Platform
Connector matrix
Registration, query execution, and deterministic context profiling are separate capabilities.
| Source | Studio | Execution | DLM boundary |
|---|---|---|---|
| Fabric SQL / Azure SQL | Current | Current · pyodbc + Azure identity | Per-dataset DLM compilation; no statistics/HLL profiling |
| PostgreSQL | Current | Current · psycopg2 | Per-dataset DLM compilation; no database-wide statistics profiling |
| StarRocks | Current | Current · MySQL protocol | Per-dataset DLM compilation; no statistics/HLL profiling |
| MySQL / MariaDB | API only | Current · pymysql | Not in Studio’s source picker |
| Trino | Registration only | Target | No driver |
| Parquet, Delta, Iceberg in storage | Current · registered as a catalog, not a connector | Current · the Engine, read in place | Statistics and cube cells per table |
Each platform query targets one selected SQL source. Cross-source federation is not implemented. The data-source test endpoint is currently a stub; use SQL Lab or a setup/admin probe for a real connection check.
A connection string is never returned by the API, and it is never stored as written. On the current path the string goes to the deployment’s secret store and only a reference to it is kept on the record; on the legacy path it is encrypted with a versioned key, and registration refuses with 503 rather than storing anything if no keyring is configured. A catalog source keeps a credential reference — a Key Vault URI — or no credential at all, where the host’s own managed or workload identity is what reaches the storage.