Response caching middleware.

For each HTTP request (the Fetcher only supports GET requests) it looks up a response in the cache using the URL as key.

If a cached response is found, it is immediately returned. If nothing is found, it executes the HTTP request and then it caches successful (200..299) responses before returning them to the caller. Unsuccessful responses and redirects (e.g. 302, 404 or 500) are never cached and always returned to the caller.

If writing to the cache fails for some reason, an error is logged but the request chain is not halted and the response is normally returned.

The entire cache layer (lookups and writes) can be bypassed by passing the cache: false option to Routemaster.Fetcher.get/2.

Caveats

At the moment the caching strategy is very simple and does not implement the features of the official Ruby client. For example, it doesn’t separately cache different representations of the same entities (i.e. API version and language headers) and as a consequence the Fetcher doesn’t yet support that kind of granularity.

Also, the entire Tesla.Env structure is compressed and cached as a Redis string, while it might be more efficient (space wise) to use a Redis Hash for different fields and rebuild the struct later.

Compressing the data in Elixir is also something to be benchmarked, since it really boils down to CPU time vs Network IO time wile exchanging payloads with the cache.