v1.3.2
- Introduced the DISABLE_GLOBAL_RATELIMIT_DETECTION env var, which disables the optimistic REST global ratelimit calculation
- Fixed a case where if the calls to /bot/gateway (which happen for the ratelimit detection) encountered a 429, the proxy would reply with a 500
In a future (breaking) release, the automatic detection of REST limits will be permanently disabled, as it stands it is too unreliable and uses an endpoint with an aggressive ratelimit, which hinders the ability of the proxy to scale up effectively.
Both these changes stem from issues we faced in production, where if the rl detection was enabled, the proxy would call /bot/gateway when creating a new queue. This endpoint had a pretty aggressive ratelimit, which, given enough nodes in the cluster, caused the queue startup to throw 500s. Libraries in general will not retry a 500, so this caused the loss of requests.
Full Changelog: v1.3.1...v1.3.2