runnerd_core/api.rs
1//! The daemon's JSON API, version 1: the request and response types every client deserializes, and the
2//! principal a transport authenticates. Plain data only: no I/O, no Docker or GitHub types, no secrets but a
3//! [`Pairing`]'s one-time link, which its `Debug` leaves out.
4//!
5//! Routes, all under the authenticated principal:
6//!
7//! | Route | Body |
8//! |---|---|
9//! | `GET /v1/status` | [`Status`] |
10//! | `POST /v1/repositories/{owner}/{name}/pause?`[`PauseQuery`] | [`Paused`]; 409 [`Busy`] while it runs a job |
11//! | `POST /v1/repositories/{owner}/{name}/resume` | none, 204 |
12//! | `POST /v1/repositories/pause?`[`PauseQuery`] | [`Paused`]; 409 [`Busy`] while any runs a job, pausing none |
13//! | `POST /v1/repositories/resume` | none, 204: every repository resumed |
14//! | `POST /v1/profiles/{name}/reseed` | none, 202; 409 while it reseeds; `cache.reseeded` when done |
15//! | `POST /v1/caches/{name}/clean` | none, 204; 409 [`Busy`] while a runner uses it |
16//! | `GET /v1/config` | [`ConfigFile`] |
17//! | `POST /v1/config/check` | TOML in, [`ConfigCheck`] |
18//! | `PUT /v1/config` | TOML in, [`ConfigCheck`]; 422 [`ConfigCheck`] when invalid |
19//! | `GET /v1/doctor` | [`Doctor`] |
20//! | `POST /v1/shutdown` | [`ShutdownRequest`] in, 202 [`ShutdownAnswer`] |
21//! | `GET /v1/dashboard` | [`Dashboard`] |
22//! | `POST /v1/dashboard/sign-out` | none, 204 |
23//! | `POST /v1/dashboard/pairings?`[`PairingQuery`] | [`Pairings`]; 409 when the dashboard is off or has no `[web] origins` |
24//! | `GET /v1/jobs?`[`JobQuery`] | [`JobPage`] |
25//! | `GET /v1/jobs/{id}` | [`Job`] |
26//! | `GET /v1/jobs/{id}/samples` | [`SampleSeries`] |
27//! | `GET /v1/jobs/{id}/logs?`[`LogQuery`] | `text/plain`, one line per log line |
28//! | `GET /v1/host/samples?`[`HostSampleQuery`] | [`HostSamples`] |
29//! | `GET /v1/stats?`[`StatsQuery`] | [`Stats`] |
30//! | `GET /v1/activity?`[`ActivityQuery`] | [`Activity`] |
31//! | `GET /v1/sizing?`[`SizingQuery`] | [`Sizing`] |
32//! | `GET /v1/insights?`[`InsightsQuery`] | [`Insights`] |
33//! | `GET /v1/digest` | [`Digest`] |
34//! | `GET /v1/savings` | [`Savings`] |
35//! | `POST /v1/notify/test?`[`NotifyTestQuery`] | [`NotifyTest`]; 404 without `[notify]` targets or with no such one |
36//! | `GET /v1/caches` | [`Caches`] |
37//! | `GET /v1/caches/history?`[`CacheHistoryQuery`] | [`CacheHistory`] |
38//! | `GET /v1/events?`[`EventQuery`] | `text/event-stream` of [`EventEnvelope`] |
39//! | `GET /v1/records` | [`Records`] |
40//! | `DELETE /v1/records/{job}` | [`RemovedRecords`]; 409 while a shell is open in it |
41//! | `DELETE /v1/records?`[`RemoveRecordsQuery`] | [`RemovedRecords`]; records with an open shell stay |
42//! | `POST /v1/records/{job}/shell?`[`ShellQuery`] | `Upgrade:` [`SHELL_PROTOCOL`], 101, then the shell's terminal; 409 when the budget is full |
43//!
44//! Errors carry an [`ErrorResponse`].
