Arty Executor

Async task executor for the Arty Runtime.

arty_executor is an internal implementation detail of the arty crate. It is responsible for every stage of task execution, including registration, polling, wake processing, cancellation, completion, telemetry, and panic handling. Applications should use arty instead of depending on this crate directly.

The purpose of the executor is to ensure that async tasks registered with the executor make progress, quickly and efficiently reacting to await statements that complete.

The executor is the internal building block through which arty implements task execution. Application code interacts with arty’s own task and join APIs and has no direct interaction with this crate.

Design tenets

The executor is single-threaded. If you want to execute tasks on multiple threads, you need to run multiple executors on different threads. The remote module provides pooled channels for the common runtime-integration scenario of returning task results to a different thread.

Tasks cannot be taken out of the executor - the only way for them to end is to either end naturally (with the future returning Poll::Ready) or for the executor to be shut down. Not only is there no “remove” function but similarly, there is no “cancel” function - once a task has started executing, the only thing that can terminate it is the task itself, by completing.

In a steady state, the executor is allocation-free, as all memory used by the executor is reused for new tasks when old ones complete.

Cloned task wakers may outlive their tasks. After completion or cancellation, an escaped waker remains callable but becomes an inert no-op and cannot delay executor shutdown.

Join handles may outlive the executor and report shutdown cancellation through JoinError. A panic in a directly registered executor task remains fail-fast.