Nodes and more than one machine

The panel can drive servers on machines other than its own. The first one is the master — it holds the database, the accounts and the screens. The others are nodes: a machine with a container engine and an agent that listens to the master.

This is an advanced thing. One machine runs more servers than most people have memory for, and a second machine adds a second thing that can fall over. Add one when the first is running out of breath, not before.

Admin → Nodes. Name the node and issue a join token — it is one-time and expires after a period you set beside it. The screen prints the command to run on the new machine.

The short life is deliberate: a join token is the right to attach a machine to your cluster. One that has been sitting in a chat for a month is not a token, it is a back door.

Once joined, the node appears in the list and servers can be created on it.

The node owns the data. A server's data directory lives on the machine the server runs on — the master does not reach into it. Which is why files and backups for a server on a node are handled through that node.

The master owns the decisions. What should run, with which image, on which port — and the accounts and permissions as well. A node does not ask what is allowed; it is not the one deciding.

The link between them is mutually authenticated TLS. Without a valid certificate from a join token, a node is not a node.

• Back up off the machine. A node's backups, stored on that node, vanish with the node. An off-box target is more of an obligation here than a choice.
• Do not walk servers between machines. Moving a server between nodes is moving its data. It can be done, but it is a move, not a switch.
• A lost node is not a lost master. The panel carries on and says which machine has gone quiet.

Next: When something does not work.