Ceph » Messengers

A user came into irc this morning have trouble getting rebooted OSDs into the cluster. Sam tracked it down to the daemons all having a nonce of 0 at all times, which meant that the monitor thought the freshly-booted OSDs were in fact old long-running daemons.
This turned out to be because the messenger doesn't fill in the nonce if the port to bind to is specified. (See Accepter.cc:140.) I can't for the life of me trace how or why this is the case, and so I assume it's a historical accident due to our mess of messenger startup.

I pushed a branch wip-set-nonce-on-bind which does precisely that. However, I didn't want to merge it:
1) I can't figure out how this bug started happening so I don't want to violate any unknown requirements without looking into it more.
2) This means that the monitors will start getting their nonce set, which they haven't had happen for literally years. No idea if that will cause trouble.