Transactions can fail after a Redpanda Broker restart in releases lower than 24.2.x

 

Problem Statement

When running client applications using transactions in Redpanda Releases lower than 24.2.x 
If a Redpanda coordinator node crashes for some reason due to the state is lost, meaning the new leader does not have the full context of the transaction and then throws errors such as
  unknown_server_error.

This is a terminal error for certain clients such as Apache Flink where the application needs to be reset.

 

Solution

This issue is fixed in all Redpanda Releases 24.2.x via #18251

Note

There has been significant changes in Transactions in 24.2.x and we recommended if using transactions to plan to upgrade to this release as soon as possible