runnerd_core/api/jobs.rs
1//! Job history as clients read it: one record per GitHub job a runner of this daemon ran, its timeline, steps,
2//! metrics and log size; the queries that list them; sample series for graphs; per-repository aggregates; cache
3//! sizes. Times are seconds since the Unix epoch, sizes are bytes, CPU is in millicores (1000 = one core busy).
4
5use serde::{Deserialize, Serialize};
6
7use super::{cpu_time, in_cores};
8use crate::config::ByteSize;
9
10mod compare;
11mod facts;
12mod log;
13mod query;
14mod timeline;
15
16pub(crate) use log::plain;
17
18#[cfg(feature = "host")]
19pub(crate) use compare::SAMPLE_GAP_INTERVALS;
20pub use compare::{
21 Baseline, BaselineRead, COMPARE_STEP_BYTES, COMPARE_STEPS_MAX, ChangeKind, CompareQuery,
22 ComparedSteps, FigureDelta, FigureKind, JobComparison, StepDelta, StepMatch, StepUsage,
23 Unavailable,
24};
25pub use facts::{
26 JobTitle, Milestone, Outcome, OutcomeKind, Phase, PhaseKind, RunComparison, SampleUsage,
27 job_title,
28};
29#[cfg(feature = "host")]
30pub(crate) use log::TITLE_SOURCE_MAX;
31pub use log::{
32 Color, FailureSummary, LINE_BYTES_MAX, LINES_DEFAULT, LINES_MAX, LineKind, LineRange, LineRead,
33 LogEvidence, LogGroup, LogLinesQuery, LogMark, LogNeedle, LogOutline, LogSearch,
34 LogSearchQuery, MarkKind, OutlineChange, OutlineParts, Outliner, RUNS_MAX, Run,
35 SEARCH_QUERY_MAX, SavedOutline, Search, StepLines, StepLog, StoredLine, StoredLog, StyledLine,
36};
37pub use query::{
38 JOB_LABELS_MAX, JOB_SEARCH_MAX, JOB_TEXT_MAX, JOBS_PAGE_BYTES, JOBS_PAGE_DEFAULT,
39 JOBS_PAGE_MAX, JobOrder, JobRange, JobSort, RecordFilter, RecordMark, SortOrder, row_bytes,
40};
41pub use timeline::{
42 BarEnd, Gap, GapKind, Lane, RunnerSpan, SlotKey, SlotTimeline, SlottedBar, SlottedRunner, Span,
43 TIMELINE_BARS_MAX, TIMELINE_DEFAULT, TIMELINE_WINDOW_MAX, TimelineBar, TimelineQuery,
44 TimelineRows,
45};
46
47/// Where a job is in its life, as this daemon saw it.
48#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
49#[serde(rename_all = "kebab-case")]
50pub enum JobState {
51 /// Waiting for a runner.
52 Queued,
53 /// A runner of this daemon took it.
54 Running,
55 /// GitHub reports it completed; `conclusion` says how.
56 Completed,
57 /// Its runner's container is gone and GitHub did not report the job completed in time.
58 Lost,
59 /// A runner of another machine took it, or it ended before any runner did, as when a newer push cancelled it.
60 /// One no runner took has no `started_at`, except in records of versions that kept the start GitHub gives it.
61 Elsewhere,
62}
63
64impl JobState {
65 /// Whether the job can still change.
66 pub fn is_final(self) -> bool {
67 matches!(
68 self,
69 JobState::Completed | JobState::Lost | JobState::Elsewhere
70 )
71 }
72}
73
74/// A job without its steps: what lists and events carry.
75#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
76pub struct JobSummary {
77 /// GitHub's job ID, also its check run ID.
78 pub id: u64,
79 /// `owner/name` on GitHub.
80 pub repository: String,
81 /// The workflow run it belongs to.
82 pub run_id: u64,
83 /// 1 for the first attempt of the run, higher for re-runs.
84 pub run_attempt: u32,
85 /// The workflow's name.
86 pub workflow: Option<String>,
87 /// The job's name in the workflow.
88 pub name: String,
89 /// The branch the run is for.
90 pub branch: Option<String>,
91 /// The job's page on GitHub.
92 pub url: Option<String>,
93 /// The workflow's `runs-on` labels.
94 pub labels: Vec<String>,
95 /// Where the job is.
96 pub state: JobState,
97 /// GitHub's conclusion once completed: `success`, `failure`, `cancelled`, `skipped`, `timed_out`, ….
98 pub conclusion: Option<String>,
99 /// The runner that ran it.
100 pub runner: Option<JobRunner>,
101 /// When each phase started.
102 pub timeline: Timeline,
103 /// CPU, memory and network use while it ran.
104 pub metrics: JobMetrics,
105 /// How much of its log is stored.
106 pub log: LogSummary,
107 /// Whether the runner failed it rather than its code (`record::runner_fault`): Sysbox lost the runner's seccomp
108 /// listener, so what its failed step printed has `function not implemented` on a mount, umount or openat2, or
109 /// `unsafe procfs detected` with `invalid argument`. Read from the log once the record is final; false for a job
110 /// that has not concluded failed, and in records from before it was kept.
111 #[serde(default)]
112 pub runner_fault: bool,
113 /// What its log says failed it, for a job that concluded failed and whose log shows one of the failures runnerd
114 /// tells apart (`record::failure_cause`): an action the runner could not download or unpack, one it could not
115 /// resolve, or `actions/checkout`. Read once the record is final; `None` for every other job, from a daemon older
116 /// than it, and where the log does not say.
