OGGBD- Frequent re-sends of events from BDA to Kafka and Firerollback (Doc ID 2846267.1)

Last updated on JANUARY 14, 2025

Applies to:

Symptoms

Noticed some frequent re-sends of events from BDA to Kafka. These often
happen very quickly, within milliseconds. Our wait for ACKs should be much
longer than that so we don't know what is causing this.
Our Consumer is expecting oracle committed data operations to come in a
complete set to KAFKA.

Example: There are 5 tables in an Oracle commit for the Financial
Transaction stream.

SCN-#1 ... processes 3/5 of the tables (a partial send of the SCN)
SCN-#2 ... processes all 5 of the tables successfully
SCN-#1... re-starts and processes all 5 of the tables successfully -- This
happens with in the milli seconds after the first send.
SCN-#2 ... gets fully re-processed as well as the BDA is trying to preserve
order
Result: Our idempotency logic ignores the re-send of SCN#1 (thinks it is a
duplicate) and accepts the partial SCN#1 as the truth (incorrectly)

What is causing the frequent resends from BDA to KAFKA?
How can we avoid this as this is causing serious business issues due to the
data miss?
---
The firerollback have seen on trail files which do not have any restartabend

Using Coordinated replicat and confirmed that the related tables are handled
in the same thread.

Cause

My Oracle Support provides customers with access to over a million knowledge articles and a vibrant support community of peers and Oracle experts.