Single‑transaction schema migration versus compensating transaction: trade‑off for rollback safety after a required‑field addition

Direct answer

An aborted FaunaDB FQL transaction most likely rolls back document data writes, but it does not automatically undo a required-field addition or reindexing. The field can persist on existing documents after the abort, and index entries created under the new schema can remain, producing orphaned entries. The history API restores document versions only; it does not reverse schema definitions or repair indexes, so it cannot by itself reconstruct the pre-change document state.

Likely behavior versus confirmed facts

Likely, not confirmed

• A required-field addition issued inside an FQL transaction may survive transaction abort because schema operations are not automatically atomic with data writes.
• Index entries written during the aborted transaction can remain afterward, causing queries on documents that lack the required field to fail or return unexpected results.
• The FaunaDB history API addresses document content versions, not schema or index state.

Confirmed general principle

Rolling back a data transaction reverses data mutations; it does not inherently reverse DDL or index maintenance performed outside the transactional data scope. Treat FaunaDB-specific behavior as version-, edition-, and region-sensitive until verified against your target release.

Verification in a non-production database

1. Add the required field inside an FQL transaction, then abort the transaction.
2. Inspect existing documents to determine whether the field persists.
3. Inspect affected indexes for entries that reference the new schema.
4. Use the history API on a sample document before and after the abort to confirm whether it returns pre-change document content or only document versions.

Compensating steps if the change persists

1. Run an explicit transaction to remove the required field from affected documents, or backfill it where the field is required for forward compatibility.
2. Drop and recreate, or rebuild, the affected indexes so they no longer contain orphaned entries from the aborted schema change.
3. Run read and write queries against a sample of documents to confirm index consistency before resuming traffic.

Missing diagnostic detail

Was the required-field addition issued as DDL inside the same FQL transaction, or through a separate schema administration call? This determines whether the operation was ever within transaction scope and changes the compensation procedure.

This draft is conservative because FaunaDB-specific DDL rollback semantics are unconfirmed in available training data. Verify against your target release before applying to production.