Mongo

RateLimiterMongo

MongoDB >=3.2

Usage

RateLimiterMongo creates unique collection for each rate limiter keyPrefix.

It supports mongodb native and mongoose packages.

See all options 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
};
let mongoConn
const dbName = 'somedb';

// For mongoose version <= 5
mongoConn = mongoose.createConnection(`mongodb://127.0.0.1:27017/${dbName}`, mongoOpts);

// For mongoose version > 5
try {
  mongoConn = await mongoose.connect(`mongodb://127.0.0.1:27017/${dbName}`);
} catch (error) {
  handleError(error);
}

const opts = {
  storeClient: mongoConn,
  points: 10, // Number of points
  duration: 1, // Per second(s)
};
  
const rateLimiterMongo = new RateLimiterMongo(opts);
rateLimiterMongo.consume(remoteAddress, 2) // consume 2 points
  .then((rateLimiterRes) => {
    // 2 points consumed
  })
  .catch((rateLimiterRes) => {
    // Not enough points to consume
  });


/* --- 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
};

const mongoConn = MongoClient.connect(
  'mongodb://localhost:27017',
  mongoOpts
);

const opts = {
  storeClient: mongoConn,
  dbName: 'somedb',
  points: 10, // Number of points
  duration: 1, // Per second(s)
};

const rateLimiterMongo = new RateLimiterMongo(opts);
rateLimiterMongo.consume(remoteAddress, 2) // consume 2 points
  .then((rateLimiterRes) => {
    // 2 points consumed
  })
  .catch((rateLimiterRes) => {
    // Not enough points to consume
  });

See all options here

Connection to Mongo takes milliseconds, so any method of rate limiter is rejected with Error, until connection established.

Note: if you use mongoose package, be careful with Operation Buffering. If Operation Buffering is enabled, mongoose connection is not established yet and your limiter consumes points before that, it may prevent creation of index by key and brake the limiter.

insuranceLimiter can be setup to avoid errors, but all changes won't be written from insuranceLimiter to RateLimiterMongo when connection established

Benchmark

Endpoint is pure NodeJS endpoint launched in node:10.5.0-jessie and mongo:3.6.5-jessie Docker containers with 4 workers

Endpoint is limited by RateLimiterMongo with config:

new RateLimiterMongo({
  mongo: mongo,
  points: 20, // Number of points
  duration: 1, // Per second(s)
});

By bombardier -c 1000 -l -d 30s -r 2000 -t 5s http://127.0.0.1:3000

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

Statistics        Avg      Stdev        Max
  Reqs/sec      1997.87     429.40    3869.98
  Latency        4.75ms     3.32ms    68.21ms
  Latency Distribution
     50%     4.15ms
     75%     5.43ms
     90%     6.95ms
     95%     8.79ms
     99%    18.96ms
  HTTP codes:
    1xx - 0, 2xx - 15000, 3xx - 0, 4xx - 45014, 5xx - 0

Heap snapshot statistic

Advanced features

MongoDB Sharding options

Range sharding by unique key index

MongoDB rate limiter creates index by key document attributes by default. It may be used for sharding.

Example options:

  const opts = {
    mongo: mongoConn,
    dbName: 'app',
    tableName: 'users-rate-limit',
    keyPrefix: '', // no need to prefix to simplify ranges
    points: 5,
    duration: 1,
  };

dbName is set to app and tableName set to users-rate-limit (in case of MongoDB it is collection name)

Prepare sharded collection and index launching next on router:

sh.addShardTag("shard0000", "OLD")
sh.addShardTag("shard0001", "NEW")

sh.enableSharding("app")
sh.shardCollection( "app.users-rate-limit", { key: 1 }, { unique: true })

We've assigned a tag OLD and NEW to examples shards.

Depending on what is used to identify user: ID, IP address, etc; you should choose correct range options. And ranges should be described with the same values types.

sh.addTagRange("app.users-rate-limit", { key: MinKey }, { key: 5000 }, "OLD")
sh.addTagRange("app.users-rate-limit", { key: 5000 }, { key: MaxKey }, "NEW")

Probably, this example doesn't provide a good shard key frequency, may be users with IDs less than 5000 don't use our app anymore. Anyway, you get the idea.

consume function should receive userId as number now and it should be routed to correct shard by configured range. userId is saved as it is, as we set keyPrefix to empty string.

Range sharding by unique compound index

Example options:

  const opts = {
    mongo: mongoConn,
    dbName: 'app',
    tableName: 'users-rate-limit',
    keyPrefix: '', // no need to prefix to simplify ranges
    points: 5,
    duration: 1,
    indexKeyPrefix: {country: 1},
  };

With option indexKeyPrefix MongoDB creates compound index {country: 1, key: 1}. (key is added by default)

Prepare sharded collection and add ranges on router:

sh.addShardTag("shard0000", "EU")
sh.addShardTag("shard0001", "AU")

sh.enableSharding("app")
sh.shardCollection( "app.users-rate-limit", { country: 1, key: 1 }, { unique: true })

sh.addTagRange("app.users-rate-limit", { country: 'Netherlands', key: MinKey }, { country: 'Netherlands', key: MaxKey }, "EU")
sh.addTagRange("app.users-rate-limit", { country: 'Australia', key: MinKey }, { country: 'Australia', key: MaxKey }, "AU")

Set additional country attribute, whenever consume or any other function is called from limiter

try {
  const key = ipAddress;
  const country = getCountryByIP(ipAddress);
  const options = {attrs: { country }};
  const rlRes = await rlTest.consume(key, 1, options);
} catch (err) {
  res.status(429).json(err);
}

As country is a part of unique index now, it always should be set.

MongoDb durability options

MongoDb saves snapshots to disk with fsync and makes journaling by default. It results to extra disk I/O.

If you already use MongoDb as data store and have high traffic like 1000 req/sec or more, you may find it useful to launch the second MongoDb instance with options:

--syncdelay 0 : disable making snapshots to disk
--nojournal : disable journal
--wiredTigerCacheSizeGB 0.25 : set minimum memory

Builtin TTL index automatically deletes expired documents. Document for one key is 68 bytes in size. MongoDb stores information for about 4 million keys in 256Mb.

Here is a small test of MongoDb with different options:

It processes 10k, 100k and 250k writes for 10k random keys for:

MongoDB default settings

10k  926ms
100k 4475ms
250k 13254ms

MongoDB fsync and journaling disabled

10k  900ms
100k 4323ms
250k 12407ms

It is about 5% faster with disabled fsync and journaling, but avoiding extra disk I/O is worth.