45
46use serde::{Deserialize, Serialize};
47
48mod admission;
49mod budget;
50mod control;
51mod events;
52mod insights;
53mod jobs;
54mod labels;
55mod records;
56mod results;
57mod savings;
58mod sizing;
59mod slots;
60mod stats;
61mod storage;
62mod tone;
63mod units;
64
65pub use admission::{Admission, AdmissionPolicy, Basis, CpuAdmission, JobPeak, cores, cpus};
66pub use budget::{
67 BudgetParts, ClosedKind, ClosedPart, FreePart, HeldKind, HeldPart, Reading, SlotPart,
68};
69pub use control::{
70 Busy, Check, CheckStatus, ConfigCheck, ConfigFile, Doctor, PauseQuery, Paused, ReplaceQuery,
71 RunningJob,
72};
73pub use events::{Event, EventEnvelope, EventQuery, RunnerFailure};
74pub use insights::{
75 AutoNote, Delivery, Digest, HostHazard, HostSpan, INSIGHT_DAYS, INSIGHT_DAYS_MAX, Insight,
76 InsightKind, Insights, InsightsQuery, LongWait, Notification, NotifyEvent, NotifyTest,
77 NotifyTestQuery, Priority, RepositoryWaits, RunLink, Severity, WaitSpread, WeekWaits,
78};
79pub use jobs::{
80 Activity, ActivityQuery, Bucket, BuildCacheCounts, BuildCacheStats, CacheHistory,
81 CacheHistoryQuery, CacheKind, CacheSize, CacheVolume, Caches, ChartLine, ChartLineKind,
82 ChartLines, Holder, HostMemory, HostSample, HostSampleQuery, HostSamples, Job, JobMetrics,
83 JobPage, JobQuery, JobRunner, JobState, JobSummary, JobTitle, Limit, LimitSource, Limits,
84 LogQuery, LogSummary, Milestone, Outcome, OutcomeKind, Phase, PhaseKind, Pressure,
85 RunComparison, Sample, SampleSeries, SampleUsage, Started, Stats, StatsQuery, StatsRow, Step,
86 Swap, Timeline, Total, host_margin, job_title,
87};
88pub use labels::{Column, View};
89pub use records::{
90 Record, RecordLimits, RecordNote, RecordRemoval, Records, RemoveRecordsQuery, RemovedRecords,
91 SHELL_PROTOCOL, ShellQuery,
92};
93pub use results::{force_pause_consequence, reseed_requested, reseeded};
94pub use savings::{
95 Period, PricedRun, PricedRunner, Pricing, PricingHistory, SELF_HOSTED_RATE, SavedWindow,
96 Savings, Sku, Usd, Visibility, billed_minutes, reads_since,
97};
98pub use sizing::{
99 Advice, AutoPlan, GRANULE, JobSizing, MemoryBound, MemoryHold, Need, PAGE_CACHE_ROOM, Peaks,
100 ProfilePlan, ProfileSizing, Resize, RevertCause, SAMPLE_INTERVAL, SLOT_DISK, Sizing,
101 SizingChange, SizingHold, SizingQuery, SizingRefused, SizingRevert, Spread, Starts, need,
102 period, sampled_note,
103};
104pub use slots::{Figure, Health, SlotState};
105pub use stats::StatsSummary;
106pub use storage::{
107 ACTIONS_CACHE, BUILD_CACHE_TOOLS, BuildCacheActivity, CacheLimit, RECORDS_NEAR_FULL,
108 RecordState, actions_cache_place, records_near_full,
109};
110pub use tone::{Tone, Toned};
111#[cfg(feature = "host")]
112pub(crate) use units::days_from_civil;
113pub use units::{ago, cores_used, cpu_time, duration, in_cores};
114
115/// Who sent a request, as the transport authenticated it. Handlers get it from one place, so a new transport only
116/// has to produce one.
117#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
118#[serde(rename_all = "kebab-case", tag = "kind")]
119pub enum Principal {
120 /// The user the daemon runs as, on the local Unix socket.
121 LocalOwner {
122 /// The Unix user ID the socket's peer runs as.
123 uid: u32,
124 },
125}
126
127/// `GET /v1/status`: what the daemon serves and runs right now.
128#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
129pub struct Status {
130 /// The daemon's version, from its `Cargo.toml`.
131 pub version: String,
132 /// Set once the daemon stops polling to shut down. [`Status::stopping_line`] says how.
133 pub shutting_down: bool,
134 /// How it stops, once it does; absent from a daemon before it said.
135 #[serde(default, skip_serializing_if = "Option::is_none")]
136 pub stopping: Option<Stopping>,
137 /// Until when GitHub's rate limit is used up, in seconds since the Unix epoch: no runner starts before.
138 /// [`Status::github_line`] says it. Absent while it is not.
139 #[serde(default, skip_serializing_if = "Option::is_none")]
140 pub github_limited_until: Option<u64>,
141 /// The memory budget and what runners use of it.
142 pub budget: Budget,
143 /// Every configured profile with its slots.
144 pub profiles: Vec<ProfileStatus>,
145 /// Every configured repository.
146 pub repositories: Vec<RepositoryStatus>,
147 /// Queued jobs the repositories' runners can take and none holds yet, at the last poll, oldest first. No event
148 /// carries the list: `job.queued` announces a job, `job.started` and `job.finished` its leaving the queue.
149 /// A repository whose last listing failed, as while GitHub rate-limits runnerd, or that is no longer polled
150 /// because runnerd is shutting down, shows no jobs here, while the history keeps its jobs queued.
151 pub queued: Vec<QueuedJob>,
152}
153
154/// The memory budget runners and the mirror share, in bytes.
155#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
156pub struct Budget {
157 /// The systemd slice runner containers run in.
158 pub slice: String,
159 /// The slice's `MemoryMax`; `None` when it is unlimited.
160 pub memory_max: Option<u64>,
161 /// Memory the running containers reserve: the runners' caps, or under measured admission the larger of each
162 /// runner's predicted peak and its last sampled memory, nothing for one whose container is gone, and the caps of
163 /// the seed containers, shells and the mirror.
