Struct Listener 

pub struct Listener<Src, Snk> { /* private fields */ }

Watches every table in a MappingSpec and keeps a graph sink current.

Implementations

impl<Src: TableSource, Snk: GraphSink> Listener<Src, Snk>

pub fn new( spec: MappingSpec, registry: TransformRegistry, source: Src, sink: Snk, state: SyncState, ) -> Self

pub fn metrics(&self) -> &Metrics

The cumulative run metrics so far (counters folded after every sync_once / poll round). See Metrics and crate::metrics::render_prometheus.

pub fn with_asset_emitter( self, emitter: Arc<dyn AssetEmitter>, uri_prefix: impl Into<String>, ) -> Self

Attach an AssetEmitter: after every successful sync round the listener emits one Airflow asset event per ingested table, with URIs {uri_prefix}{table}. See crate::airflow.

pub async fn sync_once(&mut self) -> Result<RoundReport>

One pass over all mapped tables. A failing table does not stop the others; its state is not advanced, so the next pass retries it.

pub async fn run( &mut self, interval: Duration, on_round: impl FnMut(RoundReport) -> bool, ) -> Result<()>

Poll forever. The sleep between rounds is the base interval on the happy path, but a fully-failed (no-progress) round backs the next sleep off exponentially up to interval * 10 (see Backoff); the first round that makes progress resets it. Errors are carried in the round reports handed to on_round; return false from it to stop.

Signature-compatible with the original fixed-interval run; for a custom cap use run_with_backoff.

pub async fn run_with_backoff( &mut self, backoff: Backoff, on_round: impl FnMut(RoundReport) -> bool, ) -> Result<()>

Poll forever with an explicit Backoff schedule. The happy-path cadence is backoff.interval(); consecutive no-progress rounds grow the inter-round sleep up to backoff.max(), resetting on the first round that makes progress. run is this with a interval … interval*10 default. Return false from on_round to stop.