OXYGENHUB
# cb7b3cd45dc0 oxygenhub / crates / hub / src / db / flow.rs
48.5 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 
    /// When it entered the queue (made, or run again), and when it became
167 
    /// ready to run: set at its claim (migration 0080, oxygenhub#257).
168 
    #[sqlx(default)]
169 
    pub queued_ms: Option<i64>,
170 
    #[sqlx(default)]
171 
    pub ready_ms: Option<i64>,
172 
}
173
174 
impl Job {
175 
    pub fn depends_on(&self) -> Vec<String> {
176 
        serde_json::from_str(&self.depends).unwrap_or_default()
177 
    }
178
179 
    pub fn done(&self) -> bool {
180 
        matches!(self.status.as_str(), "passed" | "failed" | "skipped" | "error" | "cancelled")
181 
    }
182 
}
183
184 
#[derive(Clone, Debug, FromRow)]
185 
pub struct LogLine {
186 
    pub seq: i64,
187 
    pub t: i64,
188 
    pub stream: String,
189 
    pub text: String,
190 
}
191
192 
/// A job as the trigger creates it.
193 
pub struct NewJob {
194 
    pub name: String,
195 
    pub depth: i64,
196 
    pub depends: Vec<String>,
197 
    pub spec: String,
198 
    pub sandbox: String,
199 
    pub runtime: String,
200 
    /// The free disk it needs, from [`learned_disk_need`].
201 
    pub need_disk_mib: Option<i64>,
202 
}
203
204 
/// Runs of a job whose peaks set its next need.
205 
const NEED_FROM_RUNS: i64 = 5;
206
207 
/// The free disk a new run of job `name` in repository `repo_id` needs
208 
/// (oxygenflow#41): the most its last runs took, plus a quarter; `None`
209 
/// for a job that has not run here yet.
210 
pub async fn learned_disk_need(pool: &Pool, repo_id: i64, name: &str) -> sqlx::Result<Option<i64>> {
211 
    let peak: Option<i64> = sqlx::query_scalar(
212 
        "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 \
213 
         WHERE r.repo_id = $1 AND j.name = $2 AND j.peak_disk_mib IS NOT NULL ORDER BY j.id DESC LIMIT $3)",
214 
    )
215 
    .bind(repo_id)
216 
    .bind(name)
217 
    .bind(NEED_FROM_RUNS)
218 
    .fetch_one(pool)
219 
    .await?;
220 
    Ok(peak.map(|p| p + p / 4 + 64))
221 
}
222
223 
/// Record why job `job_id` ran where it did.
224 
pub async fn set_placed(pool: &Pool, job_id: i64, placed: &str) -> sqlx::Result<()> {
225 
    sqlx::query("UPDATE flow_jobs SET placed = $1 WHERE id = $2").bind(placed).bind(job_id).execute(pool).await?;
226 
    Ok(())
227 
}
228
229 
/// Keep what a job took, as its runner reported.
230 
pub async fn set_peaks(pool: &Pool, job_id: i64, disk_mib: Option<u64>, memory_mib: Option<u64>) -> sqlx::Result<()> {
231 
    sqlx::query("UPDATE flow_jobs SET peak_disk_mib = $1, peak_memory_mib = $2 WHERE id = $3")
232 
        .bind(disk_mib.map(|v| v.min(i64::MAX as u64) as i64))
233 
        .bind(memory_mib.map(|v| v.min(i64::MAX as u64) as i64))
234 
        .bind(job_id)
235 
        .execute(pool)
236 
        .await?;
237 
    Ok(())
238 
}
239
240 
pub struct NewRun<'a> {
241 
    pub repo_id: i64,
242 
    pub bookmark: &'a str,
243 
    pub state: &'a str,
244 
    pub heads: &'a str,
245 
    pub pipeline_hash: &'a str,
246 
    pub actor: &'a str,
247 
    /// `push`, `proposal`, `tag:NAME`, `manual`, `schedule` or
248 
    /// `rerun:N:EVENT`.
249 
    pub trigger: &'a str,
250 
    /// Problems when the pipeline is invalid: the run is recorded in error.
251 
    pub error: Option<&'a str>,
252 
}
253
254 
pub async fn create_run(pool: &Pool, r: NewRun<'_>, jobs: &[NewJob]) -> sqlx::Result<i64> {
255 
    let mut tx = super::begin_write(pool).await?;
256 
    let now = now_ms();
257 
    let status = if r.error.is_some() { "error" } else { "queued" };
258 
    let id: i64 = sqlx::query_scalar(
259 
        "INSERT INTO flow_runs (repo_id, number, bookmark, state, heads, pipeline_hash, status, error, actor, trigger, created_ms, finished_ms) \
260 
         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",
261 
    )
262 
    .bind(r.repo_id)
263 
    .bind(r.repo_id)
264 
    .bind(r.bookmark)
265 
    .bind(r.state)
266 
    .bind(r.heads)
267 
    .bind(r.pipeline_hash)
268 
    .bind(status)
269 
    .bind(r.error.unwrap_or(""))
270 
    .bind(r.actor)
271 
    .bind(r.trigger)
272 
    .bind(now)
273 
    .bind(r.error.map(|_| now))
274 
    .fetch_one(&mut *tx)
275 
    .await?;
276 
    for j in jobs {
277 
        sqlx::query(
278 
            "INSERT INTO flow_jobs (run_id, name, depth, depends, spec, sandbox, runtime, status, need_disk_mib, approval, environment, priority, queued_ms) \
279 
             VALUES ($1, $2, $3, $4, $5, $6, $7, 'queued', $8, $9, $10, $11, $12)",
280 
        )
281 
        .bind(id)
282 
        .bind(&j.name)
283 
        .bind(j.depth)
284 
        .bind(serde_json::to_string(&j.depends).unwrap_or_else(|_| "[]".into()))
285 
        .bind(&j.spec)
286 
        .bind(&j.sandbox)
287 
        .bind(&j.runtime)
288 
        .bind(j.need_disk_mib)
289 
        .bind(if needs_approval(&j.spec) { "waiting" } else { "" })
290 
        .bind(environment_of(&j.spec))
291 
        .bind(priority_of(&j.spec))
292 
        .bind(now)
293 
        .execute(&mut *tx)
294 
        .await?;
295 
    }
296 
    tx.commit().await?;
297 
    Ok(id)
298 
}
299
300 
/// The environment a compiled job deploys to, or empty (oxygenflow#59).
