A backfill copies rows that already exist into a sink. drift sync runs one automatically the
first time a pipeline starts. Run one yourself after changing a pipeline’s query or columns,
or to rebuild an index from scratch.
drift backfill --pipeline ordersLive changes keep streaming during the backfill, and a live change to a row always wins over the backfilled copy of that row. Backfilled rows are not metered.
Rebuild into a fresh index
To change a search mapping without serving a half-built index, backfill into a new index and swap an alias when it finishes:
drift backfill --pipeline orders --index orders_v2 --swap-alias ordersSupported for Elasticsearch, OpenSearch and Typesense (aliases) and Meilisearch (index swap).
Pause and resume
drift backfill pause --pipeline orders
drift backfill resume --pipeline ordersThe keyset cursor is saved after every chunk, so a backfill also survives a restart or a deploy.
Throttle
| Flag | Default | Description |
|---|---|---|
--chunk-size | 5000 | Rows per keyset query. |
--max-rows-per-second | none | Upper bound on read rate. |
--read-from | source | A read replica URL to backfill from instead. |
Backfilling from a replica keeps the read load off your primary. Driftbase waits for the replica to catch up past each chunk’s watermark before emitting it.