiterUnion Combine 2 or more limiters to act as single
• Express middleware
• Koa middleware
• Options
• API

Benchmark

Average latency of pure NodeJS endpoint limited by (all set up on one server):

1. Memory   0.34 ms
2. Cluster  0.69 ms
3. Redis    2.45 ms
4. Mongo    4.75 ms

RateLimiterRedis benchmark

Endpoint is pure NodeJS endpoint launched in node:latest and redis:alpine Docker containers by PM2 with 4 workers

By bombardier -c 1000 -l -d 30s -r 2000 -t 5s http://127.0.0.1:8000

Test with 1000 concurrent requests with maximum 2000 requests per sec during 30 seconds

Statistics        Avg      Stdev        Max
  Reqs/sec      2015.20     511.21   14570.19
  Latency        2.45ms     7.51ms   138.41ms
  Latency Distribution
     50%     1.95ms
     75%     2.16ms
     90%     2.43ms
     95%     2.77ms
     99%     5.73ms
  HTTP codes:
    1xx - 0, 2xx - 53556, 3xx - 0, 4xx - 6417, 5xx - 0

Installation

npm i rate-limiter-flexible

Usage

RateLimiterRedis

It supports both redis and ioredis clients.

Redis client must be created with offline queue switched off.

const redis = require('redis');
const redisClient = redis.createClient({ enable_offline_queue: false });

const Redis = require('ioredis');
const redisClient = new Redis({
  options: {
    enableOfflineQueue: false
  }
});

const { RateLimiterRedis, RateLimiterMemory } = require('rate-limiter-flexible');

// It is recommended to process Redis errors and setup some reconnection strategy
redisClient.on('error', (err) => {
  
});

const opts = {
  redis: redisClient,
  keyPrefix: 'rlflx', // useful for multiple limiters
  points: 5, // Number of points
  duration: 5, // Per second(s)
  execEvenly: false,
  
  // Redis and Mongo specific
  blockOnPointsConsumed: 10, // If 10 points consumed in current duration
  blockDuration: 30, // block for 30 seconds in current process memory
  // It will be used only on Redis error as insurance
  // Can be any implemented limiter like RateLimiterMemory or RateLimiterRedis extended from RateLimiterAbstract
  insuranceLimiter: new RateLimiterMemory(
    {
      points: 1, // 1 is fair if you have 5 workers and 1 cluster
      duration: 5,
      execEvenly: false,
    })
};

const rateLimiterRedis = new RateLimiterRedis(opts);

rateLimiterRedis.consume(remoteAddress)
    .then((rateLimiterRes) => {
      // ... Some app logic here ...
      
      // Depending on results it allows to fine
      rateLimiterRedis.penalty(remoteAddress, 3)
        .then((rateLimiterRes) => {});
      // or rise number of points for current duration
      rateLimiterRedis.reward(remoteAddress, 2)
        .then((rateLimiterRes) => {});
    })
    .catch((rejRes) => {
      if (rejRes instanceof Error) {
        // Some Redis error
        // Never happen if `insuranceLimiter` set up
        // Decide what to do with it in other case
      } else {
        // Can't consume
        // If there is no error, rateLimiterRedis promise rejected with number of ms before next request allowed
        const secs = Math.round(rejRes.msBeforeNext / 1000) || 1;
        res.set('Retry-After', String(secs));
        res.status(429).send('Too Many Requests');
      }
    });

RateLimiterMongo

It supports mongodb native and mongoose packages See RateLimiterMongo benchmark here

const { RateLimiterMongo } = require('rate-limiter-flexible');
const mongoose = require('mongoose');

const mongoOpts = {
  reconnectTries: Number.MAX_VALUE, // Never stop trying to reconnect
  reconnectInterval: 100, // Reconnect every 100ms
};

mongoose.createConnection('mongodb://localhost:27017/' + RateLimiterMongo.getDbName(), mongoOpts)
  .then((mongo) => {
    const opts = {
      mongo: mongo,
      points: 10, // Number of points
      duration: 1, // Per second(s)
    };
  
    const rateLimiterMongo = new RateLimiterMongo(opts);
    // Usage is the same as for RateLimiterRedis
  });

