Comment 4 for bug 1470576

Just to add on to Pete Zaitcev's point, when node A is going to replicate to node B, it first issues a REPLICATE request to get a list of stuff to push, then invokes rsync to actually push the stuff. If B's disk gets unmounted between the REPLICATE request and the rsync push, then things will go on the root filesystem if allowed. Rsync can't be told to check mountedness.

Probably some more words in the administration guide about having root ownership on the mount points would be good. I think it's somewhere in the docs already, but repetition will make it more likely that people actually read it.