Skip to content

Share your machine with Direct

When you need your machine back, pause the runner instead of stopping its workers. You can also set the hours it stays paused, and tell local cleanup scripts which containers belong to the runner.

You need an installed, connected runner and Docker running on the same machine.

Run this as the user who runs the runner, without sudo:

Terminal window
direct-runner pause

Pausing does not stop running workers. The runner refuses new work immediately, including during the delay before Direct receives its next report. Current Direct requests reports every ten seconds; this is not a notification deadline if the connection is interrupted. Other eligible runners or the cluster may take new work, but fallback is not guaranteed for work with machine-specific requirements.

On the local page, Pause does the same thing. The page shows why the runner is paused. You can also check from a terminal:

Terminal window
direct-runner status

To take new work again:

Terminal window
direct-runner resume

Or select Resume on the local page. Pausing twice keeps the original pause; resuming twice is harmless. The commands work even when the daemon is stopped. A manual pause survives a restart or reboot until you resume it.

Add PausedDuring under Runner in your existing runner.json. For example, pause on weekday working hours and all weekend:

{
"Runner": {
"PausedDuring": "Mon-Fri 08:00-18:00; Sat,Sun 00:00-24:00"
}
}

Restart the runner after changing the setting. It uses the machine’s local time, not UTC.

Each window is [days ]HH:mm-HH:mm. Separate windows with a semicolon. Day names are Mon through Sun, case-insensitive; use a range such as Mon-Fri or a comma-separated list such as Sat,Sun. Omit the days to apply a window every day. 24:00 is allowed only as an end time. The start is included and the end is not.

A window can span midnight. Fri 22:00-06:00 pauses from Friday at 22:00 until Saturday at 06:00; the day names are the days the window starts on. A blank setting means no scheduled pauses. An invalid setting stops the runner from starting and prints what is wrong.

A scheduled pause finishes automatically at the window’s end. Resume only clears a manual pause: it does not override the schedule. The page and direct-runner status show when the scheduled pause ends.

Pausing does not keep the machine awake. Sleep/wake recovery is not a tested guarantee: do not assume a running worker survives sleep or that work resumes immediately after waking. After wake, check direct-runner status, Docker and the work’s state in Direct before relying on the machine again.

Keep cleanup scripts away from runner workers

Section titled “Keep cleanup scripts away from runner workers”

Every worker container the runner starts carries the io.cratis.direct-runner Docker label. Its value is the runner’s id. To list those containers, including stopped ones:

Terminal window
docker ps -a --filter label=io.cratis.direct-runner

If you prune stopped containers, exclude that label:

Terminal window
docker container prune --filter 'label!=io.cratis.direct-runner'

This can remove other stopped containers, so review your cleanup policy before running it. The filter protects runner workers; it does not protect your other containers.

For scripts that remove containers by id, skip the ids returned by:

Terminal window
docker ps -aq --filter label=io.cratis.direct-runner

The runner makes the same promise in return: it only manages worker containers carrying its own label value. It leaves containers without the label, or labelled for another runner, alone even when their names look like worker containers.

Workers started by an earlier runner version are not labelled. They keep running, but the updated runner no longer manages them. Once they finish, remove those containers with docker rm -f, specifying only their ids.