Thank you, Francesco. If I understand your question, you are asking what happens when a replication event process fails on one of the database mirrors. In that case an error event (containing the complete replication event and the error detail) is published and would be processed by a purpose-built Error Handler actor.

https://medium.com/nerd-for-tech/building-cloud-native-apps-part-1-393d250cf107#1aa7

All individual updates in a Context Handler actor occur within a local DBMS transaction and would have been rolled back upon the failure.

The Error Handler actor will correct the error and republish the update event, or report the problem for manual intervention.

Here is a link to a more recent piece on the topic:

Designing Microservices