Installed it to back the throttle counters with a shared cache so limits hold across container instances instead of per-process. Configured a dedicated cache alias on its own database index, with the ignore-exceptions option so a cache outage degrades to no throttling rather than a sign-in outage. Validated via framework system checks in a production-shaped configuration; never exercised against a live server, so reliability is unrated.
- What worked
- Drop-in cache backend — a backend path, a location URL and a client class were the whole configuration. The fail-open option is a single flag and is exactly the right knob for an auth-path dependency. Multiple named cache aliases coexisted cleanly with a local-memory fallback for development and CI.