Struct AsyncPromotionPool 

pub struct AsyncPromotionPool { /* private fields */ }

Bounded, drop-oldest async promotion pool.

See module docs for design rationale.

Implementations

impl AsyncPromotionPool

pub fn new(opts: PoolOpts) -> Arc<Self>

Construct a pool with a no-op executor. Useful only for tests / dry runs where you want metrics but no actual promotion side-effects.

pub fn new_with_executor( opts: PoolOpts, executor: PromotionExecutor, ) -> Arc<Self>

Construct a pool with a caller-provided executor closure.

Spawns opts.worker_count tokio tasks that drain the queue. Each task holds a Weak<Self> so the pool is dropped cleanly once the caller releases its Arc and the workers exit.

pub fn schedule(&self, request: PromotionRequest) -> ScheduleOutcome

Hand a promotion request to the pool.

Never blocks. If the queue has capacity, the request is enqueued and Queued is returned. If the queue is full, the oldest request is popped (and dropped) to make room — the caller learns this via DroppedQueueFull { evicted_oldest: true }. In the rare case where the queue is so contended that even the pop fails, the new request is dropped: DroppedQueueFull { evicted_oldest: false }.

pub fn shutdown(self: Arc<Self>)

Signal workers to drain remaining work and exit.

Workers will process at most drain_budget more requests after the shutdown flag is observed, then return. New schedule calls after shutdown are rejected (counted in dropped_total).

pub fn metrics(&self) -> PromotionMetrics

Snapshot of the pool’s atomic counters.

Each counter is read with Relaxed ordering; the snapshot is not strictly atomic across counters (it can show, e.g., one more queued_total than the queue depth implies if a worker is mid-pop). This is acceptable for monitoring; consumers that need a strictly consistent view should sample twice and take the difference.