Due to the upgraded library, it's not possible to form a cluster of nodes running a previous version of Hazelcast, i.e. existing volatile data in the cluster gets lost during the update. This usually affects all user sessions that are stored in a distributed map, and will require the users to re-login after the update.