Skip to content

Caches, history and notifications

The rest of runnerd.toml‘s keys; Configuration has the runners’ and says how sizes and durations are written.

Key Default
mirror.enabled required pull Docker Hub images through a mirror on this host
mirror.memory, mirror.cpus required the mirror’s limits; memory counts in the budget
mirror.ttl required how long it keeps an image no job pulls again
mirror.max_size "10G" emptied above this when no runner runs
pnpm.max_size, maven.max_size required the shared pnpm store and Maven repository, emptied above this when no runner runs
reseed_at "2G" slot 1’s layers become the profile’s snapshot once they changed this much and it is idle (Warm caches)
snapshot_max "10G" a larger snapshot starts its tool cache over from the image, then its Android SDK, then all of it
build.enabled true the shared build cache
build.max_size "10G" above this its least recently used entries go when no runner runs; GitHub Actions’ cache counts too
build.branches [] branches besides the default branch whose pushes, schedules and manual runs write to it
build.directory $XDG_CACHE_HOME/runnerd/build-cache its entries
build.actions_cache true it also serves GitHub Actions’ cache in place of GitHub’s
actions.enabled true runners read the actions their jobs use from runnerd’s verified action archives
actions.keep "14d" how long an archive stays after no served repository pins it any more
actions.directory $XDG_CACHE_HOME/runnerd/actions the archives
Key Default
database $XDG_STATE_HOME/runnerd/runnerd.db the job history’s file
samples "7d" how long a job’s CPU and memory samples are kept; totals and peaks outlive them
logs "30d" how long a finished job’s log is kept
events "30d" how long events are kept for Last-Event-ID resumes
jobs "90d" how long a finished job is kept
log_max "50M" the most log kept per job
Key Default
max_total "20G" the disk all records may take; runnerd doctor warns at 80%
max_count 10 how many are kept
max_age "7d" how long one is kept
min_free "10G" the job gets no record if it would leave Docker less free disk than this

The oldest records go first.

Key Default
margin 15 percent added to a sampled peak in advice and predictions
min_jobs 10 the fewest runs before a peak becomes advice or a prediction
auto false runnerd sizes each profile’s memory, slots and cpus itself (runnerd init writes true)
interval "1h" the least time between a change to a profile and one that lowers its memory or cpus
raise_interval "1h" the same for a change that only raises; at most interval
max_step 50 the most percent one change lowers memory or moves cpus
Key Default
job_actions true clients may cancel a job’s run and re-run jobs on GitHub with runnerd’s token, each confirmed before it is sent; false turns both off
Key Default
origins [] proxy addresses whose dashboard pages may write (On your phone): https://host:port, or http for loopback only
open unset the one of origins that runnerctl web opens

Notifications go nowhere without targets.

Key Default
digest_at "09:00" host-local HH:MM when targets that list digest get it
targets [] each a [[notify.targets]] table, below
Key Default
name required what logs and runnerctl notify test call it
kind required "ntfy", "discord" or "webhook"
url required the topic or webhook URL; a redirect counts as a failed send
headers {} sent with each post, such as { Authorization = "Bearer tk_…" }
events ["failed", "lost", "runner", "insights"] what it is sent; add "digest"

url and header values are secrets: the dashboard, runnerctl config show and GET /v1/config show them as "<secret>", and saving that text keeps the values. Write a new value to change one, and all values again after renaming a target. The file is mode 0600; runnerd doctor warns otherwise.