Skip to content

Watch it

Three clients read the same daemon: the dashboard in a browser, runnerctl tui in a terminal, and runnerctl for scripts. They show the same figures in the same words. Shots and output here come from a demo daemon running the example config with a third profile, small, and measured and CPU admission on.

runnerd serve serves it on 127.0.0.1:7477; --web <address> picks another loopback address, --web off serves none. A browser signs in once through a one-time link the daemon keeps in $XDG_RUNTIME_DIR/runnerd.web:

Terminal window
runnerctl web
# without runnerctl
xdg-open "$(head -n 1 "$XDG_RUNTIME_DIR/runnerd.web")"

The browser stays signed in for 30 days after its last visit. Memory and CPU are sampled every 2 s, so a shorter peak does not show. Security model says how sessions end; On another device puts it on a phone.

The overview: what needs attention as one line beside the title, four health figures, the memory budget as a lane per slot with the running jobs' step, memory and CPU, lanes for the mirror, what is free and what the host holds, then the waiting jobs with their reasons, the latest jobs, the repositories and the slice's charts.The overview: what needs attention as one line beside the title, four health figures, the memory budget as a lane per slot with the running jobs' step, memory and CPU, lanes for the mirror, what is free and what the host holds, then the waiting jobs with their reasons, the latest jobs, the repositories and the slice's charts.

The overview opens on four figures and the memory budget: a lane per slot, as wide as its share of the slice and filled to what its job uses now. What needs attention is one line beside the title that opens to the list.

The overview of an idle machine: 0 of 6 slots running, nothing waiting, 512M of 33G reserved; the lanes are the mirror's 512M, 29.5G free and 3G held by the host, and the six idle slots are chips under them.The overview of an idle machine: 0 of 6 slots running, nothing waiting, 512M of 33G reserved; the lanes are the mirror's 512M, 29.5G free and 3G held by the host, and the six idle slots are chips under them.

With no runner up, under measured admission, no slot holds any of the budget: the lanes are the mirror, what is free and what the host holds, and each slot is a chip under them, or a row on a phone.

The Jobs page: the last 30 days as five figures (861 jobs, 90% succeeded, a 9m 07s average run, an 18s average wait, a 7G sampled memory peak) and three small charts, one search field, outcome and repository chips, a Filters fold, then the rows.The Jobs page: the last 30 days as five figures (861 jobs, 90% succeeded, a 9m 07s average run, an 18s average wait, a 7G sampled memory peak) and three small charts, one search field, outcome and repository chips, a Filters fold, then the rows.

Jobs opens on its last 30 days, then one search field and the rows. Five views, 1 to 5 on the keyboard: Overview, Jobs, Insights, Capacity and Settings. Pages and keys lists what each shows.

runnerctl tui is for a host you reach over SSH. Its views and keys follow the dashboard’s.

When the daemon goes away the header says reconnecting and the event stream resumes where it stopped. cargo build -p runnerd-cli --no-default-features builds runnerctl without the TUI.

Every command prints text, or with --json the API’s response unchanged; --socket picks another socket. runnerctl lists them all.

Terminal window
$ runnerctl job 120055991526
Usage
Memory peak 6.61G
Average memory 5.13G
CPU peak 7.57 cores
Average CPU 3.33 cores
CPU time 37m 43s
Received 506.45M
Sent 56.27M
Written to disk 1.84G in 31207 writes; 64M more waited to be written back as its container ended
From the mirror 168.82M
CPU held back held at its CPU cap in 5% of its busy time, waited for a CPU 25% of its run
Build cache 30 of 36 lookups hit (83%), 101.29M served, 12.66M stored
Log 54 lines, 2.15K runnerctl logs 120055991526
memory peaks are sampled every 2 s; spikes shorter than that do not show

The job tables have the columns Outcome, Job, Repository, Branch, Queued, Waited, Ran, Memory peak, CPU and Written. Written is what the job’s runner container sent the disk under Docker’s root, its own containers’ writes included, from the job’s start to the container’s end; ≥ marks a floor, when part of the run was not read, such as an end while runnerd was stopped. CPU held back gives a share of the job’s busy time, which reads higher than the share of the whole run that auto sizing counts.

Frames set in JetBrains Mono (OFL).