The Debezium connector for Oracle uses an in-memory buffer to cache in-progress transaction changes until the transaction is either committed or rolled back. This cache can require a significant memory footprint depending on the number of events in the transaction, how many concurrent transactions are in-progress, as well as the data itself that represents each event such as large character or binary objects. This can be tough to manage for some environments that have ongoing long-running transactions.

This release introduces a significant refactoring to the connector processing code that allows for varied modes of processing LogMiner change events. Out-of-the-box the memory-based implementation remains the default, but we’ve included a new implementation based on the Infinispan distributed data store. This implementation makes it possible to track any number of in-progress transactions regardless of their size or duration.

To take advantage of the new Infinispan implementation, the following configuration options must be provided:

As this is early work, the new Infinispan buffer mode currently has a few limitations:

This should not stop you from giving it a test drive and send as much feedback as possible to us for further improvements. We expect to remove the above restrictions in the next release.

The major feature of the 1.6 release has got also new improvements. The Debezium connector for Oracle now supports incremental snapshots in the same way as the other connectors, so Oracle users now can easily add new tables to the capture list and get them snapshotted on the fly.

For the MySQL connector, there is a new incremental snapshot mode that could be used for databases that do not allow writing to the signal table. Kate Galieva invented a method based on GTID lists that could be used to provide watermarking without writing to the database. Thank you for the nice contribution!

Altogether, 49 issues were fixed for this release. A big thank you goes out to all the community members who contributed: Blake Peno, Alfusainey Jallow, Bingqin Zhou, Hossein Torabi, Katerina Galieva, Kyley Jex, Martín Pérez, Naveen Kumar KR, Patrick Chu, Pavel Strashkin, Raphael Auv, Sergei Morozov, Thiago Dantas, Tin Nguyen, Tommy Karlsson, WenChao Ke, and yangsanity.

For the upcoming 1.7 preview releases, we’re planning to focus on completing the follow-up task for the Oracle LogMiner processor and provide the support for JDBC based process too. We will explore incremental snapshotting support for MongoDB and do research on MariaDB support.

While the team’s focus is primarily on 1.7, we have recently fixed 9 issues to the 1.6 stream and released 1.6.1.Final. You can check out the release notes for details. We recommend if you’re using a release prior to 1.6, take this opportunity to upgrade and get access to the latest new features!