OXYGENHUB
# 0ee5c3fb32eb oxygenhub / crates / hub / src / db / flow.rs
46.8 KiB · Rust · last changed Newest run wins: a proposal's new run supersedes its older ones 17 days 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`, `proposal`, `tag:NAME`, `manual`, `schedule` or
14 
    /// `rerun:N:EVENT` (migration 0020).
15 
    #[sqlx(default)]
16 
    pub trigger: String,
17 
    pub bookmark: String,
18 
    pub state: String,
19 
    pub heads: String,
20 
    pub pipeline_hash: String,
21 
    pub status: String,
22 
    pub error: String,
23 
    pub actor: String,
24 
    pub created_ms: i64,
25 
    pub finished_ms: Option<i64>,
26 
    /// `urgent` when expedited (oxygenhub#244), else empty: the run follows
27 
    /// its repository's, organisation's or the site's level.
28 
    #[sqlx(default)]
29 
    pub priority: String,
30 
    /// Who expedited it and why.
31 
    #[sqlx(default)]
32 
    pub priority_note: String,
33 
}
34
35 
/// The priority levels, most urgent first (oxygenhub#244).
36 
pub const PRIORITIES: [&str; 4] = ["urgent", "high", "normal", "low"];
37
38 
/// SQL for the level a run's jobs get from where they are, as a rank
39 
/// (0 urgent … 3 low): the run's own (expedited), else its repository's,
40 
/// its organisation's, the site's (`settings.flow_priority`), else normal.
41 
macro_rules! level_rank {
42 
    () => {
43 
        "CASE COALESCE(NULLIF(r.priority, ''), NULLIF(rep.flow_priority, ''), \
44 
           NULLIF((SELECT o.flow_priority FROM organisations o WHERE o.name = rep.owner), ''), \
45 
           NULLIF((SELECT s.value FROM settings s WHERE s.key = 'flow_priority'), ''), 'normal') \
46 
         WHEN 'urgent' THEN 0 WHEN 'high' THEN 1 WHEN 'low' THEN 3 ELSE 2 END"
47 
    };
48 
}
49
50 
/// SQL for a queued job's level as a rank (0 urgent … 3 low): the level
51 
/// its run gets from where it is, lowered by the job's own priority from
52 
/// its pipeline (`low`, `normal`), never raised by it; an expedited run is
53 
/// urgent whatever its jobs say (oxygenhub#244). Needs `j` (the job), `r`
54 
/// (its run) and `rep` (its repository) in the query.
55 
pub const PRIORITY_RANK: &str = concat!(
56 
    "CASE WHEN r.priority <> '' THEN ",
57 
    level_rank!(),
58 
    " WHEN j.priority = 'low' THEN 3 WHEN j.priority = 'normal' AND (",
59 
    level_rank!(),
60 
    ") < 2 THEN 2 ELSE ",
61 
    level_rank!(),
62 
    " END"
63 
);
64
65 
/// SQL for why a job stands where it does, beside [`PRIORITY_RANK`]: an
66 
/// expedited run's note, else the pipeline's word for a job it lowered.
67 
pub const PRIORITY_NOTE: &str =
68 
    "CASE WHEN r.priority <> '' THEN r.priority_note WHEN j.priority <> '' THEN 'pipeline: ' || j.priority ELSE r.priority_note END";
69
70 
/// Raise run `run_id` to urgent, saying who and why; a run already ended
71 
/// is left alone. Whether it changed.
72 
pub async fn expedite(pool: &Pool, run_id: i64, note: &str) -> sqlx::Result<bool> {
73 
    let r = sqlx::query("UPDATE flow_runs SET priority = 'urgent', priority_note = $1 WHERE id = $2 AND status IN ('queued', 'running')")
74 
        .bind(note)
75 
        .bind(run_id)
76 
        .execute(pool)
77 
        .await?;
78 
    Ok(r.rows_affected() > 0)
79 
}
80
81 
impl Run {
82 
    /// How the run page says what started it.
83 
    pub fn started_by(&self) -> String {
84 
        match self.trigger.as_str() {
85 
            "manual" => format!("started by {}", self.actor),
86 
            "schedule" => "started on schedule".to_string(),
87 
            t if t.starts_with("tag:") => format!("tag {} pushed by {}", &t[4..], self.actor),
88 
            t if t.starts_with("rerun:") => format!("re-run of #{} by {}", self.rerun_of().unwrap_or(""), self.actor),
89 
            _ => format!("pushed by {}", self.actor),
90 
        }
91 
    }
92
93 
    /// The run a rerun repeats: `N` of `rerun:N` or `rerun:N:EVENT`.
94 
    pub fn rerun_of(&self) -> Option<&str> {
95 
        let rest = self.trigger.strip_prefix("rerun:")?;
96 
        Some(rest.split(':').next().unwrap_or(rest))
97 
    }
98
99 
    /// The trigger as people read it: `rerun:N`, without the event it repeats.
100 
    pub fn trigger_label(&self) -> String {
101 
        match self.rerun_of() {
102 
            Some(n) => format!("rerun:{n}"),
103 
            None => self.trigger.clone(),
104 
        }
105 
    }
106
107 
    pub fn head_list(&self) -> Vec<&str> {
108 
        self.heads.split('+').filter(|h| !h.is_empty()).collect()
109 
    }
110
111 
    pub fn live(&self) -> bool {
112 
        matches!(self.status.as_str(), "queued" | "running")
113 
    }
114 
}
115
116 
#[derive(Clone, Debug, Default, FromRow)]
117 
pub struct Job {
118 
    pub id: i64,
119 
    pub run_id: i64,
120 
    pub name: String,
121 
    pub depth: i64,
122 
    pub depends: String,
123 
    pub spec: String,
124 
    pub sandbox: String,
125 
    pub runtime: String,
126 
    pub status: String,
127 
    pub runner: String,
128 
    pub token_hash: String,
129 
    pub started_ms: Option<i64>,
130 
    pub finished_ms: Option<i64>,
131 
    pub exit_code: Option<i64>,
132 
    pub error: String,
133 
    pub receipt: String,
134 
    pub receipt_hash: String,
135 
    pub log_count: i64,
136 
    /// Comma-separated ids of other repositories the job token may read.
137 
    pub also_read: String,
138 
    /// The receipt attestation kept in the repository (Oxygen spec §14.6).
139 
    pub receipt_attest: String,
140 
    /// Who signed it: the runner's identity name.
141 
    pub receipt_signer: String,
142 
    /// Why a signed receipt was not kept.
143 
    pub receipt_note: String,
144 
    /// Why a queued job no runner can take waits (spec §11.4), from the
145 
    /// last placement pass; empty when it is only waiting its turn.
146 
    pub waiting: String,
147 
    /// What it took, as its runner reported (oxygenflow#41), in MiB.
148 
    pub peak_disk_mib: Option<i64>,
149 
    pub peak_memory_mib: Option<i64>,
150 
    /// The free disk it needs, learned from past runs of the same job.
151 
    pub need_disk_mib: Option<i64>,
152 
    /// Why it ran on its runner (oxygenflow#42).
153 
    pub placed: String,
154 
    /// `waiting` for a person, `approved`, or empty (oxygenflow#57).
155 
    pub approval: String,
156 
    /// Who approved or rejected it, and when.
157 
    pub approved_by: String,
158 
    pub approved_ms: Option<i64>,
159 
    /// The agent token it was decided through, if one (oxygenhub#229).
160 
    #[sqlx(default)]
161 
    pub approved_via: String,
162 
    /// The environment it deploys to, or empty (oxygenflow#59).
163 
    pub environment: String,
164 
    /// The approval attestation that released it, when sealed (#235).