117 #[serde(default, skip_serializing_if = "Option::is_none")]
118 pub failure_cause: Option<FailureCause>,
119 /// When runnerd asked GitHub to re-run its run's failed jobs because its runner failed it (ADR 0153), in seconds:
120 /// GitHub accepted, and the run went on to its next attempt. `None` for every other job, from a daemon older than
121 /// it, and where a person re-ran the run first.
122 #[serde(default, skip_serializing_if = "Option::is_none")]
123 pub rerun_at: Option<u64>,
124 /// What the runner image's probe found after the runner exited (ADR 0155): whether the calls Sysbox serves still
125 /// worked in its container when sampled. `None` where no verdict was recorded, including when collection is
126 /// disabled, a container stops early or the daemon is older. Recorded only; it decides nothing.
127 #[serde(default, skip_serializing_if = "Option::is_none")]
128 pub probe: Option<RunnerProbe>,
129 /// What became of its record, for a job the history notes one for: kept, removed or not made (ADR 0135).
130 /// `None` for a job without a record note, and from a daemon older than the mark. Not `record`, which a `Job`
131 /// that flattens this summary holds in full.
132 #[serde(default, skip_serializing_if = "Option::is_none")]
133 pub record_mark: Option<RecordMark>,
134}
135
136/// What a failed job's log says failed it (`record::failure_cause`), apart from its own code.
137#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
138#[serde(rename_all = "kebab-case")]
139pub enum FailureCause {
140 /// While it set up, the runner could not download an action's archive after its retries, or could not unpack
141 /// the archive it had.
142 ActionArchive,
143 /// While it set up, the runner could not resolve an action: GitHub's API did not say where to download it.
144 ActionResolve,
145 /// The step that failed is `actions/checkout`: the repository could not be fetched.
146 Checkout,
147 /// A cause a newer daemon tells apart.
148 #[serde(other)]
149 Unknown,
150}
151
152/// What the probe a runner's entrypoint runs as root after the runner exited found (ADR 0155): it mounts a tmpfs,
153/// a proc and a sysfs and binds one in a mount namespace of its own, then opens a file under `/proc/sys` with
154/// `openat2`, the calls Sysbox serves and a runner fault breaks.
155#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
156#[serde(rename_all = "kebab-case")]
157pub enum RunnerProbe {
158 /// Every call worked.
159 Clean,
160 /// A mount or an unmount failed; `/proc` opened.
161 Mount,
162 /// `/proc` did not open with `openat2`; the mounts worked.
163 Procfs,
164 /// Both failed.
165 MountAndProcfs,
166 /// It did not finish within its time.
167 TimedOut,
168 /// It ran but could not test: it failed before its first call.
169 Broken,
170 /// A verdict a newer daemon tells apart.
171 #[serde(other)]
172 Unknown,
173}
174
175/// `GET /v1/jobs/{id}`: a job with its steps.
176#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
177pub struct Job {
178 /// Everything but the steps.
179 #[serde(flatten)]
180 pub summary: JobSummary,
181 /// The steps as GitHub numbers them, including `Set up job` and `Complete job`.
182 pub steps: Vec<Step>,
183 /// Its record, for a failed job of a profile that records failures.
184 #[serde(default)]
185 pub record: Option<super::RecordNote>,
186 /// How its runner was admitted; `None` for one started before runnerd recorded it.
187 #[serde(default)]
188 pub admission: Option<super::Admission>,
189}
190
191impl Job {
192 /// The lines across its memory and CPU charts, as every client labels them: what admission reserved of the
193 /// budget, when the record has it, and the cap its container ran with, or for a record without it, `profile`'s
194 /// limit as configured now. A reservation of the whole cap is one line.
195 pub fn chart_lines(&self, profile: Option<&super::ProfileStatus>) -> ChartLines {
196 let limits = self.limits(profile);
197 let admission = self.admission.as_ref();
198 let name = profile.map_or("profile", |profile| profile.name.as_str());
199 ChartLines {
200 memory: chart_lines(
201 limits.memory,
202 admission.map(|admission| admission.reserved),
203 |value| ByteSize(value).to_string(),
204 name,
205 ),
206 cpu: chart_lines(
207 limits.cpu,
208 admission
209 .and_then(|admission| admission.cpu)
210 .map(|cpu| cpu.reserved),
211 super::cpus,
212 name,
213 ),
214 }
215 }
216
217 /// What its container may use of each resource, as every client scales its use against: the cap it ran with,
218 /// or, for a record without it, `profile`'s limit as configured now. Each resource falls back on its own, so a
219 /// reload that changes the profile never replaces a cap the record has.
220 pub fn limits(&self, profile: Option<&super::ProfileStatus>) -> Limits {
221 let runner = self.summary.runner.as_ref();
222 let limit = |cap: Option<u64>, now: Option<u64>| match (cap, now) {
223 (Some(value), _) => Some(Limit {
224 value,
225 source: LimitSource::Cap,
226 }),
227 (None, Some(value)) => Some(Limit {
228 value,
229 source: LimitSource::ProfileNow,
230 }),
231 (None, None) => None,
232 };
233 Limits {
234 memory: limit(
235 runner.and_then(|runner| runner.memory_cap),
236 profile.map(|profile| profile.memory),
237 ),
238 cpu: limit(
239 runner.and_then(|runner| runner.cpu_cap),
240 profile.map(|profile| u64::from(profile.cpus) * 1000),
241 ),
242 }
243 }
244}
245
246/// What a job's container may use, from `Job::limits`.
247#[derive(Debug, Clone, Copy, PartialEq, Eq)]
248pub struct Limits {
249 /// In bytes.