301 
fn environment_of(spec: &str) -> String {
302 
    serde_json::from_str::<serde_json::Value>(spec).ok().and_then(|v| v["environment"].as_str().map(String::from)).unwrap_or_default()
303 
}
304
305 
/// A compiled job's own priority (`low`, `normal`), or empty; anything
306 
/// else is not one a pipeline may give (oxygenhub#244).
307 
fn priority_of(spec: &str) -> String {
308 
    serde_json::from_str::<serde_json::Value>(spec)
309 
        .ok()
310 
        .and_then(|v| v["priority"].as_str().map(String::from))
311 
        .filter(|p| p == "low" || p == "normal")
312 
        .unwrap_or_default()
313 
}
314
315 
/// Whether a compiled job asks for approval (oxygenflow#57).
316 
fn needs_approval(spec: &str) -> bool {
317 
    serde_json::from_str::<serde_json::Value>(spec).is_ok_and(|v| !v["approval"].is_null())
318 
}
319
320 
/// The labels a compiled job and its sandbox ask a runner for (`runs-on`).
321 
fn labels_in(spec: &str, sandbox: &str) -> Vec<String> {
322 
    let mut out: Vec<String> = [spec, sandbox]
323 
        .iter()
324 
        .filter_map(|j| serde_json::from_str::<serde_json::Value>(j).ok())
325 
        .flat_map(|v| v["runs_on"].as_array().cloned().unwrap_or_default())
326 
        .filter_map(|l| l.as_str().map(String::from))
327 
        .collect();
328 
    out.sort();
329 
    out.dedup();
330 
    out
331 
}
332
333 
impl Job {
334 
    /// The runner labels this job needs.
335 
    pub fn needs(&self) -> Vec<String> {
336 
        labels_in(&self.spec, &self.sandbox)
337 
    }
338
339 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
340 
    pub fn arch(&self) -> Option<String> {
341 
        arch_in(&self.sandbox)
342 
    }
343 
}
344
345 
/// The architecture a sandbox's JSON asks for, if any.
346 
fn arch_in(sandbox: &str) -> Option<String> {
347 
    serde_json::from_str::<serde_json::Value>(sandbox).ok()?["arch"].as_str().map(String::from)
348 
}
349
350 
/// A job waiting for or held by a runner, with where it comes from.
351 
#[derive(Clone, Debug, FromRow)]
352 
pub struct QueueEntry {
353 
    pub job_id: i64,
354 
    pub run_id: i64,
355 
    pub repo_id: i64,
356 
    pub owner: String,
357 
    pub name: String,
358 
    pub visibility: String,
359 
    pub run_number: i64,
360 
    pub job: String,
361 
    pub status: String,
362 
    /// The runner holding it, when running.
363 
    pub runner: String,
364 
    pub spec: String,
365 
    pub sandbox: String,
366 
    /// When it started running.
367 
    pub started_ms: Option<i64>,
368 
    /// Its level (0 urgent … 3 low, oxygenhub#244) and, when expedited, who
369 
    /// did it and why.
370 
    pub priority_rank: i64,
371 
    pub priority_note: String,
372 
}
373
374 
impl QueueEntry {
375 
    pub fn priority(&self) -> &'static str {
376 
        PRIORITIES.get(self.priority_rank as usize).copied().unwrap_or("normal")
377 
    }
378 
}
379
380 
impl QueueEntry {
381 
    /// The runner labels the job needs.
382 
    pub fn needs(&self) -> Vec<String> {
383 
        labels_in(&self.spec, &self.sandbox)
384 
    }
385
386 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
387 
    pub fn arch(&self) -> Option<String> {
388 
        arch_in(&self.sandbox)
389 
    }
390 
}
391
392 
/// Every job of a live run that is running or queued: running ones first,
393 
/// then queued ones in the order runners claim them (hub spec §13.2).
394 
pub async fn queue(pool: &Pool) -> sqlx::Result<Vec<QueueEntry>> {
395 
    sqlx::query_as(&format!(
396 
        "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, \
397 
           CAST({PRIORITY_RANK} AS BIGINT) AS priority_rank, {PRIORITY_NOTE} AS priority_note \
398 
         FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories rep ON rep.id = r.repo_id \
399 
         WHERE j.status IN ('queued', 'running') AND j.approval != 'waiting' AND r.status IN ('queued', 'running') \
400 
         ORDER BY CASE j.status WHEN 'running' THEN 0 ELSE 1 END, \
401 
           CASE j.status WHEN 'running' THEN 0 ELSE {PRIORITY_RANK} END, \
402 
           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) \
403 
           + (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, \
404 
           j.run_id, j.depth, j.id"
405 
    ))
406 
    .fetch_all(pool)
407 
    .await
408 
}
409
410 
/// A repository's runs, newest first, numbered below `before`.
411 
pub async fn runs_by_number(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
412 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND number < $2 ORDER BY number DESC LIMIT $3")
413 
        .bind(repo_id)
414 
        .bind(before.unwrap_or(i64::MAX))
415 
        .bind(limit)
416 
        .fetch_all(pool)
417 
        .await
418 
}
419
420 
/// The latest run on this bookmark of this repository, for what a push
421 
/// changed since (a job's scope).
422 
/// The run a push to `bookmark` is scoped against: the newest that was not
423 
/// cancelled, nor a push run still going (a new push supersedes it), so
424 
/// that what a cancelled run would have checked is checked by the next.
