When a job is stuck

Adopt a job whose controller is gone, or abandon it on the box.

How you got here: Desktop app What a box needs 1. Create the controller key 2. Pair a box How a job behaves Who drives

Show previous steps 6

Install 1

Desktop app
Download the desktop app for your OS and open it.
The macOS zip holds Jumpgate.app and Jumpgate Terminal.app; Windows has jumpgate-tray.exe and jumpgate.exe; Linux has install.sh. Every download carries both Linux agents.
Show image
Illustration from the jumpgate.app demo, with made-up boxes; not a screenshot of the app.

Getting started 3

What a box needs
A Debian or Ubuntu box with systemd, reachable as root over SSH.
Setup needs root (or passwordless sudo with --sudo) and checks Linux, systemd and disk first. The node itself runs as an unprivileged account.
1. Create the controller key
Create the key that signs every command to your boxes.
jumpgate keys init stores it in the Keychain, Credential Manager, Secret Service, 1Password or an owner-only file. It never leaves your computer.
Show image
The TUI, from its test snapshot; empty rows removed.
2. Pair a box
Install the agent on a box and confirm its host key.
You confirmed the box's fingerprint, and jumpgate installed jumpgate-agent.service and a restricted tunnel user. The box now takes only signed commands.
Show image
The TUI, from its test snapshot; empty rows removed.

Durable jobs 2

How a job behaves
A job runs fixed steps on the box and asks when it must.
Each step runs as a transient systemd unit, so a running step outlives the TUI; the agent keeps a signed log, and the box refuses service actions until the job ends.
Show image
The TUI, from its test snapshot; empty rows removed.
Who drives
One controller holds a job's lease; the others watch.
A lease silent for 90 seconds is stale, and adopting it is your decision.

Its controller is gone: adopt it once its lease is stale.
No signed request can move it: stop the agent, then run sudo jumpgate agent job-abandon JOB --yes on the box. It refuses while a step of the job is still running. It ends the job as failed, with a signed log entry saying a local operator abandoned it. Start the agent again afterwards, and look at the job's log and output before you start another one.