165 
    pub approval_seal: String,
166 
}
167
168 
impl Job {
169 
    pub fn depends_on(&self) -> Vec<String> {
170 
        serde_json::from_str(&self.depends).unwrap_or_default()
171 
    }
172
173 
    pub fn done(&self) -> bool {
174 
        matches!(self.status.as_str(), "passed" | "failed" | "skipped" | "error" | "cancelled")
175 
    }
176 
}
177
178 
#[derive(Clone, Debug, FromRow)]
179 
pub struct LogLine {
180 
    pub seq: i64,
181 
    pub t: i64,
182 
    pub stream: String,
183 
    pub text: String,
184 
}
185
186 
/// A job as the trigger creates it.
187 
pub struct NewJob {
188 
    pub name: String,
189 
    pub depth: i64,
190 
    pub depends: Vec<String>,
191 
    pub spec: String,
192 
    pub sandbox: String,
193 
    pub runtime: String,
194 
    /// The free disk it needs, from [`learned_disk_need`].
195 
    pub need_disk_mib: Option<i64>,
196 
}
197
198 
/// Runs of a job whose peaks set its next need.
199 
const NEED_FROM_RUNS: i64 = 5;
200
201 
/// The free disk a new run of job `name` in repository `repo_id` needs
202 
/// (oxygenflow#41): the most its last runs took, plus a quarter; `None`
203 
/// for a job that has not run here yet.
204 
pub async fn learned_disk_need(pool: &Pool, repo_id: i64, name: &str) -> sqlx::Result<Option<i64>> {
205 
    let peak: Option<i64> = sqlx::query_scalar(
206 
        "SELECT MAX(peak_disk_mib) FROM (SELECT j.peak_disk_mib FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
207 
         WHERE r.repo_id = $1 AND j.name = $2 AND j.peak_disk_mib IS NOT NULL ORDER BY j.id DESC LIMIT $3)",
208 
    )
209 
    .bind(repo_id)
210 
    .bind(name)
211 
    .bind(NEED_FROM_RUNS)
212 
    .fetch_one(pool)
213 
    .await?;
214 
    Ok(peak.map(|p| p + p / 4 + 64))
215 
}
216
217 
/// Record why job `job_id` ran where it did.
218 
pub async fn set_placed(pool: &Pool, job_id: i64, placed: &str) -> sqlx::Result<()> {
219 
    sqlx::query("UPDATE flow_jobs SET placed = $1 WHERE id = $2").bind(placed).bind(job_id).execute(pool).await?;
220 
    Ok(())
221 
}
222
223 
/// Keep what a job took, as its runner reported.
224 
pub async fn set_peaks(pool: &Pool, job_id: i64, disk_mib: Option<u64>, memory_mib: Option<u64>) -> sqlx::Result<()> {
225 
    sqlx::query("UPDATE flow_jobs SET peak_disk_mib = $1, peak_memory_mib = $2 WHERE id = $3")
226 
        .bind(disk_mib.map(|v| v.min(i64::MAX as u64) as i64))
227 
        .bind(memory_mib.map(|v| v.min(i64::MAX as u64) as i64))
228 
        .bind(job_id)
229 
        .execute(pool)
230 
        .await?;
231 
    Ok(())
232 
}
233
234 
pub struct NewRun<'a> {
235 
    pub repo_id: i64,
236 
    pub bookmark: &'a str,
237 
    pub state: &'a str,
238 
    pub heads: &'a str,
239 
    pub pipeline_hash: &'a str,
240 
    pub actor: &'a str,
241 
    /// `push`, `proposal`, `tag:NAME`, `manual`, `schedule` or
242 
    /// `rerun:N:EVENT`.
243 
    pub trigger: &'a str,
244 
    /// Problems when the pipeline is invalid: the run is recorded in error.
245 
    pub error: Option<&'a str>,
246 
}
247
248 
pub async fn create_run(pool: &Pool, r: NewRun<'_>, jobs: &[NewJob]) -> sqlx::Result<i64> {
249 
    let mut tx = super::begin_write(pool).await?;
250 
    let now = now_ms();
251 
    let status = if r.error.is_some() { "error" } else { "queued" };
252 
    let id: i64 = sqlx::query_scalar(
253 
        "INSERT INTO flow_runs (repo_id, number, bookmark, state, heads, pipeline_hash, status, error, actor, trigger, created_ms, finished_ms) \
254 
         VALUES ($1, (SELECT COALESCE(MAX(number), 0) + 1 FROM flow_runs WHERE repo_id = $2), $3, $4, $5, $6, $7, $8, $9, $10, $11, $12) RETURNING id",
255 
    )
256 
    .bind(r.repo_id)
257 
    .bind(r.repo_id)
258 
    .bind(r.bookmark)
259 
    .bind(r.state)
260 
    .bind(r.heads)
261 
    .bind(r.pipeline_hash)
262 
    .bind(status)
263 
    .bind(r.error.unwrap_or(""))
264 
    .bind(r.actor)
265 
    .bind(r.trigger)
266 
    .bind(now)
267 
    .bind(r.error.map(|_| now))
268 
    .fetch_one(&mut *tx)
269 
    .await?;
270 
    for j in jobs {
271 
        sqlx::query(
272 
            "INSERT INTO flow_jobs (run_id, name, depth, depends, spec, sandbox, runtime, status, need_disk_mib, approval, environment, priority) \
273 
             VALUES ($1, $2, $3, $4, $5, $6, $7, 'queued', $8, $9, $10, $11)",
274 
        )
275 
        .bind(id)
276 
        .bind(&j.name)
277 
        .bind(j.depth)
278 
        .bind(serde_json::to_string(&j.depends).unwrap_or_else(|_| "[]".into()))
279 
        .bind(&j.spec)
280 
        .bind(&j.sandbox)
281 
        .bind(&j.runtime)
282 
        .bind(j.need_disk_mib)
283 
        .bind(if needs_approval(&j.spec) { "waiting" } else { "" })
284 
        .bind(environment_of(&j.spec))
285 
        .bind(priority_of(&j.spec))
286 
        .execute(&mut *tx)
287 
        .await?;
288 
    }
289 
    tx.commit().await?;
290 
    Ok(id)
291 
}
292
293 
/// The environment a compiled job deploys to, or empty (oxygenflow#59).
294 
fn environment_of(spec: &str) -> String {
295 
    serde_json::from_str::<serde_json::Value>(spec).ok().and_then(|v| v["environment"].as_str().map(String::from)).unwrap_or_default()
296 
}
297
298 
/// A compiled job's own priority (`low`, `normal`), or empty; anything
299 
/// else is not one a pipeline may give (oxygenhub#244).
300 
fn priority_of(spec: &str) -> String {
301 
    serde_json::from_str::<serde_json::Value>(spec)
302 
        .ok()
303 
        .and_then(|v| v["priority"].as_str().map(String::from))
304 
        .filter(|p| p == "low" || p == "normal")
305 
        .unwrap_or_default()
306 
}
307
308 
/// Whether a compiled job asks for approval (oxygenflow#57).
309 
fn needs_approval(spec: &str) -> bool {
310 
    serde_json::from_str::<serde_json::Value>(spec).is_ok_and(|v| !v["approval"].is_null())
311 
}
312
313 
/// The labels a compiled job and its sandbox ask a runner for (`runs-on`).
314 
fn labels_in(spec: &str, sandbox: &str) -> Vec<String> {
315 
    let mut out: Vec<String> = [spec, sandbox]
316 
        .iter()
317 
        .filter_map(|j| serde_json::from_str::<serde_json::Value>(j).ok())
318 
        .flat_map(|v| v["runs_on"].as_array().cloned().unwrap_or_default())
319 
        .filter_map(|l| l.as_str().map(String::from))
320 
        .collect();
321 
    out.sort();
322 
    out.dedup();
323 
    out
324 
}
325
326 
impl Job {
327 
    /// The runner labels this job needs.