250 pub memory: Option<Limit>,
251 /// In millicores.
252 pub cpu: Option<Limit>,
253}
254
255impl Limits {
256 /// The limits as one clause, such as `capped at 6G and 4 CPUs`, `capped at 6G; profile now 4 CPUs` or `profile
257 /// now 6G and 4 CPUs`; `None` without either.
258 pub fn line(&self) -> Option<String> {
259 let memory = self
260 .memory
261 .map(|limit| (limit.source, ByteSize(limit.value).to_string()));
262 let cpu = self
263 .cpu
264 .map(|limit| (limit.source, super::cpus(limit.value)));
265 let clause = |source: LimitSource, figures: &str| match source {
266 LimitSource::Cap => format!("capped at {figures}"),
267 LimitSource::ProfileNow => format!("profile now {figures}"),
268 };
269 match (memory, cpu) {
270 (Some((memory_source, memory)), Some((cpu_source, cpu)))
271 if memory_source == cpu_source =>
272 {
273 Some(clause(memory_source, &format!("{memory} and {cpu}")))
274 }
275 (Some((memory_source, memory)), Some((cpu_source, cpu))) => Some(format!(
276 "{}; {}",
277 clause(memory_source, &memory),
278 clause(cpu_source, &cpu)
279 )),
280 (Some((source, figure)), None) | (None, Some((source, figure))) => {
281 Some(clause(source, &figure))
282 }
283 (None, None) => None,
284 }
285 }
286}
287
288/// One resource's limit for a job's container.
289#[derive(Debug, Clone, Copy, PartialEq, Eq)]
290pub struct Limit {
291 /// Bytes or millicores.
292 pub value: u64,
293 /// Where it comes from.
294 pub source: LimitSource,
295}
296
297/// Where a job's limit comes from.
298#[derive(Debug, Clone, Copy, PartialEq, Eq)]
299pub enum LimitSource {
300 /// The cap its container ran with.
301 Cap,
302 /// Its profile's limit in the config now, for a record without its container's cap; a reload since may have
303 /// changed it.
304 ProfileNow,
305}
306
307/// One resource's lines: `value` writes a figure, such as `8.5G` or `3.6 CPUs`.
308fn chart_lines(
309 limit: Option<Limit>,
310 reserved: Option<u64>,
311 value: fn(u64) -> String,
312 profile: &str,
313) -> Vec<ChartLine> {
314 let line = |figure, kind, label: &str| ChartLine {
315 value: figure,
316 kind,
317 label: format!("{label} {}", value(figure)),
318 };
319 let mut lines = vec![];
320 match limit {
321 Some(Limit {
322 value: cap,
323 source: LimitSource::Cap,
324 }) if reserved == Some(cap) => {
325 return vec![line(cap, ChartLineKind::CapAndReserved, "cap and reserved")];
326 }
327 Some(Limit {
328 value: cap,
329 source: LimitSource::Cap,
330 }) => lines.push(line(cap, ChartLineKind::Cap, "cap")),
331 Some(Limit {
332 value: now,
333 source: LimitSource::ProfileNow,
334 }) => lines.push(line(
335 now,
336 ChartLineKind::ProfileNow,
337 &format!("{profile} now"),
338 )),
339 None => {}
340 }
341 if let Some(reserved) = reserved {
342 lines.push(line(reserved, ChartLineKind::Reserved, "reserved"));
343 }
344 lines
345}
346
347/// The lines across a job's memory and CPU charts, from `Job::chart_lines`.
348#[derive(Debug, Clone, PartialEq, Eq)]
349pub struct ChartLines {
350 /// In bytes.
351 pub memory: Vec<ChartLine>,
352 /// In millicores.
353 pub cpu: Vec<ChartLine>,
354}
355
356/// A line across a job's memory or CPU chart.
357#[derive(Debug, Clone, PartialEq, Eq)]
358pub struct ChartLine {
359 /// Bytes or millicores.
360 pub value: u64,
361 /// What it stands for.
362 pub kind: ChartLineKind,
363 /// Its label, such as `cap 8.5G` or `reserved 3.6 CPUs`.
364 pub label: String,
365}
366
367/// What a line across a job's chart stands for.
368#[derive(Debug, Clone, Copy, PartialEq, Eq)]
369pub enum ChartLineKind {
370 /// The most its container may use: its cap.
371 Cap,
372 /// Its profile's limit in the config now, for a record without its container's cap; a reload since may have
373 /// changed it.
374 ProfileNow,
375 /// What admission counted against the budget for its runner.
376 Reserved,
377 /// A reservation of the whole cap, as admission by caps makes.
378 CapAndReserved,
379}
380
381/// The runner a job ran on.
382#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
383pub struct JobRunner {
384 /// The runner's name, also its container's.
385 pub name: String,
386 /// The profile it ran in.
387 pub profile: String,
388 /// The slot it ran in, 1-based.
389 pub slot: usize,
390 /// Its container's memory cap in bytes, which a reload since may have changed in the profile. `None` in records
391 /// from before runnerd kept it.
392 pub memory_cap: Option<u64>,
393 /// Its container's CPU limit in millicores, the unit of the CPU use in `Sample` and `JobMetrics`. `None` in
394 /// records from before runnerd kept it, or when Docker reports no limit.
395 pub cpu_cap: Option<u64>,
396}
397
398/// A job's phases, in the order they happen. A runner's container starts before GitHub gives it a job.
399#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
400pub struct Timeline {
401 /// GitHub queued the job.
402 pub queued_at: u64,
403 /// The container of the runner that took it started.