164 pub memory_in_use: u64,
165 /// What pooling takes off `memory_in_use` (ADR 0099, 0130); absent while it takes nothing.
166 #[serde(default, skip_serializing_if = "Option::is_none")]
167 pub pooled: Option<Pooled>,
168 /// The growth count both budgets go by, and its brake (ADR 0130); `pooled` and `host.credit` say what it takes off
169 /// `memory_in_use`. Absent under caps admission, with `[host] pooled_count` off, and from an older daemon.
170 #[serde(default, skip_serializing_if = "Option::is_none")]
171 pub counted: Option<Counted>,
172 /// What a new runner must fit in beside them.
173 pub admission: AdmissionPolicy,
174 /// Under measured admission, whether the slice is at the high-water mark, so no runner starts.
175 pub held: bool,
176 /// The mirror's part of `memory_in_use`: its cap while it is enabled or still running, else 0.
177 pub mirror_memory: u64,
178 /// The seed containers' part of `memory_in_use`, one per profile that reseeds, by profile name.
179 pub seed_memory: Vec<SeedMemory>,
180 /// The shells' part of `memory_in_use`, one per shell open in a record, by job.
181 pub shell_memory: Vec<ShellMemory>,
182 /// What the host's memory leaves runners beside its other programs, from a recent reading of it; absent
183 /// without one.
184 #[serde(default, skip_serializing_if = "Option::is_none")]
185 pub host: Option<HostBudget>,
186 /// The CPU budget, under `[host] cpu_overcommit`. Absent without one.
187 #[serde(default, skip_serializing_if = "Option::is_none")]
188 pub cpu: Option<CpuBudget>,
189 /// The CPUs the slice may use, in millicores: its `AllowedCPUs` or the host's, less under a lower `CPUQuota`,
190 /// whether or not a CPU budget admits runners. The runners' CPU limits are quotas that may add up past it.
191 /// Absent when it could not be read.
192 #[serde(default, skip_serializing_if = "Option::is_none")]
193 pub slice_cpus: Option<u64>,
194}
195
196/// The count the budgets go by while pooling takes something off `Budget::memory_in_use` (ADR 0099).
197#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
198pub struct Pooled {
199 /// What it takes off, in bytes.
200 pub credit: u64,
201 /// `Budget::memory_in_use` less `credit`: what new runners fit beside under `MemoryMax`. The host's limit reads
202 /// `memory_in_use`.
203 pub counted: u64,
204 /// How many runners the credit counts on.
205 pub runners: u64,
206}
207
208/// How far back the growth count's brake compares what runs uses: 10 minutes (ADR 0130,
209/// `admission::growth::GROWTH_WINDOW`).
210pub const GROWTH_WINDOW: std::time::Duration = std::time::Duration::from_mins(10);
211
212/// The growth count (ADR 0130): what the runners use, and the room kept for what they may still grow by.
213#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
214pub struct Counted {
215 /// What the runners use, by their recent samples, in bytes.
216 pub used: u64,
217 /// The room it keeps for their growth, in bytes; absent while it rests (`brake`).
218 #[serde(default, skip_serializing_if = "Option::is_none")]
219 pub kept: Option<u64>,
220 /// How many of the largest growths the room covers at least.
221 pub jobs: u64,
222 /// Why it rests, while it does: the budgets then go by 0099's and 0108's credits.
223 #[serde(default, skip_serializing_if = "Option::is_none")]
224 pub brake: Option<CountBrake>,
225}
226
227/// Why the growth count rests (ADR 0130).
228#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
229#[serde(rename_all = "kebab-case", tag = "kind")]
230pub enum CountBrake {
231 /// What ran grew past the room kept within `GROWTH_WINDOW`.
232 Grew {
233 /// How much the runners present at both looks grew, together, in bytes.
234 grew: u64,
235 /// The room the earlier look kept, in bytes.
236 kept: u64,
237 /// Until when, in seconds since the Unix epoch.
238 until: u64,
239 },
240 /// The daemon started, or a reload changed what the count reads, too recently to judge growth.
241 Restarted {
242 /// Until when, in seconds since the Unix epoch.
243 until: u64,
244 },
245 /// The host is under the guard's line, or a reading is missing, now or at a look within 5 minutes.
246 Guard {
247 /// Until when, in seconds since the Unix epoch; absent for a moment the brake has not looked at yet.
248 #[serde(default, skip_serializing_if = "Option::is_none")]
249 until: Option<u64>,
250 },
251 /// A brake a newer daemon names; this client shows only that the count rests.
252 #[serde(other)]
253 Unknown,
254}
255
256/// What the host's memory leaves the runners (ADR 0055). New runners fit in the lower of `MemoryMax` and
257/// `limit`; running ones are never stopped for it.
258#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
259pub struct HostBudget {
260 /// The reading it comes from.
261 pub memory: HostMemory,
262 /// What the slice may hold now: what its runners hold without page cache plus the host's `MemAvailable`, less
263 /// a margin kept free for the host's page cache; with a ceiling, plus what the host can still grow by
264 /// (`HostMemory::growable`, ADR 0084).
