Does Cosmos DB provide an automatic rollback mechanism when a schema validation rule changes during a transaction?

Answer

No. Azure Cosmos DB does not provide an automatic rollback mechanism to revert writes that were committed via a TransactionalBatch if a container-level JSON schema validation rule is changed concurrently or immediately after the transaction.

Technical Explanation

The lack of rollback is due to the separation of the control plane and the data plane within the service architecture:

• Execution Context: A TransactionalBatch is validated against the schema version active at the moment the batch begins execution. If the documents satisfy the schema at that instant, the batch is eligible for commit.
• Control Plane vs. Data Plane: Schema modifications are control-plane operations. They are asynchronous and occur outside the scope of data-plane transactions. Because the schema update is not part of the transaction itself, it cannot trigger a transactional rollback of already-committed data.
• Post-Commit State: Once a batch is committed, the service does not perform retroactive validation. Documents that were valid under "Schema A" remain in the database even after "Schema B" is deployed, regardless of whether they violate the new rules.

Confirmed Service Constraints

• There are no service-side flags or specific API versions that enable schema-change-aware rollback.
• Schema validation is only enforced during write operations (create, replace, upsert); it does not affect existing data or read operations.
• TransactionalBatch atomicity is limited to a single logical partition key; it does not extend to container-level configuration changes.

Recommended Patterns for Schema Transitions

1. Versioned Schemas: Include a schemaVersion property in your documents. Handle validation logic within the application layer to support multiple versions during a transition period.
2. Maintenance Windows: Perform schema updates during periods of low activity or paused writes to avoid race conditions where batches are validated against inconsistent schema versions across regions.
3. Compensating Transactions: Use the Change Feed to detect documents that no longer adhere to the updated schema and trigger application-level logic to update or delete them.

Verification Method

To verify this behavior in a test environment:

1. Define a container schema requiring FieldA.
2. Initiate a TransactionalBatch to insert a document containing only FieldA.
3. Simultaneously update the container schema to require FieldB.
4. Confirm the batch commits successfully and the document persists despite missing FieldB.