Three brokers cluster on RHEL.
After a broker1 crash (June 6th), one of the SpringBoot applications (kafka-clients:2.7.0.redhat-00005) was stuck on a specific partition.
The LSO was unable to advance because of an open/hanging transaction which was not visible in the available segment logs (retention policy kicked in).

The workaround here is to delete all .snapshot files from the partition with stuck LSO before restarting each broker.
Further context and data will be provided for root cause analysis.