265 pub limit: u64,
266 /// What that leaves new runners beside what the running ones reserve, in bytes.
267 pub spare: u64,
268 /// Whether the host's tasks stall for memory (`full` pressure), so no runner starts until they stop.
269 pub stalled: bool,
270 /// What the host's limit takes off `Budget::memory_in_use` for runners that may not all grow at once (ADR 0108);
271 /// absent while it takes nothing, and from a daemon before 0.0.30.
272 #[serde(default, skip_serializing_if = "Option::is_none")]
273 pub credit: Option<HostCredit>,
274}
275
276/// What the host's limit takes off the plain count (ADR 0108): `Budget::memory_in_use` less `bytes` is what it counts,
277/// `used` of that is what the runners use, and the rest is room kept for what runs to grow into.
278#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
279pub struct HostCredit {
280 /// What it takes off, in bytes.
281 pub bytes: u64,
282 /// What the runners use, by their recent samples, in bytes.
283 pub used: u64,
284}
285
286impl HostBudget {
287 /// Whether the host itself lacks `needs` bytes for a new runner: its tasks stall, `HostMemory::room` is less, or
288 /// that room is under one more margin, so `MemAvailable` is under twice the plain margin, the line `host::stalls`
289 /// reads. There a small job fits only by what the host is about to run out of, and what closing programs frees is
290 /// what starts it. Otherwise a job the host limit holds waits for what runs, whose reservations reach the limit
291 /// (ADR 0103).
292 pub fn short_of(&self, needs: u64) -> bool {
293 let room = self.memory.room();
294 self.stalled || needs > room || room < host_margin(self.memory.grown())
295 }
296
297 /// What the host's limit counts of what runs, `reserved` being `Budget::memory_in_use`: that less the credit
298 /// (ADR 0108).
299 pub fn counts(&self, reserved: u64) -> u64 {
300 reserved.saturating_sub(self.credit.map_or(0, |credit| credit.bytes))
301 }
302
303 /// What that count is made of while a credit stands, as the host's line, a job's wait and the log say it: `it uses
304 /// 14G and keeps 9G for growth`. `None` without a credit, when it counts every reservation.
305 pub fn growth_clause(&self, reserved: u64) -> Option<String> {
306 let credit = self.credit?;
307 let counts = self.counts(reserved);
308 Some(format!(
309 "it uses {} and keeps {} for growth",
310 crate::config::ByteSize(credit.used.min(counts)),
311 crate::config::ByteSize(counts.saturating_sub(credit.used))
312 ))
313 }
314}
315
316/// Why a job waits for the host's limit, needing `needs` bytes with `free` left of it, by the latest reading: for what
317/// runs, or for a host that lacks the room, whose programs it names. While the host's limit takes a credit, what runs
318/// is what it uses and the room it keeps for growth, not its reservations (ADR 0108); `reserved` is
319/// `Budget::memory_in_use`.
320fn host_reason(needs: u64, free: u64, host: Option<&HostBudget>, reserved: u64) -> String {
321 let size = |bytes: u64| crate::config::ByteSize(bytes).to_string();
322 if let Some(host) = host.filter(|host| !host.short_of(needs)) {
323 if let Some(growth) = host.growth_clause(reserved) {
324 return format!(
325 "The host has {}, but what runs may still grow into it: {growth}; needs {}, {} is free",
326 host.memory.available_line(),
327 size(needs),
328 size(free)
329 );
330 }
331 return format!(
332 "The host has {}, but what runs may still grow into it: needs {}; {} of the {} it leaves the slice is free",
333 host.memory.available_line(),
334 size(needs),
335 size(free),
336 size(host.limit)
337 );
338 }
339 let reason = if host.is_some_and(|host| host.stalled) {
340 "The host is short of memory: its programs stall waiting for it, so nothing starts"
341 .to_owned()
342 } else {
343 format!(
344 "The host is short of memory: needs {}; the host has {} to spare",
345 size(needs),
346 size(free)
347 )
348 };
349 match host {
350 Some(host) => format!("{reason}; {}", host.memory.outside_line()),
351 None => reason,
352 }
353}
354
355/// Hundredths of a percent as a percentage: `1.5%`, `12%`.
356pub fn percent(hundredths: u32) -> String {
357 let whole = hundredths / 100;
358 match hundredths % 100 / 10 {
359 0 => format!("{whole}%"),
360 tenth => format!("{whole}.{tenth}%"),
361 }
362}
363
364/// The CPU budget runners share, in millicores.
365#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
366pub struct CpuBudget {
367 /// The CPUs the slice may use: its `AllowedCPUs` or the host's, less under a lower `CPUQuota`.
368 pub cpus: u64,
369 /// Those times `cpu_overcommit`: what the runners' reservations must fit.
370 pub budget: u64,
371 /// What the runners reserve: each its predicted average CPU under measured admission, else its `cpus`.
372 pub in_use: u64,
373 /// Whether a queued job waits for CPU that memory and a slot have room for.
