Pular para o conteúdo
Português (BR)

PersistenceQuery

Este conteúdo não está disponível em sua língua ainda.

Defined in: src/persistence/query/PersistenceQuery.ts:32

Read-side query layer for the journal. Designed for projections — processes that materialise a read-model by sweeping events out of the write-side journal and feeding them into a user handler.

Two flavours of query:

  • current* — one-shot snapshot of events currently in the journal at call time. Resolves to a Promise. Use this for a batch backfill or a self-contained report.
  • events* — continuous live stream. Polls the journal at pollIntervalMs for new events. Yields an AsyncIterable so consumers can for await (const ev of stream) .... The stream stays open until the consumer breaks out of the loop or calls return() on the iterator.

Delivery guarantees: at-least-once. A projection that fails mid-event must accept that the event will be redelivered after restart. Handlers therefore have to be idempotent.

Push where the journal can push (#42). A live query subscribes to journal.events when the journal exposes one, and polls at pollIntervalMs when it does not. So an in-process journal delivers within a tick while a cross-process one (Cassandra, Postgres) — where an in-process bus could not see another node’s writes anyway — keeps the polling loop. Which of the two is in play is invisible to the consumer: the iteration contract is the same either way.

allPersistenceIds(options?): AsyncIterable<string>

Defined in: src/persistence/query/PersistenceQuery.ts:139

Live stream of every persistence id the journal has ever seen, plus each new one as it first appears. The stream never completes on its own — break out of the loop or call return() on the iterator.

This is the fan-out primitive: start a per-entity projection as its entity shows up, instead of polling currentPersistenceIds and diffing the result yourself.

Once per stream, not once per journal. A fresh subscription re-emits every id, so a consumer that must not act twice across a restart needs its own checkpoint — the same at-least-once posture the event queries have.

Memory. “Once per stream” is enforced with an in-process set of the ids emitted so far, so a stream over a journal with a million ids holds a million strings for as long as it runs. A lexicographic watermark would be cheaper and wrong: ids are not created in sorted order, so an id that sorts below the mark would be skipped forever, and a fan-out projection that silently never starts is the worst failure this API could have. The catch-up sweep is paged, so the peak is the set plus one page — and a one-shot enumeration that needs no set at all is currentPersistenceIdsPaginated.

LiveQueryOptions

AsyncIterable<string>


currentEventsByPersistenceId<E>(persistenceId, fromSeq, toSeq?): Promise<PersistentEvent<E>[]>

Defined in: src/persistence/query/PersistenceQuery.ts:51

One-shot read of every event for persistenceId whose sequenceNr >= fromSeq (and <= toSeq if given). Resolves once with the events known at call time.

E

string

number

number

Promise<PersistentEvent<E>[]>


currentEventsByTag<E>(filter, fromOffset): Promise<TaggedEvent<E>[]>

Defined in: src/persistence/query/PersistenceQuery.ts:78

One-shot read of every event matching filter whose offset is >= fromOffset. See TagFilter for the operator semantics.

E

TagFilter

Offset

Promise<TaggedEvent<E>[]>


currentPersistenceIds(): Promise<string[]>

Defined in: src/persistence/query/PersistenceQuery.ts:95

Snapshot of every persistence id known to the journal, as one array. Resolves once.

Kept, and not deprecated: for a journal with a handful of ids this is simply the convenient shape, and “small journal” is a permanent case rather than a legacy one. It is, however, the only method here with no bound on what it materialises — at a million ids that is a million strings in one allocation. Past the point where the array itself is a cost, use currentPersistenceIdsPaginated, which walks the same data one page at a time.

Promise<string[]>


currentPersistenceIdsPaginated(options?): AsyncIterable<string>

Defined in: src/persistence/query/PersistenceQuery.ts:112

Cursor-paginated snapshot of the persistence ids currently in the journal, ascending, yielded one at a time. Fetches pageSize ids per round-trip, so peak memory is one page rather than the whole set. Completes when the journal is exhausted — unlike allPersistenceIds it does not wait for new ids.

The cursor is a persistence id, not an opaque token. Resume a partial walk by passing the last id you handled as afterPersistenceId. Encoding it would have bought per-backend freedom that no backend here needs — every one of them enumerates ids through an index keyed on the id itself — at the price of a checkpoint no operator can read and no other process can construct.

PaginationOptions

AsyncIterable<string>


eventsByPersistenceId<E>(persistenceId, fromSeq, options?): AsyncIterable<PersistentEvent<E>>

Defined in: src/persistence/query/PersistenceQuery.ts:40

Live stream of every event for persistenceId whose sequenceNr >= fromSeq. Past events are emitted first (chronological by sequenceNr), then new events as they are appended. The stream never completes on its own — break out of the loop or call return() on the iterator to stop polling.

E

string

number

LiveQueryOptions

AsyncIterable<PersistentEvent<E>>


eventsByTag<E>(filter, fromOffset, options?): AsyncIterable<TaggedEvent<E>>

Defined in: src/persistence/query/PersistenceQuery.ts:68

Live stream of every event matching filter whose offset is >= fromOffset. Yields events ordered by (timestamp, persistenceId, sequenceNr). See Offset for offset semantics — the stream emits the offset alongside the event so the consumer can persist progress.

filter accepts either a single tag string (back-compat shortcut for { all: [tag] }) or a TagFilter object that combines all (intersect), any (union), and not (exclusion) operators.

E

TagFilter

Offset

LiveQueryOptions

AsyncIterable<TaggedEvent<E>>