Struct DeadlineScheduler 

pub struct DeadlineScheduler<DS: DeadlineStore> { /* private fields */ }

A background task that polls DeadlineStore::due_now and dispatches deadline commands to the owning processes via a caller-supplied function.

Obtain via EngineContext::run_deadline_scheduler and drive by spawning DeadlineScheduler::run in a Tokio task.

Dispatch function

The dispatch function receives a fired Deadline and returns a future that dispatches the appropriate timeout command to the process. The function is responsible for resuming the correct workflow and calling execute. After the future completes, the scheduler cancels the deadline from the store regardless of the dispatch outcome (to prevent re-firing).

use std::time::Duration;

let scheduler = ctx.run_deadline_scheduler(
    |deadline| async move {
        tracing::warn!(
            deadline_id = %deadline.deadline_id(),
            label = %deadline.label(),
            "deadline fired",
        );
        Ok(())
    },
    100,
    Duration::from_secs(30),
);
tokio::spawn(async move { scheduler.run().await });

Implementations

impl<DS: DeadlineStore> DeadlineScheduler<DS>

pub async fn run(self)

Run the deadline poll loop until the shutdown token is cancelled.

Cancellation is observed between deadlines and during the idle sleep, so a deadline already being dispatched runs to completion. A deadline left undispatched stays registered and fires on the next start — it is due, so the next due_now returns it again.

impl<DS: DeadlineStore> DeadlineScheduler<DS>

pub fn with_heartbeat(self, heartbeat: Arc<AtomicI64>) -> Self

Attach a liveness heartbeat to this scheduler.

The scheduler will store the current UTC Unix timestamp (seconds) into heartbeat at the end of every poll cycle.

pub fn with_shutdown(self, shutdown: CancellationToken) -> Self

Attach a graceful-shutdown token.

Cancelling it makes DeadlineScheduler::run return at the next deadline boundary or immediately out of its idle sleep, so the caller can close the event store once the scheduler has stopped writing to it.