Hand your own servers over to the node group

When you start Observe, every server behind your load balancer joins the node group as an adopted, protected member: the seed you named and your other servers. They count toward Min and Max, and Hetscale never removes them, so the node group never runs fewer servers than you had. This guide hands them over to the node group one at a time. You take each one out of service, Hetscale starts a server of its own from the current release wherever Min needs one, and in the end only the seed stays yours.

What do I set up first?

1. Go Live with Min equal to the number of servers behind the load balancer today, your own and the seed (Go Live). Then every server you hand over is replaced by one of the node group’s own.
2. Watch the timeline for a few days. The servers Hetscale adds when load rises should become ready and serve before you take any of yours away.

How do I hand over one server?

Never the seed. For one of your other servers:

1. Take it out of the load balancer in the Hetzner Console, at the load balancer’s targets. About two minutes later, once two reads a minute apart no longer find it among the targets, Hetscale stops counting it as capacity and starts a server of its own from the current release in its place. The timeline records “Not behind the load balancer”, and you get a mail at warning level: “‹name› is no longer a target of the load balancer: it gets no traffic and no longer counts as capacity.” Hetscale never adds your own servers back to the load balancer. Do this only when your other servers can carry the traffic until the new server serves.
2. Wait at least the node group’s drain window, 60 seconds unless you changed it, so that open requests finish. As of September 2026, according to Hetzner’s support, a load balancer keeps a removed server’s open connections alive for up to 5 minutes and sends it no new ones.
3. Turn its delete protection off in the Hetzner Console, if the Delete protection step of Go Live turned it on, then delete the server there.
4. Hetscale notices that the server left. The timeline records it, and you get a mail at warning level: “Your server ‹name› is no longer in the Hetzner project, so it no longer counts toward the group’s capacity.” The server Hetscale started in its place at step 1 stays; nothing else is started for it.
5. Wait until the server started in its place serves, as the node group’s Servers page shows, before you hand over the next one.

What do I do when only the seed is left?

Lower Min in the node group’s Settings. A minimum of 2 keeps one spare beside the seed, so the node group keeps serving while a server is replaced (What happens with a minimum of one server?). From then on, the servers Hetscale created are the ones that leave when load drops, and the seed stays.

Can Hetscale manage the seed too?

The seed stays yours and protected: scale-in never selects it, and healing starts a server beside it instead of replacing it. To let Hetscale remove it when load falls and replace it when it fails, promote it with Promote to managed on the Servers page. That is offered only for the seed of a Live node group whose Rehearsal passed, and only after you turn its delete protection off in the Hetzner Console. A removed seed is deleted, a new server costs today’s Hetzner price rather than the price the seed has now, and there is no way back (Quickstart, step 4). A seed Coolify, Forge or Ploi deploys is not offered for promotion.

What does this not cover?

• Leaving Hetscale and removing what it created: Leaving Hetscale.
• What your own servers cost compared with the node group’s: the day-0 report in Observe and the calculator give your numbers.