Amazon RDS for Oracle

Preview

Amazon RDS for Oracle 19c passed the connector's qualification suites on 9 October 2026 in both architectures, on RDS SE2 License Included. As a non-CDB: the setup script below, the doctor's checks, capture in online and archive-only mode, and dictionary builds. With the CDB architecture, from release 0.1.3: the same setup script, checks and capture, mined from inside the tenant database in range mode. A real Kafka Connect worker against each, killed and restarted mid-stream, then delivered every committed row. A log purged by the retention setting, Multi-AZ failover and releases from 21c have not been run on RDS yet, so treat RDS support as a preview.

The connector mines an RDS for Oracle database with LogMiner exactly as it mines one on premises: LogMiner reads the online and archived redo logs on the database server, so nothing has to reach the files. What differs is how the database is prepared, because RDS gives the master user no SYSDBA, no ALTER SYSTEM and no ALTER DATABASE.

Topic	On premises	Amazon RDS for Oracle
Grants on SYS objects	GRANT SELECT ON V_$... TO user as SYSDBA	rdsadmin.rdsadmin_util.grant_sys_object as the master user
Minimal supplemental logging	ALTER DATABASE ADD SUPPLEMENTAL LOG DATA	rdsadmin.rdsadmin_util.alter_supplemental_logging(p_action => 'ADD')
ARCHIVELOG mode	ALTER DATABASE ARCHIVELOG	On while automated backups are on (backup retention of 1 day or more)
How long archived redo stays	Your deletion job (RMAN)	archivelog retention hours, 0 by default
Initialisation parameters	ALTER SYSTEM	The instance's DB parameter group
Online redo log size	ALTER DATABASE ADD LOGFILE	rdsadmin.rdsadmin_util.add_logfile and drop_logfile

The doctor detects RDS (the RDSADMIN schema exists) and gives every fix in the form above.

Moving from Confluent's Oracle CDC Source connector on an RDS non-CDB? The cutover in Migrating from Confluent Oracle CDC Source was run on RDS for Oracle 19c on 12 October 2026.

Architecture

The preview covers Oracle Database 19c in both architectures. With the CDB architecture, which every RDS release from 21c uses and 19c offers, the client always connects to the tenant database as a local user and cannot reach CDB$ROOT. From release 0.1.3 the connector mines that tenant database in range mode (cdc.mining.mode=auto chooses it): LogMiner chooses the logs, and the connector still lists them and checks them for gaps. Two things differ from a non-CDB:

• Rows written before a DDL that the connector has not mined yet stop the task with CDC-6001 instead of being replayed, because LogMiner cannot use a dictionary from the redo inside a PDB. Keep the connector's lag short where DDL is frequent.
• Dictionary builds are switched off.

Run the setup script below connected to the tenant database. Releases from 21c have not been through the qualification suites yet.

The setup script

java -jar oracle-cdc-doctor-cli.jar setup-sql --platform rds --user cdc > setup-rds.sql

The script carries a <change-me> placeholder password: put the real one in before you run it, and keep the edited file out of version control. Review it, then run it as the master user. It:

1. creates the capture user (cdc by default; a local user, no C## prefix);
2. grants CREATE SESSION, LOGMINING, EXECUTE_CATALOG_ROLE, SELECT ANY TABLE, FLASHBACK ANY TABLE, SELECT ANY DICTIONARY and SELECT ANY TRANSACTION;
3. grants EXECUTE on DBMS_LOGMNR and DBMS_LOGMNR_D, and SELECT on every fixed view the connector reads, through grant_sys_object (names as in DBA_OBJECTS, grantee in upper case);
4. turns on minimal supplemental logging;
5. sets archivelog retention hours and commits.

Supplemental logging for each captured table is still a statement per table, run by its owner or the master user:

ALTER TABLE <owner>.<table> ADD SUPPLEMENTAL LOG DATA (ALL) COLUMNS;

Archived redo retention

RDS deletes archived redo from the instance once it is older than archivelog retention hours. The default is 0, so logs disappear soon after they are archived and a connector that stops for longer than the online logs last finds its redo gone (CDC-2002). The script sets the retention to the journal threshold plus a day of downtime; raise it if the connector may be stopped for longer, and size the instance's storage for that much redo:

BEGIN
  rdsadmin.rdsadmin_util.set_configuration(
    name  => 'archivelog retention hours',
    value => '48');
END;
/
COMMIT;

The commit is required. To read the setting back, as the master user:

SET SERVEROUTPUT ON
EXEC rdsadmin.rdsadmin_util.show_configuration;

Doctor rule DOC-23 reads the same setting. It is blocking at 0 and a warning below cdc.txjournal.threshold.ms plus the planned maximum downtime (--max-downtime, default 24 hours). The capture user usually cannot read rdsadmin.rds_configuration; the finding is then information, and running oracle-cdc-doctor check with the master user's credentials gives the full answer.

Connecting

Point the connector at the instance endpoint, port 1521 by default, and the database name you gave RDS as the service:

{
  "cdc.database.host": "<instance>.<id>.<region>.rds.amazonaws.com",
  "cdc.database.port": "1521",
  "cdc.database.service": "<database name>",
  "cdc.database.user": "cdc"
}

Leave cdc.database.pdbs empty: a non-CDB has no pluggable databases.