Queue Keeps Growing

Retry timed out requests server side so the work already sitting in the queue eventually produces a response the caller accepts on the second attempt

Apply admission control at the front door: reject arrivals immediately once queue wait exceeds the caller deadline, and drop queued items already past it

Increase the queue capacity so the surge is buffered rather than rejected, and let the backlog drain once the arrival rate falls under the service rate

Raise the worker thread count so that more requests run concurrently, giving the saturated CPUs more work in flight to interleave between blocking calls and stalls