328 
    pub fn needs(&self) -> Vec<String> {
329 
        labels_in(&self.spec, &self.sandbox)
330 
    }
331
332 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
333 
    pub fn arch(&self) -> Option<String> {
334 
        arch_in(&self.sandbox)
335 
    }
336 
}
337
338 
/// The architecture a sandbox's JSON asks for, if any.
339 
fn arch_in(sandbox: &str) -> Option<String> {
340 
    serde_json::from_str::<serde_json::Value>(sandbox).ok()?["arch"].as_str().map(String::from)
341 
}
342
343 
/// A job waiting for or held by a runner, with where it comes from.
344 
#[derive(Clone, Debug, FromRow)]
345 
pub struct QueueEntry {
346 
    pub job_id: i64,
347 
    pub run_id: i64,
348 
    pub repo_id: i64,
349 
    pub owner: String,
350 
    pub name: String,
351 
    pub visibility: String,
352 
    pub run_number: i64,
353 
    pub job: String,
354 
    pub status: String,
355 
    /// The runner holding it, when running.
356 
    pub runner: String,
357 
    pub spec: String,
358 
    pub sandbox: String,
359 
    /// When it started running.
360 
    pub started_ms: Option<i64>,
361 
    /// Its level (0 urgent … 3 low, oxygenhub#244) and, when expedited, who
362 
    /// did it and why.
363 
    pub priority_rank: i64,
364 
    pub priority_note: String,
365 
}
366
367 
impl QueueEntry {
368 
    pub fn priority(&self) -> &'static str {
369 
        PRIORITIES.get(self.priority_rank as usize).copied().unwrap_or("normal")
370 
    }
371 
}
372
373 
impl QueueEntry {
374 
    /// The runner labels the job needs.
375 
    pub fn needs(&self) -> Vec<String> {
376 
        labels_in(&self.spec, &self.sandbox)
377 
    }
378
379 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
380 
    pub fn arch(&self) -> Option<String> {
381 
        arch_in(&self.sandbox)
382 
    }
383 
}
384
385 
/// Every job of a live run that is running or queued: running ones first,
386 
/// then queued ones in the order runners claim them (hub spec §13.2).
387 
pub async fn queue(pool: &Pool) -> sqlx::Result<Vec<QueueEntry>> {
388 
    sqlx::query_as(&format!(
389 
        "SELECT j.id AS job_id, j.run_id, r.repo_id, rep.owner, rep.name, rep.visibility, r.number AS run_number, j.name AS job, j.status, j.runner, j.spec, j.sandbox, j.started_ms, \
390 
           CAST({PRIORITY_RANK} AS BIGINT) AS priority_rank, {PRIORITY_NOTE} AS priority_note \
391 
         FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories rep ON rep.id = r.repo_id \
392 
         WHERE j.status IN ('queued', 'running') AND j.approval != 'waiting' AND r.status IN ('queued', 'running') \
393 
         ORDER BY CASE j.status WHEN 'running' THEN 0 ELSE 1 END, \
394 
           CASE j.status WHEN 'running' THEN 0 ELSE {PRIORITY_RANK} END, \
395 
           CASE j.status WHEN 'running' THEN 0 ELSE ROW_NUMBER() OVER (PARTITION BY r.repo_id, j.status ORDER BY j.run_id, j.depth, j.id) \
396 
           + (SELECT COUNT(*) FROM flow_jobs x JOIN flow_runs xr ON xr.id = x.run_id WHERE x.status = 'running' AND xr.repo_id = r.repo_id) END, \
397 
           j.run_id, j.depth, j.id"
398 
    ))
399 
    .fetch_all(pool)
400 
    .await
401 
}
402
403 
/// A repository's runs, newest first, numbered below `before`.
404 
pub async fn runs_by_number(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
405 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND number < $2 ORDER BY number DESC LIMIT $3")
406 
        .bind(repo_id)
407 
        .bind(before.unwrap_or(i64::MAX))
408 
        .bind(limit)
409 
        .fetch_all(pool)
410 
        .await
411 
}
412
413 
/// The latest run on this bookmark of this repository, for what a push
414 
/// changed since (a job's scope).
415 
/// The run a push to `bookmark` is scoped against: the newest that was not
416 
/// cancelled, nor a push run still going (a new push supersedes it), so
417 
/// that what a cancelled run would have checked is checked by the next.
418 
pub async fn latest_on(pool: &Pool, repo_id: i64, bookmark: &str) -> sqlx::Result<Option<Run>> {
419 
    sqlx::query_as(
420 
        "SELECT * FROM flow_runs WHERE repo_id = $1 AND bookmark = $2 AND heads != '' AND status != 'cancelled'
421 
         AND NOT (trigger = 'push' AND status IN ('queued', 'running')) ORDER BY id DESC LIMIT 1",
422 
    )
423 
    .bind(repo_id)
424 
    .bind(bookmark)
425 
    .fetch_optional(pool)
426 
    .await
427 
}
428
429 
pub async fn runs(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
430 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND id < $2 ORDER BY id DESC LIMIT $3")
431 
        .bind(repo_id)
432 
        .bind(before.unwrap_or(i64::MAX))
433 
        .bind(limit)
434 
        .fetch_all(pool)
435 
        .await
436 
}
437
438 
/// A repository's run by its number there.
439 
pub async fn run(pool: &Pool, repo_id: i64, number: i64) -> sqlx::Result<Option<Run>> {
440 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND number = $2").bind(repo_id).bind(number).fetch_optional(pool).await
441 
}
442
443 
pub async fn run_by_id(pool: &Pool, id: i64) -> sqlx::Result<Option<Run>> {
444 
    sqlx::query_as("SELECT * FROM flow_runs WHERE id = $1").bind(id).fetch_optional(pool).await
445 
}
446
447 
pub async fn jobs(pool: &Pool, run_id: i64) -> sqlx::Result<Vec<Job>> {
448 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1 ORDER BY depth, name").bind(run_id).fetch_all(pool).await
449 
}
450
451 
pub async fn job(pool: &Pool, id: i64) -> sqlx::Result<Option<Job>> {
452 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE id = $1").bind(id).fetch_optional(pool).await
453 
}
454
455 
/// The running job a job token belongs to, while it may still be used.
456 
pub async fn job_by_token(pool: &Pool, token_hash: &str, max_age_ms: i64) -> sqlx::Result<Option<Job>> {
457 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE token_hash = $1 AND status = 'running' AND started_ms >= $2")
458 
        .bind(token_hash)
459 
        .bind(now_ms() - max_age_ms)
460 
        .fetch_optional(pool)
461 
        .await
462 
}
463
464 
/// Take the oldest queued job whose dependencies all passed and whose tier
465 
/// the runner offers, and mark it running for `runner` with a token.
466 
/// Let a job's token also read these repositories (its pinned subrepos).
467 
pub async fn set_also_read(pool: &Pool, job_id: i64, repo_ids: &[i64]) -> sqlx::Result<()> {
468 
    let list: Vec<String> = repo_ids.iter().map(|i| i.to_string()).collect();
469 
    sqlx::query("UPDATE flow_jobs SET also_read = $1 WHERE id = $2").bind(list.join(",")).bind(job_id).execute(pool).await?;
470 
    Ok(())
471 
}
472
473 
impl Job {
474 
    pub fn also_read_ids(&self) -> Vec<i64> {
475 
        self.also_read.split(',').filter_map(|x| x.trim().parse().ok()).collect()
476 
    }
477 
}
478
479 
/// What a runner can take on: free memory (MiB) and vCPUs, `None` for no
480 
/// limit, and the rootfs images it holds, `None` when it does not say.
