Bare metal & on-prem
Install your product on any Linux machine with one command, whether it sits in a customer's data center or on your own bare metal. After that, every new version you ship updates the machine automatically, over outbound HTTPS, with no SSH, VPN, Ansible, or Kubernetes.
Architecture diagram. Alien manages 3 remote environments: machine-1 (customer-1 server room), https 443, outbound; machine-2 (customer-1 server room), https 443, outbound; machine-3 (your bare metal), https 443, outbound.
The join command installs the machine agent, a systemd service (alien-machine) that connects out to Alien, pulls each release, and runs your workloads. You need no SSH access, no VPN, and no inventory file.
To turn it on, add machines to the platforms your alien.toml allows:
platforms = ["machines"]What you get
- One command per machine. The join command carries a token you can revoke from the dashboard or the CLI.
- Outbound only. Machines reach Alien over HTTPS. No inbound port from the internet, no VPN.
- Every release, everywhere. A release reaches every machine in the deployment, on the deployment's release channel. You can pin one deployment to a release.
- Updates that wait for your work. Alien updates one machine at a time, gives your process up to the stop grace period you set to finish, and moves on only after the new version is healthy. If it never becomes healthy, the update fails after the startup timeout (10 minutes by default) and the machines after it keep the previous release.
- One place for status. Health, versions, and logs for every machine, for you and, if you choose, for the customer.
- Keeps running offline. A machine cut off from Alien keeps running its last release, and comes back up on it after a reboot.
- One definition for every environment. The same
alien.tscan also deploy your product into a customer's AWS, Google Cloud, or Azure account, if they ask for that later.
What your product can run on a machine
Your alien.ts defines what runs on each machine. A typical bare metal product is one long-running process that needs the whole machine:
const agent = new alien.Daemon("agent")
.code({ type: "image", image: "ghcr.io/acme/agent:2.1.0" })
.cluster("hosts")
.runtime({ privileged: true, networkMode: "host", pidNamespace: "host" })
.stopGracePeriod(3600)
.build()- Daemons run one copy on every machine. A daemon can get full access to the machine: root privileges, the machine's own network and processes, and its directories.
- Stateless containers run as many copies as you ask for, and Alien spreads them across the machines.
Not supported today: GPUs, Sandboxes, stateful containers, Workers, and Builds. Data resources such as Storage and Postgres point at services that already exist. See Workloads.
Requirements
- Linux booted with systemd, on
x86_64orarm64, and root orsudo. containerd,wireguard-tools, andnftables, and the WireGuard kernel module. On Debian and Ubuntu, the machine agent installs the packages.- Outbound HTTPS to Alien.
- UDP
51820open between the machines of one deployment, for their encrypted mesh.
Machine requirements lists the capacity checks the join command runs.
Limits
- No GPUs, Sandboxes, stateful containers, Workers, or Builds.
- No air-gapped mode. Air-gapped sites run on Kubernetes.
- A machine belongs to one deployment at a time.