Honor carrier quotas, absorb bursts safely, and keep internal queues from turning throttling into an outage.
429 Is a Traffic Signal, Not a Surprise
Carrier throttling is part of the contract whether it is well documented or not. A 429 response, Retry-After header, or burst quota metric is the carrier telling you to slow the system down now, not after another hundred retries.
Backpressure Must Reach the Queue
If workers slow down but the enqueue side keeps filling the queue at the same rate, you are only moving the outage around. Good backpressure propagates to schedulers, bulk-job orchestration, and the internal APIs that create work.
Carrier Reality
A carrier may publish a per-minute shipment quota while silently applying lower burst limits per token, IP, or account region. If you only tune for the published quota, the real limiter still trips first.
Throttle Proactively, Retry Selectively
A token bucket or leaky bucket keeps you under the carrier's steady-state limit. Exponential backoff handles the exceptional case. You need both. Otherwise you alternate between hammering the carrier and waiting uselessly for the queue to clear itself.