481 
#[derive(Clone, Debug, Default)]
482 
pub struct Room {
483 
    pub memory_mib: Option<u32>,
484 
    pub cpus: Option<u32>,
485 
    pub images: Option<std::collections::HashSet<String>>,
486 
    /// The runner's architecture (`x86_64`, `riscv64`, …), empty when unknown.
487 
    pub arch: String,
488 
    /// Each image's architecture as runners have reported it (`flow_images`).
489 
    pub image_archs: std::collections::HashMap<String, String>,
490 
    /// Jobs this runner declined, not to be offered to it again (spec §11.5).
491 
    pub declined: std::collections::HashSet<i64>,
492 
    /// Images the hub holds that this runner would fetch when it lacks
493 
    /// them (oxygenflow#39); empty for a runner that does not fetch.
494 
    pub fetchable: std::collections::HashSet<String>,
495 
    /// Its free disk now, from its claim's facts (oxygenflow#40); `None`
496 
    /// when it does not say.
497 
    pub free_disk_mib: Option<u64>,
498 
}
499
500 
impl Room {
501 
    /// Whether a job needing `need` MiB of disk fits in its free disk.
502 
    pub fn disk_fits(&self, need: Option<i64>) -> bool {
503 
        match (need, self.free_disk_mib) {
504 
            (Some(n), Some(free)) => n.max(0) as u64 <= free,
505 
            _ => true,
506 
        }
507 
    }
508 
}
509
510 
impl Room {
511 
    /// Whether a job's sandbox (its stored JSON) fits: its memory and vCPUs,
512 
    /// and its rootfs image among the runner's (oxygenhub#73).
513 
    pub fn fits(&self, sandbox_json: &str) -> bool {
514 
        let v: serde_json::Value = serde_json::from_str(sandbox_json).unwrap_or_default();
515 
        let need = |k: &str| v[k].as_u64().unwrap_or(0);
516 
        let rootfs = v["rootfs"].as_str().unwrap_or("");
517 
        // The architecture: the sandbox's own ask, else the image's as read
518 
        // or learned from its passes (oxygenhub#173); one nobody has read
519 
        // or run yet matches (OxygenFlow spec §11.3).
520 
        let wanted = v["arch"].as_str().map(String::from).or_else(|| self.image_archs.get(rootfs).cloned());
521 
        let arch_ok = self.arch.is_empty() || wanted.as_deref().is_none_or(|w| w == self.arch);
522 
        arch_ok
523 
            && self.memory_mib.is_none_or(|m| need("memory_mib") <= m as u64)
524 
            && self.cpus.is_none_or(|c| need("cpus") <= c as u64)
525 
            && (rootfs.is_empty() || self.images.as_ref().is_none_or(|have| have.contains(rootfs)) || self.fetchable.contains(rootfs))
526 
    }
527 
}
528
529 
/// Remember which architecture each image is for, as a runner reports it.
530 
pub async fn record_image_archs(pool: &Pool, archs: &std::collections::BTreeMap<String, String>) -> sqlx::Result<()> {
531 
    for (hash, arch) in archs.iter().take(500) {
532 
        if hash.len() != 64 || !flow_core::arch_ok(arch) {
533 
            continue;
534 
        }
535 
        sqlx::query("INSERT INTO flow_images (hash, arch, seen_ms) VALUES ($1, $2, $3) ON CONFLICT (hash) DO UPDATE SET arch = excluded.arch, seen_ms = excluded.seen_ms")
536 
        .bind(hash.to_ascii_lowercase())
537 
        .bind(arch)
538 
        .bind(now_ms())
539 
        .execute(pool)
540 
        .await?;
541 
    }
542 
    Ok(())
543 
}
544
545 
/// Give every image nobody has read the architecture of the runners its
546 
/// jobs passed on (oxygenhub#173), for the jobs before the hub learned at
547 
/// each pass. Never overrides a known architecture.
548 
pub async fn learn_archs_from_runs(pool: &Pool) -> sqlx::Result<u64> {
549 
    let rootfs = crate::db::json_text("j.sandbox", "rootfs");
550 
    let r = sqlx::query(&format!(
551 
        "INSERT INTO flow_images (hash, arch, seen_ms)
552 
         SELECT lower({rootfs}), MIN(r.arch), $1 FROM flow_jobs j JOIN flow_runners r ON r.id = j.runner_id
553 
         WHERE j.status = 'passed' AND r.arch != '' AND length({rootfs}) = 64
554 
         GROUP BY lower({rootfs})
555 
         ON CONFLICT (hash) DO NOTHING"
556 
    ))
557 
    .bind(now_ms())
558 
    .execute(pool)
559 
    .await?;
560 
    Ok(r.rows_affected())
561 
}
562
563 
/// Every image's architecture the hub knows.
564 
pub async fn image_archs(pool: &Pool) -> sqlx::Result<std::collections::HashMap<String, String>> {
565 
    let rows: Vec<(String, String)> = sqlx::query_as("SELECT hash, arch FROM flow_images").fetch_all(pool).await?;
566 
    Ok(rows.into_iter().collect())
567 
}
568
569 
pub async fn claim(pool: &Pool, runtimes: &[String], runner: &str, token_hash: &str) -> sqlx::Result<Option<Job>> {
570 
    claim_within(pool, runtimes, runner, token_hash, Room::default(), &Default::default(), &[]).await
571 
}
572
573 
/// The first queued job a runner with these tiers, this room and this
574 
/// scope may take, now running under `runner`. Repositories take turns
575 
/// (fair share, oxygenhub#244): a job's turn is its place in its own
576 
/// repository's queue plus that repository's running jobs, lowest first,
577 
/// oldest first among equals, so one repository's burst can't hold up the
578 
/// rest. A scoped runner never sees
579 
/// another scope's jobs (oxygenhub#41).
580 
pub async fn claim_within(
581 
    pool: &Pool,
582 
    runtimes: &[String],
583 
    runner: &str,
584 
    token_hash: &str,
585 
    room: Room,
586 
    scope: &super::runners::Scope,
587 
    labels: &[String],
588 
) -> sqlx::Result<Option<Job>> {
589 
    claim_ranked(pool, runtimes, runner, token_hash, room, scope, labels, &|_| false).await
590 
}
591
592 
/// [`claim_within`], passing over a job `elsewhere` says a better runner
593 
/// should take (oxygenflow#42).
594 
#[allow(clippy::too_many_arguments)]
595 
pub async fn claim_ranked(
596 
    pool: &Pool,
597 
    runtimes: &[String],
598 
    runner: &str,
599 
    token_hash: &str,
600 
    room: Room,
601 
    scope: &super::runners::Scope,
602 
    labels: &[String],
603 
    elsewhere: &(dyn Fn(&Job) -> bool + Sync),
604 
) -> sqlx::Result<Option<Job>> {
605 
    let mut tx = super::begin_write(pool).await?;
606 
    let queued: Vec<Job> = sqlx::query_as(&format!(
607 
        "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 \
608 
         WHERE j.status = 'queued' AND j.approval != 'waiting' AND r.status IN ('queued', 'running') \
609 
         AND ($1 = '' OR rep.owner = $2) AND ($3 IS NULL OR r.repo_id = $4) \
610 
         AND NOT ($5 = '' AND $6 IS NULL AND rep.flow = 'own') \
611 
         ORDER BY {PRIORITY_RANK}, ROW_NUMBER() OVER (PARTITION BY r.repo_id ORDER BY j.run_id, j.depth, j.id) \
612 
           + (SELECT COUNT(*) FROM flow_jobs x JOIN flow_runs xr ON xr.id = x.run_id WHERE x.status = 'running' AND xr.repo_id = r.repo_id), \
613 
           j.run_id, j.depth, j.id LIMIT 200"
614 
    ))
615 
    .bind(&scope.owner)
616 
    .bind(&scope.owner)
617 
    .bind(scope.repo_id)
618 
    .bind(scope.repo_id)
619 
    // A repository on its own runners only (#225): the site's runners skip it.
