OXYGENHUB
# 36e5a9c15a84 oxygenhub / crates / hub / src / db / flow.rs
31.1 KiB · Rust · last changed Runners decline, and the queue says why a job waits 1 month ago source blame history raw
1 
//! OxygenFlow runs, jobs and logs (hub spec §13.2).
2
3 
use super::{now_ms, Pool};
4 
use sqlx::FromRow;
5
6 
#[derive(Clone, Debug, FromRow)]
7 
pub struct Run {
8 
    pub id: i64,
9 
    pub repo_id: i64,
10 
    /// The run's number within its repository, from 1 (migration 0019).
11 
    #[sqlx(default)]
12 
    pub number: i64,
13 
    /// `push`, `manual` or `rerun:N` (migration 0020).
14 
    #[sqlx(default)]
15 
    pub trigger: String,
16 
    pub bookmark: String,
17 
    pub state: String,
18 
    pub heads: String,
19 
    pub pipeline_hash: String,
20 
    pub status: String,
21 
    pub error: String,
22 
    pub actor: String,
23 
    pub created_ms: i64,
24 
    pub finished_ms: Option<i64>,
25 
}
26
27 
impl Run {
28 
    /// How the run page says what started it.
29 
    pub fn started_by(&self) -> String {
30 
        match self.trigger.as_str() {
31 
            "manual" => format!("started by {}", self.actor),
32 
            "schedule" => "started on schedule".to_string(),
33 
            t if t.starts_with("tag:") => format!("tag {} pushed by {}", &t[4..], self.actor),
34 
            t if t.starts_with("rerun:") => format!("re-run of #{} by {}", self.rerun_of().unwrap_or(""), self.actor),
35 
            _ => format!("pushed by {}", self.actor),
36 
        }
37 
    }
38
39 
    /// The run a rerun repeats: `N` of `rerun:N` or `rerun:N:EVENT`.
40 
    pub fn rerun_of(&self) -> Option<&str> {
41 
        let rest = self.trigger.strip_prefix("rerun:")?;
42 
        Some(rest.split(':').next().unwrap_or(rest))
43 
    }
44
45 
    /// The trigger as people read it: `rerun:N`, without the event it repeats.
46 
    pub fn trigger_label(&self) -> String {
47 
        match self.rerun_of() {
48 
            Some(n) => format!("rerun:{n}"),
49 
            None => self.trigger.clone(),
50 
        }
51 
    }
52
53 
    pub fn head_list(&self) -> Vec<&str> {
54 
        self.heads.split('+').filter(|h| !h.is_empty()).collect()
55 
    }
56
57 
    pub fn live(&self) -> bool {
58 
        matches!(self.status.as_str(), "queued" | "running")
59 
    }
60 
}
61
62 
#[derive(Clone, Debug, Default, FromRow)]
63 
pub struct Job {
64 
    pub id: i64,
65 
    pub run_id: i64,
66 
    pub name: String,
67 
    pub depth: i64,
68 
    pub depends: String,
69 
    pub spec: String,
70 
    pub sandbox: String,
71 
    pub runtime: String,
72 
    pub status: String,
73 
    pub runner: String,
74 
    pub token_hash: String,
75 
    pub started_ms: Option<i64>,
76 
    pub finished_ms: Option<i64>,
77 
    pub exit_code: Option<i64>,
78 
    pub error: String,
79 
    pub receipt: String,
80 
    pub receipt_hash: String,
81 
    pub log_count: i64,
82 
    /// Comma-separated ids of other repositories the job token may read.
83 
    pub also_read: String,
84 
    /// The receipt attestation kept in the repository (Oxygen spec §14.6).
85 
    pub receipt_attest: String,
86 
    /// Who signed it: the runner's identity name.
87 
    pub receipt_signer: String,
88 
    /// Why a signed receipt was not kept.
89 
    pub receipt_note: String,
90 
    /// Why a queued job no runner can take waits (spec §11.4), from the
91 
    /// last placement pass; empty when it is only waiting its turn.
92 
    pub waiting: String,
93 
}
94
95 
impl Job {
96 
    pub fn depends_on(&self) -> Vec<String> {
97 
        serde_json::from_str(&self.depends).unwrap_or_default()
98 
    }
99
100 
    pub fn done(&self) -> bool {
101 
        matches!(self.status.as_str(), "passed" | "failed" | "skipped" | "error" | "cancelled")
102 
    }
103 
}
104
105 
#[derive(Clone, Debug, FromRow)]
106 
pub struct LogLine {
107 
    pub seq: i64,
108 
    pub t: i64,
109 
    pub stream: String,
110 
    pub text: String,
111 
}
112
113 
/// A job as the trigger creates it.
114 
pub struct NewJob {
115 
    pub name: String,
116 
    pub depth: i64,
117 
    pub depends: Vec<String>,
118 
    pub spec: String,
119 
    pub sandbox: String,
120 
    pub runtime: String,
121 
}
122
123 
pub struct NewRun<'a> {
124 
    pub repo_id: i64,
125 
    pub bookmark: &'a str,
126 
    pub state: &'a str,
127 
    pub heads: &'a str,
128 
    pub pipeline_hash: &'a str,
129 
    pub actor: &'a str,
130 
    /// `push`, `manual` or `rerun:N`.