374 pub held: bool,
375}
376
377impl CpuBudget {
378 /// The budget's line as clients show it beside memory, such as `CPU 4.5 of 32 CPUs reserved` or, while a job
379 /// waits for it, `CPU 30 of 32 CPUs reserved; a job waits for CPU`.
380 pub fn line(&self) -> String {
381 let mut line = format!(
382 "CPU {} of {} CPUs reserved",
383 cores(self.in_use),
384 cores(self.budget)
385 );
386 if self.held {
387 line.push_str("; a job waits for CPU");
388 }
389 line
390 }
391}
392
393impl Budget {
394 /// How runners are admitted, as clients add it to the budget's line: nothing under caps, else `by predicted
395 /// peaks`, and why nothing starts while the slice is at the high-water mark.
396 pub fn admission_note(&self) -> Option<&'static str> {
397 match (self.admission, self.held) {
398 (AdmissionPolicy::Caps, _) => None,
399 (AdmissionPolicy::Measured, false) => Some("by predicted peaks"),
400 (AdmissionPolicy::Measured, true) => Some(
401 "by predicted peaks; the slice is past 90% of its MemoryMax, so no runner starts",
402 ),
403 }
404 }
405}
406
407/// A shell container's memory cap: its record's profile's limit when the shell opened.
408#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
409pub struct ShellMemory {
410 /// The job whose record the shell runs.
411 pub job: u64,
412 /// Its cap in bytes.
413 pub memory: u64,
414}
415
416/// A seed container's memory cap: its profile's limit when the reseed started, or, for a requested reseed whose
417/// profile drained, the room it holds until it starts.
418#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
419pub struct SeedMemory {
420 /// The profile it reseeds.
421 pub profile: String,
422 /// Its cap in bytes.
423 pub memory: u64,
424}
425
426/// A runner profile and what runs in its slots.
427#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
428pub struct ProfileStatus {
429 /// The name under `[profiles]` in `runnerd.toml`.
430 pub name: String,
431 /// The image its runners run, such as `runnerd-runner:base`.
432 pub image: String,
433 /// CPUs per slot.
434 pub cpus: u32,
435 /// Bytes per slot.
436 pub memory: u64,
437 /// Set while slot 1's layers become the profile's snapshot.
438 pub reseeding: bool,
439 /// Index 0 is slot 1.
440 pub slots: Vec<SlotStatus>,
441}
442
443/// One slot of a profile: room for one job.
444#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
445pub struct SlotStatus {
446 /// 1-based.
447 pub slot: usize,
448 /// `None` when the slot is free.
449 pub runner: Option<RunnerStatus>,
450}
451
452/// A runner registered for one job, in its container.
453#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
454pub struct RunnerStatus {
455 /// `owner/name` the runner is registered on.
456 pub repository: String,
457 /// The runner's name, also its container's.
458 pub name: String,
459 /// The runner's ID on GitHub.
460 pub runner_id: u64,
461 /// The Docker container ID.
462 pub container: String,
463 /// Whether it runs a job, as GitHub or its log reports it: its log names the job before GitHub lists it busy.
464 pub busy: bool,
465 /// Seconds since the Unix epoch.
466 pub started_at: u64,
467 /// What it takes of the budget: its cap under caps; under measured admission the larger of its predicted peak
468 /// and its last sampled memory.
469 pub reserved: u64,
470 /// A warm runner: idle, and no queued job of its repository waits for it; it takes the next one GitHub hands
471 /// out.
472 pub warm: bool,
473}
474
475/// A configured repository and its queue.
476#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
477pub struct RepositoryStatus {
478 /// `owner/name` on GitHub.
479 pub name: String,
480 /// Its pools in config order, one per `[[repositories]]` entry: a job runs in the one whose labels match it
481 /// (ADR 0058).
482 pub pools: Vec<PoolStatus>,
483 /// Paused repositories get no new runners; their running jobs finish.
484 pub paused: bool,
485 /// Its jobs in `Status::queued`: queued ones none of its runners took yet, those no pool serves too.
486 pub queued_jobs: usize,
487}
488
489/// One of a repository's pools: its runners in one profile.
490#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
491pub struct PoolStatus {
492 /// The profile its jobs run in.
493 pub profile: String,
494 /// The labels its runners register with.
495 pub labels: Vec<String>,
496 /// Its jobs in `Status::queued`: those its labels match.
497 pub queued_jobs: usize,
498}
499
500/// A queued job runners of this daemon can take, as GitHub listed it at the last poll.
501#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
502pub struct QueuedJob {
503 /// GitHub's job ID.
504 pub id: u64,
505 /// The repository whose runners can take it, `owner/name`.
506 pub repository: String,
507 /// The profile of its repository's pool whose labels match it; `None` when no pool's do.
508 #[serde(default, skip_serializing_if = "Option::is_none")]
509 pub profile: Option<String>,
510 /// The workflow run it belongs to.
511 pub run_id: u64,
512 /// 1 for the first attempt of the run, higher for re-runs.
513 pub run_attempt: u32,
514 /// The workflow's name.
515 pub workflow: Option<String>,
516 /// The job's name in the workflow.
517 pub name: String,
518 /// The branch the run is for.
519 pub branch: Option<String>,
520 /// The job's page on GitHub.
521 pub url: Option<String>,
522 /// When GitHub queued it, in seconds since the Unix epoch; 0 if GitHub did not say.
523 pub queued_at: u64,
524 /// Why no runner has it yet, as the last plan saw it. `None` before the daemon has planned with it, and while
525 /// it shuts down.
526 #[serde(default, skip_serializing_if = "Option::is_none")]
527 pub waiting: Option<Waiting>,
528}
529
530/// The slot of an idle runner of `job`'s repository, in its pool's profile when known, such as `jvm 2`.
531fn starting_for(job: &QueuedJob, status: &Status) -> Option<String> {
532 status
533 .profiles
534 .iter()
535 .filter(|profile| job.profile.as_ref().is_none_or(|own| *own == profile.name))
536 .find_map(|profile| {
537 profile
538 .slots
539 .iter()
540 .find(|slot| {
541 slot.runner
542 .as_ref()
543 .is_some_and(|runner| runner.repository == job.repository && !runner.busy)
544 })
545 .map(|slot| format!("{}\u{a0}{}", profile.name, slot.slot))
546 })
547}
548
549impl QueuedJob {
550 /// Where it runs, as every client names it: its repository, and its pool's profile when the repository has
551 /// several pools in `status`, such as `halyard/ledger · light`.
552 pub fn place(&self, status: &Status) -> String {
553 let pools = status
554 .repositories
555 .iter()
556 .find(|repository| repository.name == self.repository)
557 .map_or(0, |repository| repository.pools.len());
558 match self.profile.as_deref().filter(|_| pools > 1) {
559 Some(profile) => format!("{}\u{a0}· {profile}", self.repository),
560 None => self.repository.clone(),
561 }
562 }
563}
564
565impl RepositoryStatus {
566 /// The profile of its first pool, which every client colours it by.
567 pub fn first_profile(&self) -> &str {
568 self.pools.first().map_or("", |pool| pool.profile.as_str())
569 }
570
571 /// Its pools' profiles, as every client lists them on one line: `jvm` or `jvm + light`.
572 pub fn profiles(&self) -> String {
573 self.pools
574 .iter()
575 .map(|pool| pool.profile.as_str())
576 .collect::<Vec<_>>()
577 .join(" + ")
578 }
579
580 /// The profile of its pool whose labels match a job asking for `labels`, as the daemon picks it.
581 pub fn pool_for(&self, labels: &[String]) -> Option<String> {
582 self.pools
583 .iter()
584 .find(|pool| crate::config::labels_match(labels, &pool.labels))
585 .map(|pool| pool.profile.clone())
586 }
587}
588
589/// What failed, that holds a repository's runners back for a while.
590#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
591#[serde(rename_all = "kebab-case")]
592pub enum HoldCause {
593 /// Its runners ended without connecting, so without taking a job (ADR 0104).
594 NoJob,
595 /// Its runners connected and ended without a job, one after the other.
596 IdleExits,
597 /// GitHub did not register its runner.
598 Register,
599 /// Docker did not start its runner's container.
600 Start,
601}
602
603/// Why a queued job has no runner of its own yet. Jobs are served oldest first; each takes the first of these that
604/// holds for it.
605#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
606#[serde(rename_all = "kebab-case", tag = "kind")]
607pub enum Waiting {
608 /// Its repository is paused.
609 Paused,
610 /// Runners of its repository failed, so none starts before `until` (seconds since the Unix epoch).
611 Backoff {
612 /// When runners may start again.
613 until: u64,
614 /// What failed.
615 cause: HoldCause,
616 },
617 /// A runner started for it and waits for GitHub to hand it the job: a runner takes any job its labels match,
618 /// so not necessarily this one.
619 RunnerStarting,
620 /// Slot 1 of its profile is becoming the profile's snapshot, so none of the profile's slots starts.
621 Reseeding {
622 /// The profile.
623 profile: String,
624 },
625 /// Its profile's image is being pulled, so none of the profile's slots starts.
626 PullingImage {
627 /// The profile.
628 profile: String,
629 },
630 /// Every slot of its profile holds a runner.
631 NoSlot {
632 /// The profile.
633 profile: String,
634 },
635 /// What its runner would reserve does not fit the memory budget left. Under measured admission, `free` is 0
636 /// while the slice is at the high-water mark (`Budget::held`).
637 Memory {
638 /// What its runner would reserve, in bytes.
639 needs: u64,
640 /// What the budget has left, in bytes.
641 free: u64,
642 },
643 /// What its runner would reserve does not fit what the host's memory has to spare beside its other programs
644 /// (`HostBudget::spare`), though it fits the slice's budget. `free` is 0 while the host's tasks stall for memory
645 /// (`HostBudget::stalled`).
646 HostMemory {
647 /// What its runner would reserve, in bytes.
648 needs: u64,
649 /// What the host has to spare, in bytes.
650 free: u64,
651 },
652 /// What its runner would reserve does not fit the CPU budget left, in millicores.
