Operations
Backups, metrics, upgrades and the failure people hit.
Day two
- Backups — The state is a SQLite file. Backing up the deployment means backing up that file.
- Prometheus metrics — Exported by the server.
- Admin API over the mesh — The admin API can be restricted to the mesh itself, so it is not reachable from the public internet at all.
- Configuration as a file — The deployment can be described in a file rather than clicked together.
- Several servers in one client — Work and home, or a customer's mesh and your own. One connected at a time.
Devices that stop showing up
The usual cause is the reflector: if UDP 8788 is not reachable, enrollment and TLS both look healthy while peers silently never connect. Check that before anything else, because every other symptom points the wrong way.
Disposable devices
Short-lived machines can be enrolled and reaped without leaving stale peers in everyone's list.