Abstract: Implementations are provided herein for non-disruptive upgrade including rollback capabilities for a distributed file system within a cluster of nodes. To continue availability of the file system to external clients during the upgrade process, nodes can be upgraded piecemeal, for example, in one implementation, one node at a time. When a node is undergoing certain stages of the upgrade process, external clients can be directed toward the remaining nodes of the file system that are not currently being upgraded, including already upgraded nodes, to perform client activity. During the upgrade process, a first subset of nodes can be running in an upgraded state while a second subset of nodes can be in a non-upgraded state, both providing access to external clients in a seamless manner. During the upgrade process, an administrator can decide to rollback any upgrades and return the distributed file system to its previous version (e.g.