425 
pub async fn latest_on(pool: &Pool, repo_id: i64, bookmark: &str) -> sqlx::Result<Option<Run>> {
426 
    sqlx::query_as(
427 
        "SELECT * FROM flow_runs WHERE repo_id = $1 AND bookmark = $2 AND heads != '' AND status != 'cancelled'
428 
         AND NOT (trigger = 'push' AND status IN ('queued', 'running')) ORDER BY id DESC LIMIT 1",
429 
    )
430 
    .bind(repo_id)
431 
    .bind(bookmark)
432 
    .fetch_optional(pool)
433 
    .await
434 
}
435
436 
pub async fn runs(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
437 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND id < $2 ORDER BY id DESC LIMIT $3")
438 
        .bind(repo_id)
439 
        .bind(before.unwrap_or(i64::MAX))
440 
        .bind(limit)
441 
        .fetch_all(pool)
442 
        .await
443 
}
444
445 
/// A repository's run by its number there.
446 
pub async fn run(pool: &Pool, repo_id: i64, number: i64) -> sqlx::Result<Option<Run>> {
447 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND number = $2").bind(repo_id).bind(number).fetch_optional(pool).await
448 
}
449
450 
pub async fn run_by_id(pool: &Pool, id: i64) -> sqlx::Result<Option<Run>> {
451 
    sqlx::query_as("SELECT * FROM flow_runs WHERE id = $1").bind(id).fetch_optional(pool).await
452 
}
453
454 
pub async fn jobs(pool: &Pool, run_id: i64) -> sqlx::Result<Vec<Job>> {
455 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1 ORDER BY depth, name").bind(run_id).fetch_all(pool).await
456 
}
457
458 
pub async fn job(pool: &Pool, id: i64) -> sqlx::Result<Option<Job>> {
459 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE id = $1").bind(id).fetch_optional(pool).await
460 
}
461
462 
/// The running job a job token belongs to, while it may still be used.
463 
pub async fn job_by_token(pool: &Pool, token_hash: &str, max_age_ms: i64) -> sqlx::Result<Option<Job>> {
464 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE token_hash = $1 AND status = 'running' AND started_ms >= $2")
465 
        .bind(token_hash)
466 
        .bind(now_ms() - max_age_ms)
467 
        .fetch_optional(pool)
468 
        .await
469 
}
470
471 
/// Take the oldest queued job whose dependencies all passed and whose tier
472 
/// the runner offers, and mark it running for `runner` with a token.
473 
/// Let a job's token also read these repositories (its pinned subrepos).
474 
pub async fn set_also_read(pool: &Pool, job_id: i64, repo_ids: &[i64]) -> sqlx::Result<()> {
475 
    let list: Vec<String> = repo_ids.iter().map(|i| i.to_string()).collect();
476 
    sqlx::query("UPDATE flow_jobs SET also_read = $1 WHERE id = $2").bind(list.join(",")).bind(job_id).execute(pool).await?;
477 
    Ok(())
478 
}
479
480 
impl Job {
481 
    pub fn also_read_ids(&self) -> Vec<i64> {
482 
        self.also_read.split(',').filter_map(|x| x.trim().parse().ok()).collect()
483 
    }
484 
}
485
486 
/// What a runner can take on: free memory (MiB) and vCPUs, `None` for no
487 
/// limit, and the rootfs images it holds, `None` when it does not say.
488 
#[derive(Clone, Debug, Default)]
489 
pub struct Room {
490 
    pub memory_mib: Option<u32>,
491 
    pub cpus: Option<u32>,
492 
    pub images: Option<std::collections::HashSet<String>>,
493 
    /// The runner's architecture (`x86_64`, `riscv64`, …), empty when unknown.
494 
    pub arch: String,
495 
    /// Each image's architecture as runners have reported it (`flow_images`).
496 
    pub image_archs: std::collections::HashMap<String, String>,
497 
    /// Jobs this runner declined, not to be offered to it again (spec §11.5).
498 
    pub declined: std::collections::HashSet<i64>,
499 
    /// Images the hub holds that this runner would fetch when it lacks
500 
    /// them (oxygenflow#39); empty for a runner that does not fetch.
501 
    pub fetchable: std::collections::HashSet<String>,
502 
    /// Its free disk now, from its claim's facts (oxygenflow#40); `None`
503 
    /// when it does not say.
504 
    pub free_disk_mib: Option<u64>,
505 
}
506
507 
impl Room {
508 
    /// Whether a job needing `need` MiB of disk fits in its free disk.
509 
    pub fn disk_fits(&self, need: Option<i64>) -> bool {
510 
        match (need, self.free_disk_mib) {
511 
            (Some(n), Some(free)) => n.max(0) as u64 <= free,
512 
            _ => true,
513 
        }
514 
    }
515 
}
516
517 
impl Room {
518 
    /// Whether a job's sandbox (its stored JSON) fits: its memory and vCPUs,
519 
    /// and its rootfs image among the runner's (oxygenhub#73).
520 
    pub fn fits(&self, sandbox_json: &str) -> bool {
521 
        let v: serde_json::Value = serde_json::from_str(sandbox_json).unwrap_or_default();
522 
        let need = |k: &str| v[k].as_u64().unwrap_or(0);
523 
        let rootfs = v["rootfs"].as_str().unwrap_or("");
524 
        // The architecture: the sandbox's own ask, else the image's as read
525 
        // or learned from its passes (oxygenhub#173); one nobody has read
526 
        // or run yet matches (OxygenFlow spec §11.3).
