When an index that is red/yellow - perhaps due to the node that holds the other shards having an outage - is deleted we only delete the currently available shards along with the index. However if the missing node comes back we then recover whatever shards existed on it via dangling indices.

It seems that if a user requests an index to be deleted, having those other shards come back is aberrant behaviour and instead we should make note that some shards are not available, and if/when they come back they are then also deleted (maybe add a timeout value to that as well?).

Sure, the user could turn dangling indices off, but that seems incomplete and still leaves the data on disk when it's obviously not wanted - presuming that node comes back of course.