131 
    pub trigger: &'a str,
132 
    /// Problems when the pipeline is invalid: the run is recorded in error.
133 
    pub error: Option<&'a str>,
134 
}
135
136 
pub async fn create_run(pool: &Pool, r: NewRun<'_>, jobs: &[NewJob]) -> sqlx::Result<i64> {
137 
    let mut tx = super::begin_write(pool).await?;
138 
    let now = now_ms();
139 
    let status = if r.error.is_some() { "error" } else { "queued" };
140 
    let id: i64 = sqlx::query_scalar(
141 
        "INSERT INTO flow_runs (repo_id, number, bookmark, state, heads, pipeline_hash, status, error, actor, trigger, created_ms, finished_ms) \
142 
         VALUES (?, (SELECT COALESCE(MAX(number), 0) + 1 FROM flow_runs WHERE repo_id = ?), ?, ?, ?, ?, ?, ?, ?, ?, ?, ?) RETURNING id",
143 
    )
144 
    .bind(r.repo_id)
145 
    .bind(r.repo_id)
146 
    .bind(r.bookmark)
147 
    .bind(r.state)
148 
    .bind(r.heads)
149 
    .bind(r.pipeline_hash)
150 
    .bind(status)
151 
    .bind(r.error.unwrap_or(""))
152 
    .bind(r.actor)
153 
    .bind(r.trigger)
154 
    .bind(now)
155 
    .bind(r.error.map(|_| now))
156 
    .fetch_one(&mut *tx)
157 
    .await?;
158 
    for j in jobs {
159 
        sqlx::query(
160 
            "INSERT INTO flow_jobs (run_id, name, depth, depends, spec, sandbox, runtime, status) VALUES (?, ?, ?, ?, ?, ?, ?, 'queued')",
161 
        )
162 
        .bind(id)
163 
        .bind(&j.name)
164 
        .bind(j.depth)
165 
        .bind(serde_json::to_string(&j.depends).unwrap_or_else(|_| "[]".into()))
166 
        .bind(&j.spec)
167 
        .bind(&j.sandbox)
168 
        .bind(&j.runtime)
169 
        .execute(&mut *tx)
170 
        .await?;
171 
    }
172 
    tx.commit().await?;
173 
    Ok(id)
174 
}
175
176 
/// The labels a compiled job and its sandbox ask a runner for (`runs-on`).
177 
fn labels_in(spec: &str, sandbox: &str) -> Vec<String> {
178 
    let mut out: Vec<String> = [spec, sandbox]
179 
        .iter()
180 
        .filter_map(|j| serde_json::from_str::<serde_json::Value>(j).ok())
181 
        .flat_map(|v| v["runs_on"].as_array().cloned().unwrap_or_default())
182 
        .filter_map(|l| l.as_str().map(String::from))
183 
        .collect();
184 
    out.sort();
185 
    out.dedup();
186 
    out
187 
}
188
189 
impl Job {
190 
    /// The runner labels this job needs.
191 
    pub fn needs(&self) -> Vec<String> {
192 
        labels_in(&self.spec, &self.sandbox)
193 
    }
194
195 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
196 
    pub fn arch(&self) -> Option<String> {
197 
        arch_in(&self.sandbox)
198 
    }
199 
}
200
201 
/// The architecture a sandbox's JSON asks for, if any.
202 
fn arch_in(sandbox: &str) -> Option<String> {
203 
    serde_json::from_str::<serde_json::Value>(sandbox).ok()?["arch"].as_str().map(String::from)
204 
}
205
206 
/// A job waiting for or held by a runner, with where it comes from.
207 
#[derive(Clone, Debug, FromRow)]
208 
pub struct QueueEntry {
209 
    pub job_id: i64,
210 
    pub run_id: i64,
211 
    pub repo_id: i64,
212 
    pub owner: String,
213 
    pub name: String,
214 
    pub visibility: String,
215 
    pub run_number: i64,
216 
    pub job: String,
217 
    pub status: String,
218 
    /// The runner holding it, when running.
219 
    pub runner: String,
220 
    pub spec: String,
221 
    pub sandbox: String,
222 
}
223
224 
impl QueueEntry {
225 
    /// The runner labels the job needs.
226 
    pub fn needs(&self) -> Vec<String> {
227 
        labels_in(&self.spec, &self.sandbox)
228 
    }
229
230 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
231 
    pub fn arch(&self) -> Option<String> {
232 
        arch_in(&self.sandbox)
233 
    }
234 
}
235
236 
/// Every job of a live run that is running or queued: running ones first,
237 
/// then queued ones in the order runners claim them (hub spec §13.2).
238 
pub async fn queue(pool: &Pool) -> sqlx::Result<Vec<QueueEntry>> {
239 
    sqlx::query_as(
240 
        "SELECT j.id AS job_id, j.run_id, r.repo_id, p.owner, p.name, p.visibility, r.number AS run_number, j.name AS job, j.status, j.runner, j.spec, j.sandbox \
241 
         FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories p ON p.id = r.repo_id \
242 
         WHERE j.status IN ('queued', 'running') AND r.status IN ('queued', 'running') \
243 
         ORDER BY CASE j.status WHEN 'running' THEN 0 ELSE 1 END, j.run_id, j.depth, j.id",
244 
    )
245 
    .fetch_all(pool)
246 
    .await
247 
}
248
249 
/// A repository's runs, newest first, numbered below `before`.
