IndexedDB persistence

Use this backend for a durable store in the browser. It keeps the whole log in memory, like persistence.memory, and writes every change to an IndexedDB database so a later page finds it again.

EventStore.persistence.indexeddb<TMap>(
  options: IndexedDbEventStoreOptions
): Promise<TypedEventStore<TMap>>

EventStore.persistence.indexeddb.destroy(
  options: IndexedDbEventStoreOptions
): Promise<void>

interface IndexedDbEventStoreOptions {
  name: string;
}

The optional TMap schema map is described in EventStore; it defaults to an unconstrained map.

import * as EventStore from "@jolly-pixel/event-store";

using store = await EventStore.persistence.indexeddb({
  name: "jolly-events:my-workspace"
});

The factory opens or creates the database name, with an events store keyed by eventId and a tags store keyed by [assetId, name], and loads both before it resolves. Reads never touch IndexedDB afterwards. The indexedDB global is only reached when the factory runs, so importing the package in Node.js stays safe. Node.js tests can install the IndexedDB globals with fake-indexeddb/auto.

Writes

append, tags.create, tags.delete and compact issue their IndexedDB request in a transaction opened before they return, then update the in-memory log. A call that changes nothing, such as deleting a missing tag, writes nothing. Nothing waits on a timer or a promise first, so a page that reloads right after an append finds the event once IndexedDB has committed it. Another connection to the same database reads it as soon as that transaction ends, even if the writing store is never closed.

When IndexedDB aborts a write, quota errors included, the in-memory log keeps the change but the store stops writing: every later change fails with the error IndexedDB reported and leaves the in-memory log as it was. Reads keep working.

The store closes its connection when another connection deletes or upgrades the database. Later writes then fail with IndexedDB's InvalidStateError.

Ids and versions

A reopened store continues from the highest eventId and from each asset's highest eventVersion it loaded. Two stores open on the same database at the same time do not see each other's appends: give each database one writer.

Compaction

compact deletes the superseded events from the database in the same transaction pattern as other writes. The reclaim option has no effect: the browser manages the space of the database.

Lifetime

close() discards the in-memory log and closes the connection. IndexedDB still commits the writes already issued. destroy deletes the database name. An open store on it closes its connection first, then fails on its next write.