527 
        let wanted = v["arch"].as_str().map(String::from).or_else(|| self.image_archs.get(rootfs).cloned());
528 
        let arch_ok = self.arch.is_empty() || wanted.as_deref().is_none_or(|w| w == self.arch);
529 
        arch_ok
530 
            && self.memory_mib.is_none_or(|m| need("memory_mib") <= m as u64)
531 
            && self.cpus.is_none_or(|c| need("cpus") <= c as u64)
532 
            && (rootfs.is_empty() || self.images.as_ref().is_none_or(|have| have.contains(rootfs)) || self.fetchable.contains(rootfs))
533 
    }
534 
}
535
536 
/// Remember which architecture each image is for, as a runner reports it.
537 
pub async fn record_image_archs(pool: &Pool, archs: &std::collections::BTreeMap<String, String>) -> sqlx::Result<()> {
538 
    for (hash, arch) in archs.iter().take(500) {
539 
        if hash.len() != 64 || !flow_core::arch_ok(arch) {
540 
            continue;
541 
        }
542 
        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")
543 
        .bind(hash.to_ascii_lowercase())
544 
        .bind(arch)
545 
        .bind(now_ms())
546 
        .execute(pool)
547 
        .await?;
548 
    }
549 
    Ok(())
550 
}
551
552 
/// Give every image nobody has read the architecture of the runners its
553 
/// jobs passed on (oxygenhub#173), for the jobs before the hub learned at
554 
/// each pass. Never overrides a known architecture.
555 
pub async fn learn_archs_from_runs(pool: &Pool) -> sqlx::Result<u64> {
556 
    let rootfs = crate::db::json_text("j.sandbox", "rootfs");
557 
    let r = sqlx::query(&format!(
558 
        "INSERT INTO flow_images (hash, arch, seen_ms)
559 
         SELECT lower({rootfs}), MIN(r.arch), $1 FROM flow_jobs j JOIN flow_runners r ON r.id = j.runner_id
560 
         WHERE j.status = 'passed' AND r.arch != '' AND length({rootfs}) = 64
561 
         GROUP BY lower({rootfs})
562 
         ON CONFLICT (hash) DO NOTHING"
563 
    ))
564 
    .bind(now_ms())
565 
    .execute(pool)
566 
    .await?;
567 
    Ok(r.rows_affected())
568 
}
569
570 
/// Every image's architecture the hub knows.
571 
pub async fn image_archs(pool: &Pool) -> sqlx::Result<std::collections::HashMap<String, String>> {
572 
    let rows: Vec<(String, String)> = sqlx::query_as("SELECT hash, arch FROM flow_images").fetch_all(pool).await?;
573 
    Ok(rows.into_iter().collect())
574 
}
575
576 
pub async fn claim(pool: &Pool, runtimes: &[String], runner: &str, token_hash: &str) -> sqlx::Result<Option<Job>> {
577 
    claim_within(pool, runtimes, runner, token_hash, Room::default(), &Default::default(), &[]).await
578 
}
579
580 
/// The first queued job a runner with these tiers, this room and this
581 
/// scope may take, now running under `runner`. Repositories take turns
582 
/// (fair share, oxygenhub#244): a job's turn is its place in its own
583 
/// repository's queue plus that repository's running jobs, lowest first,
584 
/// oldest first among equals, so one repository's burst can't hold up the
585 
/// rest. A scoped runner never sees
586 
/// another scope's jobs (oxygenhub#41).
587 
pub async fn claim_within(
588 
    pool: &Pool,
589 
    runtimes: &[String],
590 
    runner: &str,
591 
    token_hash: &str,
592 
    room: Room,
593 
    scope: &super::runners::Scope,
594 
    labels: &[String],
595 
) -> sqlx::Result<Option<Job>> {
596 
    claim_ranked(pool, runtimes, runner, token_hash, room, scope, labels, &|_| false).await
597 
}
598
599 
/// [`claim_within`], passing over a job `elsewhere` says a better runner
600 
/// should take (oxygenflow#42).
601 
#[allow(clippy::too_many_arguments)]
602 
pub async fn claim_ranked(
603 
    pool: &Pool,
604 
    runtimes: &[String],
605 
    runner: &str,
606 
    token_hash: &str,
607 
    room: Room,
608 
    scope: &super::runners::Scope,
609 
    labels: &[String],
610 
    elsewhere: &(dyn Fn(&Job) -> bool + Sync),
611 
) -> sqlx::Result<Option<Job>> {
612 
    let mut tx = super::begin_write(pool).await?;
613 
    let queued: Vec<Job> = sqlx::query_as(&format!(
614 
        "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 \
615 
         WHERE j.status = 'queued' AND j.approval != 'waiting' AND r.status IN ('queued', 'running') \
616 
         AND ($1 = '' OR rep.owner = $2) AND ($3 IS NULL OR r.repo_id = $4) \
617 
         AND NOT ($5 = '' AND $6 IS NULL AND rep.flow = 'own') \
618 
         ORDER BY {PRIORITY_RANK}, ROW_NUMBER() OVER (PARTITION BY r.repo_id ORDER BY j.run_id, j.depth, j.id) \
619 
           + (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), \
620 
           j.run_id, j.depth, j.id LIMIT 200"
621 
    ))
622 
    .bind(&scope.owner)
623 
    .bind(&scope.owner)
624 
    .bind(scope.repo_id)
625 
    .bind(scope.repo_id)
626 
    // A repository on its own runners only (#225): the site's runners skip it.
627 
    .bind(&scope.owner)
628 
    .bind(scope.repo_id)
629 
    .fetch_all(&mut *tx)
630 
    .await?;
631 
    for j in queued {
632 
        if !runtimes.iter().any(|r| r == &j.runtime)
633 
            || !room.fits(&j.sandbox)
634 
            || !room.disk_fits(j.need_disk_mib)
635 
            || room.declined.contains(&j.id)
636 
        {
637 
            continue;
638 
        }
639 
        // Every label the job and its sandbox ask for (`runs-on`).
640 
        if !j.needs().iter().all(|l| labels.contains(l)) {
641 
            continue;
642 
        }
643 
        let deps = j.depends_on();
644 
        // When it became ready: queued, its dependencies passed, approved.