250 
pub async fn runs_by_number(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
251 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = ? AND number < ? ORDER BY number DESC LIMIT ?")
252 
        .bind(repo_id)
253 
        .bind(before.unwrap_or(i64::MAX))
254 
        .bind(limit)
255 
        .fetch_all(pool)
256 
        .await
257 
}
258
259 
/// The latest run on this bookmark of this repository, for what a push
260 
/// changed since (a job's scope).
261 
/// The run a push to `bookmark` is scoped against: the newest that was not
262 
/// cancelled, nor a push run still going (a new push supersedes it), so
263 
/// that what a cancelled run would have checked is checked by the next.
264 
pub async fn latest_on(pool: &Pool, repo_id: i64, bookmark: &str) -> sqlx::Result<Option<Run>> {
265 
    sqlx::query_as(
266 
        "SELECT * FROM flow_runs WHERE repo_id = ? AND bookmark = ? AND heads != '' AND status != 'cancelled'
267 
         AND NOT (trigger = 'push' AND status IN ('queued', 'running')) ORDER BY id DESC LIMIT 1",
268 
    )
269 
    .bind(repo_id)
270 
    .bind(bookmark)
271 
    .fetch_optional(pool)
272 
    .await
273 
}
274
275 
pub async fn runs(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
276 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = ? AND id < ? ORDER BY id DESC LIMIT ?")
277 
        .bind(repo_id)
278 
        .bind(before.unwrap_or(i64::MAX))
279 
        .bind(limit)
280 
        .fetch_all(pool)
281 
        .await
282 
}
283
284 
/// A repository's run by its number there.
285 
pub async fn run(pool: &Pool, repo_id: i64, number: i64) -> sqlx::Result<Option<Run>> {
286 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = ? AND number = ?").bind(repo_id).bind(number).fetch_optional(pool).await
287 
}
288
289 
pub async fn run_by_id(pool: &Pool, id: i64) -> sqlx::Result<Option<Run>> {
290 
    sqlx::query_as("SELECT * FROM flow_runs WHERE id = ?").bind(id).fetch_optional(pool).await
291 
}
292
293 
pub async fn jobs(pool: &Pool, run_id: i64) -> sqlx::Result<Vec<Job>> {
294 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = ? ORDER BY depth, name").bind(run_id).fetch_all(pool).await
295 
}
296
297 
pub async fn job(pool: &Pool, id: i64) -> sqlx::Result<Option<Job>> {
298 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE id = ?").bind(id).fetch_optional(pool).await
299 
}
300
301 
/// The running job a job token belongs to, while it may still be used.
302 
pub async fn job_by_token(pool: &Pool, token_hash: &str, max_age_ms: i64) -> sqlx::Result<Option<Job>> {
303 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE token_hash = ? AND status = 'running' AND started_ms >= ?")
304 
        .bind(token_hash)
305 
        .bind(now_ms() - max_age_ms)
306 
        .fetch_optional(pool)
307 
        .await
308 
}
309
310 
/// Take the oldest queued job whose dependencies all passed and whose tier
311 
/// the runner offers, and mark it running for `runner` with a token.
312 
/// Let a job's token also read these repositories (its pinned subrepos).
313 
pub async fn set_also_read(pool: &Pool, job_id: i64, repo_ids: &[i64]) -> sqlx::Result<()> {
314 
    let list: Vec<String> = repo_ids.iter().map(|i| i.to_string()).collect();
315 
    sqlx::query("UPDATE flow_jobs SET also_read = ? WHERE id = ?").bind(list.join(",")).bind(job_id).execute(pool).await?;
316 
    Ok(())
317 
}
318
319 
impl Job {
320 
    pub fn also_read_ids(&self) -> Vec<i64> {
321 
        self.also_read.split(',').filter_map(|x| x.trim().parse().ok()).collect()
322 
    }
323 
}
324
325 
/// What a runner can take on: free memory (MiB) and vCPUs, `None` for no
326 
/// limit, and the rootfs images it holds, `None` when it does not say.
327 
#[derive(Clone, Debug, Default)]
328 
pub struct Room {
329 
    pub memory_mib: Option<u32>,
330 
    pub cpus: Option<u32>,
331 
    pub images: Option<std::collections::HashSet<String>>,
332 
    /// The runner's architecture (`x86_64`, `riscv64`, …), empty when unknown.
333 
    pub arch: String,
334 
    /// Each image's architecture as runners have reported it (`flow_images`).
335 
    pub image_archs: std::collections::HashMap<String, String>,
336 
    /// Jobs this runner declined, not to be offered to it again (spec §11.5).
337 
    pub declined: std::collections::HashSet<i64>,
338 
}
339
340 
impl Room {
341 
    /// Whether a job's sandbox (its stored JSON) fits: its memory and vCPUs,
342 
    /// and its rootfs image among the runner's (oxygenhub#73).