620 
    .bind(&scope.owner)
621 
    .bind(scope.repo_id)
622 
    .fetch_all(&mut *tx)
623 
    .await?;
624 
    for j in queued {
625 
        if !runtimes.iter().any(|r| r == &j.runtime)
626 
            || !room.fits(&j.sandbox)
627 
            || !room.disk_fits(j.need_disk_mib)
628 
            || room.declined.contains(&j.id)
629 
        {
630 
            continue;
631 
        }
632 
        // Every label the job and its sandbox ask for (`runs-on`).
633 
        if !j.needs().iter().all(|l| labels.contains(l)) {
634 
            continue;
635 
        }
636 
        let deps = j.depends_on();
637 
        if !deps.is_empty() {
638 
            let mut passed = 0i64;
639 
            for d in &deps {
640 
                let st: Option<String> = sqlx::query_scalar("SELECT status FROM flow_jobs WHERE run_id = $1 AND name = $2")
641 
                    .bind(j.run_id)
642 
                    .bind(d)
643 
                    .fetch_optional(&mut *tx)
644 
                    .await?;
645 
                if st.as_deref() == Some("passed") {
646 
                    passed += 1;
647 
                }
648 
            }
649 
            if passed != deps.len() as i64 {
650 
                continue;
651 
            }
652 
        }
653 
        if elsewhere(&j) {
654 
            continue;
655 
        }
656 
        let now = now_ms();
657 
        let done = sqlx::query(
658 
            "UPDATE flow_jobs SET status = 'running', runner = $1, token_hash = $2, started_ms = $3, waiting = '' WHERE id = $4 AND status = 'queued'",
659 
        )
660 
        .bind(runner)
661 
        .bind(token_hash)
662 
        .bind(now)
663 
        .bind(j.id)
664 
        .execute(&mut *tx)
665 
        .await?;
666 
        if done.rows_affected() == 1 {
667 
            sqlx::query("UPDATE flow_runs SET status = 'running' WHERE id = $1 AND status = 'queued'")
668 
                .bind(j.run_id)
669 
                .execute(&mut *tx)
670 
                .await?;
671 
            tx.commit().await?;
672 
            return job(pool, j.id).await;
673 
        }
674 
    }
675 
    tx.commit().await?;
676 
    Ok(None)
677 
}
678
679 
/// How long a decline keeps a job from its runner when the runner's
680 
/// profile stays the same: a reason such as memory in use passes.
681 
pub const DECLINE_HOLDS_MS: i64 = 10 * 60_000;
682
683 
/// The jobs runner `runner_id` declined with this profile, recently.
684 
pub async fn declined_by(pool: &Pool, runner_id: i64, profile: &str) -> sqlx::Result<std::collections::HashSet<i64>> {
685 
    let rows: Vec<(i64,)> = sqlx::query_as("SELECT job_id FROM flow_declines WHERE runner_id = $1 AND profile = $2 AND ms > $3")
686 
        .bind(runner_id)
687 
        .bind(profile)
688 
        .bind(now_ms() - DECLINE_HOLDS_MS)
689 
        .fetch_all(pool)
690 
        .await?;
691 
    Ok(rows.into_iter().map(|(j,)| j).collect())
692 
}
693
694 
/// A decline: one runner's reason for handing a job back.
695 
#[derive(Clone, Debug, sqlx::FromRow)]
696 
pub struct DeclineRow {
697 
    pub job_id: i64,
698 
    pub runner_id: i64,
699 
    pub runner: String,
700 
    pub reason: String,
701 
    pub ms: i64,
702 
}
703
704 
/// Runner `runner_id` hands back job `job_id`, running on it: the decline
705 
/// is kept (one per job and runner, the latest), and the job is queued
706 
/// again. Returns every decline the job has had, or `None` when the job
707 
/// was not running on that runner.
708 
pub async fn decline(
709 
    pool: &Pool,
710 
    job_id: i64,
711 
    runner_id: i64,
712 
    runner: &str,
713 
    profile: &str,
714 
    reason: &str,
715 
) -> sqlx::Result<Option<Vec<DeclineRow>>> {
716 
    let mut tx = super::begin_write(pool).await?;
717 
    let back = sqlx::query(
718 
        "UPDATE flow_jobs SET status = 'queued', runner = '', token_hash = '', started_ms = NULL, runner_id = NULL \
719 
         WHERE id = $1 AND status = 'running' AND runner_id = $2",
720 
    )
721 
    .bind(job_id)
722 
    .bind(runner_id)
723 
    .execute(&mut *tx)
724 
    .await?;
725 
    if back.rows_affected() != 1 {
726 
        tx.rollback().await?;
727 
        return Ok(None);
728 
    }
729 
    sqlx::query(
730 
        "INSERT INTO flow_declines (job_id, runner_id, runner, profile, reason, ms) VALUES ($1, $2, $3, $4, $5, $6) \
731 
         ON CONFLICT (job_id, runner_id) DO UPDATE SET runner = excluded.runner, profile = excluded.profile, reason = excluded.reason, ms = excluded.ms",
732 
    )
733 
    .bind(job_id)
734 
    .bind(runner_id)
735 
    .bind(runner)
736 
    .bind(profile)
737 
    .bind(reason)
738 
    .bind(now_ms())
739 
    .execute(&mut *tx)
740 
    .await?;
741 
    let all: Vec<DeclineRow> =
742 
        sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = $1 ORDER BY ms")
743 
            .bind(job_id)
744 
            .fetch_all(&mut *tx)
745 
            .await?;
746 
    tx.commit().await?;
747 
    Ok(Some(all))
748 
}
749
750 
/// The declines of job `job_id`, oldest first.
751 
pub async fn declines_of_job(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<DeclineRow>> {
752 
    sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = $1 ORDER BY ms")
753 
        .bind(job_id)
754 
        .fetch_all(pool)
755 
        .await
756 
}
757
758 
/// A runner's latest declines, newest first, with where the job is.
759 
#[derive(Clone, Debug, sqlx::FromRow)]
760 
pub struct RunnerDecline {
761 
    pub job_id: i64,
762 
    pub job: String,
763 
    pub run_number: i64,
764 
    pub owner: String,
765 
    pub name: String,
766 
    pub reason: String,
767 
    pub ms: i64,
768 
}
769
770 
pub async fn declines_of_runner(pool: &Pool, runner_id: i64, limit: i64) -> sqlx::Result<Vec<RunnerDecline>> {
771 
    sqlx::query_as(
772 
        "SELECT d.job_id, j.name AS job, r.number AS run_number, rep.owner, rep.name, d.reason, d.ms \
773 
         FROM flow_declines d JOIN flow_jobs j ON j.id = d.job_id JOIN flow_runs r ON r.id = j.run_id \
774 
         JOIN repositories rep ON rep.id = r.repo_id WHERE d.runner_id = $1 ORDER BY d.ms DESC LIMIT $2",
775 
    )
776 
    .bind(runner_id)
777 
    .bind(limit)
778 
    .fetch_all(pool)
779 
    .await
780 
}
781
782 
/// Queued jobs whose dependencies have passed, in their live runs, with
783 
/// their repositories (owner, id): the ones a runner could take now.
