@jonhoo I think all the threads have been addressed.

I thing that I have just realized is that we do not allow to Worker::run a terminated worker, and we drop the processing workers we have spawned, but Worker::run_one is still possible since the coordinating worker is not dropped.

Imagine the situation where the server told us to stop the consumption (we get this message in the heartbeat task on the coordinator), and we do so and mark the coordinator as terminated. What they can do then is take this worker and call run_one in a loop mimicking this way the Worker::run logic, but without HEARTBEATing and - most importantly - breaking the server instruction we've received previously. I think this is not something we want to be possible, wdyt?