343 
    pub fn fits(&self, sandbox_json: &str) -> bool {
344 
        let v: serde_json::Value = serde_json::from_str(sandbox_json).unwrap_or_default();
345 
        let need = |k: &str| v[k].as_u64().unwrap_or(0);
346 
        let rootfs = v["rootfs"].as_str().unwrap_or("");
347 
        // The architecture: the sandbox's own ask, else the image's as known;
348 
        // an unknown one matches, as before (OxygenFlow spec §11.3).
349 
        let wanted = v["arch"].as_str().map(String::from).or_else(|| self.image_archs.get(rootfs).cloned());
350 
        let arch_ok = self.arch.is_empty() || wanted.as_deref().is_none_or(|w| w == self.arch);
351 
        arch_ok
352 
            && self.memory_mib.is_none_or(|m| need("memory_mib") <= m as u64)
353 
            && self.cpus.is_none_or(|c| need("cpus") <= c as u64)
354 
            && (rootfs.is_empty() || self.images.as_ref().is_none_or(|have| have.contains(rootfs)))
355 
    }
356 
}
357
358 
/// Remember which architecture each image is for, as a runner reports it.
359 
pub async fn record_image_archs(pool: &Pool, archs: &std::collections::BTreeMap<String, String>) -> sqlx::Result<()> {
360 
    for (hash, arch) in archs.iter().take(500) {
361 
        if hash.len() != 64 || !flow_core::arch_ok(arch) {
362 
            continue;
363 
        }
364 
        sqlx::query("INSERT INTO flow_images (hash, arch, seen_ms) VALUES (?, ?, ?) ON CONFLICT (hash) DO UPDATE SET arch = excluded.arch, seen_ms = excluded.seen_ms")
365 
        .bind(hash.to_ascii_lowercase())
366 
        .bind(arch)
367 
        .bind(now_ms())
368 
        .execute(pool)
369 
        .await?;
370 
    }
371 
    Ok(())
372 
}
373
374 
/// Every image's architecture the hub knows.
375 
pub async fn image_archs(pool: &Pool) -> sqlx::Result<std::collections::HashMap<String, String>> {
376 
    let rows: Vec<(String, String)> = sqlx::query_as("SELECT hash, arch FROM flow_images").fetch_all(pool).await?;
377 
    Ok(rows.into_iter().collect())
378 
}
379
380 
pub async fn claim(pool: &Pool, runtimes: &[String], runner: &str, token_hash: &str) -> sqlx::Result<Option<Job>> {
381 
    claim_within(pool, runtimes, runner, token_hash, Room::default(), &Default::default(), &[]).await
382 
}
383
384 
/// The oldest runnable job that fits in `room`.
385 
/// The first queued job a runner with these tiers, this room and this
386 
/// scope may take, now running under `runner`. A scoped runner never sees
387 
/// another scope's jobs (oxygenhub#41).
388 
pub async fn claim_within(
389 
    pool: &Pool,
390 
    runtimes: &[String],
391 
    runner: &str,
392 
    token_hash: &str,
393 
    room: Room,
394 
    scope: &super::runners::Scope,
395 
    labels: &[String],
396 
) -> sqlx::Result<Option<Job>> {
397 
    let mut tx = super::begin_write(pool).await?;
398 
    let queued: Vec<Job> = sqlx::query_as(
399 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories rep ON rep.id = r.repo_id \
400 
         WHERE j.status = 'queued' AND r.status IN ('queued', 'running') \
401 
         AND (? = '' OR rep.owner = ?) AND (? IS NULL OR r.repo_id = ?) \
402 
         ORDER BY j.run_id, j.depth, j.id LIMIT 200",
403 
    )
404 
    .bind(&scope.owner)
405 
    .bind(&scope.owner)
406 
    .bind(scope.repo_id)
407 
    .bind(scope.repo_id)
408 
    .fetch_all(&mut *tx)
409 
    .await?;
410 
    for j in queued {
411 
        if !runtimes.iter().any(|r| r == &j.runtime) || !room.fits(&j.sandbox) || room.declined.contains(&j.id) {
412 
            continue;
413 
        }
414 
        // Every label the job and its sandbox ask for (`runs-on`).
415 
        if !j.needs().iter().all(|l| labels.contains(l)) {
416 
            continue;
417 
        }
418 
        let deps = j.depends_on();
419 
        if !deps.is_empty() {
420 
            let mut passed = 0i64;
421 
            for d in &deps {
422 
                let st: Option<String> = sqlx::query_scalar("SELECT status FROM flow_jobs WHERE run_id = ? AND name = ?")
423 
                    .bind(j.run_id)
424 
                    .bind(d)
425 
                    .fetch_optional(&mut *tx)
426 
                    .await?;
427 
                if st.as_deref() == Some("passed") {
428 
                    passed += 1;
429 
                }
430 
            }
431 
            if passed != deps.len() as i64 {
432 
                continue;
433 
            }
434 
        }
435 
        let now = now_ms();
436 
        let done = sqlx::query(
437 
            "UPDATE flow_jobs SET status = 'running', runner = ?, token_hash = ?, started_ms = ? WHERE id = ? AND status = 'queued'",
438 
        )
439 
        .bind(runner)
440 
        .bind(token_hash)
441 
        .bind(now)
442 
        .bind(j.id)
443 
        .execute(&mut *tx)
444 
        .await?;
445 
        if done.rows_affected() == 1 {
446 
            sqlx::query("UPDATE flow_runs SET status = 'running' WHERE id = ? AND status = 'queued'")
447 
                .bind(j.run_id)
448 
                .execute(&mut *tx)
449 
                .await?;
450 
            tx.commit().await?;
451 
            return job(pool, j.id).await;
452 
        }
453 
    }
454 
    tx.commit().await?;
455 
    Ok(None)
456 
}
457
458 
/// How long a decline keeps a job from its runner when the runner's
459 
/// profile stays the same: a reason such as memory in use passes.
460 
pub const DECLINE_HOLDS_MS: i64 = 10 * 60_000;
461
462 
/// The jobs runner `runner_id` declined with this profile, recently.
463 
pub async fn declined_by(pool: &Pool, runner_id: i64, profile: &str) -> sqlx::Result<std::collections::HashSet<i64>> {
464 
    let rows: Vec<(i64,)> = sqlx::query_as("SELECT job_id FROM flow_declines WHERE runner_id = ? AND profile = ? AND ms > ?")
465 
        .bind(runner_id)
466 
        .bind(profile)
467 
        .bind(now_ms() - DECLINE_HOLDS_MS)
468 
        .fetch_all(pool)
469 
        .await?;
470 
    Ok(rows.into_iter().map(|(j,)| j).collect())
471 
}
472
473 
/// A decline: one runner's reason for handing a job back.
474 
#[derive(Clone, Debug, sqlx::FromRow)]
475 
pub struct DeclineRow {
476 
    pub job_id: i64,
477 
    pub runner_id: i64,
478 
    pub runner: String,
479 
    pub reason: String,
480 
    pub ms: i64,
481 
}
482
483 
/// Runner `runner_id` hands back job `job_id`, running on it: the decline
484 
/// is kept (one per job and runner, the latest), and the job is queued
485 
/// again. Returns every decline the job has had, or `None` when the job
486 
/// was not running on that runner.
487 
pub async fn decline(
488 
    pool: &Pool,
489 
    job_id: i64,
490 
    runner_id: i64,
491 
    runner: &str,
492 
    profile: &str,
493 
    reason: &str,
494 
) -> sqlx::Result<Option<Vec<DeclineRow>>> {
495 
    let mut tx = super::begin_write(pool).await?;
496 
    let back = sqlx::query(
497 
        "UPDATE flow_jobs SET status = 'queued', runner = '', token_hash = '', started_ms = NULL, runner_id = NULL \
498 
         WHERE id = ? AND status = 'running' AND runner_id = ?",
499 
    )
500 
    .bind(job_id)
501 
    .bind(runner_id)
502 
    .execute(&mut *tx)
503 
    .await?;
504 
    if back.rows_affected() != 1 {
505 
        tx.rollback().await?;
506 
        return Ok(None);
507 
    }
508 
    sqlx::query(
509 
        "INSERT INTO flow_declines (job_id, runner_id, runner, profile, reason, ms) VALUES (?, ?, ?, ?, ?, ?) \
510 
         ON CONFLICT (job_id, runner_id) DO UPDATE SET runner = excluded.runner, profile = excluded.profile, reason = excluded.reason, ms = excluded.ms",
511 
    )
512 
    .bind(job_id)
513 
    .bind(runner_id)
514 
    .bind(runner)
515 
    .bind(profile)
516 
    .bind(reason)
517 
    .bind(now_ms())
518 
    .execute(&mut *tx)
519 
    .await?;
520 
    let all: Vec<DeclineRow> =
521 
        sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = ? ORDER BY ms")
522 
            .bind(job_id)
523 
            .fetch_all(&mut *tx)
524 
            .await?;
525 
    tx.commit().await?;
526 
    Ok(Some(all))
527 
}
528
529 
/// The declines of job `job_id`, oldest first.
530 
pub async fn declines_of_job(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<DeclineRow>> {
531 
    sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = ? ORDER BY ms")
532 
        .bind(job_id)
533 
        .fetch_all(pool)
534 
        .await
535 
}
536
537 
/// A runner's latest declines, newest first, with where the job is.
538 
#[derive(Clone, Debug, sqlx::FromRow)]
539 
pub struct RunnerDecline {
540 
    pub job_id: i64,
541 
    pub job: String,
542 
    pub run_number: i64,
543 
    pub owner: String,
544 
    pub name: String,
545 
    pub reason: String,
546 
    pub ms: i64,
547 
}
548
549 
pub async fn declines_of_runner(pool: &Pool, runner_id: i64, limit: i64) -> sqlx::Result<Vec<RunnerDecline>> {
550 
    sqlx::query_as(
551 
        "SELECT d.job_id, j.name AS job, r.number AS run_number, rep.owner, rep.name, d.reason, d.ms \
552 
         FROM flow_declines d JOIN flow_jobs j ON j.id = d.job_id JOIN flow_runs r ON r.id = j.run_id \
553 
         JOIN repositories rep ON rep.id = r.repo_id WHERE d.runner_id = ? ORDER BY d.ms DESC LIMIT ?",
554 
    )
555 
    .bind(runner_id)
556 
    .bind(limit)
557 
    .fetch_all(pool)
558 
    .await
559 
}
560
561 
/// Queued jobs whose dependencies have passed, in their live runs, with
562 
/// their repositories (owner, id): the ones a runner could take now.
