Redis

Requires a redis datastore to supply the connection.

Description

Each counter is a key that expires on its own once its period is over, so no cleanup is needed. Each counter is updated atomically on the server using the server’s clock, so concurrent requests from several instances never lose a count. In cluster mode the counters for one request can live on different servers, which costs a round trip to each of them in parallel.

Also accepts the policies and options shared by every store.

Example

YAML
JSON

datastores:
  - name: redis
    id: redis-0
    mode: cluster
    servers:
      - host: redis-0.internal
        port: 6379
      - host: redis-1.internal
        port: 6379
      - host: redis-2.internal
        port: 6379

ratelimit:
  name: redis
  datastore: redis-0
  policies:
    - id: standard
      rules:
        - segment:
            - user
          limit: 50
          period: 1

{
  "datastores": [
    {
      "name": "redis",
      "id": "redis-0",
      "mode": "cluster",
      "servers": [
        {
          "host": "redis-0.internal",
          "port": 6379
        },
        {
          "host": "redis-1.internal",
          "port": 6379
        },
        {
          "host": "redis-2.internal",
          "port": 6379
        }
      ]
    }
  ],
  "ratelimit": {
    "name": "redis",
    "datastore": "redis-0",
    "policies": [
      {
        "id": "standard",
        "rules": [
          {
            "segment": [
              "user"
            ],
            "limit": 50,
            "period": 1
          }
        ]
      }
    ]
  }
}

Configuration

Parameter Description Type Required Default

Parameter	Description	Type	Required	Default
`datastore`	The ID of a redis datastore to use	string	Yes	None
`keyprefix`	A prefix for every key. Changing it resets every counter	string	No	ratelimit:

datastore

The ID of a redis datastore to use

string

Yes

None

keyprefix

A prefix for every key. Changing it resets every counter

string

ratelimit:

Deployment Recommendations

The cluster mode of the datastore should be used for high availability. The ring mode doesn’t replicate, so losing one server resets every counter stored on it.

Configure redis with a maxmemory-policy of noeviction. An evicted key resets the limit for whoever it belonged to. If you use calendar periods of a day or longer, also turn on persistence (AOF) so a restart of redis doesn’t reset everyone’s quota. Replication is asynchronous, so a failover can lose the last few increments.

Use a separate redis cluster for rate limiting and caching so the memory pressure and policies of the two concerns don’t conflict.