In part 1 we built the minimal io_uring loop: setup, mmaps, SQE/CQE draining, accept/recv/send via opcode dispatch.This was the first step to understanding how the kernel interface works but the dispatch logic was basically hand coded, a state machine that would not scale. This second part is all about introducing the asynchronous model to await data pushed by the kernel. For simplicity, in this article scope we will simply return the number of bytes received, in later parts it will be described how to adapt this to return the actual request data and parse it. As usual, the full source code can be found at .... Why not just use Task? The direct option would be to just return a Task, completed by the dispatcher when a CQE arrives, to avoid the allocations this would cause for every asynchronous read we are going with the zero allocation option. ValueTask over a reusable source, the high performance path, the source is a single object that lives as long as the TCP connection.…