firestore-read-cache

Create a module-scope cache with createReadCache or createCachedDocReader from @delfrance/data/admin/cache, keyed by document path or query predicate values. Module scope matters because Cloud Functions gen2 instances persist across invocations, so the cache survives between requests.

Never cache reads inside runTransaction, read-modify-write flows, or OAuth token lookups. A cached tx.get drops the document from the transaction read set, a cached read-modify-write decides on stale data, and a cached token can resurrect a rotated single-use refresh token.

Call the reader's invalidate method with the same key arguments immediately after the write, as exchangeAndPersist does after merging user_id onto the integracao document. This only clears the local instance; other warm instances rely on the TTL or an isFresh predicate.

No, the cache is process-scoped and instances do not coordinate. The TTL is the staleness bound for other instances, and isFresh re-validation on each hit handles fields with tighter freshness requirements without cross-instance messaging.

Inject a controllable clock via the now option when creating the cache, then advance the fake time past ttlMs and assert a second read occurs. Also call __resetAllReadCaches in beforeEach and afterEach so cached state never leaks between tests in one process.

Set negativeTtlMs to 0 whenever an absent result drives an irreversible decision, such as resolveIntegracaoByUserId returning no-account which deterministically acks and drops notifications. Caching that miss would drop events for a seller who connected seconds earlier.