Use nsd and/or unbound, they're more robust and faster than bind.

Good to know, hadn't really had a chance to explore those yet but will.

No software is entirely bug-free, so I do like having them relaunch on failure rather than stay dead.

Before systemd, Nagios was employed to do that and alert you when it happens. It's still a good idea to have proper monitoring set up, be it Nagios/Icinga, Zabbix or something you know will fit the bill. In good fashion of Unix philosophy, a combination of components works really well. This repository has most of them: https://github.com/monitoringsucks/tool-repos