litellm - 💡(How to fix) Fix Cooldown TTL doesn't distinguish 429-rate-limit from 429-quota-exhausted (question / possible bug)

hey, posting this as a question first since i may have misread the code. happy to be told i'm wrong.

reading litellm/router.py at the current main, the Router class uses cooldown_time (a single TTL) and allowed_fails (a count). when a deployment exceeds allowed_fails, it goes into CooldownCache for cooldown_time. operators can configure both per deployment via cooldown_time kwarg or environment.

the failure mode i hit and that prompted this question: a free-tier provider (groq, in this case) returned 429 with body containing "you exceeded your monthly quota". my router (a homegrown thing modeled on this pattern) treated it the same as a transient 429-rate-limit, retried every 60s for hours.

reading the code i couldn't find a path where the body of a 429 is inspected to disambiguate "wait 60 seconds" from "wait until the period rolls over". cooldown_time is uniform.

questions:

1. is there a config knob i missed that does this distinction by body keyword?
2. if not, is "user should configure cooldown_time = 86400 per provider that has monthly quota" the intended workaround?
3. would you accept a PR that adds an optional body-keyword classifier (off by default, opt-in via config) for users who route across mixed free-tier providers?

context: i ran into this hard enough to write a small library that does the disambiguation: https://github.com/eleata/resilient-llm-router

my version is alpha and not feature-equivalent to litellm. if option 3 above sounds useful upstream, i can extract the parser as a small PR rather than recommend my lib.

happy to be wrong about any of the above. let me know.