10 posts tagged with "release"

Security fix: The Docker image now ships Jackson 2.21.6 in the Kafka libraries, the bundled Apicurio Registry AVRO converter, and the Strimzi Kubernetes Config Provider, resolving GHSA-r7wm-3cxj-wff9, a high-severity jackson-core issue where the async parser could bypass the maxNumberLength limit (#171).

Docker image enhancements: The bundled Apicurio Registry AVRO converter has been updated to 3.3.3.

Dependency updates: Bouncy Castle 1.85 (#170).

Bug fix: Fixed an ArrayIndexOutOfBoundsException: Index -1 out of bounds in the RedoMiner redo log parser that could crash and hang the connector. The cause was the evaluation order of the hasNext() loop condition, which read a record length before checking that the block offset was valid (#168). Thanks to @regme for the report and the fix.

Faster key detection: A single data dictionary query now detects a table's primary key or, when there is none, the unique index on NOT NULL columns with the fewest columns — replacing several round trips per table (#163). Partitioned unique indexes are accepted when all of their partitions are usable.

Maverick Synergy is no longer bundled: The Maverick Synergy SSH libraries (LGPL-3.0) are now a provided dependency and are not shipped in the standalone jar or the Docker image. If you use a2.ssh.provider=maverick, download the jars and add them to the JVM CLASSPATH — see Reading via SSH. The default sshj provider and mina are unaffected.

Grant scripts: redo_miner_fixed_tables.sql and redo_miner_fixed_tables_cdb.sql now grant SELECT on V$ARCHIVE_DEST, which is required by the a2.archive.log.dest.id parameter introduced in v2.15.5.

Internal improvements: Transaction bookkeeping for JMX metrics has been simplified, and LimitedSizeQueue was rewritten as a thread-safe bounded deque (#166). Exposed JMX attributes are unchanged.

Contributing: Pull requests now require accepting the Contributor License Agreement (individual or corporate), checked automatically by a CLA bot. CONTRIBUTING.md was reworked and a pull request template added (#164).

Java 25: oracdc now requires Java 25, completing the planned JVM transition (#161). The --enable-preview flag is no longer needed — only --enable-native-access=ALL-UNNAMED must be passed to the JVM.

New parameter a2.archive.log.dest.id: Selects which archived redo log destination (V$ARCHIVE_DEST.DEST_ID) to read logs from when multiple archive log destinations are configured (#160). Default — 1.

Bug fix: Fixed the binding of UPDATE operations materialized as a delete row piece (OP:11.3) plus overwrite row piece (OP:11.6) pair when parsing redo records (#157).

Dependency updates: HikariCP 7.1.0, PostgreSQL JDBC 42.7.13, Log4j 2.26.1, Agrona 2.5.0.

Security fix: The Docker image no longer bundles the vulnerable jline jars, resolving GHSA-47qp-hqvx-6r3f and GHSA-2r2c-cx56-8933.

Kubernetes config provider: The Docker image now bundles the Strimzi Kafka Kubernetes Config Provider, allowing connector configurations to reference values stored in Kubernetes Secrets and ConfigMaps.

Docker image enhancements: The bundled Apicurio Registry AVRO converter has been updated to 3.3.0.

Logging migrated to Log4j 2: SLF4J has been replaced with Apache Log4j 2 across all Java sources (#150).

Performance: Faster redo block checksum calculation — unnecessary memcpy calls and allocations have been removed from the hot path (#151).

Bug fix: Fixed a NullPointerException in the TaskBase class (#152).

Docker image enhancements: The Docker image now bundles an Apicurio Registry AVRO converter alongside the Confluent Schema Registry converters.

License change: oracdc is now licensed under the GNU Affero General Public License v3.0 (AGPL-3.0) (#148). License headers across the Java sources, the LICENSE file, and the Contributor License Agreement have been updated accordingly.

Bug fix: Resolved an ArrayIndexOutOfBoundsException triggered by incorrect (Conditional rather than Always) supplemental logging, where NULL columns were missed during redo parsing (#147).

Bug fix: Resolved an issue with handling of guard columns in LabWare LIMS schemas (#144).

Internal refactoring: Pseudo-columns processing has been reorganized into a dedicated Kafka-runtime module (KafkaPseudoColumnsProcessor), simplifying the core CDC engine and improving separation between the Oracle parsing layer and the Kafka Connect runtime.

Off-heap memory upgrade: Chronicle Queue has been replaced with Java's native Foreign Function and Memory API (MemorySegment) for off-heap transaction storage. This removes the Chronicle Queue dependency and leverages JDK-native memory management.

Apache Mina SSH support: New SSH provider option (a2.ssh.provider=mina) using Apache Mina SSHD, alongside existing SSHJ and Maverick Synergy providers. Includes a new a2.ssh.connect.timeout parameter (default 30,000 ms).

Performance improvements: Agrona-based optimizations for internal data structures.

Bug fixes:

• Fixed incorrect type cast in SinkTable for tables with LOB columns (#142)
• Fixed OraCdcRedoMinerConnector error with Oracle 11g large UPDATE statements (#139)
• Fixed wrong Date generation for test redo records
• Corrected "the message value must only be a record schema" error
• Dockerfile fix for CVE-2025-67030

Security fix: Dockerfile fix for GHSA-72hv-8253-57qq and CVE-2026-1605.

Flexible topic naming: New strategy for specifying a Kafka topic by table name - KafkaFlexibleTopicNameMapper. See the RedoMiner Connector documentation for details.

Advanced compression: Correct processing of compressed KDOIRP vector.

Bug fixes

Code cleanup: split code into core CDC code and code specific to the Kafka Connect ecosystem.

Bug fixes