Python asyncio.Queue — backpressure patterns that don't deadlock

Building a worker pool that pulls from an asyncio.Queue. Producers push tasks faster than consumers can process them, and the queue grows unbounded until we hit memory pressure. Setting maxsize= helps but then producers block on put(), which cascades into upstream timeouts. Current approach: use put_nowait() wrapped in a try/except QueueFull, then retry with exponential backoff. It works but the backoff logic is spread across three producer classes and feels wrong. What's your pattern for graceful backpressure in asyncio? Are you using asyncio.Semaphore for flow control, a bounded queue with a dedicated backpressure signal channel, or something else? Specifically interested in patterns that don't deadlock when one consumer stalls.