653 Cpu {
654 /// What its runner would reserve.
655 needs: u64,
656 /// What the CPU budget has left.
657 free: u64,
658 },
659 /// An older job has waited long enough for CPU that no younger one takes CPU before it.
660 BehindCpu {
661 /// The older job's ID.
662 job: u64,
663 },
664 /// An older job has waited long enough for memory that no younger one starts before it (ADR 0096).
665 BehindMemory {
666 /// The older job's ID.
667 job: u64,
668 },
669 /// An older job is about to have a runner start for it, which takes the slot, memory or CPU this one would
670 /// otherwise start with: what holds it is that start, not what runs now.
671 BehindStart {
672 /// The older job's ID.
673 job: u64,
674 },
675 /// No pool of its repository has every label it asks for, so no runner of this daemon can take it (ADR 0058).
676 NoPool {
677 /// The labels it asks for.
678 labels: Vec<String>,
679 },
680 /// A reason from a newer daemon than this client knows, so a reason added later does not keep the client from
681 /// reading the status.
682 #[serde(other)]
683 Unknown,
684}
685
686/// What a [`Waiting`] reason waits on, which clients draw it by.
687#[derive(Debug, Clone, Copy, PartialEq, Eq)]
688pub enum WaitKind {
689 /// Something the owner can act on: a pause, runners that failed to take a job, or labels no pool serves.
690 Act,
691 /// Something that clears by itself shortly: a runner starting, a reseed, an image pull, an older job's start.
692 Soon,
693 /// Room: a slot, memory or CPU that a running job holds.
694 Full,
695 /// A reason this client does not know (`Waiting::Unknown`), drawn neutral.
696 Other,
697}
698
699impl Waiting {
700 /// What it waits on.
701 pub fn kind(&self) -> WaitKind {
702 match self {
703 Waiting::Paused | Waiting::Backoff { .. } | Waiting::NoPool { .. } => WaitKind::Act,
704 Waiting::Unknown => WaitKind::Other,
705 Waiting::RunnerStarting
706 | Waiting::Reseeding { .. }
707 | Waiting::PullingImage { .. }
708 | Waiting::BehindStart { .. } => WaitKind::Soon,
709 Waiting::NoSlot { .. }
710 | Waiting::Memory { .. }
711 | Waiting::HostMemory { .. }
712 | Waiting::Cpu { .. }
713 | Waiting::BehindCpu { .. }
714 | Waiting::BehindMemory { .. } => WaitKind::Full,
715 }
716 }
717
718 /// Why `job` waits, in the owner's words with the numbers that hold it, as every client says it: such as
719 /// `All 2 node slots are busy` or `Needs 9G of memory; 2.5G of the budget is free`. `status` is the one `job`
720 /// is queued in, which names the slot of a starting runner and the older job one waits behind.
721 pub fn reason(&self, job: &QueuedJob, status: &Status, now: u64) -> String {
722 let size = |bytes: u64| crate::config::ByteSize(bytes).to_string();
723 let older_name = |id: u64| {
724 status
725 .queued
726 .iter()
727 .find(|candidate| candidate.id == id)
728 .map_or_else(|| "an older job".to_owned(), |older| older.name.clone())
729 };
730 match self {
731 Waiting::Paused => format!("{} is paused", job.repository),
732 Waiting::Unknown => "Waits for a reason this client does not know yet".to_owned(),
733 Waiting::NoPool { labels } if labels.is_empty() => {
734 "It asks for no labels, so no profile serves it".to_owned()
735 }
736 Waiting::NoPool { labels } => format!(
737 "No profile serves the {} {}",
738 if labels.len() == 1 { "label" } else { "labels" },
739 labels.join(", ")
740 ),
741 Waiting::Backoff { until, cause } => format!(
742 "{}; retrying in {}",
743 match cause {
744 HoldCause::NoJob => "Runners failed to take a job",
745 HoldCause::IdleExits => "Runners ended without a job after waiting for one",
746 HoldCause::Register => "Runners cannot register with GitHub",
747 HoldCause::Start => "Runners cannot start",
748 },
749 duration(until.saturating_sub(now))
750 ),
751 Waiting::RunnerStarting => match starting_for(job, status) {
752 Some(slot) => format!("A runner is starting for it in {slot}"),
753 None => "A runner is starting for it".to_owned(),
754 },
755 Waiting::Reseeding { profile } => format!("{profile} is rebuilding its snapshot"),
756 Waiting::PullingImage { profile } => format!("Pulling {profile}'s image"),
757 Waiting::NoSlot { profile } => {
758 let slots = status
759 .profiles
760 .iter()
761 .find(|candidate| &candidate.name == profile)
762 .map_or(0, |profile| profile.slots.len());
763 if slots == 1 {
764 format!("The {profile} slot is busy")
765 } else {
766 format!("All {slots} {profile} slots are busy")
767 }
768 }
769 Waiting::Memory { .. } if status.budget.held => {
770 "The slice is past 90% of its MemoryMax, so nothing starts".to_owned()
771 }
772 Waiting::Memory { needs, free } => format!(
773 "Needs {} of memory; {} of the budget is free",
774 size(*needs),
775 size(*free)
776 ),
777 Waiting::HostMemory { needs, free } => host_reason(
778 *needs,
779 *free,
780 status.budget.host.as_ref(),
781 status.budget.memory_in_use,
782 ),
783 Waiting::Cpu { needs, free } => {
784 format!(
785 "Needs {}; {} are free",
786 cpus(*needs),
787 admission::free_cores(*free)
788 )
789 }
790 Waiting::BehindCpu { job: older } => {
791 format!("Behind {}, which waits for CPU", older_name(*older))
792 }
793 Waiting::BehindMemory { job: older } => {
794 format!("Behind {}, which waits for memory", older_name(*older))
795 }
796 Waiting::BehindStart { job: older } => {
797 format!(
798 "Next after {}, whose runner starts first",
799 older_name(*older)
800 )
801 }
802 }
803 }
804}
805
806/// `GET /v1/dashboard`: the dashboard's `[web]` settings, as the daemon read them when it started.