645 
        let mut ready = j.queued_ms.unwrap_or(0).max(j.approved_ms.unwrap_or(0));
646 
        if !deps.is_empty() {
647 
            let mut passed = 0i64;
648 
            for d in &deps {
649 
                let st: Option<(String, Option<i64>)> =
650 
                    sqlx::query_as("SELECT status, finished_ms FROM flow_jobs WHERE run_id = $1 AND name = $2")
651 
                        .bind(j.run_id)
652 
                        .bind(d)
653 
                        .fetch_optional(&mut *tx)
654 
                        .await?;
655 
                if let Some((status, finished)) = st {
656 
                    if status == "passed" {
657 
                        passed += 1;
658 
                        ready = ready.max(finished.unwrap_or(0));
659 
                    }
660 
                }
661 
            }
662 
            if passed != deps.len() as i64 {
663 
                continue;
664 
            }
665 
        }
666 
        if elsewhere(&j) {
667 
            continue;
668 
        }
669 
        let now = now_ms();
670 
        let done = sqlx::query(
671 
            "UPDATE flow_jobs SET status = 'running', runner = $1, token_hash = $2, started_ms = $3, waiting = '', ready_ms = $5 \
672 
             WHERE id = $4 AND status = 'queued'",
673 
        )
674 
        .bind(runner)
675 
        .bind(token_hash)
676 
        .bind(now)
677 
        .bind(j.id)
678 
        .bind(if ready > 0 { ready.min(now) } else { now })
679 
        .execute(&mut *tx)
680 
        .await?;
681 
        if done.rows_affected() == 1 {
682 
            sqlx::query("UPDATE flow_runs SET status = 'running' WHERE id = $1 AND status = 'queued'")
683 
                .bind(j.run_id)
684 
                .execute(&mut *tx)
685 
                .await?;
686 
            tx.commit().await?;
687 
            return job(pool, j.id).await;
688 
        }
689 
    }
690 
    tx.commit().await?;
691 
    Ok(None)
692 
}
693
694 
/// How long a decline keeps a job from its runner when the runner's
695 
/// profile stays the same: a reason such as memory in use passes.
696 
pub const DECLINE_HOLDS_MS: i64 = 10 * 60_000;
697
698 
/// The jobs runner `runner_id` declined with this profile, recently.
699 
pub async fn declined_by(pool: &Pool, runner_id: i64, profile: &str) -> sqlx::Result<std::collections::HashSet<i64>> {
700 
    let rows: Vec<(i64,)> = sqlx::query_as("SELECT job_id FROM flow_declines WHERE runner_id = $1 AND profile = $2 AND ms > $3")
701 
        .bind(runner_id)
702 
        .bind(profile)
703 
        .bind(now_ms() - DECLINE_HOLDS_MS)
704 
        .fetch_all(pool)
705 
        .await?;
706 
    Ok(rows.into_iter().map(|(j,)| j).collect())
707 
}
708
709 
/// A decline: one runner's reason for handing a job back.
710 
#[derive(Clone, Debug, sqlx::FromRow)]
711 
pub struct DeclineRow {
712 
    pub job_id: i64,
713 
    pub runner_id: i64,
714 
    pub runner: String,
715 
    pub reason: String,
716 
    pub ms: i64,
717 
}
718
719 
/// Runner `runner_id` hands back job `job_id`, running on it: the decline
720 
/// is kept (one per job and runner, the latest), and the job is queued
721 
/// again. Returns every decline the job has had, or `None` when the job
722 
/// was not running on that runner.
723 
pub async fn decline(
724 
    pool: &Pool,
725 
    job_id: i64,
726 
    runner_id: i64,
727 
    runner: &str,
728 
    profile: &str,
729 
    reason: &str,
730 
) -> sqlx::Result<Option<Vec<DeclineRow>>> {
731 
    let mut tx = super::begin_write(pool).await?;
732 
    let back = sqlx::query(
733 
        "UPDATE flow_jobs SET status = 'queued', runner = '', token_hash = '', started_ms = NULL, runner_id = NULL \
734 
         WHERE id = $1 AND status = 'running' AND runner_id = $2",
735 
    )
736 
    .bind(job_id)
737 
    .bind(runner_id)
738 
    .execute(&mut *tx)
739 
    .await?;
740 
    if back.rows_affected() != 1 {
741 
        tx.rollback().await?;
742 
        return Ok(None);
743 
    }
744 
    sqlx::query(
745 
        "INSERT INTO flow_declines (job_id, runner_id, runner, profile, reason, ms) VALUES ($1, $2, $3, $4, $5, $6) \
746 
         ON CONFLICT (job_id, runner_id) DO UPDATE SET runner = excluded.runner, profile = excluded.profile, reason = excluded.reason, ms = excluded.ms",
747 
    )
748 
    .bind(job_id)
749 
    .bind(runner_id)
750 
    .bind(runner)
751 
    .bind(profile)
752 
    .bind(reason)
753 
    .bind(now_ms())
754 
    .execute(&mut *tx)
755 
    .await?;
756 
    let all: Vec<DeclineRow> =
757 
        sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = $1 ORDER BY ms")
758 
            .bind(job_id)
759 
            .fetch_all(&mut *tx)
760 
            .await?;
761 
    tx.commit().await?;
762 
    Ok(Some(all))
763 
}
764
765 
/// The declines of job `job_id`, oldest first.
766 
pub async fn declines_of_job(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<DeclineRow>> {
767 
    sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = $1 ORDER BY ms")
768 
        .bind(job_id)
769 
        .fetch_all(pool)
770 
        .await
771 
}
772
773 
/// A runner's latest declines, newest first, with where the job is.
