In Zeebe cluster, leader node performs append only logs and has the responsibility for replicating the logs for follower nodes based Raft Consensus Algorithm. Raft uses a two-phase approach for altering cluster membership.
So there are situations where leader node can fail, in that moment one of the follower node will get elected as leader node and it will handle the inconsistency.
How newly elected leader node in Zeebe cluster handles the inconsistency interms of log replication?
Is Zeebe workflow is a Cloud based State machine like AWS step function?
Because all the business logic was executed/invoked by handlers/workers in Zeebe process. Also Raft Consensus Algorithms deals with about managing distributed state machines for replicating the shared state of the every activity or client requests(logs).
Raft Consensus Algorithm from wiki source: