Expand description
The daemon’s JSON API, version 1: the request and response types every client deserializes, and the
principal a transport authenticates. Plain data only: no I/O, no Docker or GitHub types, no secrets but a
Pairing’s one-time link, which its Debug leaves out.
Routes, all under the authenticated principal:
| Route | Body |
|---|---|
GET /v1/status | Status |
POST /v1/repositories/{owner}/{name}/pause?PauseQuery | Paused; 409 Busy while it runs a job |
POST /v1/repositories/{owner}/{name}/resume | none, 204 |
POST /v1/repositories/pause?PauseQuery | Paused; 409 Busy while any runs a job, pausing none |
POST /v1/repositories/resume | none, 204: every repository resumed |
POST /v1/profiles/{name}/reseed | none, 202; 409 while it reseeds; cache.reseeded when done |
POST /v1/caches/{name}/clean | none, 204; 409 Busy while a runner uses it |
GET /v1/config | ConfigFile |
POST /v1/config/check | TOML in, ConfigCheck |
PUT /v1/config | TOML in, ConfigCheck; 422 ConfigCheck when invalid |
GET /v1/doctor | Doctor; 503 ErrorResponse while no run has finished |
GET /v1/attention | Attention; 404 from a daemon before 0.2 |
POST /v1/shutdown | ShutdownRequest in, 202 ShutdownAnswer |
GET /v1/dashboard | Dashboard |
POST /v1/dashboard/sign-out | none, 204 |
POST /v1/dashboard/pairings?PairingQuery | Pairings; 409 when the dashboard is off or has no [web] origins |
GET /v1/jobs?JobQuery | JobPage |
GET /v1/jobs/{id} | Job |
GET /v1/jobs/{id}/samples | SampleSeries |
GET /v1/jobs/{id}/logs?LogQuery | text/plain, one line per log line |
GET /v1/host/samples?HostSampleQuery | HostSamples |
GET /v1/stats?StatsQuery | Stats |
GET /v1/activity?ActivityQuery | Activity |
GET /v1/sizing?SizingQuery | Sizing |
GET /v1/insights?InsightsQuery | Insights |
GET /v1/digest | Digest |
GET /v1/savings | Savings |
POST /v1/notify/test?NotifyTestQuery | NotifyTest; 404 without [notify] targets or with no such one |
GET /v1/caches | Caches |
GET /v1/caches/history?CacheHistoryQuery | CacheHistory |
GET /v1/events?EventQuery | text/event-stream of EventEnvelope |
GET /v1/records | Records |
DELETE /v1/records/{job} | RemovedRecords; 409 while a shell is open in it |
DELETE /v1/records?RemoveRecordsQuery | RemovedRecords; records with an open shell stay |
POST /v1/records/{job}/shell?ShellQuery | Upgrade: SHELL_PROTOCOL, 101, then the shell’s terminal; 409 when the budget is full |
Errors carry an ErrorResponse.
Structs§
- Action
Archives - The action archives runnerd keeps, as its store last published them; in
super::Statusandsuper::Caches. - Action
Preview - The 409 a request without
force, or with aconfirmfor something else, gets: what the action would do, said before anything is sent to GitHub. Has themessageof anErrorResponse, so a client that reads only that still gets it. - Action
Query POST /v1/jobs/{id}/cancelandPOST /v1/jobs/{id}/rerunquery parameters.- Action
Result POST /v1/jobs/{id}/cancelandPOST /v1/jobs/{id}/rerunonce GitHub was asked: how it took it. The job’s record is not changed: GitHub’s listings bring the run’s new state and a re-run’s new jobs.- Action
Target - What an action acts on: the job as the history records it, and the run’s attempt as GitHub said it.
- Activity
GET /v1/activity: every job runnerd’s runners started or finished in a window, by when that happened rather than when the job was queued, so a job that waited or ran long counts in the window it started or ended in.- Activity
Query GET /v1/activityquery parameters.- Admission
- How a job’s runner was admitted.
- Archive
Failure - Something the store could not do, which jobs make up for by downloading their actions themselves.
- Attention
GET /v1/attention: what needs attention now, most urgent first.- Attention
Item - One thing that needs attention.
- Auto
Plan - What auto sizing will do: its rules, the changes its next look would make as the history and config are now, and the profiles a recent change holds.
- Budget
- The memory budget runners and the mirror share, in bytes.
- Budget
Parts - The budget as a bar draws it, from
Status::budget_parts. - Build
Cache Activity - What the build cache did for one job, from
BuildCacheCounts::activity. - Build
Cache Counts - What the shared build cache served one runner and took from it (ADR 0052), counted by the cache itself.
- Build
Cache Stats - The shared build cache (ADR 0052): its size now, and what it served the jobs of the last 7 days.
- Busy
- The body of a 409 refused because runners use what the request would change. Has the
messageof anErrorResponse, so a client that reads only that still gets it. - Cache
History GET /v1/caches/history: every measurement, oldest first.- Cache
History Query GET /v1/caches/historyquery parameters.- Cache
Limit - A cache volume against its limit, from
CacheVolume::limit_status. - Cache
Size - One volume’s size at one measurement.
- Cache
Volume - One cache volume and its size.
- Caches
GET /v1/caches: every cache volume of the configured profiles, as last measured. Docker measures volumes slowly, so runnerd measures them at start and whenever no runner runs after jobs ended.- Cause
Wait - What one cause held a profile’s jobs for.
- Chart
Line - A line across a job’s memory or CPU chart.
- Chart
Lines - The lines across a job’s memory and CPU charts, from
Job::chart_lines. - Check
- One thing the host needs to run runners. Its
Displayis the linerunnerd doctorprints. - Check
Count - How many checks carry one
CheckStatus::label. - Closed
Part - Room past the bar’s blocks that no runner may start in.
- Commits
- btrfs’s transaction commits over the stretches the daemon read them.
- Compare
Query GET /v1/jobs/{id}/comparequery parameters.- Compared
Steps - A job’s steps as far as a comparison reads them: the first
COMPARE_STEPS_MAXin order, while their names, each cut to its first 800 bytes, add up to at mostCOMPARE_STEP_BYTES. Steps pair by these names. - Config
Check POST /v1/config/checkandPUT /v1/config: whatrunnerd check-configprints for a config.PUTrefuses an invalid one with 422 and this body, which has anErrorResponse’smessage.- Config
File GET /v1/config: the config the daemon runs with. Parsetextwithconfig::Config::parsefor its fields.- Counted
- The growth count (ADR 0130): what the runners use, and the room kept for what they may still grow by.
- CpuAdmission
- What a runner reserved of the CPU budget. CPU in millicores, times in seconds.
- CpuBudget
- The CPU budget runners share, in millicores.
- Dashboard
GET /v1/dashboard: the dashboard’s[web]settings, as the daemon read them when it started.- Delivery
- What one target did with a message.
- Digest
GET /v1/digest: the last 24 hours.- Disk
Writes - The disk-writes section of
Insights: what the jobs that ended in the window wrote, the workflows and jobs that wrote most, and what the drive wrote while the daemon read it. - Disk
Written - What a job’s runner container wrote to the disks under Docker’s root, from its cgroup’s
io.stat: what its processes, the Docker Engine inside it and that engine’s containers sent the disks from the job’s start to the container’s last read. Bytes still in the page cache then go to the disk after, on the runner slice’s account. - Doctor
GET /v1/doctor: whatrunnerd doctorchecks, for the config the daemon runs with.- Doctor
Query GET /v1/doctorquery parameters.- Doctor
Summary - The doctor’s checks laid out for a reader, from
Doctor::summary. - Drive
Share - One owner’s share of what the drive wrote.
- Drive
Shares - Whose the drive’s bytes were, over the stretches the daemon read the runner slice, or over all it read the drive when it never read the slice: a stretch that missed one figure leaves the others’ shares be.
- Drive
Writes - What the disks under Docker’s root wrote while the daemon read them in a window: from
/sys/class/block, the runner slice’sio.stat, which keeps what its containers wrote after they are gone, the daemon’s own/proc/self/io, and btrfs’s commit count. - Error
Response - The body of every error response.
- Event
Envelope - One event as
GET /v1/eventssends it: the SSEdatafield. The SSEeventfield isevent.kind()and the SSEidfield isid. - Event
Query GET /v1/eventsquery parameters. Resume with theLast-Event-IDheader.- Failure
Summary - What a failed job’s log says of its failure. Each part is absent where the evidence is.
- Figure
- One of the overview’s four health figures, worded once for every client: the web dashboard lays them out as tiles, the TUI on one line.
- Figure
Delta - One figure of this job beside the baseline’s.
- Free
Part - What the bar leaves free past every block, less what the host holds back.
- Gap
- A stretch of a lane without a job.
- Health
- Whether CI keeps up now, from
Status::health. - Held
Part - Memory the budget holds for something other than a slot.
- Holder
- A program outside the slice and the memory its processes hold.
- Host
Budget - What the host’s memory leaves the runners (ADR 0055). New runners fit in the lower of
MemoryMaxandlimit; running ones are never stopped for it. - Host
Credit - What the host’s limit takes off the plain count (ADR 0108):
Budget::memory_in_uselessbytesis what it counts,usedof that is what the runners use, and the rest is room kept for what runs to grow into. - Host
Figure - One figure of the host’s reading, as
Budget::host_figuresgives it:Host available23G of 64G. - Host
Hazard - What the host’s memory came to while CI ran over a window, from the readings the budget’s events stored (ADR 0108):
how long
MemAvailablestood under each line, the least it andSwapFreefell to, and how much of that came while the host’s limit took a credit. Times are seconds, sizes bytes. - Host
Memory - The host’s memory at one moment, beside the slice’s: what admission fits new runners in besides
MemoryMax(ADR 0055). Read every few seconds; not kept in the history. - Host
Sample - One sample of the whole slice runners run in.
- Host
Sample Query GET /v1/host/samplesquery parameters.- Host
Samples GET /v1/host/samples: the slice’s use over time, oldest first.- Host
Span - How long the host stood under a line while CI ran, in seconds.
- Insight
- One thing the history says, about one job or one repository.
- Insight
Figure - The number an insight is about, short, for a figure beside its sentence:
6m 10soverfrom 4m 40s. - Insights
GET /v1/insights: what the history says about the window.- Insights
Heading - What the insights page says over its list, as
Insights::headingwords it:5 insights, 4 warnings, from the last 14 days,1 insight from the last 14 days,no insights from the last 14 days. A client may set the figure apart; together, as itsDisplaywrites them, they read as one line. - Insights
Query GET /v1/insightsquery parameters.- Job
GET /v1/jobs/{id}: a job with its steps.- JobComparison
GET /v1/jobs/{id}/compare: a job beside its baseline.- JobList
GET /v1/jobs: one page of jobs (ADR 0135). AJobPagewith fields more, so a client that reads aJobPagereads it too, and one that reads aJobListfrom an older daemon finds them absent.- JobMetrics
- What a job used, over the samples taken while it ran (every few seconds).
Noneuntil a sample exists. - JobOrder
- An order a list of jobs can be read in: what it is sorted by, and which way.
- JobPage
- One page of jobs, as
GET /v1/jobsanswered before it searched and sorted: whatJobListholds. - JobQuery
GET /v1/jobsquery parameters. Jobs come newest first (by ID, which grows with time), or bysort(ADR 0135).JobQuery::checksays which queries the daemon refuses.- JobRunner
- The runner a job ran on.
- JobSizing
- The jobs of one name in one workflow of one repository.
- JobSummary
- A job without its steps: what lists and events carry.
- JobTitle
- A job’s name as every client writes it, in parts for a page that sets the workflow apart.
- JobWrites
- What the jobs that ended in a window wrote, each its runner container’s
DiskWritten. - Labelled
Figure - One figure of a group:
Median a job120M, orLookups hit, 7 days75%of 400. - Lane
- One slot as it was: the profile and slot its runners had when they ran, whatever the config holds now.
- Limit
- One resource’s limit for a job’s container.
- Limits
- What a job’s container may use, from
Job::limits. - Line
Range - Lines from
fromup to, not including,to. - Line
Read - Stored lines to read: up to
limitfrom linefrom(past a gap pruning left, too), each cut to its firstprefixbytes, stopping after the line that brings the text read tobytes. - Listing
Failure - Why runnerd cannot list a repository’s queued jobs (ADR 0132): what class of failure GitHub’s answer was, never its text. The engine keeps it until a listing answers or a reload removes the repository; a restart forgets it.
- LogGroup
- A foldable group of lines, from its
##[group]to its##[endgroup]line. - LogLines
Query GET /v1/jobs/{id}/log/linesquery parameters. The response isapplication/x-ndjson: oneStyledLineper line, in order.- LogMark
- An error or warning line.
- LogNeedle
- A query, ready to find.
- LogOutline
GET /v1/jobs/{id}/log/outline: a job’s log as structure.- LogQuery
GET /v1/jobs/{id}/logsquery parameters. The response istext/plain, one log line per line.- LogSearch
GET /v1/jobs/{id}/log/search: one page of the lines whose message holds a query.- LogSearch
Query GET /v1/jobs/{id}/log/searchquery parameters.- LogSummary
- A job’s stored log.
- Long
Wait - A job that waited.
- Memory
Hold - A bound of the config holding a profile’s
memoryaway from what its jobs need: the config check refuses a profile past it, so the advice goes no further thanbound, and auto sizing writes it as is (ADR 0086, 0088). Under a raise it says only that no raise pastboundis possible, not that the jobs fit. - Milestone
- A moment of a job’s life.
- Need
- What a set of jobs needs, as the advice sizes for it, from
Sizing::need. - Notification
- What a webhook target is posted, as JSON; an ntfy target gets its title, message, priority and link.
- Notify
Test POST /v1/notify/test: what each target did with a test message.- Notify
Test Query POST /v1/notify/testquery parameters.- Outcome
- A job’s or a step’s outcome: its word and what it means.
- Outline
Parts - What the store keeps of a job’s outline: its saved state,
Nonefor a log no outline was kept of, and its groups, marks and section lines, each by line. - Outliner
- Builds a job’s outline a line at a time. The store resumes one from its saved state for each batch of lines and saves it after, in the transaction that stores the lines.
- Pairing
- A link that signs one device in to the dashboard on
origin, and nowhere else, withinlifetime. - Pairing
Query POST /v1/dashboard/pairings: which of[web] originsto make a link for.- Pairings
POST /v1/dashboard/pairings: fresh one-time links that sign a device in to the dashboard.- Pause
Query POST /v1/repositories/{owner}/{name}/pauseandPOST /v1/repositories/pausequery parameters.- Paused
POST /v1/repositories/{owner}/{name}/pauseandPOST /v1/repositories/pause: the repository, or every one, is paused, and stays paused when the daemon restarts.- Peaks
- The peaks of a set of jobs, each the highest of its samples.
- Phase
- A stretch of a job’s life.
- Pool
Status - One of a repository’s pools: its runners in one profile.
- Pooled
- The count the budgets go by while pooling takes something off
Budget::memory_in_use(ADR 0099). - Pressure
- One line of a pressure stall file, in hundredths of a percent: 250 is 2.5% of the time.
- Preview
Facts - The facts of an
ActionPreview: what the route found for the job. - Priced
Run - A job of the history as pricing reads it: one a runner of this machine took, completed or lost. Times are milliseconds.
- Priced
Runner - A window’s jobs priced at one runner.
- Pricing
- The runner each profile’s jobs are priced at, from the config running now.
- Pricing
History - What pricing reads of the history.
- Profile
Queue - Why one profile’s jobs queued over a window, from what each plan that held them said (ADR 0154). Unlike the rest
of insights, its times are milliseconds, so shares of short waits come out right. A wait runs from GitHub queueing
a job to its start, less what passed while no runnerd ran, summed over the jobs counted; the causes and the two
parts no cause covers add up to
waited_ms. - Profile
Sizing - One profile: its limits now, its jobs’ peaks and waits, and the advice.
- Profile
Status - A runner profile and what runs in its slots.
- Queued
Job - A queued job runners of this daemon can take, as GitHub listed it at the last poll.
- Reading
- A figure and the words that follow it, such as
8Gandof 24G reserved in runnerd.slice, for a client that sets the figure apart. - Record
- One record: the runner of a failed job as the job left it.
- Record
Limits - The limits on records, from
[records]in the config. - Record
Note - What the job history keeps about a failed job’s record, after the record is gone too.
- Records
GET /v1/records: every record, oldest first, and the limits they are kept within.- Remove
Records Query DELETE /v1/recordsquery parameters: exactly one of them.- Removed
Records DELETE /v1/records…: what was removed.- Replace
Query PUT /v1/configquery parameters.- Repository
Status - A configured repository and its queue.
- Repository
Waits - A repository’s waits this week against the week before.
- Rerun
Refused - Why runnerd holds a repository’s automatic re-runs of its runner faults (ADR 0153): GitHub refused one in a way
that asking again soon cannot change, such as a token without Actions: write. Held for an hour, or until the config
changes, and forgotten at a restart;
RepositoryStatus::rerun_refusedcarries it. - Resize
- A profile’s new
memory,slotsand, when auto sizing changed them,cpus. - Run
- A run of a line’s text in one style.
- RunComparison
- A job’s latest runs and what they say together, from
Job::compare. - RunLink
- One run of a job, to open in the history or on GitHub.
- Runner
Failure - A runner that could not be registered or started (
runner.failed). - Runner
Span - A runner container in a slot: up from its start to its removal, with a job or without, as a warm runner waits. Its life is not its job’s run.
- Runner
Status - A runner registered for one job, in its container.
- Running
Job - A runner of this daemon that holds a slot, and the job it runs, if it took one.
- Sample
- One sample of a container’s use.
- Sample
Series GET /v1/jobs/{id}/samples: what the job’s container used over time, from its start to its removal. Samples older thanhistory.samplesare deleted; the job’s metrics stay.- Sample
Usage - What a container used over a stretch of its run, from
SampleSeries::usage_between. - Saved
Outline - What the store keeps of an outline between batches of lines: one row per job.
- Saved
Window - The jobs of one window and what they would have cost.
- Savings
GET /v1/savings: what this machine’s job minutes would have cost on GitHub-hosted runners.- Search
- One page of a search, fed the log in the batches it asks for:
while let Some(read) = search.wants() { search.read(lines(read)) }, thenSearch::finish. - Seed
Memory - A seed container’s memory cap: its profile’s limit when the reseed started, or, for a requested reseed whose profile drained, the room it holds until it starts.
- Shell
Memory - A shell container’s memory cap: its record’s profile’s limit when the shell opened.
- Shell
Query POST /v1/records/{job}/shellquery parameters.- Shutdown
Answer - What
POST /v1/shutdownanswers: how the daemon stops now, which a stop asked for earlier may have made more urgent than the request. - Shutdown
Request POST /v1/shutdown.- Sizing
GET /v1/sizing: what the job history says each configured profile needs.- Sizing
Change - A profile’s
memoryandslotsas auto sizing changed them in the config file. - Sizing
Hold - A profile auto sizing changes nothing in before
until, and lowers nothing in beforelower_until. - Sizing
Query GET /v1/sizingquery parameters.- Sizing
Refused - A raise of a profile’s
cpusorslotsits jobs ask for that auto sizing cannot make, as when it does not fit the CPU budget or Docker’s disk. - Sizing
Revert - A lowering of a profile’s
memoryby auto sizing that went too far, which the advice takes back (ADR 0070). - Slot
Form - One way to read a slot’s block, from
SlotReading::forms. - SlotKey
- A profile’s slot, as a runner had it.
- Slot
Part - A slot’s block of the budget.
- Slot
Reading - What a slot’s block reads under its name, from
Status::slot_reading: what its job uses of what the block is, or without a sample of its job, what the slot is doing and the block’s size. - Slot
Status - One slot of a profile: room for one job.
- Slot
Timeline GET /v1/timeline: what each slot ran over a window.- Slotted
Bar - A job’s bar and the slot its runner had.
- Slotted
Runner - A runner container’s span and its slot.
- Span
- A stretch of time, in seconds since the epoch: from
fromup toto. - Spread
- Percentiles of a set of values, by nearest rank: each is one of the values, and with fewer than 100 values
p99is the largest. - Started
- A job that started: when, and how long it had waited since it was queued.
- Starts
- A profile’s jobs by whether their runner was already up when GitHub queued them (ADR 0050).
- Stats
GET /v1/stats: finished jobs per repository and time bucket, for trends. Sums, so clients can average.- Stats
Query GET /v1/statsquery parameters.- Stats
Row - The finished jobs of one repository in one bucket.
- Stats
Summary - Finished jobs summed over a time, every repository of the query together, from
Stats::daily. - Status
GET /v1/status: what the daemon serves and runs right now.- Step
- One step of a job, as GitHub reports it.
- Step
Delta - One step beside the baseline’s step of the same name.
- Step
Lines - A stored step and its lines.
- Step
Usage - What a step’s samples say, as the store reads them over its window: the samples after its start up to its end.
- Stored
Line - A stored line as a
LineReadgives it: its number, and its text, cut to the bytes the read asked for. - Stored
Log - What the store has of a job’s log besides its outline.
- Styled
Line - One stored line of a job’s log, styled.
- Swap
- A host’s swap at one reading: zram and zswap’s device included, as
/proc/meminfocounts it. - Timeline
- A job’s phases, in the order they happen. A runner’s container starts before GitHub gives it a job.
- Timeline
Bar - One job on a timeline.
- Timeline
Query GET /v1/timelinequery parameters.- Timeline
Rows - What the store reads for a timeline, before
SlotTimeline::assemblelays it out. - Toned
- Text and what it means, for a client to colour.
- Total
- One figure of what a job used, worded for every client.
- Usd
- An amount in US dollars, as whole micro-dollars, so no float carries money.
- Wait
Figure - One of a repository’s wait figures: its label, such as
Warm median, and its value,–without jobs to take it over. - Wait
Spread - How long some jobs waited.
- Week
Waits - The waits of the jobs that started in a week, from queued to started, leaving out the time before the history’s start and the time no runnerd ran (ADR 0081).
- Write
Section - A part of the disk-writes section, under its title.
- Write
Top - A workflow, or a job, and what its jobs in the window wrote.
Enums§
- Action
Answer - What GitHub answered an action, as runnerd classifies it.
- Action
Outcome - How GitHub took an action.
- Action
Refusal - What class of refusal an
ActionOutcome::Refusedwas. Never GitHub’s own words. - Admission
Policy [host] admission, as a job’s record and the budget name it.- Advice
- What to set a profile’s
memoryandslotsto.memoryis the largest sampled peak plus the margin, or plusPAGE_CACHE_ROOMwhen that is more, rounded up to 256M: the cap kills a job that goes over it, so it is sized for the heaviest run, not a percentile. - Archive
Failure Kind - What went wrong with an action archive.
- Attention
Kind - Where an attention item comes from.
- Auto
Note - What auto sizing does about a job near its cap.
- BarEnd
- How a bar or a runner’s span ends.
- Baseline
- What a job is compared with.
- Baseline
Read - The baseline as the store found it.
- Basis
- What a runner reserved goes by, in the order measured admission falls back.
- Bucket
- How
GET /v1/statsgroups jobs in time. - Cache
Kind - What a cache volume holds.
- Change
Kind - How a figure changed from the baseline.
- Chart
Line Kind - What a line across a job’s chart stands for.
- Check
Status - How a
Checkcame out. - Closed
Kind - What closes room past the bar’s blocks.
- Color
- A colour a run of text was printed in, from the closed set a log is drawn with. Bright and dark shades are one colour; white, default and 256-colour or RGB colours beyond the first 16 have none.
- Column
- A column of a table of jobs or of records, as the history,
runnerctl jobsand the records name it. - Count
Brake - Why the growth count rests (ADR 0130).
- Doctor
State - How current the daemon’s doctor checks are (ADR 0132). Times are seconds since the Unix epoch.
- Event
- Everything that happens, tagged by
type. - Failure
Cause - What a failed job’s log says failed it (
record::failure_cause), apart from its own code. - Feature
- What a client reads from the daemon that a runnerd older than the client may not serve: a route it lacks, which
the client says is not supported rather than empty (
ClientError::is_unsupported,JobList::understood). - Figure
Kind - What a figure measures.
- Fix
- What clears an attention item. A client reads a fix of a kind or view it does not know as none
(
AttentionItem::fix). - GapKind
- What a lane’s stretch without a job means.
- Health
Kind - Which health figure a
Figureis. - Held
Kind - What holds a part of the budget beside the slots.
- Hold
Cause - What failed, that holds a repository’s runners back for a while.
- Host
Wait - Whether queued jobs wait for the host’s memory, as
Status::host_waitanswers it. - Insight
Kind - What an insight is, tagged by
kind. - JobAction
- What an owner asks GitHub to do with a job.
- JobActions
- Whether a daemon cancels and re-runs jobs, as
GET /v1/statussays. - JobPeak
- Which of a job’s own peaks
Basis::Jobgoes by. - JobRange
- How far back a list of jobs reaches, by when GitHub queued them, as every client offers it.
- JobSort
- What a sorted list of jobs is sorted by.
- JobState
- Where a job is in its life, as this daemon saw it.
- Limit
Source - Where a job’s limit comes from.
- Line
Kind - What a line is, from the command the runner prefixed it with.
- Listing
Problem - What a failed listing ran into.
- LogEvidence
- How much of a job’s log there is to read, in order of precedence.
- Mark
Kind - What a marked line says.
- Measure
- Whether a job’s
DiskWrittenis all it wrote. A job no reading measured has no figure at all. - Memory
Bound - Which bound of the config holds a profile’s
memory. - Notify
Event - What a notification target is sent, and the events
[notify]targets choose from. - Outcome
Kind - What kind of outcome a job’s or a step’s is, for each client to colour as it colours that meaning.
- Outline
Change - What a line changes in the outline, for the store to keep.
- Period
- Which window a
SavedWindowcovers. - Phase
Kind - The phases of a job, in the order they happen.
- Principal
- Who sent a request, as the transport authenticated it. Handlers get it from one place, so a new transport only has to produce one.
- Priority
- How urgent a notification is, as ntfy ranks it.
- Profile
Plan - What auto sizing does next with one profile at a given time, from
AutoPlan::profile_plan. - Queue
Shown - Whether a status shows a repository’s queued jobs, or none because they could not be listed (ADR 0150).
- Record
Filter JobQuery::record: jobs by what became of their record.- Record
Mark - What became of a failed job’s record, as a list marks it (
JobSummary::record_mark): what the history notes, whichRecordNote::statereads in full on the job’s page. Docker’s images are the inventory (GET /v1/records); a record removed outside runnerd reads removed once runnerd lists the images and finds it gone. - Record
Removal - Why a record was removed.
- Record
State - What became of a failed job’s record, from
RecordNote::state. - Revert
Cause - What showed a lowering went too far.
- RunSeen
- The run as GitHub said it when the preview was made.
- Runner
Probe - What the probe a runner’s entrypoint runs as root after the runner exited found (ADR 0155): it mounts a tmpfs,
a proc and a sysfs and binds one in a mount namespace of its own, then opens a file under
/proc/syswithopenat2, the calls Sysbox serves and a runner fault breaks. - Section
- A page inside a
Viewof several, as every client names it. - Severity
- How much an insight matters.
- Share
Owner - Whose a share of what the drive wrote was.
- Sku
- A GitHub-hosted Linux runner, by the billing SKU GitHub’s pricing page and usage report name it with, as a
profile’s
priced_astakes it. Windows, macOS and GPU runners are not here: runnerd runs Linux containers. - Slot
State - What a slot is doing, from
Status::slot_state. - Sort
Order - Which way a sorted list goes.
- StepLog
- Which lines a step printed, as far as the log tells.
- Step
Match - How a step of one job was paired with the other’s.
- Stopping
- How a stopping daemon treats running jobs (ADR 0091).
- Subject
- What an attention item is about, for a client to link.
- Superseded
- Why a request that came with a confirmation got a new preview instead, and nothing was sent.
- Tone
- What a shown fact means for the reader. On the wire in kebab-case, as an attention item carries it.
- Unavailable
- Why a job has no baseline.
- Unknown
Cause - Why runnerd cannot tell what GitHub did. Not on the wire: the result’s
messagewords it. - View
- A group of pages of the web dashboard, a group of views of
runnerctl tui, in the order every client lists them: what runs now, the jobs, what the history says, the machine’s room and the daemon’s settings (ADR 0132). - Visibility
- A repository’s visibility as GitHub gives it.
- Wait
Cause - What a queued job’s wait is counted under in the history, one per
Waitingreason or a few of them together (ADR 0154): each plan that holds a job adds the time to the next plan to its cause. Stored byWaitCause::word, so a cause a later runnerd adds reads asWaitCause::Other. - Wait
Kind - What a
Waitingreason waits on, which clients draw it by. - Waiting
- Why a queued job has no runner of its own yet. Jobs are served oldest first; each takes the first of these that holds for it.
Constants§
- ACTIONS_
CACHE - What doctor and the Caches page name GitHub Actions’ cache.
- ATTENTION_
MAX - The most items an
Attentionlists; the rest are counted inomitted. - BUILD_
CACHE_ TOOLS - The tools whose caches the build cache serves, as clients name them over its counts (ADR 0052, 0083). The image points each at it but Nx, which fails its run when the cache does not answer.
- COMPARE_
STEPS_ MAX - The most steps of each job a comparison reads, in order of their numbers.
- COMPARE_
STEP_ BYTES - The most bytes of step names of each job a comparison reads, each name read from its first 800 bytes, the most a title is read from; a first step is read whatever its name’s length.
- CONFIRM_
SECS - How long a preview’s confirmation can be sent back, in seconds.
- GRANULE
- Advice rounds memory up to a multiple of this: 256M.
- GROWTH_
WINDOW - How far back the growth count’s brake compares what runs uses: 10 minutes (ADR 0130,
admission::growth::GROWTH_WINDOW). - INSIGHT_
DAYS - The window
GET /v1/insightsreads by default, in days. - INSIGHT_
DAYS_ MAX - The longest window
GET /v1/insightsreads, in days. - JOBS_
HEADING - What a count of jobs, or a list of them on a page, is headed: the history’s figures, a profile’s facts, the waits
table. Not the name of
View::Jobs, though it reads the same. - JOBS_
PAGE_ BYTES - The text a page holds at most, in bytes, counted over each job’s repository, workflow, name, branch, URL and
labels as the list shows them (
row_bytes): a page ends before the job that would pass it. A job’s text is cut toJOB_TEXT_MAXandJOB_LABELS_MAX, so even a page’s first job is far below it. - JOBS_
PAGE_ DEFAULT - Jobs a page holds unless
limitsays otherwise. - JOBS_
PAGE_ MAX - Jobs a page holds at most.
- JOB_
ACTIONS_ CHECK - The doctor check that says whether the token can cancel and re-run jobs. Advisory: the attention list leaves it out whatever its status, since only GitHub’s answer to a real request says for sure.
- JOB_
LABELS_ MAX - The most bytes of a job’s labels, as stored (a JSON array of strings), a list shows; past them it shows none,
and the job is in
JobList::shortened. - JOB_
SEARCH_ MAX - The longest search,
JobQuery::q, in bytes. - JOB_
TEXT_ MAX - The most bytes of a job’s repository, workflow, name, branch and URL a list shows, each, to the last whole
character; past them, the text is cut and the job is in
JobList::shortened. - LINES_
DEFAULT GET /v1/jobs/{id}/log/lines: lines a page holds unlesslimitsays otherwise.- LINES_
MAX GET /v1/jobs/{id}/log/lines: lines a page holds at most.- LINE_
BYTES_ MAX - The most of a stored line’s text a styled line shows, in bytes; the rest is counted in
StyledLine::omitted. - NOT_
SUPPORTED - What a client says where an older daemon lacks a route or a field it reads: never an empty answer in its place.
- PAGE_
CACHE_ ROOM - What a cap keeps at least over the largest peak, for page cache: 1G. Peaks leave page cache out, and a container whose memory past them leaves the kernel less than about 500M of page cache to reclaim at its cap reclaims badly, which throttles its sockets and slows its downloads (ADR 0055).
- RECORDS_
NEAR_ FULL - The share of
[records] max_totalpast which the records are near it, in percent: new ones then remove the oldest to fit. - RUNS_
MAX - The most runs a styled line holds; the text after them is counted in
StyledLine::omitted. - SAMPLE_
INTERVAL - Seconds between two samples of a running job’s container. Clients without a
SizingorSampleSeriesthat says so, such as a job’s figures inrunnerctl, word their sampling caveat from it. - SEARCH_
QUERY_ MAX - The longest query, in bytes.
- SELF_
HOSTED_ RATE - What a minute of a job costs on this machine, in micro-dollars, subtracted from each hosted rate. GitHub postponed its $0.002 a minute charge for self-hosted runners; this is the one place to set it if it comes back.