404 pub container_started_at: Option<u64>,
405 /// The runner started the job.
406 pub started_at: Option<u64>,
407 /// The job finished, by GitHub's clock; or when its container was found gone.
408 pub finished_at: Option<u64>,
409 /// The container and the runner's registration were removed.
410 pub cleaned_at: Option<u64>,
411 /// Seconds of its wait that passed while no runnerd ran, as the daemon that started after measured them. Waits
412 /// for a slot, as sizing, insights and the history's figures count them, leave them out; its own wait does not.
413 #[serde(default)]
414 pub unseen: u64,
415 /// When runnerd first listed it queued (ADR 0092). `None` for a job no listing showed queued, such as one a
416 /// warm runner took first, and for one an older runnerd recorded.
417 #[serde(default)]
418 pub seen_at: Option<u64>,
419}
420
421impl Timeline {
422 /// When the runner that took it began to start for it: its container's start, or the job's queueing when that
423 /// container waited for jobs before, as a warm runner does. `None` until a runner took it.
424 pub fn runner_from(&self) -> Option<u64> {
425 self.container_started_at
426 .map(|started| started.max(self.queued_at))
427 }
428}
429
430/// What a job used, over the samples taken while it ran (every few seconds). `None` until a sample exists.
431#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
432pub struct JobMetrics {
433 /// CPU time, in milliseconds.
434 pub cpu_time: Option<u64>,
435 /// The highest CPU use of a sample interval, in millicores.
436 pub cpu_peak: Option<u64>,
437 /// CPU time over wall time, in millicores.
438 pub cpu_average: Option<u64>,
439 /// The highest memory use without page cache.
440 pub memory_peak: Option<u64>,
441 /// The average memory use without page cache.
442 pub memory_average: Option<u64>,
443 /// Bytes the container received.
444 pub network_received: Option<u64>,
445 /// Bytes the container sent.
446 pub network_sent: Option<u64>,
447 /// Of the bytes received, those the Docker Hub mirror served.
448 pub mirror_bytes: u64,
449 /// What the shared build cache served it and took from it.
450 pub build_cache: BuildCacheCounts,
451 /// Times the kernel held its container's sockets' receive windows small because the container's memory was
452 /// under pressure (`memory.events` `sock_throttled`): reclaim went badly, mostly because the host was short of
453 /// memory, and its downloads slowed. `None` when the kernel does not count it, or in records from before.
454 #[serde(default, skip_serializing_if = "Option::is_none")]
455 pub sock_throttled: Option<u64>,
456 /// The periods of its CPU quota, 100 ms each, in which it had work: its busy time (`cpu.stat` `nr_periods`).
457 /// `None` where the kernel keeps no CPU quota's figures, or in records from before.
458 #[serde(default, skip_serializing_if = "Option::is_none")]
459 pub cpu_periods: Option<u64>,
460 /// Of `cpu_periods`, those in which it used up its quota and was held until the next (`nr_throttled`): its CPU
461 /// cap held it back.
462 #[serde(default, skip_serializing_if = "Option::is_none")]
463 pub cpu_throttled: Option<u64>,
464 /// Milliseconds in which a task of it could run and waited for a CPU (`cpu.pressure` `some`): held by its cap,
465 /// or by the other runners and programs on the host's CPUs. `None` without pressure stall information.
466 #[serde(default, skip_serializing_if = "Option::is_none")]
467 pub cpu_stalled: Option<u64>,
468 /// What its runner container wrote to the disks under Docker's root (ADR 0143). `None` where no reading measured
469 /// it, as without the io controller, and in records from before.
470 #[serde(default, skip_serializing_if = "Option::is_none")]
471 pub disk: Option<super::DiskWritten>,
472}
473
474/// One figure of what a job used, worded for every client.
475#[derive(Debug, Clone, PartialEq, Eq)]
476pub struct Total {
477 /// What it measures: `Memory peak`, `Received`.
478 pub label: &'static str,
479 /// The figure with its unit.
480 pub value: String,
481}
482
483impl JobMetrics {
484 /// What the job used, over its whole life: memory, CPU and network, each where the daemon measured it. The
485 /// peaks are of the samples taken every few seconds, so the figures' note says that
486 /// ([`sampled_note`](super::sampled_note)).
487 pub fn totals(&self) -> Vec<Total> {
488 let mut totals = Vec::new();
489 let mut add = |label: &'static str, value: Option<String>| {
490 totals.extend(value.map(|value| Total { label, value }));
491 };
492 let bytes = |bytes: u64| ByteSize(bytes).to_string();
493 add("Memory peak", self.memory_peak.map(bytes));
494 add("Average memory", self.memory_average.map(bytes));
495 add("CPU peak", self.cpu_peak.map(in_cores));
496 add("Average CPU", self.cpu_average.map(in_cores));
497 add("CPU time", self.cpu_time.map(cpu_time));
498 add("Received", self.network_received.map(bytes));
499 add("Sent", self.network_sent.map(bytes));
500 add("Written to disk", self.disk.map(|disk| disk.line()));
501 // Counted from the mirror's own log, so a job without network samples still has it.
502 add(
503 "From the mirror",
504 (self.network_received.is_some() || self.mirror_bytes > 0)
505 .then(|| bytes(self.mirror_bytes)),
506 );
507 add(
508 "Throttled for memory pressure",
509 self.sock_throttled
510 .filter(|times| *times > 0)
511 .map(|times| format!("{times} times")),
512 );
513 totals
514 }
515}
516
517/// What the shared build cache served one runner and took from it (ADR 0052), counted by the cache itself.
518#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
519pub struct BuildCacheCounts {
520 /// Lookups that found an entry.