784 
pub async fn ready_queued(pool: &Pool) -> sqlx::Result<Vec<(Job, String, i64)>> {
785 
    let queued: Vec<Job> = sqlx::query_as(
786 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
787 
         WHERE j.status = 'queued' AND j.approval != 'waiting' AND r.status IN ('queued', 'running') ORDER BY j.run_id, j.depth, j.id LIMIT 500",
788 
    )
789 
    .fetch_all(pool)
790 
    .await?;
791 
    let mut out = Vec::new();
792 
    for j in queued {
793 
        let mut ready = true;
794 
        for d in j.depends_on() {
795 
            let st: Option<String> = sqlx::query_scalar("SELECT status FROM flow_jobs WHERE run_id = $1 AND name = $2")
796 
                .bind(j.run_id)
797 
                .bind(&d)
798 
                .fetch_optional(pool)
799 
                .await?;
800 
            if st.as_deref() != Some("passed") {
801 
                ready = false;
802 
                break;
803 
            }
804 
        }
805 
        if !ready {
806 
            continue;
807 
        }
808 
        let repo: Option<(String, i64)> =
809 
            sqlx::query_as("SELECT rep.owner, rep.id FROM flow_runs r JOIN repositories rep ON rep.id = r.repo_id WHERE r.id = $1")
810 
                .bind(j.run_id)
811 
                .fetch_optional(pool)
812 
                .await?;
813 
        if let Some((owner, id)) = repo {
814 
            out.push((j, owner, id));
815 
        }
816 
    }
817 
    Ok(out)
818 
}
819
820 
pub async fn set_waiting(pool: &Pool, job_id: i64, waiting: &str) -> sqlx::Result<()> {
821 
    sqlx::query("UPDATE flow_jobs SET waiting = $1 WHERE id = $2 AND waiting <> $3")
822 
        .bind(waiting)
823 
        .bind(job_id)
824 
        .bind(waiting)
825 
        .execute(pool)
826 
        .await?;
827 
    Ok(())
828 
}
829
830 
/// Append records; returns how many were kept (the rest exceed the cap).
831 
pub async fn append_log(pool: &Pool, job_id: i64, records: &[flow_core::api::Record], cap: i64) -> sqlx::Result<usize> {
832 
    let mut tx = super::begin_write(pool).await?;
833 
    let count: i64 = sqlx::query_scalar("SELECT log_count FROM flow_jobs WHERE id = $1").bind(job_id).fetch_one(&mut *tx).await?;
834 
    let room = (cap - count).max(0) as usize;
835 
    let keep = &records[..records.len().min(room)];
836 
    for (i, r) in keep.iter().enumerate() {
837 
        sqlx::query("INSERT INTO flow_logs (job_id, seq, t, stream, text) VALUES ($1, $2, $3, $4, $5)")
838 
            .bind(job_id)
839 
            .bind(count + i as i64 + 1)
840 
            .bind(r.t as i64)
841 
            .bind(&r.stream)
842 
            .bind(&r.text)
843 
            .execute(&mut *tx)
844 
            .await?;
845 
    }
846 
    sqlx::query("UPDATE flow_jobs SET log_count = log_count + $1 WHERE id = $2")
847 
        .bind(keep.len() as i64)
848 
        .bind(job_id)
849 
        .execute(&mut *tx)
850 
        .await?;
851 
    tx.commit().await?;
852 
    Ok(keep.len())
853 
}
854
855 
pub async fn log(pool: &Pool, job_id: i64, after: i64, limit: i64) -> sqlx::Result<Vec<LogLine>> {
856 
    sqlx::query_as("SELECT seq, t, stream, text FROM flow_logs WHERE job_id = $1 AND seq > $2 ORDER BY seq LIMIT $3")
857 
        .bind(job_id)
858 
        .bind(after)
859 
        .bind(limit)
860 
        .fetch_all(pool)
861 
        .await
862 
}
863
864 
/// Jobs running since before `ms` (the oldest first).
865 
/// Jobs still running at `now` past their own time limit (`default_s` for
866 
/// one claimed without) plus `grace_ms`.
867 
pub async fn running_past_limit(pool: &Pool, now: i64, default_s: u64, grace_ms: i64) -> sqlx::Result<Vec<Job>> {
868 
    sqlx::query_as(
869 
        "SELECT * FROM flow_jobs WHERE status = 'running' \
870 
         AND COALESCE(started_ms, 0) + (CASE WHEN timeout_s > 0 THEN timeout_s ELSE $1 END) * 1000 + $2 < $3 ORDER BY started_ms",
871 
    )
872 
    .bind(default_s as i64)
873 
    .bind(grace_ms)
874 
    .bind(now)
875 
    .fetch_all(pool)
876 
    .await
877 
}
878
879 
/// Record the limit a claimed job runs under.
880 
pub async fn set_job_timeout(pool: &Pool, job_id: i64, timeout_s: u64) -> sqlx::Result<()> {
881 
    sqlx::query("UPDATE flow_jobs SET timeout_s = $1 WHERE id = $2").bind(timeout_s as i64).bind(job_id).execute(pool).await?;
882 
    Ok(())
883 
}
884
885 
pub async fn running_since_before(pool: &Pool, ms: i64) -> sqlx::Result<Vec<Job>> {
886 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND COALESCE(started_ms, 0) < $1 ORDER BY started_ms")
887 
        .bind(ms)
888 
        .fetch_all(pool)
889 
        .await
890 
}
891
892 
/// The jobs running on the runner with this id.
893 
pub async fn running_on(pool: &Pool, runner_id: i64) -> sqlx::Result<Vec<Job>> {
894 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND runner_id = $1 ORDER BY id").bind(runner_id).fetch_all(pool).await
895 
}
896
897 
pub struct Outcome<'a> {
898 
    pub status: &'a str,
899 
    pub exit_code: Option<i64>,
900 
    pub error: &'a str,
901 
    pub receipt: &'a str,
902 
    pub receipt_hash: &'a str,
903 
}
904
905 
/// Record a job's end, skip the jobs that can no longer run, and settle the
906 
/// run. Returns the jobs whose status changed (for their checks).
907 
pub async fn finish(pool: &Pool, job_id: i64, o: Outcome<'_>) -> sqlx::Result<Vec<Job>> {
908 
    let mut tx = super::begin_write(pool).await?;
909 
    let now = now_ms();
910 
    let j: Job = sqlx::query_as("SELECT * FROM flow_jobs WHERE id = $1").bind(job_id).fetch_one(&mut *tx).await?;
911 
    sqlx::query("UPDATE flow_jobs SET status = $1, exit_code = $2, error = $3, receipt = $4, receipt_hash = $5, finished_ms = $6, token_hash = '' WHERE id = $7")
912 
        .bind(o.status)
913 
        .bind(o.exit_code)
914 
        .bind(o.error)
915 
        .bind(o.receipt)
916 
        .bind(o.receipt_hash)
917 
        .bind(now)
918 
        .bind(job_id)
919 
        .execute(&mut *tx)
920 
        .await?;
921 
    // A pass teaches what its image is for, when nobody has read it
922 
    // (oxygenhub#173): the architecture of the runner it passed on.
923 
    if o.status == "passed" {
924 
        let rootfs = serde_json::from_str::<serde_json::Value>(&j.sandbox)
925 
            .ok()
926 
            .and_then(|v| v["rootfs"].as_str().map(str::to_ascii_lowercase))
927 
            .unwrap_or_default();
928 
        if rootfs.len() == 64 {
929 
            sqlx::query(
930 
                "INSERT INTO flow_images (hash, arch, seen_ms) SELECT $1, r.arch, $2 FROM flow_jobs j JOIN flow_runners r ON r.id = j.runner_id \
931 
                 WHERE j.id = $3 AND r.arch != '' ON CONFLICT (hash) DO NOTHING",
932 
            )
933 
            .bind(&rootfs)
934 
            .bind(now)
935 
            .bind(job_id)
936 
            .execute(&mut *tx)
937 
            .await?;
938 
        }
939 
    }
940 
    let mut changed = vec![job_id];
941 
    // Dependents of anything that did not pass are skipped, transitively.
