Skip to content

Release v19.122.0

Choose a tag to compare

@github-actions github-actions released this 03 Nov 23:18
74f9276

Reactors and reducers can now handle events in batches: the kernel delivers events to clients as pages with one result per partition, and the .NET client lets a reactor or reducer declare a batch method with [Observes<TEvent>].

Added

  • Reactors can declare batch methods: a method taking an EventBatch and marked [Observes<TEvent>] once per event type is called once per page with every event of those types, in sequence order, alongside the reactor's per-event methods. See Handling events in batches.
  • Reducers can declare one batch method that receives all of a partition's events and the current read model, and returns the new read model. See Reducing events in batches.
  • BatchContext tells a batch method where a page came from (Live, CatchUp, Replay or Retry), which partitions it holds, its sequence range, deadline and cancellation token.
  • Batch methods report a single partition with Fail or Defer on the EventBatch; a deferred partition is delivered again after the given delay without being failed or counting towards quarantine.
  • The reactor and reducer gRPC contracts gain page delivery, observation origin and deferral, so other clients can receive pages and answer per partition. Existing clients keep receiving single-partition deliveries.

Changed

  • Registration fails with a named error when an event type would reach more than one method of a reactor, when a batch method lacks [Observes<TEvent>] or returns events, or when a reducer mixes a batch method with per-event methods.