521 pub hits: u64,
522 /// Lookups that found none.
523 pub misses: u64,
524 /// Bytes of the entries it served.
525 pub read_bytes: u64,
526 /// Bytes of the entries it stored.
527 pub written_bytes: u64,
528 /// Entries it was sent and did not store: the job may not write (a pull request, another branch), or the
529 /// cache was full.
530 pub refused: u64,
531 /// Whether the job could write: a push, scheduled or manually started run of a branch that may write.
532 pub writer: bool,
533}
534
535impl BuildCacheCounts {
536 /// Hits and misses.
537 pub fn lookups(&self) -> u64 {
538 self.hits + self.misses
539 }
540
541 /// Hits in percent of lookups, `None` without any.
542 pub fn hit_percent(&self) -> Option<u64> {
543 let lookups = self.lookups();
544 (lookups > 0).then(|| self.hits * 100 / lookups)
545 }
546
547 /// Adds `other`'s counts; `writer` stays.
548 pub fn add(&mut self, other: &BuildCacheCounts) {
549 self.hits += other.hits;
550 self.misses += other.misses;
551 self.read_bytes += other.read_bytes;
552 self.written_bytes += other.written_bytes;
553 self.refused += other.refused;
554 }
555}
556
557/// A job's stored log.
558#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
559pub struct LogSummary {
560 /// Lines stored.
561 pub lines: u64,
562 /// Bytes stored.
563 pub bytes: u64,
564 /// Set once the log reached `history.log_max` and later lines were dropped.
565 pub truncated: bool,
566}
567
568/// One step of a job, as GitHub reports it.
569#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
570pub struct Step {
571 /// GitHub's step number; numbers may skip.
572 pub number: u32,
573 /// The name GitHub shows, such as `Set up job` or `Run docker run --rm hello-world`.
574 pub name: String,
575 /// `pending`, `queued`, `in_progress` or `completed`.
576 pub status: String,
577 /// Once completed: `success`, `failure`, `skipped`, `cancelled`, ….
578 pub conclusion: Option<String>,
579 /// When it started.
580 pub started_at: Option<u64>,
581 /// When it completed.
582 pub completed_at: Option<u64>,
583}
584
585/// `GET /v1/jobs` query parameters. Jobs come newest first (by ID, which grows with time), or by `sort` (ADR 0135).
586/// [`JobQuery::check`] says which queries the daemon refuses.
587#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
588pub struct JobQuery {
589 /// Only this `owner/name`.
590 pub repository: Option<String>,
591 /// Only jobs in this state.
592 pub state: Option<JobState>,
593 /// Only jobs with this conclusion, such as `failure`.
594 pub conclusion: Option<String>,
595 /// Only jobs of the workflow with this name, such as `CI`.
596 pub workflow: Option<String>,
597 /// Only jobs with this name in their workflow, such as `build`.
598 pub name: Option<String>,
599 /// Only jobs queued at or after this time.
600 pub since: Option<u64>,
601 /// Only jobs queued before this time.
602 pub until: Option<u64>,
603 /// Only jobs with a smaller ID: the `next` of the previous page. Only in the default order, newest first.
604 pub before: Option<u64>,
605 /// At most this many jobs; default 50, at most 500. A page also ends once it holds
606 /// [`JOBS_PAGE_BYTES`] of text.
607 pub limit: Option<u32>,
608 /// Only jobs whose name, workflow or branch holds this text, as written: 1 to [`JOB_SEARCH_MAX`] bytes, ASCII
609 /// letters in either case. The history keeps no commit, so a commit is not searched.
610 #[serde(default, skip_serializing_if = "Option::is_none")]
611 pub q: Option<String>,
612 /// Only jobs of runs for this branch.
613 #[serde(default, skip_serializing_if = "Option::is_none")]
614 pub branch: Option<String>,
615 /// Only jobs whose record this says became of.
616 #[serde(default, skip_serializing_if = "Option::is_none")]
617 pub record: Option<RecordFilter>,
618 /// Sort by this; a job whose figure is unknown comes last either way, ties by ID. `queued`, newest first, is
619 /// the default order, which pages with `before`; any other order is one page, the first `limit` jobs.
620 #[serde(default, skip_serializing_if = "Option::is_none")]
621 pub sort: Option<JobSort>,
622 /// Which way; default largest or latest first.
623 #[serde(default, skip_serializing_if = "Option::is_none")]
624 pub order: Option<SortOrder>,
625 /// Refused: a sorted list is one page, which a search or the filters narrow. Read only to say so.
626 #[serde(default, skip_serializing_if = "Option::is_none")]
627 pub cursor: Option<String>,
628}
629
630/// One page of jobs, as `GET /v1/jobs` answered before it searched and sorted: what [`JobList`] holds.
631#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
632pub struct JobPage {
633 /// Newest first, or in the query's sort.
634 pub jobs: Vec<JobSummary>,
635 /// In the default order: pass as `before` for the next page; `None` on the last page, and in any other order.
636 pub next: Option<u64>,
637}
638
639/// `GET /v1/jobs`: one page of jobs (ADR 0135). A [`JobPage`] with fields more, so a client that reads a `JobPage`
640/// reads it too, and one that reads a `JobList` from an older daemon finds them absent.
641#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
642pub struct JobList {
643 /// The jobs and, in the default order, `next`.
644 #[serde(flatten)]
645 pub page: JobPage,
646 /// In any order but the default: how many jobs the filters admit, of which the page shows the first
647 /// ([`JobList::shown_words`]).
