Per-ractor (or process) semaphore

[nateberkopec]

How do I ensure a limit of tasks on a per-process basis? Storing a Semaphore in a class variable or constant and then using MY_SEMAPHORE.async results in fibers being used across threads. e.g., for the url-fetch-in-parallel example, how do we ensure only 10 tasks are created in the current Ractor?

[ioquatix]

If you're trying to enforce a concurrency limit in a non-homogeneous environment (e.g. multiple threads, Ractors, or processes), it's impossible to do so using an in-memory Async::Semaphore shared via a class variable or constant. These constructs only work within a single event loop, and each Process, Thread or Ractor has its own.

In environments like Falcon, where your app might run across multiple schedulers or even processes, any meaningful concurrency limit must be enforced using an external coordination mechanism—something that all units of concurrency can observe and respect.

Examples include:

• Shared memory semaphores (e.g., via FFI or native extensions).
• UNIX pipes or sockets.
• Redis-based counters or distributed semaphores.

In short: Async::Semaphore is local to one event loop. If your runtime isn't, you need to coordinate outside of it.

[ioquatix]

See https://github.com/socketry/async-examples/tree/main/shared-semaphore for an example.