- SHELL_
PROTOCOL - The
Upgradeprotocol ofPOST /v1/records/{job}/shell: after101 Switching Protocolsthe connection carries the shell’s terminal, raw bytes both ways. - SIBLINGS_
MAX - The most other jobs of the run a confirmation names.
- SLOT_
DISK - The least share of Docker’s disk, in bytes, auto sizing leaves each slot when it adds slots under measured admission: 2G. Every slot’s Docker store shares a quarter of Docker’s disk; at a 2G share the stores measured on a busy host keep their images between jobs, and at 1G, the least a share can be, every store over 1G is emptied after each job. What a store keeps does not set it: a store fills the share it gets (ADR 0090).
- TIMELINE_
BARS_ MAX - The most job bars a timeline holds, and the most runner spans: past it, the latest are kept and
SlotTimeline::cut_beforesays from when. - TIMELINE_
DEFAULT - The window a timeline shows unless
sincesays otherwise: the last 24 hours, in seconds. - TIMELINE_
WINDOW_ MAX - The longest window a timeline shows, in seconds: 7 days, the samples the history keeps by default.
Functions§
- actions_
cache_ place - Where jobs’ GitHub Actions cache is (ADR 0082), with
[caches.build] actions_cacheatserved. - actions_
in_ flight - The 409 while another action on the same run is being sent to GitHub, or one with the same confirmation.
- actions_
no_ run - The 422 for a job whose run the history did not record.
- actions_
not_ configured - The 403 for a job of a repository the config no longer has.
- actions_
off - What both routes answer, 403, while
[github] job_actionsis off. - ago
- How long ago
atwas:just nowunder ten seconds, then asdurationwrites it,1h 05m ago. - attention_
items - Everything that needs attention now, most urgent first: from
status, the doctor’s last run asdoctorsays it, andinsightswhen they could be read, atnow(seconds since the Unix epoch). Each item stands on evidence that holds now; a source without it is left out, and doctor checks that are not current are one item saying so. - billed_
minutes - How many minutes GitHub bills a job that ran from
startedtofinished(ms): each partial minute rounded up, and at least one. - cores
- Millicores as a number of CPUs to a tenth, rounded up:
2.5,32. - cores_
used - Millicores in use as cores, to a hundredth rounded up so use never reads as none:
0.25,1.5,12,0.01. - cpu_
time - CPU time given in milliseconds:
4.6sto a tenth under a minute, then asdurationwrites it. - cpus
- Millicores as
2.5 CPUsor1 CPU, to a tenth rounded up. - duration
- Seconds as a stopwatch writes them, its width holding as it counts:
45s,1m 12s,2h 05m,3d 4h. - force_
pause_ consequence - What forcing a pause does to
jobsrunning jobs:cancels its run on GitHub and removes its runner, or for several,cancels their runs on GitHub and removes their runners. - host_
margin - The plain margin
MemAvailablekeeps beside the runners on a host withtotalmemory: the larger ofMARGIN_MINandtotal / MARGIN_SHARE(ADR 0055). - in_
cores - Millicores in use as
cores_usedwrites them, with the unit:1.5 cores,1 core. - job_
title workflow / job, or the job’s name alone; for a job whose name the daemon does not know yet, as for one its runner took before the daemon saw it queued,job <id>, ora jobwithout an ID.- need
- What a set of jobs whose largest sampled peak is
peakneeds, withmarginpercent: the peak plus the margin, or plusPAGE_CACHE_ROOMwhen that is more, rounded up toGRANULE. The largest, not a percentile: the cap kills a job that goes over it, so a cap below a peak a job reached kills it again. - percent
- Hundredths of a percent as a percentage:
1.5%,12%. - period
- A setting’s length in words, as a config’s periods read:
5 minutes,24 hours,90 s. - reads_
since - From when, in milliseconds,
Savings::ofreads the history atnow(ms): the earlier of the month’s start and 30 days before. - records_
check - The doctor’s
recordscheck for records takingbytesof[records] max_total: a warning with what to run when they are near it (records_near_full), else how much they take. - records_
near_ full - Whether records taking
bytesare nearmax_total(RECORDS_NEAR_FULL). - reseed_
outcome - The line of a reseed’s
output, ascache.reseededcarries it, that says what it did: the one that started the snapshot or some of its layers over from the image, else its last. Empty for no output. - reseed_
requested - A reseed of
profilethe daemon accepted:jvm: reseed requested; it runs once the profile's slots are free. - reseeded
- A reseed of
profiledone, ascache.reseededreports it:jvm: reseededorjvm: reseed failed: …. - row_
bytes - The bytes of a job’s text the page budget counts, as the store counts them in its columns.
- sampled_
note - The caveat on a sampled peak, as every client words it:
memory peaks are sampled every 2 s; spikes shorter than that do not show.