Optimizing the Cache Storage
Last updated: 2 minutes read.
Tyk creates the API cache in Redis, as it gives high performance and low latency. By default, the cache will use the same database that is used to store the API keys, minimizing the deployment footprint.
For multi-data center deployments, the Data Planes have a locally deployed Redis. This enables them to have a localised cache close to the traffic-serving Gateways.
The cache key is used as the Redis key, for quick lookups.
For high-traffic systems that make heavy use of caching, it can make sense to use separate Redis databases for cache storage and for API keys, at the expense of increased deployment footprint.
Configuring a separate cache
To enable a separate cache server, you must deploy additional Redis instance(s) and apply additional configuration within your Tyk Gateway’s tyk.conf
configuration file.
You must
- set
enable_separate_cache_store
totrue
- provide additional Redis connection information in the
cache_storage
section
For example:
{
"enable_separate_cache_store": true,
"cache_storage": {
"type": "redis",
"host": "",
"port": 0,
"addrs": [
"localhost:6379"
],
"username": "",
"password": "",
"database": 0,
"optimisation_max_idle": 3000,
"optimisation_max_active": 5000,
"enable_cluster": false
}
}
The configuration of the separate Redis Cache is the same (and uses the same underlying driver) as the regular configuration, so Redis Cluster is fully supported. If you set enable_cluster
to false
, you only need to set one entry in addrs
.
Note
Prior to Tyk Gateway v2.9.3, hosts
was used instead of addrs
; since v2.9.3 hosts
has been deprecated.