I'm working with Ray async actors and I want to understand exactly what happens—at a deep technical level—when a synchronous method is called on such an actor. I know that calling a synchronous method will block, but am looking for clarity on how exactly this blocking occurs within the actor's architecture. As I see it, there are two possibilities: The event loop in the actor thread is preempted The Ray async actor runs on a thread that hosts an asyncio event loop. When the synchronous method is called, Ray's dispatcher invokes the sync method directly on that same thread, effectively pausing the event loop and running the sync method. Once the sync method finishes, the event loop resumes. This means the sync method blocks the thread by interrupting the event loop's execution. The event loop itself executes the synchronous method The synchronous method is somehow wrapped as a coroutine/Future/Task and queued into the event loop's task queue. The event loop picks it up and executes it within its normal flow. The sync method’s blocking behavior causes the event loop itself to freeze until the sync method completes. This means the blocking happens "inside" the event loop rather than as an external interruption. I'm looking for either an precise step-by-step explanation of what happens, or a reference to docs or the right parts of the source code. Continue reading...
