This is a cache for the feature flipper that has proven itself over the past year on multiple production deployments of Adobe services.
The API is identical to the client API
No caching is currently done for GET /set/:set_id/features?user_id=\<user id\>
On a cache miss, if there is no data to return, the cache will call API Gateway synchronously. On subsequent requests, if the TTL of the cached data has not expired, the cache returns the data it has. If the TTL has expired it will additionally make an async request to API Gateway to refresh its data. If that request fails it will keep the last good set of data it has, otherwise it will refresh the cache.
The above behavior is identical for the GET /sets and
GET /set/:set_id/features APIs. The only variation is on the TTLs.
TTL is the time for cache data to live before attempting to refresh it.
These numbers are based on our usage patterns and scale at which we operate our feature flipper and should probably be configurable. Currently they're hardcoded.
Sets: 30 minutes
Set aliases: 1 hour
Features in a set: 10 minutes
We typically follow 12-factor. The cache is configured with two environment variables.
PORT the port to bind to
FEATURE_FLIPPER_URI the partial URL that the
client API paths will be appended to. An example is
https://abcde12345.execute-api.us-west-2.amazonaws.com/prod
A Docker image is available at phylake/feature-flipper-cache. The Docker tag
is a stripped down version of the git tag.
https://github.com/adobe-platform/feature-flipper/tree/cache-v1.0.0
equals
phylake/feature-flipper-cache:1.0.0