Tristan, can you clarify what do we want to communicate to users for this bug?
The key here is the fact that the store is unshared and therefore, without implementing "tombstones" (i.e. markers for removed keys), it is impossible to have the expected behaviour described in the issue. So we should say that we do not support this configuration (clustered cache, unshared cachestore, preload=true).
@Tristan, this is not related to preload=true, in any case the stale entry is returned on this node as it is not marked as deleted!
@Divya this is a documentation bug. We should add a note to the loaders/stores documentation: If a clustered cache is configured with an unshared cachestore (shared = false), on node join, stale entries which might have been removed from the cluster might still be present in the stores and will reappear.
I think the text in docs looks good. VERIFIED.