The stale gns0 reference was actually not involved in this. When the Yahoo cluster site was decommissioned in July this year, that gns0 reference was overlooked. It's been like this for nearly 6 months. It's fixed now, thanks!
This particular problem was caused by a partial hardware failure. One drive in the volume had an imminent failure and was operating painfully slowly. The system livelocked while ZFS was trying to rebuild onto a spare. Among other things nginx was accepting connections and immediately dropping them, which caused the ssl handshake error. The geo-dns rotation system was looking for accepting connections only.