The product is the shape of the tables, not the copy
MongoDB collections do not have a fixed schema to diff. Conversion Studio infers a model from the documents, then lets you decide which nested fields become child tables and which stay as JSON. MySQL is a complete target on the tested type-mapping matrix, and CDC is available on this path.
Four stages, each reviewable
Nothing writes to MySQL until the mapping is the one you will live with.
Read the collections
Explorer relabels Data and Columns to Documents and Schema. Conversion Studio samples documents, counts records, and lifts nested structure into a canonical model. There is no tabular schema reader.
Draft MySQL tables
Generate a proposed relational schema. Nested documents can become parent and child tables, or selected fields can stay as JSON. Review the model, the field map, the ER diagram, and the DDL.
Choose flatten, child table, or JSON
Field overrides: ignore, force JSON, rename, or set a target logical type. Structural choices for nested data: default, flatten, child table, or JSON. Re-preview until the shape is deliberate.
Load with CDC or a full copy
Run with live progress, written and rejected counts, rejected samples, and per-entity validation. CDC is implemented for MongoDB sources, so the cutover can be a snapshot plus change passes.
What is automated, and what is not
| Aspect | Status on this path |
|---|---|
| Source modelling | Automated from documents. MongoDB has no structural schema reader, so there is no generic export dialect for collections. |
| Target schema generation | Automated. MySQL is a complete Conversion Studio target with the tested type-mapping matrix. |
| Schema comparison | Not available for MongoDB. Compare on this engine is data-only. After the move, compare MySQL to MySQL. |
| Change data capture | Available for MongoDB sources. Same product defaults as other CDC engines: snapshot first, then change passes. This shortens cutover; it is not a permanent replication pipeline. |
| Nested documents | You choose: default, flatten, child table, or JSON. That decision is the migration, more than the copy job. |
| Application query changes | Not automated. Drivers, aggregations, and document-shaped APIs have to be rewritten. Risk assessment can still show which repositories touch related names once you have tables. |
Relational operations start on day one
Once the data is relational, schema and data comparison, sync scripts, and drift snapshots apply to MySQL environments. They do not apply back to the MongoDB source as a structural diff.
Approvals, promotion, rollback generation, and the hash-chained audit trail cover the MySQL estate from the first change after cutover.
During a phased cutover both engines can stay connected. Credentials stay on the Executor. Production reads remain audited.