Hi everyone,
We are facing performance issues with the Debezium Oracle connector. We’ve noticed a growing "behind source" lag, which eventually leads to an exception stating that the requested SCN is no longer available because the redo logs have been deleted. Please note that our redo log retention is set to at least 24 hours.
It seems that Debezium registers an open transaction and holds it, but never receives the "end of transaction" signal. Consequently, Debezium reports a long-running transaction even though our DBA has confirmed there are no such transactions active in the database.
Our Environment:
Deployment: Debezium running within a Spring Boot application.
Redo Log Size: 2GB, multiple groups.
Traffic: Average of 300 log switches per day, with peaks of up to 35 switches per hour.
Observed behaviour across versions:
3.2.2.Final: No major issues recorded (only occasional warnings).
3.3.1.Final: Issues appeared after several months of stable run.
3.3.2.Final / 3.4.0.Final / 3.4.1.Final: Issues appear within a few days.
What are the recommended settings for this kind of load and to prevent these "ghost" long-running transactions from stalling the connector?
Thanks in advance for any advice!