The versioning scheme used is x.y.z, where a change in y constitutes major release; if only z changes, this is minor release.

In general, it is recommended to run the latest releases. If extra care is needed (continuous release model cannot catch all problems), run x.y.z-2 (or x.y-1.latest-2 if z is 0).

Indexing across minor releases

Just install the new version (and deploy the web application), add/change indexer/configuration options (if any) and continue.

Indexing across major releases

This usually means that index version changed which means the data need to be reindexed from scratch. To do so, delete everything under data root and run the indexer. Usually, this is costly operation in terms of time so it is recommended to use file system snapshotting/cloning capabilities to run this process in the background to avoid service downtime.