942 
    loop {
943 
        let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(j.run_id).fetch_all(&mut *tx).await?;
944 
        let blocked: Vec<i64> = all
945 
            .iter()
946 
            .filter(|x| x.status == "queued")
947 
            .filter(|x| {
948 
                x.depends_on()
949 
                    .iter()
950 
                    .any(|d| all.iter().any(|y| &y.name == d && matches!(y.status.as_str(), "failed" | "skipped" | "error" | "cancelled")))
951 
            })
952 
            .map(|x| x.id)
953 
            .collect();
954 
        if blocked.is_empty() {
955 
            break;
956 
        }
957 
        for id in blocked {
958 
            sqlx::query("UPDATE flow_jobs SET status = 'skipped', finished_ms = $1 WHERE id = $2")
959 
                .bind(now)
960 
                .bind(id)
961 
                .execute(&mut *tx)
962 
                .await?;
963 
            changed.push(id);
964 
        }
965 
    }
966 
    let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(j.run_id).fetch_all(&mut *tx).await?;
967 
    if all.iter().all(Job::done) {
968 
        let status = if all.iter().all(|x| x.status == "passed") { "passed" } else { "failed" };
969 
        sqlx::query("UPDATE flow_runs SET status = $1, finished_ms = $2 WHERE id = $3")
970 
            .bind(status)
971 
            .bind(now)
972 
            .bind(j.run_id)
973 
            .execute(&mut *tx)
974 
            .await?;
975 
    }
976 
    tx.commit().await?;
977 
    let mut out = Vec::new();
978 
    for id in changed {
979 
        if let Some(x) = job(pool, id).await? {
980 
            out.push(x);
981 
        }
982 
    }
983 
    Ok(out)
984 
}
985
986 
/// Cancel a run (oxygenflow#32): its jobs not yet done end as cancelled,
987 
/// and so does the run. A runner learns of it when the hub next refuses
988 
/// its log lines, and stops the job. Returns the jobs whose status changed.
989 
pub async fn cancel_run(pool: &Pool, run_id: i64, why: &str) -> sqlx::Result<Vec<Job>> {
990 
    let now = now_ms();
991 
    let mut tx = pool.begin().await?;
992 
    let ids: Vec<(i64,)> = sqlx::query_as("SELECT id FROM flow_jobs WHERE run_id = $1 AND status IN ('queued', 'running')")
993 
        .bind(run_id)
994 
        .fetch_all(&mut *tx)
995 
        .await?;
996 
    sqlx::query(
997 
        "UPDATE flow_jobs SET status = 'cancelled', error = $1, finished_ms = $2, token_hash = ''
998 
         WHERE run_id = $3 AND status IN ('queued', 'running')",
999 
    )
1000 
    .bind(why)
1001 
    .bind(now)
1002 
    .bind(run_id)
1003 
    .execute(&mut *tx)
1004 
    .await?;
1005 
    sqlx::query("UPDATE flow_runs SET status = 'cancelled', finished_ms = $1 WHERE id = $2 AND status IN ('queued', 'running')")
1006 
        .bind(now)
1007 
        .bind(run_id)
1008 
        .execute(&mut *tx)
1009 
        .await?;
1010 
    tx.commit().await?;
1011 
    let mut out = Vec::new();
1012 
    for (id,) in ids {
1013 
        if let Some(x) = job(pool, id).await? {
1014 
            out.push(x);
1015 
        }
1016 
    }
1017 
    Ok(out)
1018 
}
1019
1020 
/// Run job `name` of finished run `run_id` again, with every job that
1021 
/// depends on it (oxygenflow#31): back to the queue as if new, their logs,
1022 
/// artifacts and reports gone, and the run running again. The jobs they
1023 
/// depend on keep their results. The jobs reset; none when there is no
1024 
/// such job.
1025 
pub async fn rerun_job(pool: &Pool, run_id: i64, name: &str) -> sqlx::Result<Vec<Job>> {
1026 
    let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(run_id).fetch_all(pool).await?;
1027 
    if !all.iter().any(|j| j.name == name) {
1028 
        return Ok(Vec::new());
1029 
    }
1030 
    let mut again = vec![name.to_string()];
1031 
    loop {
1032 
        let more: Vec<String> = all
1033 
            .iter()
1034 
            .filter(|j| !again.contains(&j.name) && j.depends_on().iter().any(|d| again.contains(d)))
1035 
            .map(|j| j.name.clone())
1036 
            .collect();
1037 
        if more.is_empty() {
1038 
            break;
1039 
        }
1040 
        again.extend(more);
1041 
    }
1042 
    let ids: Vec<i64> = all.iter().filter(|j| again.contains(&j.name)).map(|j| j.id).collect();
1043 
    let mut tx = pool.begin().await?;
1044 
    for id in &ids {
1045 
        for table in ["flow_logs", "flow_artifacts", "flow_declines", "test_reports", "coverage_reports", "reports"] {
1046 
            sqlx::query(&format!("DELETE FROM {table} WHERE job_id = $1")).bind(id).execute(&mut *tx).await?;
1047 
        }
1048 
        sqlx::query(
1049 
            "UPDATE flow_jobs SET status = 'queued', runner = '', token_hash = '', started_ms = NULL, finished_ms = NULL,
1050 
             exit_code = NULL, error = '', receipt = '', receipt_hash = '', log_count = 0, also_read = '', receipt_attest = '',
1051 
             receipt_signer = '', receipt_note = '', runner_id = NULL, waiting = '', placed = '', peak_disk_mib = NULL,
1052 
             peak_memory_mib = NULL, approval = CASE WHEN approval = '' THEN '' ELSE 'waiting' END, approved_by = '',
1053 
             approved_ms = NULL, approval_seal = '', approved_via = '' WHERE id = $1",
1054 
        )
1055 
        .bind(id)
1056 
        .execute(&mut *tx)
1057 
        .await?;
1058 
    }
1059 
    sqlx::query("UPDATE flow_runs SET status = 'running', finished_ms = NULL WHERE id = $1").bind(run_id).execute(&mut *tx).await?;
1060 
    tx.commit().await?;
1061 
    let mut out = Vec::new();
1062 
    for id in ids {
1063 
        if let Some(x) = job(pool, id).await? {
1064 
            out.push(x);
1065 
        }
1066 
    }
1067 
    Ok(out)
1068 
}
1069
1070 
/// The runs of `bookmark` started by `trigger` still going, older than run
1071 
/// `number`: those a newer run supersedes (#267). With `running_too`
1072 
/// false, only those still queued.
1073 
pub async fn live_runs_before(
1074 
    pool: &Pool,
1075 
    repo_id: i64,
1076 
    bookmark: &str,
1077 
    trigger: &str,
1078 
    number: i64,
1079 
    running_too: bool,
1080 
) -> sqlx::Result<Vec<Run>> {
1081 
    sqlx::query_as(
1082 
        "SELECT * FROM flow_runs WHERE repo_id = $1 AND bookmark = $2 AND number < $3 AND trigger = $4
1083 
         AND (status = 'queued' OR ($5 AND status = 'running')) ORDER BY id",
1084 
    )
1085 
    .bind(repo_id)
1086 
    .bind(bookmark)
1087 
    .bind(number)
1088 
    .bind(trigger)
1089 
    .bind(running_too)
1090 
    .fetch_all(pool)
1091 
    .await
1092 
}
1093
1094 
/// The newest run of exactly this state (heads joined with `+`, sorted).