648 #[serde(default, skip_serializing_if = "Option::is_none")]
649 pub total: Option<u64>,
650 /// The jobs whose name, workflow, branch, repository, URL or labels were longer than a list shows
651 /// ([`JOB_TEXT_MAX`] bytes each, [`JOB_LABELS_MAX`] bytes of labels) and are cut here; their own page has them
652 /// whole.
653 #[serde(default, skip_serializing_if = "Vec::is_empty")]
654 pub shortened: Vec<u64>,
655 /// The query as the daemon read it, its limit and direction filled in; `None` from a daemon that predates
656 /// search and sorting, which ignored them ([`JobList::understood`]).
657 #[serde(default, skip_serializing_if = "Option::is_none")]
658 pub applied: Option<JobQuery>,
659}
660
661/// `GET /v1/jobs/{id}/logs` query parameters. The response is `text/plain`, one log line per line.
662#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
663pub struct LogQuery {
664 /// Start at this line, 0-based.
665 pub from: Option<u64>,
666 /// Keep the response open and send lines as they are stored, until the job finishes.
667 #[serde(default)]
668 pub follow: bool,
669}
670
671/// One sample of a container's use.
672#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
673pub struct Sample {
674 /// When it was taken.
675 pub at: u64,
676 /// CPU use since the previous sample, in millicores.
677 pub cpu: u64,
678 /// Memory use without page cache.
679 pub memory: u64,
680 /// Bytes received since the container started.
681 pub network_received: u64,
682 /// Bytes sent since the container started.
683 pub network_sent: u64,
684}
685
686/// `GET /v1/jobs/{id}/samples`: what the job's container used over time, from its start to its removal. Samples
687/// older than `history.samples` are deleted; the job's metrics stay.
688#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
689pub struct SampleSeries {
690 /// The job.
691 pub job: u64,
692 /// Seconds between samples.
693 pub interval: u64,
694 /// Oldest first.
695 pub samples: Vec<Sample>,
696}
697
698/// One sample of the whole slice runners run in.
699#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
700pub struct HostSample {
701 /// When it was taken.
702 pub at: u64,
703 /// CPU use of the slice since the previous sample, in millicores.
704 pub cpu: u64,
705 /// Memory use of the slice less its inactive page cache, as `docker stats` counts it: a conservative number,
706 /// since active page cache still counts, which the high-water mark of measured admission compares to
707 /// `MemoryMax`.
708 pub memory: u64,
709}
710
711/// The host's memory at one moment, beside the slice's: what admission fits new runners in besides `MemoryMax`
712/// (ADR 0055). Read every few seconds; not kept in the history.
713#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
714pub struct HostMemory {
715 /// When it was read.
716 pub at: u64,
717 /// `MemTotal` in `/proc/meminfo`, in bytes.
718 pub total: u64,
719 /// `MemAvailable`: what the host could give without swapping, page cache it can drop included, in bytes.
720 pub available: u64,
721 /// The slice's memory without page cache and reclaimable slab: what the runners hold that `MemAvailable` leaves
722 /// out.
723 pub slice: u64,
724 /// `/proc/pressure/memory`'s `some` line: the share of time at least one task waited for memory. `None` on a
725 /// kernel without pressure stall information.
726 pub some: Option<Pressure>,
727 /// Its `full` line: the share of time every task that could run waited for memory.
728 pub full: Option<Pressure>,
729 /// The programs outside the slice that hold the most memory, largest first: at most a few, none under a floor.
730 /// Read less often than the rest. Empty when none is that large.
731 #[serde(default, skip_serializing_if = "Vec::is_empty")]
732 pub holders: Vec<Holder>,
733 /// `[host] memory_ceiling`: the most memory the host can grow to, in bytes, for a VM whose hypervisor adds
734 /// memory as the guest asks for it (ADR 0084). From the config, not from `/proc`; absent unless it is set.
735 #[serde(default, skip_serializing_if = "Option::is_none")]
736 pub ceiling: Option<u64>,
737 /// `SwapTotal` and `SwapFree`, from the same read of `/proc/meminfo`; absent on a host without swap (ADR 0101).
738 /// Admission counts none of it.
739 #[serde(default, skip_serializing_if = "Option::is_none")]
740 pub swap: Option<Swap>,
741}
742
743/// A host's swap at one reading: zram and zswap's device included, as `/proc/meminfo` counts it.
744#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
745pub struct Swap {
746 /// `SwapTotal`, in bytes. Never 0: a host without swap has none.
747 pub total: u64,
748 /// `SwapFree`, in bytes.
749 pub free: u64,
750}
751
752impl Swap {
753 /// What is in use, in bytes.
754 pub fn used(&self) -> u64 {
755 self.total.saturating_sub(self.free)
756 }
757}
758
759/// The margin kept free at least: 2G.
760const MARGIN_MIN: u64 = 2 << 30;
761
762/// The margin as a share of the host's memory, as a divisor: a sixteenth, 4G of 64G.
763const MARGIN_SHARE: u64 = 16;
764
765/// The plain margin `MemAvailable` keeps beside the runners on a host with `total` memory: the larger of `MARGIN_MIN`
766/// and `total / MARGIN_SHARE` (ADR 0055).
767pub fn host_margin(total: u64) -> u64 {
768 MARGIN_MIN.max(total / MARGIN_SHARE)
769}
770
771impl HostMemory {
772 /// What the host has for a new runner this moment, in bytes: `MemAvailable` and what it can still grow by, less
773 /// the plain margin. Unlike `HostBudget::spare` it does not count what runners reserve past what they hold, which
774 /// is memory they may yet take, not memory taken.
