# Diagnose ingestion failures after an upgrade

When an importer starts failing after a deployment change, identify the failing request before attributing it to a release. Coordinator DB does not publish a compatibility promise for another product's version numbers. Diagnose against the documented formats, schema and request limits of the installed build.

## Reduce the reproduction

Save the status code, exception code, query identifier, destination schema and a redacted sample that still fails. Separate an ordinary SQL body from an attached upload: ordinary HTTP SQL requests require Content-Length and have a 16 MiB ceiling, while attached INSERT uploads have a different size contract.

Check the registered format spelling, field names, nullability and numeric ranges. Format names use snake_case. Place statement `SETTINGS` before `FORMAT`; everything after the format name is input data.

## Compare with a small valid batch

Create an isolated test table and send one known-good row. Add the failing fields gradually until the conversion or request error is clear. Do not repeatedly replay a production batch whose outcome is unknown. Reconcile committed rows with stable identifiers first.

Keep the original deployment details in your incident record and test a correction with the same reduced input before resuming normal ingestion.

See [HTTP limits](https://coordinatordb.com/docs/concepts/features/interfaces/http), [format reference](https://coordinatordb.com/docs/reference/formats) and [INSERT INTO](https://coordinatordb.com/docs/reference/statements/insert-into).