A node running as a service no longer opens browser tabs to sign in
TL;DR: A node started by launchd, systemd or a Windows service could open a new sign-in tab each time it restarted, and finishing one enrolled the machine as a second node. From v0.0.134 it logs the sign-in link instead and waits.
If your node runs in the background and couldn’t open its saved enrollment, it could fill your browser with sign-in tabs, one every time it restarted. Signing in through any of them enrolled the same machine as a new, second node. From v0.0.134, a node running as a service never opens a browser, and a node that can’t open its enrollment waits for it instead of replacing it.
What went wrong
A node running in the background that couldn’t open its saved enrollment started a browser sign-in. It waited 15 minutes for the sign-in code to expire, exited, was restarted, and opened another tab. This repeated for as long as the problem lasted. Completing any of those sign-ins enrolled the machine as a second node, separate from the one you already had.
What is true now
- No browser without a terminal. Under launchd, systemd, a Windows service or a container, the node writes the sign-in link and code to its log instead of opening a browser. Running
install.sh,edge-node loginoredge-nodein a terminal still opens the browser, as before. - Retries wait instead of restarting. If a sign-in fails while the node runs in the background, it tries again after a wait that doubles from 1 minute up to 30 minutes. The retries stop when the service stops.
- An enrollment it can’t open is waited for, not replaced. The node doesn’t start and doesn’t sign in again. It logs why every 15 minutes and checks the file every minute. Once the file opens again, the node carries on as the same node. If you remove the file, the node starts without an enrollment.
edge-node loginrefuses in that state. It points you at--force, which enrolls the machine as a new node.
The most common reason a node couldn’t open its enrollment was a change of hostname. That is fixed in the same release, as described in the entry on keeping your enrollment when the hostname changes. The troubleshooting section of the operator guide covers both new log messages.
What you need to do
Update your node to v0.0.134 or later.