774 
#[derive(Clone, Debug, sqlx::FromRow)]
775 
pub struct RunnerDecline {
776 
    pub job_id: i64,
777 
    pub job: String,
778 
    pub run_number: i64,
779 
    pub owner: String,
780 
    pub name: String,
781 
    pub reason: String,
782 
    pub ms: i64,
783 
}
784
785 
pub async fn declines_of_runner(pool: &Pool, runner_id: i64, limit: i64) -> sqlx::Result<Vec<RunnerDecline>> {
786 
    sqlx::query_as(
787 
        "SELECT d.job_id, j.name AS job, r.number AS run_number, rep.owner, rep.name, d.reason, d.ms \
788 
         FROM flow_declines d JOIN flow_jobs j ON j.id = d.job_id JOIN flow_runs r ON r.id = j.run_id \
789 
         JOIN repositories rep ON rep.id = r.repo_id WHERE d.runner_id = $1 ORDER BY d.ms DESC LIMIT $2",
790 
    )
791 
    .bind(runner_id)
792 
    .bind(limit)
793 
    .fetch_all(pool)
794 
    .await
795 
}
796
797 
/// Queued jobs whose dependencies have passed, in their live runs, with
798 
/// their repositories (owner, id): the ones a runner could take now.
799 
pub async fn ready_queued(pool: &Pool) -> sqlx::Result<Vec<(Job, String, i64)>> {
800 
    let queued: Vec<Job> = sqlx::query_as(
801 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
802 
         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",
803 
    )
804 
    .fetch_all(pool)
805 
    .await?;
806 
    let mut out = Vec::new();
807 
    for j in queued {
808 
        let mut ready = true;
809 
        for d in j.depends_on() {
810 
            let st: Option<String> = sqlx::query_scalar("SELECT status FROM flow_jobs WHERE run_id = $1 AND name = $2")
811 
                .bind(j.run_id)
812 
                .bind(&d)
813 
                .fetch_optional(pool)
814 
                .await?;
815 
            if st.as_deref() != Some("passed") {
816 
                ready = false;
817 
                break;
818 
            }
819 
        }
820 
        if !ready {
821 
            continue;
822 
        }
823 
        let repo: Option<(String, i64)> =
824 
            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")
825 
                .bind(j.run_id)
826 
                .fetch_optional(pool)
827 
                .await?;
828 
        if let Some((owner, id)) = repo {
829 
            out.push((j, owner, id));
830 
        }
831 
    }
832 
    Ok(out)
833 
}
834
835 
pub async fn set_waiting(pool: &Pool, job_id: i64, waiting: &str) -> sqlx::Result<()> {
836 
    sqlx::query("UPDATE flow_jobs SET waiting = $1 WHERE id = $2 AND waiting <> $3")
837 
        .bind(waiting)
838 
        .bind(job_id)
839 
        .bind(waiting)
840 
        .execute(pool)
841 
        .await?;
842 
    Ok(())
843 
}
844
845 
/// Append records; returns how many were kept (the rest exceed the cap).
846 
pub async fn append_log(pool: &Pool, job_id: i64, records: &[flow_core::api::Record], cap: i64) -> sqlx::Result<usize> {
847 
    let mut tx = super::begin_write(pool).await?;
848 
    let count: i64 = sqlx::query_scalar("SELECT log_count FROM flow_jobs WHERE id = $1").bind(job_id).fetch_one(&mut *tx).await?;
849 
    let room = (cap - count).max(0) as usize;
850 
    let keep = &records[..records.len().min(room)];
851 
    for (i, r) in keep.iter().enumerate() {
852 
        sqlx::query("INSERT INTO flow_logs (job_id, seq, t, stream, text) VALUES ($1, $2, $3, $4, $5)")
853 
            .bind(job_id)
854 
            .bind(count + i as i64 + 1)
855 
            .bind(r.t as i64)
856 
            .bind(&r.stream)
857 
            .bind(&r.text)
858 
            .execute(&mut *tx)
859 
            .await?;
860 
    }
861 
    sqlx::query("UPDATE flow_jobs SET log_count = log_count + $1 WHERE id = $2")
862 
        .bind(keep.len() as i64)
863 
        .bind(job_id)
864 
        .execute(&mut *tx)
865 
        .await?;
866 
    tx.commit().await?;
867 
    Ok(keep.len())
868 
}
869
870 
pub async fn log(pool: &Pool, job_id: i64, after: i64, limit: i64) -> sqlx::Result<Vec<LogLine>> {
871 
    sqlx::query_as("SELECT seq, t, stream, text FROM flow_logs WHERE job_id = $1 AND seq > $2 ORDER BY seq LIMIT $3")
872 
        .bind(job_id)
873 
        .bind(after)
874 
        .bind(limit)
875 
        .fetch_all(pool)
876 
        .await
877 
}
878
879 
/// Jobs running since before `ms` (the oldest first).
880 
/// Jobs still running at `now` past their own time limit (`default_s` for
881 
/// one claimed without) plus `grace_ms`.
882 
pub async fn running_past_limit(pool: &Pool, now: i64, default_s: u64, grace_ms: i64) -> sqlx::Result<Vec<Job>> {
883 
    sqlx::query_as(
884 
        "SELECT * FROM flow_jobs WHERE status = 'running' \
885 
         AND COALESCE(started_ms, 0) + (CASE WHEN timeout_s > 0 THEN timeout_s ELSE $1 END) * 1000 + $2 < $3 ORDER BY started_ms",
886 
    )
887 
    .bind(default_s as i64)
888 
    .bind(grace_ms)
889 
    .bind(now)
890 
    .fetch_all(pool)
891 
    .await
892 
}
893
894 
/// Record the limit a claimed job runs under.
895 
pub async fn set_job_timeout(pool: &Pool, job_id: i64, timeout_s: u64) -> sqlx::Result<()> {
896 
    sqlx::query("UPDATE flow_jobs SET timeout_s = $1 WHERE id = $2").bind(timeout_s as i64).bind(job_id).execute(pool).await?;
897 
    Ok(())
898 
}
899
900 
pub async fn running_since_before(pool: &Pool, ms: i64) -> sqlx::Result<Vec<Job>> {
901 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND COALESCE(started_ms, 0) < $1 ORDER BY started_ms")
902 
        .bind(ms)
903 
        .fetch_all(pool)
904 
        .await
905 
}
906
907 
/// The jobs running on the runner with this id.