1095 
pub async fn latest_for_heads(pool: &Pool, repo_id: i64, heads: &str) -> sqlx::Result<Option<Run>> {
1096 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND heads = $2 ORDER BY id DESC LIMIT 1")
1097 
        .bind(repo_id)
1098 
        .bind(heads)
1099 
        .fetch_optional(pool)
1100 
        .await
1101 
}
1102
1103 
/// Record what became of job `job_id`'s signed receipt.
1104 
pub async fn set_receipt(pool: &Pool, job_id: i64, attest: &str, signer: &str, note: &str) -> sqlx::Result<()> {
1105 
    sqlx::query("UPDATE flow_jobs SET receipt_attest = $1, receipt_signer = $2, receipt_note = $3 WHERE id = $4")
1106 
        .bind(attest)
1107 
        .bind(signer)
1108 
        .bind(note)
1109 
        .bind(job_id)
1110 
        .execute(pool)
1111 
        .await?;
1112 
    Ok(())
1113 
}
1114
1115 
/// A job's artifact, kept as a blob of the run's repository.
1116 
#[derive(Clone, Debug, FromRow)]
1117 
pub struct Artifact {
1118 
    pub job_id: i64,
1119 
    pub name: String,
1120 
    pub path: String,
1121 
    pub blob: String,
1122 
    pub size: i64,
1123 
    /// BLAKE3 of the bytes.
1124 
    pub content: String,
1125 
    pub created_ms: i64,
1126 
}
1127
1128 
pub async fn add_artifact(pool: &Pool, a: &Artifact) -> sqlx::Result<()> {
1129 
    sqlx::query(
1130 
        "INSERT OR REPLACE INTO flow_artifacts (job_id, name, path, blob, size, content, created_ms) VALUES ($1, $2, $3, $4, $5, $6, $7)",
1131 
    )
1132 
    .bind(a.job_id)
1133 
    .bind(&a.name)
1134 
    .bind(&a.path)
1135 
    .bind(&a.blob)
1136 
    .bind(a.size)
1137 
    .bind(&a.content)
1138 
    .bind(a.created_ms)
1139 
    .execute(pool)
1140 
    .await?;
1141 
    Ok(())
1142 
}
1143
1144 
/// Every artifact blob a repository's runs keep, for garbage collection to
1145 
/// keep too (#126): the database is the only thing that names them.
1146 
pub async fn artifact_blobs(pool: &Pool, repo_id: i64) -> sqlx::Result<Vec<String>> {
1147 
    sqlx::query_scalar(
1148 
        "SELECT DISTINCT a.blob FROM flow_artifacts a JOIN flow_jobs j ON j.id = a.job_id JOIN flow_runs r ON r.id = j.run_id WHERE r.repo_id = $1",
1149 
    )
1150 
    .bind(repo_id)
1151 
    .fetch_all(pool)
1152 
    .await
1153 
}
1154
1155 
/// Job `job_id`'s artifacts, by name.
1156 
pub async fn artifacts(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<Artifact>> {
1157 
    sqlx::query_as("SELECT * FROM flow_artifacts WHERE job_id = $1 ORDER BY name").bind(job_id).fetch_all(pool).await
1158 
}
1159
1160 
/// Where an artifact with given content came from: the run and job that
1161 
/// built it, and the job's signed receipt when one was kept.
1162 
#[derive(Clone, Debug, FromRow)]
1163 
pub struct Provenance {
1164 
    pub run_number: i64,
1165 
    pub job: String,
1166 
    pub name: String,
1167 
    pub receipt_attest: String,
1168 
    pub receipt_signer: String,
1169 
}
1170
1171 
/// The build of the bytes whose BLAKE3 is `content` in repository
1172 
/// `repo_id`: a passed job, one with a kept signed receipt first, newest
1173 
/// run first.
1174 
pub async fn provenance(pool: &Pool, repo_id: i64, content: &str) -> sqlx::Result<Option<Provenance>> {
1175 
    sqlx::query_as(
1176 
        "SELECT r.number AS run_number, j.name AS job, a.name AS name, j.receipt_attest, j.receipt_signer \
1177 
         FROM flow_artifacts a JOIN flow_jobs j ON j.id = a.job_id JOIN flow_runs r ON r.id = j.run_id \
1178 
         WHERE r.repo_id = $1 AND a.content = $2 AND j.status = 'passed' \
1179 
         ORDER BY (j.receipt_attest != '') DESC, r.number DESC LIMIT 1",
1180 
    )
1181 
    .bind(repo_id)
1182 
    .bind(content)
1183 
    .fetch_optional(pool)
1184 
    .await
1185 
}
1186
1187 
/// Approve job `id` if it is still waiting for a person (oxygenflow#57);
1188 
/// whether it was.
1189 
pub async fn approve_job(pool: &Pool, id: i64, who: &str, seal: &str, via: &str) -> sqlx::Result<bool> {
1190 
    let r = sqlx::query(
1191 
        "UPDATE flow_jobs SET approval = 'approved', approved_by = $1, approved_ms = $2, approval_seal = $3, approved_via = $5 \
1192 
         WHERE id = $4 AND status = 'queued' AND approval = 'waiting'",
1193 
    )
1194 
    .bind(who)
1195 
    .bind(now_ms())
1196 
    .bind(seal)
1197 
    .bind(id)
1198 
    .bind(via)
1199 
    .execute(pool)
1200 
    .await?;
1201 
    Ok(r.rows_affected() > 0)
1202 
}
1203
1204 
/// Note who rejected job `id` (it is then finished as cancelled).
1205 
pub async fn note_rejection(pool: &Pool, id: i64, who: &str, via: &str) -> sqlx::Result<()> {
1206 
    sqlx::query("UPDATE flow_jobs SET approved_by = $1, approved_ms = $2, approved_via = $4 WHERE id = $3")
1207 
        .bind(who)
1208 
        .bind(now_ms())
1209 
        .bind(id)
1210 
        .bind(via)
1211 
        .execute(pool)
1212 
        .await?;
1213 
    Ok(())
1214 
}
1215
1216 
/// Jobs waiting for a person, in runs still going.
1217 
pub async fn waiting_for_approval(pool: &Pool) -> sqlx::Result<Vec<Job>> {
1218 
    sqlx::query_as(
1219 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
1220 
         WHERE j.status = 'queued' AND j.approval = 'waiting' AND r.status IN ('queued', 'running')",
1221 
    )
1222 
    .fetch_all(pool)
1223 
    .await
1224 
}
1225
1226 
/// One deployment: a job for an environment that passed.
1227 
#[derive(Clone, Debug, FromRow)]
1228 
pub struct Deployment {
1229 
    pub environment: String,
1230 
    pub job: String,
1231 
    pub run_number: i64,
1232 
    pub state: String,
1233 
    pub bookmark: String,
1234 
    pub actor: String,
1235 
    pub approved_by: String,
1236 
    pub finished_ms: i64,
1237 
}
1238
1239 
/// A repository's deployments, newest first (oxygenflow#59).
1240 
pub async fn deployments(pool: &Pool, repo_id: i64, limit: i64) -> sqlx::Result<Vec<Deployment>> {
1241 
    sqlx::query_as(
1242 
        "SELECT j.environment, j.name AS job, r.number AS run_number, r.state, r.bookmark, r.actor, j.approved_by, \
1243 
         COALESCE(j.finished_ms, 0) AS finished_ms FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
1244 
         WHERE r.repo_id = $1 AND j.environment != '' AND j.status = 'passed' ORDER BY j.finished_ms DESC LIMIT $2",
1245 
    )
1246 
    .bind(repo_id)
1247 
    .bind(limit)
1248 
    .fetch_all(pool)
1249 
    .await
1250 
}