Use Redis as a shared metrics data store to coordinate back off in the aggregate and to track collective throughput (and the delta between functional baseline and when you’re exceeding counterparty limits). Make workers aware of allowance state, and responsive to it and limits.

Via this mechanism, you should be able to pause your worker fleet as it scales out as well as regulate its request rate while monitoring on health of the steady state interface between your workers and other systems.