775 pub fn room(&self) -> u64 {
776 self.available
777 .saturating_add(self.growable())
778 .saturating_sub(host_margin(self.grown()))
779 }
780
781 /// The host's memory as admission counts it, in bytes: `MemTotal`, or the ceiling when that is more (ADR 0084).
782 pub fn grown(&self) -> u64 {
783 self.ceiling
784 .map_or(self.total, |ceiling| ceiling.max(self.total))
785 }
786
787 /// What the host can still grow by before it reaches its ceiling, in bytes: 0 without one, or once `MemTotal`
788 /// reached it.
789 pub fn growable(&self) -> u64 {
790 self.grown().saturating_sub(self.total)
791 }
792
793 /// What the host holds outside the slice's runners, in bytes: its other programs, tmpfs and the kernel, without
794 /// page cache it can drop.
795 pub fn outside(&self) -> u64 {
796 self.total
797 .saturating_sub(self.available)
798 .saturating_sub(self.slice)
799 }
800
801 /// What holds the host's memory outside the slice, as the clients say it: `programs outside CI hold 30G` and,
802 /// with `holders`, `, most java 9.3G and qemu-system-x86 5.8G`.
803 pub fn outside_line(&self) -> String {
804 let size = |bytes: u64| ByteSize(bytes).to_string();
805 let line = format!("programs outside CI hold {}", size(self.outside()));
806 let named: Vec<String> = self
807 .holders
808 .iter()
809 .map(|holder| format!("{} {}", holder.name, size(holder.memory)))
810 .collect();
811 match named.as_slice() {
812 [] => line,
813 [one] => format!("{line}, most {one}"),
814 [rest @ .., last] => format!("{line}, most {} and {last}", rest.join(", ")),
815 }
816 }
817}
818
819/// A program outside the slice and the memory its processes hold.
820#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
821pub struct Holder {
822 /// Its processes' name, as `/proc/<pid>/comm` says it: never a command line, argument or path.
823 pub name: String,
824 /// What its processes hold that the host can't drop as page cache (`RssAnon` and `RssShmem`), in bytes.
825 pub memory: u64,
826}
827
828/// One line of a pressure stall file, in hundredths of a percent: 250 is 2.5% of the time.
829#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
830pub struct Pressure {
831 /// Over the last 10 seconds.
832 pub avg10: u32,
833 /// Over the last 60 seconds.
834 pub avg60: u32,
835}
836
837/// `GET /v1/host/samples` query parameters.
838#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
839pub struct HostSampleQuery {
840 /// Only samples taken at or after this time; default the last hour, at most the last 24 hours.
841 pub since: Option<u64>,
842}
843
844/// `GET /v1/host/samples`: the slice's use over time, oldest first.
845#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
846pub struct HostSamples {
847 /// Seconds between samples.
848 pub interval: u64,
849 /// Oldest first.
850 pub samples: Vec<HostSample>,
851}
852
853/// How `GET /v1/stats` groups jobs in time.
854#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
855#[serde(rename_all = "kebab-case")]
856pub enum Bucket {
857 /// By UTC hour.
858 Hour,
859 /// By UTC day.
860 #[default]
861 Day,
862}
863
864/// `GET /v1/activity` query parameters.
865#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
866pub struct ActivityQuery {
867 /// The start of the window; default 24 hours ago. It ends now.
868 pub since: Option<u64>,
869}
870
871/// `GET /v1/activity`: every job runnerd's runners started or finished in a window, by when that happened rather
872/// than when the job was queued, so a job that waited or ran long counts in the window it started or ended in.
873#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
874pub struct Activity {
875 /// The start of the window.
876 pub since: u64,
877 /// When it was read: the end of the window.
878 pub until: u64,
879 /// Each job that started in the window, oldest start first.
880 pub started: Vec<Started>,
881 /// The jobs that finished in the window: completed, or lost with their runner.
882 pub finished: u64,
883 /// Of those, the ones that failed, timed out or were lost, as `GET /v1/stats` counts failures.
884 pub failed: u64,
885 /// Of the failed, the ones their runner failed rather than their own steps: lost with their runner, or failed
886 /// after the kernel killed a process of theirs for memory at their cap.
887 pub runner_failed: u64,
888}
889
890impl Activity {
891 /// The 95th percentile of the waits of the jobs that started, by nearest rank as sizing's are; `None` when none
892 /// started.
893 pub fn wait_p95(&self) -> Option<u64> {
894 let mut waits: Vec<u64> = self.started.iter().map(|start| start.waited).collect();
895 waits.sort_unstable();
896 let rank = (waits.len() * 95).div_ceil(100);
897 waits.get(rank.checked_sub(1)?).copied()
898 }
899
900 /// The longest wait of the jobs that started in each of the 24 hours before `until`, oldest hour first; 0 for
901 /// an hour without a start. Counted back from `until`, so the hours are the day the other figures count.
902 pub fn wait_hours(&self) -> [u64; 24] {
903 const HOUR: u64 = 3600;
904 let since = self.until.saturating_sub(24 * HOUR);
905 let mut hours = [0u64; 24];
906 for start in self
907 .started
908 .iter()
909 .filter(|start| (since..=self.until).contains(&start.at))
910 {
911 // A start exactly 24 hours ago is in the oldest hour.
912 let back =
913 usize::try_from((self.until - start.at) / HOUR).map_or(23, |back| back.min(23));
914 hours[23 - back] = hours[23 - back].max(start.waited);
915 }
916 hours
917 }
918}
919
920/// A job that started: when, and how long it had waited since it was queued.
921#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
922pub struct Started {
923 /// When it started.