563 
pub async fn ready_queued(pool: &Pool) -> sqlx::Result<Vec<(Job, String, i64)>> {
564 
    let queued: Vec<Job> = sqlx::query_as(
565 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
566 
         WHERE j.status = 'queued' AND r.status IN ('queued', 'running') ORDER BY j.run_id, j.depth, j.id LIMIT 500",
567 
    )
568 
    .fetch_all(pool)
569 
    .await?;
570 
    let mut out = Vec::new();
571 
    for j in queued {
572 
        let mut ready = true;
573 
        for d in j.depends_on() {
574 
            let st: Option<String> = sqlx::query_scalar("SELECT status FROM flow_jobs WHERE run_id = ? AND name = ?")
575 
                .bind(j.run_id)
576 
                .bind(&d)
577 
                .fetch_optional(pool)
578 
                .await?;
579 
            if st.as_deref() != Some("passed") {
580 
                ready = false;
581 
                break;
582 
            }
583 
        }
584 
        if !ready {
585 
            continue;
586 
        }
587 
        let repo: Option<(String, i64)> =
588 
            sqlx::query_as("SELECT rep.owner, rep.id FROM flow_runs r JOIN repositories rep ON rep.id = r.repo_id WHERE r.id = ?")
589 
                .bind(j.run_id)
590 
                .fetch_optional(pool)
591 
                .await?;
592 
        if let Some((owner, id)) = repo {
593 
            out.push((j, owner, id));
594 
        }
595 
    }
596 
    Ok(out)
597 
}
598
599 
pub async fn set_waiting(pool: &Pool, job_id: i64, waiting: &str) -> sqlx::Result<()> {
600 
    sqlx::query("UPDATE flow_jobs SET waiting = ? WHERE id = ? AND waiting <> ?")
601 
        .bind(waiting)
602 
        .bind(job_id)
603 
        .bind(waiting)
604 
        .execute(pool)
605 
        .await?;
606 
    Ok(())
607 
}
608
609 
/// Append records; returns how many were kept (the rest exceed the cap).
610 
pub async fn append_log(pool: &Pool, job_id: i64, records: &[flow_core::api::Record], cap: i64) -> sqlx::Result<usize> {
611 
    let mut tx = super::begin_write(pool).await?;
612 
    let count: i64 = sqlx::query_scalar("SELECT log_count FROM flow_jobs WHERE id = ?").bind(job_id).fetch_one(&mut *tx).await?;
613 
    let room = (cap - count).max(0) as usize;
614 
    let keep = &records[..records.len().min(room)];
615 
    for (i, r) in keep.iter().enumerate() {
616 
        sqlx::query("INSERT INTO flow_logs (job_id, seq, t, stream, text) VALUES (?, ?, ?, ?, ?)")
617 
            .bind(job_id)
618 
            .bind(count + i as i64 + 1)
619 
            .bind(r.t as i64)
620 
            .bind(&r.stream)
621 
            .bind(&r.text)
622 
            .execute(&mut *tx)
623 
            .await?;
624 
    }
625 
    sqlx::query("UPDATE flow_jobs SET log_count = log_count + ? WHERE id = ?")
626 
        .bind(keep.len() as i64)
627 
        .bind(job_id)
628 
        .execute(&mut *tx)
629 
        .await?;
630 
    tx.commit().await?;
631 
    Ok(keep.len())
632 
}
633
634 
pub async fn log(pool: &Pool, job_id: i64, after: i64, limit: i64) -> sqlx::Result<Vec<LogLine>> {
635 
    sqlx::query_as("SELECT seq, t, stream, text FROM flow_logs WHERE job_id = ? AND seq > ? ORDER BY seq LIMIT ?")
636 
        .bind(job_id)
637 
        .bind(after)
638 
        .bind(limit)
639 
        .fetch_all(pool)
640 
        .await
641 
}
642
643 
/// Jobs running since before `ms` (the oldest first).
644 
/// Jobs still running at `now` past their own time limit (`default_s` for
645 
/// one claimed without) plus `grace_ms`.
646 
pub async fn running_past_limit(pool: &Pool, now: i64, default_s: u64, grace_ms: i64) -> sqlx::Result<Vec<Job>> {
647 
    sqlx::query_as(
648 
        "SELECT * FROM flow_jobs WHERE status = 'running' \
649 
         AND COALESCE(started_ms, 0) + (CASE WHEN timeout_s > 0 THEN timeout_s ELSE ? END) * 1000 + ? < ? ORDER BY started_ms",
650 
    )
651 
    .bind(default_s as i64)
652 
    .bind(grace_ms)
653 
    .bind(now)
654 
    .fetch_all(pool)
655 
    .await
656 
}
657
658 
/// Record the limit a claimed job runs under.