807#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
808pub struct Dashboard {
809 /// `[web] open`: the one of `[web] origins` that `runnerctl web` opens the dashboard on; the loopback address
810 /// when none.
811 #[serde(default, skip_serializing_if = "Option::is_none")]
812 pub open: Option<String>,
813}
814
815/// `POST /v1/dashboard/pairings`: which of `[web] origins` to make a link for.
816#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
817pub struct PairingQuery {
818 /// One of `[web] origins`; each of them when none.
819 #[serde(default, skip_serializing_if = "Option::is_none")]
820 pub origin: Option<String>,
821}
822
823/// `POST /v1/dashboard/pairings`: fresh one-time links that sign a device in to the dashboard.
824#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
825pub struct Pairings {
826 /// One per origin asked for.
827 pub links: Vec<Pairing>,
828}
829
830/// A link that signs one device in to the dashboard on `origin`, and nowhere else, within `lifetime`.
831#[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
832pub struct Pairing {
833 /// One of `[web] origins`.
834 pub origin: String,
835 /// The link with its one-time code: a credential, which `Debug` leaves out.
836 pub url: String,
837 /// Seconds it works for from when it was made.
838 pub lifetime: u64,
839}
840
841impl std::fmt::Debug for Pairing {
842 fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
843 formatter
844 .debug_struct("Pairing")
845 .field("origin", &self.origin)
846 .field("lifetime", &self.lifetime)
847 .finish_non_exhaustive()
848 }
849}
850
851/// `POST /v1/shutdown`.
852#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
853pub struct ShutdownRequest {
854 /// Kill running jobs and delete their runners instead of waiting for them.
855 #[serde(default)]
856 pub force: bool,
857}
858
859/// What `POST /v1/shutdown` answers: how the daemon stops now, which a stop asked for earlier may have made more
860/// urgent than the request.
861#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
862pub struct ShutdownAnswer {
863 /// How it stops.
864 pub stopping: Stopping,
865}
866
867/// How a stopping daemon treats running jobs (ADR 0091).
868#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
869#[serde(rename_all = "kebab-case")]
870pub enum Stopping {
871 /// `runnerd stop`: it waits for running jobs, then removes every runner.
872 WaitsForJobs,
873 /// A signal: it removes idle runners and leaves running jobs to the next start.
874 LeavesJobs,
875 /// `runnerd stop --force` or a second signal: it kills running jobs.
876 KillsJobs,
877 /// A kind of stop from a newer daemon than this client knows, so one added later does not keep the client from
878 /// reading the status.
879 #[serde(other)]
880 Unknown,
881}
882
883impl Stopping {
884 /// What it does, as every client says it: `shutting down: waits for running jobs`.
885 pub fn line(self) -> &'static str {
886 match self {
887 Stopping::WaitsForJobs => "shutting down: waits for running jobs",
888 Stopping::LeavesJobs => "shutting down: leaves running jobs to the next start",
889 Stopping::KillsJobs => "shutting down: kills running jobs",
890 Stopping::Unknown => "shutting down",
891 }
892 }
893}
894
895impl Status {
896 /// What a stopping daemon does, as every client says it ([`Stopping::line`]); `shutting down` from a daemon
897 /// that does not say how; `None` while it runs.
898 pub fn stopping_line(&self) -> Option<&'static str> {
899 match (self.stopping, self.shutting_down) {
900 (Some(stopping), _) => Some(stopping.line()),
901 (None, true) => Some("shutting down"),
902 (None, false) => None,
903 }
904 }
905
906 /// Why no runner starts while GitHub's rate limit is used up, as every client says it; `None` while it is not.
907 pub fn github_line(&self) -> Option<&'static str> {
908 self.github_limited_until
909 .map(|_| "GitHub's rate limit is used up: no runner starts until it resets")
910 }
911}
912
913/// The body of every error response.
914#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
915pub struct ErrorResponse {
916 /// What went wrong, for people to read.
917 pub message: String,
918}