Machines
What a ship is, the six classes, where your machines appear in the dashboard and what each line there means.
A ship is one whole machine
A ship is one physical machine that runs GitHub Actions jobs for your organisation and for nobody else. All of its cores, threads and memory are reserved for you, busy or idle.
Jobs do not run directly on the machine. Each job gets a new virtual machine on it, of the size its label asks for, and that virtual machine is deleted when the job ends. The ship is what you rent by the month. A runner size is how much of the ship one job books while it runs.
The six classes
A class is a size of machine. A bigger class has more cores and more memory.
| Class | Cores | Threads | Memory |
|---|---|---|---|
| Corvette | 6 | 12 | 64 GB |
| Frigate | 8 | 16 | 64 GB |
| Destroyer | 12 | 24 | 64 GB |
| Cruiser | 16 | 32 | 128 GB |
| Dreadnought | 48 | 96 | 256 GB |
| Carrier | 10 | 20 | 64 GB, and 24 GB of GPU memory |
- Cores are physical processor cores.
- Threads are what jobs book. Each core runs two threads, and one thread is what a metered runner sells as one vCPU. A Frigate, written 8 cores · 16 threads · 64 GB, compares with a 16 vCPU metered runner. It does not have 16 cores.
- Memory usually decides how many jobs run at once. See How many jobs run at once.
The pricing page gives each class its monthly price. The machines listed there are example stock, and the Carrier’s price is proposed. Runners are Linux x64 and do not use a GPU; see Platform support.
Place
You choose where a ship is when you choose the ship: France, Germany, Canada or US East. A place is where the machine sits. Your workflows, secrets and logs stay with GitHub wherever the ship is.
The dashboard does not show a machine’s place.
How a machine is added
Self-service sign-up and ordering are not open, and the pricing page takes no payment. Today we set a ship up with you: we order the machine, connect it to your organisation, and configure its GitHub App and runner pools. Contact us to start, or to add another ship.
Rental hardware is ordered once it is paid for. The example ships on the pricing page are marked as ready in hours or in 1 to 3 days; the time for your ship is agreed when you order it.
The part that happens in your GitHub organisation is in the quickstart.
Machines in the dashboard
Sign in and open Machines under Infrastructure. The page lists the machines of your organisation, online ones first.
Each machine has one row:
| On the row | Meaning |
|---|---|
| Name, then Registered, the platform and the runtime | Which machine this is. A ship reads Linux · Incus. |
| Online or Offline | Whether the machine is connected and reporting. |
| Readiness line | What the machine can do now. The states are in the next section. |
| CPU, memory and disk allowance | How much of the machine work may book. A reserve is kept for the machine itself, so the allowance can be lower than the hardware. |
| Active job | The work the machine is running now, when it reports one. |
| Can run | The kinds of work the machine takes. |
| Available for new work | The CPU, memory and disk still free for the next job. |
| Idle since | How long the machine has had no work, followed by the reason. |
| Refused here recently | Work that named this machine and could not be placed on it. |
| Advanced | The machine’s ID, when it was last verified, the hardware it reports and what is in use on it now. |
If your organisation has no machine yet, the page says No machines connected yet and links to the pricing page.
Ready, busy and the other states
| The row says | Meaning | What to do |
|---|---|---|
| Runtime ready | The machine is connected and can take a job. | Nothing. |
| Running a workload | The machine is busy with work and may still have room for more. | Nothing. Read Available for new work for what is left. |
| Capacity limited | The machine is connected and work on it continues. New work must fit the CPU, memory and disk that remain. | Jobs that do not fit wait in the queue. If this is the usual state, read Usage. |
| Templates unavailable | The image runners start from is not ready on this machine. | Contact us. |
| Needs attention | Something on the machine must be checked before it takes more work. | Contact us. |
| Runtime disconnected | The machine is not reporting. The row is tagged Offline. | Jobs wait in GitHub while it is away. Contact us if it stays offline. |
Idle and refused work
An idle machine is not a fault. The price is the same whether it works or waits, and its idle time is yours. The row says Idle since with the time, then the first reason it has: most often, nothing has asked for it.
A refusal is a piece of work that named this machine and could not be placed on it, for example because the machine had no room left at that moment. The row lists the three most recent under Refused here recently, newest first, each with what waited, why and when. Earlier ones are folded beneath. A refused job is not failed: it waits and is tried again. See Queueing and capacity.
Usage counts placed work and refusals over time.
How a job is placed, and what happens when the machine is full
A job reaches your machine by its runs-on label. Battleship starts a runner for it when the job’s size fits the CPU, memory and disk the machine has left, and the job books that much until it ends.
When the size does not fit, the job stays queued in GitHub and starts when a running job frees enough room. Nothing is charged for waiting, and the job is not sent to a runner outside your machines. The order in which waiting jobs start is in Queueing and capacity.
Stop using a machine
To take work off a machine, restore the previous runs-on in the jobs that name its labels. They run on GitHub-hosted runners again from the next run. To make sure nothing else reaches it, remove the repositories from the runner group or uninstall the GitHub App, as described in Remove access.
To end the rental, contact us. Move your jobs back first: a job that still names the machine’s labels waits in GitHub’s queue until GitHub cancels it.