659 
pub async fn set_job_timeout(pool: &Pool, job_id: i64, timeout_s: u64) -> sqlx::Result<()> {
660 
    sqlx::query("UPDATE flow_jobs SET timeout_s = ? WHERE id = ?").bind(timeout_s as i64).bind(job_id).execute(pool).await?;
661 
    Ok(())
662 
}
663
664 
pub async fn running_since_before(pool: &Pool, ms: i64) -> sqlx::Result<Vec<Job>> {
665 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND COALESCE(started_ms, 0) < ? ORDER BY started_ms")
666 
        .bind(ms)
667 
        .fetch_all(pool)
668 
        .await
669 
}
670
671 
/// The jobs running on the runner with this id.
672 
pub async fn running_on(pool: &Pool, runner_id: i64) -> sqlx::Result<Vec<Job>> {
673 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND runner_id = ? ORDER BY id").bind(runner_id).fetch_all(pool).await
674 
}
675
676 
pub struct Outcome<'a> {
677 
    pub status: &'a str,
678 
    pub exit_code: Option<i64>,
679 
    pub error: &'a str,
680 
    pub receipt: &'a str,
681 
    pub receipt_hash: &'a str,
682 
}
683
684 
/// Record a job's end, skip the jobs that can no longer run, and settle the
685 
/// run. Returns the jobs whose status changed (for their checks).
686 
pub async fn finish(pool: &Pool, job_id: i64, o: Outcome<'_>) -> sqlx::Result<Vec<Job>> {
687 
    let mut tx = super::begin_write(pool).await?;
688 
    let now = now_ms();
689 
    let j: Job = sqlx::query_as("SELECT * FROM flow_jobs WHERE id = ?").bind(job_id).fetch_one(&mut *tx).await?;
690 
    sqlx::query("UPDATE flow_jobs SET status = ?, exit_code = ?, error = ?, receipt = ?, receipt_hash = ?, finished_ms = ?, token_hash = '' WHERE id = ?")
691 
        .bind(o.status)
692 
        .bind(o.exit_code)
693 
        .bind(o.error)
694 
        .bind(o.receipt)
695 
        .bind(o.receipt_hash)
696 
        .bind(now)
697 
        .bind(job_id)
698 
        .execute(&mut *tx)
699 
        .await?;
700 
    let mut changed = vec![job_id];
701 
    // Dependents of anything that did not pass are skipped, transitively.
702 
    loop {
703 
        let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = ?").bind(j.run_id).fetch_all(&mut *tx).await?;
704 
        let blocked: Vec<i64> = all
705 
            .iter()
706 
            .filter(|x| x.status == "queued")
707 
            .filter(|x| {
708 
                x.depends_on()
709 
                    .iter()
710 
                    .any(|d| all.iter().any(|y| &y.name == d && matches!(y.status.as_str(), "failed" | "skipped" | "error")))
711 
            })
712 
            .map(|x| x.id)
713 
            .collect();
714 
        if blocked.is_empty() {
715 
            break;
716 
        }
717 
        for id in blocked {
718 
            sqlx::query("UPDATE flow_jobs SET status = 'skipped', finished_ms = ? WHERE id = ?")
719 
                .bind(now)
720 
                .bind(id)
721 
                .execute(&mut *tx)
722 
                .await?;
723 
            changed.push(id);
724 
        }
725 
    }
726 
    let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = ?").bind(j.run_id).fetch_all(&mut *tx).await?;
727 
    if all.iter().all(Job::done) {
728 
        let status = if all.iter().all(|x| x.status == "passed") { "passed" } else { "failed" };
729 
        sqlx::query("UPDATE flow_runs SET status = ?, finished_ms = ? WHERE id = ?")
730 
            .bind(status)
731 
            .bind(now)
732 
            .bind(j.run_id)
733 
            .execute(&mut *tx)
734 
            .await?;
735 
    }
736 
    tx.commit().await?;
737 
    let mut out = Vec::new();
738 
    for id in changed {
739 
        if let Some(x) = job(pool, id).await? {
740 
            out.push(x);
741 
        }
742 
    }
743 
    Ok(out)
744 
}
745
746 
/// Cancel a run (oxygenflow#32): its jobs not yet done end as cancelled,
747 
/// and so does the run. A runner learns of it when the hub next refuses
748 
/// its log lines, and stops the job. Returns the jobs whose status changed.
