When something goes down

The panel decides and the workers carry it out, so the two fail in different ways. Players notice little when the panel is gone. A node that goes down takes its servers with it.

The panel goes offline

Everything that already runs keeps running:

  • Servers and proxies keep their players. Each worker runs its servers without the panel.
  • Every gate keeps the routes it last got, so players still join through port 25565.
  • Proxies keep the list of servers they last got and keep sending players to them.

What waits for the panel:

  • Starting, stopping and restarting servers, from the panel or from a release.
  • Restarting a static server that crashed. Atlas leaves a crashed server down until the panel is back.
  • Consoles, files, SFTP and the activity feed.
  • New servers in a proxy's list. A server that stops in the meantime stays on the list.
  • /atlas link and /deploy.
  • Changes to other servers in the plugin API. /atlas servers and the metadata of other servers stay as they were.

Each worker tries to reach the panel again, waiting one second at first and at most thirty. Once it's back in, the worker reports every server it runs. Atlas then restarts the static servers that stopped while the panel was away and removes the dynamic ones, the same way it handles any server that stops by itself.

Updating the panel restarts it, and the same holds for those few seconds.

A node goes offline

The panel counts a node as offline after three missed heartbeats, 45 seconds. Needs attention on the dashboard shows it, with the time of its last heartbeat and how many servers were on it. Its servers show offline next to their node, and anything you try on them says the node is offline.

Players on that node's servers and proxies lose their connection. The other gates stop sending players to its proxies once their own node reports it can't reach the offline node any more, which takes up to three minutes.

Atlas doesn't move a node's servers somewhere else. A static server's files live on its node, and a dynamic server is new each time anyway, so start fresh ones in the group if you need them before the node is back.

The node comes back

The worker starts again with the machine and reconnects by itself. It picks up the servers that are still running. A server that stopped in the meantime, for example because the machine restarted, counts as stopped by itself: static servers start again and dynamic ones are removed. A server the panel has no record of is removed from the node.

If the machine is gone for good, so are the files of its static servers.

The worker restarts

Restarting or updating the worker leaves the servers and the gate running. The worker picks up the running servers when it starts, and players notice nothing.