-
Type:
Task
-
Resolution: Fixed
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: Kafka
Description
There has been some questions around delivery guarantee of data to and from the topic with respect to exactly once, at least once, etc. We should add this delivery information explicitly to the documentation for both source and sink.
Answered by Robert in the forums here: https://developer.mongodb.com/community/forums/t/kafka-connector-delivery-guarantees/8757
For sink default is At Least Once. If there is an error when processing data from a topic the connector will retry the write. However, if the data on the topic contains a unique attribute, it is possible to achieve exactly once semantics by configuring the Sink connector to use upserts and the DocumentIdAdder strategy. The sink connector can not support at most once.
For source default is At least once. There is a risk of duplicate messages if you use copy.existing. Note that change stream events are idempotent so the need to support other delivery guarantees are not applicable.
Scope of changes
All versions