## Check for existing issues - [x] I have searched the existing issues and checked that my issue is not a duplicate.

## What happened?

`Router._time_to_sleep_before_retry` only honors the upstream `Retry-After` header when a model group has a **single** deployment. As soon as a model group has **more than one** deployment (and there is ≥1 "healthy" deployment), it returns `0` and retries **instantly**, discarding `Retry-After`:

```python # litellm/router.py :: _time_to_sleep_before_retry ## base case - single deployment if all_deployments is not None and len(all_deployments) == 1: pass # -> reads Retry-After from the exception headers elif ( healthy_deployments is not None and isinstance(healthy_deployments, list) and len(healthy_deployments) > 0 ): return 0 # -> instant retry, Retry-After ignored ```

The assumption is that with multiple deployments the router can immediately fail over to a *different, healthy* deployment. But when the deployments in the group all point to the **same upstream / share the same rate-limit or token quota**, they are all throttled at once. The instant retries then just hammer the throttled…