749 
pub async fn cancel_run(pool: &Pool, run_id: i64, why: &str) -> sqlx::Result<Vec<Job>> {
750 
    let now = now_ms();
751 
    let mut tx = pool.begin().await?;
752 
    let ids: Vec<(i64,)> = sqlx::query_as("SELECT id FROM flow_jobs WHERE run_id = ? AND status IN ('queued', 'running')")
753 
        .bind(run_id)
754 
        .fetch_all(&mut *tx)
755 
        .await?;
756 
    sqlx::query(
757 
        "UPDATE flow_jobs SET status = 'cancelled', error = ?, finished_ms = ?, token_hash = ''
758 
         WHERE run_id = ? AND status IN ('queued', 'running')",
759 
    )
760 
    .bind(why)
761 
    .bind(now)
762 
    .bind(run_id)
763 
    .execute(&mut *tx)
764 
    .await?;
765 
    sqlx::query("UPDATE flow_runs SET status = 'cancelled', finished_ms = ? WHERE id = ? AND status IN ('queued', 'running')")
766 
        .bind(now)
767 
        .bind(run_id)
768 
        .execute(&mut *tx)
769 
        .await?;
770 
    tx.commit().await?;
771 
    let mut out = Vec::new();
772 
    for (id,) in ids {
773 
        if let Some(x) = job(pool, id).await? {
774 
            out.push(x);
775 
        }
776 
    }
777 
    Ok(out)
778 
}
779
780 
/// The push runs of `bookmark` still going, older than run `number`.
781 
pub async fn live_push_runs_before(pool: &Pool, repo_id: i64, bookmark: &str, number: i64) -> sqlx::Result<Vec<Run>> {
782 
    sqlx::query_as(
783 
        "SELECT * FROM flow_runs WHERE repo_id = ? AND bookmark = ? AND number < ? AND trigger = 'push'
784 
         AND status IN ('queued', 'running') ORDER BY id",
785 
    )
786 
    .bind(repo_id)
787 
    .bind(bookmark)
788 
    .bind(number)
789 
    .fetch_all(pool)
790 
    .await
791 
}
792
793 
/// The newest run of exactly this state (heads joined with `+`, sorted).
794 
pub async fn latest_for_heads(pool: &Pool, repo_id: i64, heads: &str) -> sqlx::Result<Option<Run>> {
795 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = ? AND heads = ? ORDER BY id DESC LIMIT 1")
796 
        .bind(repo_id)
797 
        .bind(heads)
798 
        .fetch_optional(pool)
799 
        .await
800 
}
801
802 
/// Record what became of job `job_id`'s signed receipt.
803 
pub async fn set_receipt(pool: &Pool, job_id: i64, attest: &str, signer: &str, note: &str) -> sqlx::Result<()> {
804 
    sqlx::query("UPDATE flow_jobs SET receipt_attest = ?, receipt_signer = ?, receipt_note = ? WHERE id = ?")
805 
        .bind(attest)
806 
        .bind(signer)
807 
        .bind(note)
808 
        .bind(job_id)
809 
        .execute(pool)
810 
        .await?;
811 
    Ok(())
812 
}
813
814 
/// A job's artifact, kept as a blob of the run's repository.
815 
#[derive(Clone, Debug, FromRow)]
816 
pub struct Artifact {
817 
    pub job_id: i64,
818 
    pub name: String,
819 
    pub path: String,
820 
    pub blob: String,
821 
    pub size: i64,
822 
    /// BLAKE3 of the bytes.
823 
    pub content: String,
824 
    pub created_ms: i64,
825 
}
826
827 
pub async fn add_artifact(pool: &Pool, a: &Artifact) -> sqlx::Result<()> {
828 
    sqlx::query("INSERT OR REPLACE INTO flow_artifacts (job_id, name, path, blob, size, content, created_ms) VALUES (?, ?, ?, ?, ?, ?, ?)")
829 
        .bind(a.job_id)
830 
        .bind(&a.name)
831 
        .bind(&a.path)
832 
        .bind(&a.blob)
833 
        .bind(a.size)
834 
        .bind(&a.content)
835 
        .bind(a.created_ms)
836 
        .execute(pool)
837 
        .await?;
838 
    Ok(())
839 
}
840
841 
/// Job `job_id`'s artifacts, by name.
842 
pub async fn artifacts(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<Artifact>> {
843 
    sqlx::query_as("SELECT * FROM flow_artifacts WHERE job_id = ? ORDER BY name").bind(job_id).fetch_all(pool).await
844 
}
845
846 
/// Where an artifact with given content came from: the run and job that
847 
/// built it, and the job's signed receipt when one was kept.
848 
#[derive(Clone, Debug, FromRow)]
849 
pub struct Provenance {
850 
    pub run_number: i64,
851 
    pub job: String,
852 
    pub name: String,
853 
    pub receipt_attest: String,
854 
    pub receipt_signer: String,
855 
}
856
857 
/// The build of the bytes whose BLAKE3 is `content` in repository
858 
/// `repo_id`: a passed job, one with a kept signed receipt first, newest
859 
/// run first.
860 
pub async fn provenance(pool: &Pool, repo_id: i64, content: &str) -> sqlx::Result<Option<Provenance>> {
861 
    sqlx::query_as(
862 
        "SELECT r.number AS run_number, j.name AS job, a.name AS name, j.receipt_attest, j.receipt_signer \
863 
         FROM flow_artifacts a JOIN flow_jobs j ON j.id = a.job_id JOIN flow_runs r ON r.id = j.run_id \
864 
         WHERE r.repo_id = ? AND a.content = ? AND j.status = 'passed' \
865 
         ORDER BY (j.receipt_attest != '') DESC, r.number DESC LIMIT 1",
866 
    )
867 
    .bind(repo_id)
868 
    .bind(content)
869 
    .fetch_optional(pool)
870 
    .await
871 
}