Usage

Documentation / Your machines

Usage

Where the dashboard shows how much of your machines your jobs use, and how to read it before you change size.

Where to look

Sign in and open Overview under Infrastructure. It is the history of your machines. Choose a range of 24 hours, 7 days, 30 days or all time, one machine or all of them, and the kind of work. For CI, choose Actions jobs.

The Machines page beside it shows the present moment: what each machine is running and what it has free. See Machines in the dashboard.

What the history shows

FigureMeaning
Your machines onlineThe share of the time your machines were connected and reporting.
Work placedHow many pieces of work were placed on your machines. With Actions jobs chosen, these are your jobs.
Refusal attemptsHow many times work that named a machine could not be placed on it. Retries are counted, so one waiting job can add several.
Memory booked by your workThe memory your running jobs held, in GiB. A job books the memory of its runner size for as long as it runs.
CPU booked by your workThe vCPU your running jobs held, in cores.

Each figure also has a chart over the range. Under the charts, Resources gives the mean of each booked figure with the machine-hours that reported it, and Refusals lists refusals by reason, with attempts and the number of distinct pieces of work.

Three things to keep in mind when reading it:

  • Booked is not used. Booked memory and CPU are what your jobs reserved through their runner sizes. They are not a measurement of what the jobs consumed.
  • Values are means while a machine reported. With all machines chosen they are per reporting machine, not totals.
  • A gap is unknown. Time without a report is shown as unknown, not as offline or idle.

Times per repository, workflow and job

Your organisation’s dashboard shows usage by machine. For one repository, workflow or job, use GitHub: every job’s page in the Actions tab shows when it was queued, when it started and how long it ran, and the workflow’s run list shows the same over time. The time between queued and started is the time the job waited for a runner.

How queueing shows up

  • In GitHub, a job stays in the queued state for longer before it starts.
  • In the history, Refusal attempts rises and Refusals names the reason.
  • On the Machines page, the row reads Capacity limited, Available for new work is small, and recent refusals are listed.

Some queueing at the busiest hour is the normal price of a flat bill. Queueing and capacity explains the order in which waiting jobs start.

A bigger ship, or another one?

Look at a range that covers your busy days, with Actions jobs chosen.

What you seeWhat it suggests
Booked memory and CPU well under the machine’s allowance, few refusalsThe machine has room. Nothing to change.
Booked memory close to the allowance at busy hours, CPU well under itMemory is the limit. First try a smaller runner size for jobs that pass on it. If the jobs need the memory, choose a class with more.
One job needs more vCPU or memory than the machine can give a single runnerA bigger class. A second machine of the same class would not help that job.
Many jobs of ordinary size waiting at once, for long stretchesMore room for jobs at the same time: a bigger class or a second ship.
Refusals and queueing in one short burst a dayProbably nothing. The jobs wait a little and the price does not move.

The machine’s allowance is on its row on the Machines page. The classes are in Machines. Contact us to change class or add a ship.

Cache and the bill

A ship has one flat monthly price. Jobs, minutes and waiting do not change it, so there is no usage to count for the ship itself.

The price rule for cache is:

  • What a ship keeps on its own disk is never counted and never charged.
  • The shared cache is the store two or more of your ships share with each other. With one ship there is no shared cache and nothing to pay.
  • 10 GB of the shared cache is always free. Above that the price is $0.03 per GB a month for what you store.

The dashboard has no bill, cache meter or cache setting today. The history described on this page is where usage can be seen. Licensing and pricing states the same rule with the price of a machine.

The source repository is private. Source access is by request.

Search concepts, guides and command examples.