This is caused by all of the tservers that host a replica of that tablet being leader blacklisted (so we can't move the leaders anywhere).

The cluster balancer currently handles leader blacklisting without considering data moves, since users normally expect leader blacklists to take effect quickly. In this case, we would have to move a tablet off of the leader blacklisted set to another node and then move the leader on to that, which we don't currently support.

We probably won't add support for this in the near future because the usual use case for leader blacklisting is temporarily taking down a node or set of nodes in the same region/zone, in which case the nodes in other regions are able to take the leaders. You could use a data blacklist to move the actual data off the nodes.