908 
pub async fn running_on(pool: &Pool, runner_id: i64) -> sqlx::Result<Vec<Job>> {
909 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND runner_id = $1 ORDER BY id").bind(runner_id).fetch_all(pool).await
910 
}
911
912 
pub struct Outcome<'a> {
913 
    pub status: &'a str,
914 
    pub exit_code: Option<i64>,
915 
    pub error: &'a str,
916 
    pub receipt: &'a str,
917 
    pub receipt_hash: &'a str,
918 
}
919
920 
/// Record a job's end, skip the jobs that can no longer run, and settle the
921 
/// run. Returns the jobs whose status changed (for their checks).
922 
pub async fn finish(pool: &Pool, job_id: i64, o: Outcome<'_>) -> sqlx::Result<Vec<Job>> {
923 
    let mut tx = super::begin_write(pool).await?;
924 
    let now = now_ms();
925 
    let j: Job = sqlx::query_as("SELECT * FROM flow_jobs WHERE id = $1").bind(job_id).fetch_one(&mut *tx).await?;
926 
    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")
927 
        .bind(o.status)
928 
        .bind(o.exit_code)
929 
        .bind(o.error)
930 
        .bind(o.receipt)
931 
        .bind(o.receipt_hash)
932 
        .bind(now)
933 
        .bind(job_id)
934 
        .execute(&mut *tx)
935 
        .await?;
936 
    // A pass teaches what its image is for, when nobody has read it
937 
    // (oxygenhub#173): the architecture of the runner it passed on.
938 
    if o.status == "passed" {
939 
        let rootfs = serde_json::from_str::<serde_json::Value>(&j.sandbox)
940 
            .ok()
941 
            .and_then(|v| v["rootfs"].as_str().map(str::to_ascii_lowercase))
942 
            .unwrap_or_default();
943 
        if rootfs.len() == 64 {
944 
            sqlx::query(
945 
                "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 \
946 
                 WHERE j.id = $3 AND r.arch != '' ON CONFLICT (hash) DO NOTHING",
947 
            )
948 
            .bind(&rootfs)
949 
            .bind(now)
950 
            .bind(job_id)
951 
            .execute(&mut *tx)
952 
            .await?;
953 
        }
954 
    }
955 
    let mut changed = vec![job_id];
956 
    // Dependents of anything that did not pass are skipped, transitively.
957 
    loop {
958 
        let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(j.run_id).fetch_all(&mut *tx).await?;
959 
        let blocked: Vec<i64> = all
960 
            .iter()
961 
            .filter(|x| x.status == "queued")
962 
            .filter(|x| {
963 
                x.depends_on()
964 
                    .iter()
965 
                    .any(|d| all.iter().any(|y| &y.name == d && matches!(y.status.as_str(), "failed" | "skipped" | "error" | "cancelled")))
966 
            })
967 
            .map(|x| x.id)
968 
            .collect();
969 
        if blocked.is_empty() {
970 
            break;
971 
        }
972 
        for id in blocked {
973 
            sqlx::query("UPDATE flow_jobs SET status = 'skipped', finished_ms = $1 WHERE id = $2")
974 
                .bind(now)
975 
                .bind(id)
976 
                .execute(&mut *tx)
977 
                .await?;
978 
            changed.push(id);
979 
        }
980 
    }
981 
    let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(j.run_id).fetch_all(&mut *tx).await?;
982 
    if all.iter().all(Job::done) {
983 
        let status = if all.iter().all(|x| x.status == "passed") { "passed" } else { "failed" };
984 
        sqlx::query("UPDATE flow_runs SET status = $1, finished_ms = $2 WHERE id = $3")
985 
            .bind(status)
986 
            .bind(now)
987 
            .bind(j.run_id)
988 
            .execute(&mut *tx)
989 
            .await?;
990 
    }
991 
    tx.commit().await?;
992 
    let mut out = Vec::new();
993 
    for id in changed {
994 
        if let Some(x) = job(pool, id).await? {
995 
            out.push(x);
996 
        }
997 
    }
998 
    Ok(out)
999 
}
1000
1001 
/// Cancel a run (oxygenflow#32): its jobs not yet done end as cancelled,
1002 
/// and so does the run. A runner learns of it when the hub next refuses
1003 
/// its log lines, and stops the job. Returns the jobs whose status changed.
1004 
pub async fn cancel_run(pool: &Pool, run_id: i64, why: &str) -> sqlx::Result<Vec<Job>> {
1005 
    let now = now_ms();
1006 
    let mut tx = pool.begin().await?;
1007 
    let ids: Vec<(i64,)> = sqlx::query_as("SELECT id FROM flow_jobs WHERE run_id = $1 AND status IN ('queued', 'running')")
1008 
        .bind(run_id)
1009 
        .fetch_all(&mut *tx)
1010 
        .await?;
1011 
    sqlx::query(
1012 
        "UPDATE flow_jobs SET status = 'cancelled', error = $1, finished_ms = $2, token_hash = ''
1013 
         WHERE run_id = $3 AND status IN ('queued', 'running')",
1014 
    )
1015 
    .bind(why)
1016 
    .bind(now)
1017 
    .bind(run_id)
1018 
    .execute(&mut *tx)
1019 
    .await?;
1020 
    sqlx::query("UPDATE flow_runs SET status = 'cancelled', finished_ms = $1 WHERE id = $2 AND status IN ('queued', 'running')")
1021 
        .bind(now)
1022 
        .bind(run_id)
1023 
        .execute(&mut *tx)
1024 
        .await?;
1025 
    tx.commit().await?;
1026 
    let mut out = Vec::new();
1027 
    for (id,) in ids {
1028 
        if let Some(x) = job(pool, id).await? {
1029 
            out.push(x);
1030 
        }
1031 
    }
1032 
    Ok(out)
1033 
}
1034
1035 
/// Run job `name` of finished run `run_id` again, with every job that
1036 
/// depends on it (oxygenflow#31): back to the queue as if new, their logs,
1037 
/// artifacts and reports gone, and the run running again. The jobs they
1038 
/// depend on keep their results. The jobs reset; none when there is no
1039 
/// such job.
