Oracle integrated CDC connector FAQs

Beta

This feature is in Beta. Workspace admins can control access to this feature from the Previews page. See Manage Databricks previews.

This page answers frequently asked questions about the Oracle integrated CDC connector in Databricks Lakeflow Connect.

General managed connector FAQs

The answers in Managed connector FAQs apply to all managed connectors in Lakeflow Connect. Keep reading for connector-specific FAQs.

What extraction method does the connector use?

The connector uses Oracle LogMiner in uncommitted, non-continuous mode and connects over JDBC. It does not use XStream, Oracle GoldenGate, or direct access to redo log files.

In uncommitted mode, LogMiner reads changes before they commit, which lets replication start immediately and keeps processing and memory (PGA) load on the source database low. The pipeline stages the raw changes and applies a transaction to the destination only after it reads the corresponding commit. In committed mode, which the connector does not use, LogMiner buffers whole transactions in the source database's PGA memory, which large transactions can exhaust.

Isn't LogMiner deprecated?

No. Oracle continues to support LogMiner as a built-in database feature for querying online and archived redo logs through SQL. Oracle deprecated the CONTINUOUS_MINE option of DBMS_LOGMNR.START_LOGMNR in Oracle Database 12c and removed it in 19c, but the connector uses LogMiner in uncommitted mode and does not rely on continuous mining, so this change does not affect ingestion.

If the pipeline fails, does ingestion resume without data loss?

Yes. Databricks keeps track of what the connector has extracted from the source and applied in the destination. If anything happens, Databricks can resume at that point as long as the archive logs remain on the source database. If the archive logs are purged before the pipeline runs, resumption is not possible, and the target tables require a full refresh.

Does the connector capture time zones for date and time columns?

The connector preserves time zone information for TIMESTAMP WITH TIME ZONE and TIMESTAMP WITH LOCAL TIME ZONE columns when the fractional-seconds precision is 6 or less. DATE columns map to TIMESTAMP, and plain TIMESTAMP columns are ingested as strings, both without time zone conversion. See Oracle integrated CDC connector reference.

Which supplemental logging method should I choose?

Enable supplemental logging at the table level, or at the container (PDB) or database level. When configured at schema or container level, the tables inherit the configured supplemental log level.

• Primary-key supplemental logging is the minimum, and is enough for most tables, including tables that have no primary key. When a table has no primary key, it records all columns except LOB and LONG types, and the connector treats them as a bundled key when writing to the destination Delta table.
• Full supplemental logging is required only on tables whose primary-key or unique-key columns are updated by UPDATE statements. Without it, the pipeline fails and requests a full refresh when it encounters such an update. Given this method records every column in redo logs, so it uses more disk space.

Databricks recommends replicating tables that have a primary key. See Enable supplemental logging.

How often can I schedule the ingestion pipeline to run?

The pipeline runs in triggered mode. To ingest data on a recurring schedule, run it from a Lakeflow Jobs task, scheduled frequently enough to process changes before archive logs are purged on the source. See Common pipeline maintenance tasks and Schedule recurring updates.

Does the connector support multi-tenant (CDB) databases?

Yes. For a multi-tenant database, create a common user in CDB$ROOT and use the CDB$ROOT service name in the Unity Catalog connection. Multi-tenant Amazon RDS for Oracle instances are not supported. See Multi-tenant (CDB) databases.

Can I ingest from a physical standby or read-only instance?

No. LogMiner-based change capture requires a primary Oracle instance.

Does the connector support Logical Standby using Active Data Guard?

No. The connector replicates only from the primary database. A Logical Standby database already uses LogMiner to apply changes from the primary and restricts LogMiner access for non-DBA users, which prevents additional LogMiner-based replication.

Does the connector support read replicas?

No, not at this time.

Does the connector support on-premises Oracle?

Yes, for source databases reachable over AWS Direct Connect, Azure ExpressRoute, or VPN with sufficient bandwidth.

Does the connector support Oracle E-Business Suite (EBS)?

Yes, if you have access to the underlying database. The connector reads from the database directly, not from the application layer.

Does the connector support Oracle applications such as CCS, WACS, and Fusion?

Not officially. The connector connects directly to the database and reads changes with LogMiner, so it is not application-aware. If an application exposes its database port, the connector might replicate the underlying tables, but it does not account for application-level relationships between them, so related tables can become inconsistent. For this reason, Databricks does not formally support Oracle Customer Cloud Service (CCS), Work and Asset Cloud Service (WACS), Oracle Fusion, or similar applications.

Can I ingest from Oracle views?

No, not at this time. The connector ingests from tables only.