Do nothing, but allow Chef to continue. Hopefully the next round will pick up from there and find the bootstrap key. Without this, setting the run list to "role[ceph-mon]", "role[ceph-osd]" before the first run would never recover.
This depends on ceph.git commit e0a636f90701620670e747571cc7ad27efac958f.
The list of recipes a node has is only updated after a successful run, and in this case the first run using single_mon fails because ceph.conf gets no mon addresses. Use a role called "ceph-mon" instead. That should have, in its runlist, ceph::run_mon and ceph::single_mon.
MON_ID becomes id. OSD_ID becomes id.
…around. (It doesn't know that that instance used to exist. That's fine. Just ugly.)