After a fresh deployment of our application on a kubernetes cluster we always will end up with a hazelcast network partitioning problem (split brain).

All three replicas starting at the same time. Hazelcast is configured to use the jhipster-registry (i.e. Eureka) for discovery. At the time, the application tries to initialize hazelcast, no instance is registered in the registry which gives us three independet hazelcast nodes.

The documentation says that hazelcast will detect this and merge all my maps. But this does not happen.

BTW