// Or with native mongodb package

const { MongoClient } = require('mongodb');

const mongoOpts = {
  useNewUrlParser: true,
  reconnectTries: Number.MAX_VALUE, // Never stop trying to reconnect
  reconnectInterval: 100, // Reconnect every 100ms
};

MongoClient.connect(
  'mongodb://localhost:27017',
  mongoOpts
).then((mongo) => {
  const opts = {
    mongo: mongo,
    points: 10, // Number of points
    duration: 1, // Per second(s)
  };
  
  const rateLimiterMongo = new RateLimiterMongo(opts);
      
  // Usage is the same as for RateLimiterRedis
  rateLimiterMongo.consume(remoteAddress)
  .then(() => {})
  .catch(() => {});
});

RateLimiterCluster

Note: it doesn't work with PM2 yet

RateLimiterCluster performs limiting using IPC. Each request is sent to master process, which handles all the limits, then master send results back to worker.

See RateLimiterCluster benchmark and detailed description here

const cluster = require('cluster');
const numCPUs = require('os').cpus().length;
const { RateLimiterClusterMaster, RateLimiterCluster } = require('rate-limiter-flexible');

if (cluster.isMaster) {
  // Doesn't require any options, it is only storage and messages handler
  new RateLimiterClusterMaster();

  for (let i = 0; i < numCPUs; i++) {
    cluster.fork();
  }
} else {
  const rateLimiter = new RateLimiterCluster({
    keyPrefix: 'myclusterlimiter', // Must be unique for each limiter
    points: 100,
    duration: 1,
    timeoutMs: 3000 // Promise is rejected, if master doesn't answer for 3 secs
  });
  
  // Usage is the same as for RateLimiterRedis
}  
    

RateLimiterMemory

It manages limits in current process memory, so keep it in mind when use it in cluster

const rateLimiter = new RateLimiterMemory( // It will be used only on Redis error as insurance
{
  keyPrefix: 'rlflx',
  points: 1, // 1 is fair if you have 5 workers and 1 cluster, all workers will limit it to 5 in sum
  duration: 5,
  execEvenly: false,
});
    
// Usage is the same as for RateLimiterRedis
// Except: it never rejects Promise with Error    
    

RateLimiterUnion

Combine 2 or more rate limiters to act as single

Any rate limiters from this rate-limiter-flexible can be united

Useful for authorization, which must be protected from password brute force

For example, not more than once per second and only 5 points per minute

keyPrefix is necessary as resolved and rejected results depend on it

const limiter1 = new RateLimiterMemory({
  keyPrefix: 'limit1',
  points: 1,
  duration: 1,
});
const limiter2 = new RateLimiterMemory({
  keyPrefix: 'limit2',
  points: 5,
  duration: 60,
});
const rateLimiterUnion = new RateLimiterUnion(limiter1, limiter2);

rateLimiterUnion.consume(remoteAddress)
  .then((res) => {
    // Returns object with 2 RateLimiterRes objects
    res['limit1'].remainingPoints;
    res['limit2'].remainingPoints;
  })
  .catch((rej) => {
    /* Returns object with RateLimiterRes objects only for rejected limiters
    * For example:
    * { limit1: RateLimiterRes { ... } }
    * 
    * It may be Error if you use Redis, Mongo or Cluster without insurance 
    * { limit2: Error }
    */
  });

Express middleware

const rateLimiterMiddleware = (req, res, next) => {
  rateLimiter.consume(req.connection.remoteAddress)
    .then(() => {
      next();
    })
    .catch((rejRes) => {
      res.status(429).send('Too Many Requests');
    });
};

Koa middleware

app.use(async (ctx, next) => {
  try {
    await rateLimiter.consume(ctx.ip)
    next()
  } catch (rejRes) {
    ctx.status = 429
    ctx.body = 'Too Many Requests'
  }
})

Options

• keyPrefix Default: 'rlflx' If you need to create several limiters for different purpose

• points Default: 4 Maximum number of points can be consumed over duration

• duration Default: 1 Number of seconds before points are reset

• execEvenly Default: false Delay action to be executed evenly over duration First action in duration is executed without delay. All next allowed actions in current duration are delayed by formula msBeforeDurationEnd / (remainingPoints + 2) It allows to cut off load peaks. Note: it isn't recommended to use it for long duration, as it may delay action for too long

Options specific to Redis and Mongo

• blockOnPointsConsumed Default: 0 Against DDoS attacks. Blocked key isn't checked by requesting Redis. Blocking works in current process memory. Redis is quite fast, however, it may be significantly slowed down on dozens of thousands requests.

• blockDuration Default: 0 Block key for blockDuration seconds, if blockOnPointsConsumed or more points are consumed

• insuranceLimiter Default: undefined Instance of RateLimiterAbstract extended object to store limits, when Redis comes up with any error. Additional RateLimiterRedis or RateLimiterMemory can be used as insurance. Be careful when use RateLimiterMemory in cluster or in distributed app. It may result to floating number of allowed actions. If an action with a same key is launched on one worker several times in sequence, limiter will reach out of points soon. Omit it if you want strictly use Redis and deal with errors from it

Options specific to Cluster

• timeoutMs Default: 5000 Timeout for communication between worker and master over IPC. If master doesn't response in time, promise is rejected with Error

API

RateLimiterRes object

Both Promise resolve and reject returns object of RateLimiterRes class if there is no any error. Object attributes:

RateLimiterRes = {
    msBeforeNext: 250, // Number of milliseconds before next action can be done
    remainingPoints: 0, // Number of remaining points in current duration 
    consumedPoints: 5, // Number of consumed points in current duration 
    isFirstInDuration: false, // action is first in current duration 
}

rateLimiter.consume(key, points = 1)

Returns Promise, which:

• resolved with RateLimiterRes when point(s) is consumed, so action can be done
• rejected only for Redis and Mongo if insuranceLimiter isn't setup: when some error happened, where reject reason rejRes is Error object
• rejected only for RateLimiterCluster if insuranceLimiter isn't setup: when timeoutMs exceeded, where reject reason rejRes is Error object
• rejected when there is no points to be consumed, where reject reason rejRes is RateLimiterRes object
• rejected when key is blocked (if block strategy is set up), where reject reason rejRes is RateLimiterRes object

Arguments:

• key is usually IP address or some unique client id
• points number of points consumed. default: 1

rateLimiter.penalty(key, points = 1)

Fine key by points number of points for one duration.

Note: Depending on time penalty may go to next durations

Returns Promise, which:

• resolved with RateLimiterRes
• rejected only for Redis and Mongo if insuranceLimiter isn't setup: when some error happened, where reject reason rejRes is Error object
• rejected only for RateLimiterCluster if insuranceLimiter isn't setup: when timeoutMs exceeded, where reject reason rejRes is Error object

rateLimiter.reward(key, points = 1)

Reward key by points number of points for one duration.

Note: Depending on time reward may go to next durations

Returns Promise, which:

• resolved with RateLimiterRes
• rejected only for RateLimiterRedis if insuranceLimiter isn't setup: when some Redis error happened, where reject reason rejRes is Error object
• rejected only for RateLimiterCluster if insuranceLimiter isn't setup: when timeoutMs exceeded, where reject reason rejRes is Error object

Contribution

Make sure you've launched npm run eslint before creating PR, all errors have to be fixed.

You can try to run npm run eslint-fix to fix some issues.

Appreciated, feel free!