ingestr Step 2 of 3
Data migration with ingestr
Choose the migration method
Start with the source's change behaviour, not the tool's feature list.
| Method | Choose it when | Watch for |
|---|---|---|
| Full load | The data is small, mostly static, or cheap to rebuild, and downtime is acceptable. | Writes during the copy can create an inconsistent target unless you use a consistent snapshot or pause writes. |
| Cursor-based incremental | The source exposes a reliable timestamp, version, partition, or monotonic ID and periodic freshness is enough. | Hard deletes and updates that do not advance the cursor can be missed. Late-arriving records need a lookback or a repair job. |
| Log-based CDC | You need low downtime, exact inserts and updates, or hard deletes from a supported database log or change feed. | You need source privileges, log retention, a stable key, monitoring for lag, and a plan for schema changes. CDC is not automatically zero latency. |
| Hybrid load plus CDC | A bulk copy is faster, but applications must keep writing while the copy runs. | The snapshot boundary and CDC start position must overlap correctly, and replays must be idempotent. |
If the table is small enough to rebuild in seconds, a full load may be the safest choice. Complexity is not a virtue by itself. If the source is a busy production database and the cutover window is short, a consistent initial load followed by CDC is more appropriate. If the source is a SaaS API, a cursor-based incremental load is usually the practical option because transaction logs are not available.