sync

Define a managed database with worker.database, then implement a worker.sync execute function that fetches pages from the API and returns upsert changes with a hasMore flag. Deploy with ntn workers deploy and the sync runs on a schedule.

Use replace mode when the dataset is small or the API lacks change tracking; the runtime auto-deletes records missing from each full cycle. Use a backfill plus incremental delta pair when the API exposes updated_at fields, events, or cursors.

Store the pagination position in the sync state, such as an opaque cursor, page number, or keyset timestamp, and return it as nextState with hasMore true. The runtime calls the sync again with that state until the dataset is exhausted.

Yes, run ntn workers exec <key> --local to execute the sync on your machine with .env loaded, passing -d with prior nextState to test subsequent pages. OAuth-based syncs cannot run locally and require deploy plus preview instead.

If the API emits delete events, emit delete changes in the delta sync. Otherwise, replace mode's mark-and-sweep removes records absent from a full cycle, so periodic manual backfills clean up stale entries.

Eventually consistent APIs may not have indexed recent changes when the cursor advances past them. Apply a consistency buffer by never advancing the delta cursor closer than 10-15 seconds to the current time.