924 pub at: u64,
925 /// Seconds it waited, leaving out the time before the history's start and the time no runnerd ran (ADR 0081).
926 pub waited: u64,
927}
928
929/// `GET /v1/stats` query parameters.
930#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
931pub struct StatsQuery {
932 /// Only this `owner/name`.
933 pub repository: Option<String>,
934 /// Only jobs queued at or after this time; default 30 days ago. At most 744 buckets before `until` (or now)
935 /// are counted: 31 days of hours, two years of days.
936 pub since: Option<u64>,
937 /// Only jobs queued before this time.
938 pub until: Option<u64>,
939 /// The bucket size; default a day.
940 #[serde(default)]
941 pub bucket: Bucket,
942}
943
944/// `GET /v1/stats`: finished jobs per repository and time bucket, for trends. Sums, so clients can average.
945#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
946pub struct Stats {
947 /// The bucket size.
948 pub bucket: Bucket,
949 /// By bucket start, then repository.
950 pub rows: Vec<StatsRow>,
951}
952
953/// The finished jobs of one repository in one bucket.
954#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
955pub struct StatsRow {
956 /// The bucket's start.
957 pub start: u64,
958 /// `owner/name`.
959 pub repository: String,
960 /// Jobs completed or lost.
961 pub jobs: u64,
962 /// Of those, concluded `success`.
963 pub succeeded: u64,
964 /// Of those, concluded `failure` or `timed_out`, or lost.
965 pub failed: u64,
966 /// Of those, concluded `cancelled`.
967 pub cancelled: u64,
968 /// Seconds from queued to started, summed, leaving out the time before the history's start and the time no runnerd
969 /// ran (ADR 0081).
970 pub queue_seconds: u64,
971 /// Seconds from started to finished, summed.
972 pub run_seconds: u64,
973 /// CPU time, in milliseconds, summed.
974 pub cpu_time: u64,
975 /// The highest memory peak of a job.
976 pub memory_peak: u64,
977 /// Bytes received, summed.
978 pub network_received: u64,
979 /// Bytes from the mirror, summed.
980 pub mirror_bytes: u64,
981}
982
983/// What a cache volume holds.
984#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
985#[serde(rename_all = "kebab-case", tag = "kind")]
986pub enum CacheKind {
987 /// A profile's read-only snapshot.
988 Snapshot {
989 /// The profile.
990 profile: String,
991 },
992 /// A slot's layer changes and Docker Engine data.
993 Slot {
994 /// The profile.
995 profile: String,
996 /// 1-based.
997 slot: usize,
998 },
999 /// The pnpm store all slots share.
1000 Pnpm,
1001 /// The Maven repository all slots share.
1002 Maven,
1003 /// The Docker Hub mirror's storage.
1004 Mirror,
1005 /// The build cache every runner shares: a directory of the daemon's, not a Docker volume.
1006 Build,
1007}
1008
1009/// One cache volume and its size.
1010#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
1011pub struct CacheVolume {
1012 /// The Docker volume.
1013 pub name: String,
1014 /// What it holds.
1015 #[serde(flatten)]
1016 pub kind: CacheKind,
1017 /// Its size at `measured_at`.
1018 pub bytes: u64,
1019 /// The size above which runnerd empties it, or for a snapshot cuts it back at its next reseed, if it has one.
1020 pub limit: Option<u64>,
1021}
1022
1023/// `GET /v1/caches`: every cache volume of the configured profiles, as last measured. Docker measures volumes
1024/// slowly, so runnerd measures them at start and whenever no runner runs after jobs ended.
1025#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
1026pub struct Caches {
1027 /// When the sizes were measured; `None` before the first measurement.
1028 pub measured_at: Option<u64>,
1029 /// By name.
1030 pub volumes: Vec<CacheVolume>,
1031 /// The shared build cache now, when it is on.
1032 pub build: Option<BuildCacheStats>,
1033 /// The action archives runnerd keeps for its runners, when `[caches.actions]` is on (ADR 0147); absent from an
1034 /// older daemon.
1035 #[serde(default, skip_serializing_if = "Option::is_none")]
1036 pub actions: Option<super::ActionArchives>,
1037}
1038
1039/// The shared build cache (ADR 0052): its size now, and what it served the jobs of the last 7 days.
1040#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
1041pub struct BuildCacheStats {
1042 /// The entries' bytes.
1043 pub bytes: u64,
1044 /// How many entries it holds.
1045 pub entries: u64,
1046 /// `[caches.build] max_size`.
1047 pub limit: u64,
1048 /// The runners of the last 7 days, summed.
1049 pub week: BuildCacheCounts,
1050 /// `[caches.build] actions_cache`: whether it also serves GitHub Actions' cache (ADR 0082). `false` from a
1051 /// daemon older than it, which never served it.
1052 #[serde(default)]
1053 pub actions_cache: bool,
1054}
1055
1056/// `GET /v1/caches/history` query parameters.
1057#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
1058pub struct CacheHistoryQuery {
1059 /// Only measurements at or after this time; default 30 days ago.
1060 pub since: Option<u64>,
1061}
1062
1063/// One volume's size at one measurement.
1064#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
1065pub struct CacheSize {
1066 /// When it was measured.
1067 pub at: u64,
1068 /// The Docker volume.
1069 pub name: String,
1070 /// Its size.
1071 pub bytes: u64,
1072}
1073
1074/// `GET /v1/caches/history`: every measurement, oldest first.
1075#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
1076pub struct CacheHistory {
1077 /// Oldest first.
1078 pub sizes: Vec<CacheSize>,
1079}