Installation

The connector ships as a plugin ZIP in the Confluent Hub component layout: one directory, osodevops-kafka-connect-oracle-cdc-<version>/, holding manifest.json, lib/ (the connector, its core library and their dependencies), doc/ (licence and notice files) and etc/ (an example configuration). The Oracle JDBC driver (ojdbc11) is bundled under Oracle's Free Use Terms and Conditions, reproduced in doc/licenses/. Jars the Connect worker provides itself, such as the Connect API and the Kafka clients, are not bundled.

Releases are on the GitHub release page; the current one is 0.1.2. Each release attaches:

File	Contents
osodevops-kafka-connect-oracle-cdc-<version>.zip	The plugin ZIP
osodevops-kafka-connect-oracle-cdc-<version>.spdx.json	Software bill of materials for the ZIP (SPDX)
oracle-cdc-doctor-<version>-cli.jar	The preflight checker as a single jar
SHA256SUMS	SHA-256 checksums of the files above

Download the ZIP and check it against the release's checksums:

VERSION=0.1.2
BASE=https://github.com/osodevops/kafka-connect-oracle-cdc/releases/download/v$VERSION
curl -fsSLO "$BASE/osodevops-kafka-connect-oracle-cdc-$VERSION.zip"
curl -fsSLO "$BASE/SHA256SUMS"
sha256sum --check --ignore-missing SHA256SUMS

The jars are also on Maven Central under the group sh.oso (oracle-cdc-core, kafka-connect-oracle-cdc and oracle-cdc-doctor), signed with the OSO release key. A Connect worker needs the ZIP rather than the jars, because the ZIP carries the dependencies and the Oracle JDBC driver.

To build the ZIP from source instead, use JDK 17 or 21; the repository's Maven wrapper fetches Maven itself (see building from source):

git clone https://github.com/osodevops/kafka-connect-oracle-cdc.git
cd kafka-connect-oracle-cdc
./mvnw -q package -DskipTests -DskipE2E
ls kafka-connect-oracle-cdc/target/*-kafka-connect-plugin.zip

The build names the file kafka-connect-oracle-cdc-<version>-kafka-connect-plugin.zip; its contents are the same as the release ZIP.

The same build produces oracle-cdc-doctor/target/oracle-cdc-doctor-<version>-cli.jar, the doctor as one runnable jar.

Install the plugin

Platform	Steps
Apache Kafka Connect	Unzip into a directory listed in the worker's plugin.path and restart the worker
Confluent Platform	confluent-hub install with the path of the ZIP, or unzip it into plugin.path
Strimzi	Add the ZIP as a plugin artefact in the build section of the KafkaConnect resource, or build an image with the plugin under /opt/kafka/plugins, as the Strimzi lab does
Amazon MSK Connect	Upload the ZIP to S3 and create a custom plugin from it

After the restart, GET /connector-plugins on the worker lists sh.oso.connect.oracle.OracleCdcSourceConnector.

Worker settings

• Distributed mode is recommended; the connector stores its position in Kafka Connect's offsets.
• For exactly-once delivery, every worker needs exactly.once.source.support=enabled (distributed mode only), and the brokers' transaction.max.timeout.ms must be longer than the connector's Kafka transactions.
• Give the connector broker access with cdc.kafka.bootstrap.servers (and cdc.kafka.* for security settings). It needs it to create its internal topics, to journal long transactions, to keep schema versions and to read signals (see the ops topic reference).
• Size the worker heap for cdc.buffer.memory.max.bytes (256 MiB by default) on top of the worker's usual needs, and give cdc.buffer.spill.dir a volume with room for cdc.buffer.spill.max.bytes (10 GiB by default).
• The connector runs one task. A tasks.max above one is accepted with a warning and ignored.

Compatibility

Component	Supported	What the test suites run today
Oracle Database	19c and later (the doctor refuses older releases), single instance, CDB or non-CDB	Oracle Database Free 23.26.3, CDB with three PDBs. 19c and 21c are listed in the support policy for 1.0 and are not yet qualified
Kafka Connect	3.6 and later (the offsets REST API used in the runbooks needs 3.6)	Apache Kafka 3.9.1 in the integration tests and the Docker Compose lab; Strimzi with Kafka 4 in the Kubernetes lab
Java	17 and 21	Builds and unit tests on 17 and 21
Converters	JSON, and Avro with the name adjustment modes	The JSON converter with and without schemas, and Apicurio Registry 3.3.3's Avro converter. Protobuf and other converters are not tested
Platforms	Apache Kafka Connect, Strimzi, Confluent Platform 7.6 and later, Amazon MSK Connect	Apache Kafka Connect and Strimzi. Exactly-once delivery on MSK Connect has not been verified

Amazon RDS for Oracle 19c non-CDB is a preview: it passed the qualification suites, and its setup differs. Not supported yet: Oracle RAC (a database with more than one enabled redo thread is refused at start with CDC-5001), the RDS CDB architecture, Autonomous Database, and capture from a standby database (accepted in archive-only mode but not qualified).