Description
In some Oracle RAC environments using the Oracle XStream CDC Source Connector for Confluent Platform, customers observe a severe and persistent throughput degradation after a RAC instance failover:
- Under normal conditions, the connector commits reasonable messages per commit.
- After failing over XStream components from one RAC node to the other, commit batches suddenly drop to very low messages per commit and remain low (e.g., 900–1,000 msgs/commit) for a long time.
- Throughput returns to normal only after a connector task restart
This article explains the confirmed root cause and the required Oracle-side fixes to prevent this behavior.
Applies To
Confluent Platform
Confluent Cloud Fully Managed Connector
Oracle Xstream Source Connector
Oracle Database Version 19.30 and 19.31
Cause
The persistent throughput drop after RAC failover is caused by a defect in Oracle Database 19c’s memory and flow‑control handling for Xstream components under specific workloads and RAC failover patterns. It is not a defect in the Confluent Oracle XStream CDC connector itself.
The connector’s commit behavior only exposes the underlying issue: when XStream capture/outbound and Streams Pool are stressed by large SCN gaps and failover transitions, Oracle XStream 19c (pre‑patch) can enter a degraded flow‑control state and never fully recover until processes are restarted or the database is patched.
Resolution
- Workaround : A possible immediate workaround for the connector which is experiencing low throughput rate is to recycle the connector including task. This will bring the connector process the data with normal throughput.
Permanent Fix : As this is a defect in Oracle Database 19c version below patches are available to be applied for specific database version which can be downloaded from Oracle support with the help of DBA and applied to permanently fix the issue.
Oracle Database Version Patch Details 19.30 Merge patch 39465348 19.31 Merge patch 39623089