1040 
pub async fn rerun_job(pool: &Pool, run_id: i64, name: &str) -> sqlx::Result<Vec<Job>> {
1041 
    let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(run_id).fetch_all(pool).await?;
1042 
    if !all.iter().any(|j| j.name == name) {
1043 
        return Ok(Vec::new());
1044 
    }
1045 
    let mut again = vec![name.to_string()];
1046 
    loop {
1047 
        let more: Vec<String> = all
1048 
            .iter()
1049 
            .filter(|j| !again.contains(&j.name) && j.depends_on().iter().any(|d| again.contains(d)))
1050 
            .map(|j| j.name.clone())
1051 
            .collect();
1052 
        if more.is_empty() {
1053 
            break;
1054 
        }
1055 
        again.extend(more);
1056 
    }
1057 
    let ids: Vec<i64> = all.iter().filter(|j| again.contains(&j.name)).map(|j| j.id).collect();
1058 
    let mut tx = pool.begin().await?;
1059 
    for id in &ids {
1060 
        for table in ["flow_logs", "flow_artifacts", "flow_declines", "test_reports", "coverage_reports", "reports"] {
1061 
            sqlx::query(&format!("DELETE FROM {table} WHERE job_id = $1")).bind(id).execute(&mut *tx).await?;
1062 
        }
1063 
        sqlx::query(
1064 
            "UPDATE flow_jobs SET status = 'queued', runner = '', token_hash = '', started_ms = NULL, finished_ms = NULL,
1065 
             exit_code = NULL, error = '', receipt = '', receipt_hash = '', log_count = 0, also_read = '', receipt_attest = '',
1066 
             receipt_signer = '', receipt_note = '', runner_id = NULL, waiting = '', placed = '', peak_disk_mib = NULL,
1067 
             peak_memory_mib = NULL, approval = CASE WHEN approval = '' THEN '' ELSE 'waiting' END, approved_by = '',
1068 
             approved_ms = NULL, approval_seal = '', approved_via = '', queued_ms = $2, ready_ms = NULL WHERE id = $1",
1069 
        )
1070 
        .bind(id)
1071 
        .bind(now_ms())
1072 
        .execute(&mut *tx)
1073 
        .await?;
1074 
    }
1075 
    sqlx::query("UPDATE flow_runs SET status = 'running', finished_ms = NULL WHERE id = $1").bind(run_id).execute(&mut *tx).await?;
1076 
    tx.commit().await?;
1077 
    let mut out = Vec::new();
1078 
    for id in ids {
1079 
        if let Some(x) = job(pool, id).await? {
1080 
            out.push(x);
1081 
        }
1082 
    }
1083 
    Ok(out)
1084 
}
1085
1086 
/// The runs of `bookmark` started by `trigger` still going, older than run
1087 
/// `number`: those a newer run supersedes (#267). With `running_too`
1088 
/// false, only those still queued.
1089 
pub async fn live_runs_before(
1090 
    pool: &Pool,
1091 
    repo_id: i64,
1092 
    bookmark: &str,
1093 
    trigger: &str,
1094 
    number: i64,
1095 
    running_too: bool,
1096 
) -> sqlx::Result<Vec<Run>> {
1097 
    sqlx::query_as(
1098 
        "SELECT * FROM flow_runs WHERE repo_id = $1 AND bookmark = $2 AND number < $3 AND trigger = $4
1099 
         AND (status = 'queued' OR ($5 AND status = 'running')) ORDER BY id",
1100 
    )
1101 
    .bind(repo_id)
1102 
    .bind(bookmark)
1103 
    .bind(number)
1104 
    .bind(trigger)
1105 
    .bind(running_too)
1106 
    .fetch_all(pool)
1107 
    .await
1108 
}
1109
1110 
/// The newest run of exactly this state (heads joined with `+`, sorted).
1111 
pub async fn latest_for_heads(pool: &Pool, repo_id: i64, heads: &str) -> sqlx::Result<Option<Run>> {
1112 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND heads = $2 ORDER BY id DESC LIMIT 1")
1113 
        .bind(repo_id)
1114 
        .bind(heads)
1115 
        .fetch_optional(pool)
1116 
        .await
1117 
}
1118
1119 
/// Record what became of job `job_id`'s signed receipt.
1120 
pub async fn set_receipt(pool: &Pool, job_id: i64, attest: &str, signer: &str, note: &str) -> sqlx::Result<()> {
1121 
    sqlx::query("UPDATE flow_jobs SET receipt_attest = $1, receipt_signer = $2, receipt_note = $3 WHERE id = $4")
1122 
        .bind(attest)
1123 
        .bind(signer)
1124 
        .bind(note)
1125 
        .bind(job_id)
1126 
        .execute(pool)
1127 
        .await?;
1128 
    Ok(())
1129 
}
1130
1131 
/// A job's artifact, kept as a blob of the run's repository.
1132 
#[derive(Clone, Debug, FromRow)]
1133 
pub struct Artifact {
1134 
    pub job_id: i64,
1135 
    pub name: String,
1136 
    pub path: String,
1137 
    pub blob: String,
1138 
    pub size: i64,
1139 
    /// BLAKE3 of the bytes.
1140 
    pub content: String,
1141 
    pub created_ms: i64,
1142 
}
1143
1144 
pub async fn add_artifact(pool: &Pool, a: &Artifact) -> sqlx::Result<()> {
1145 
    sqlx::query(
1146 
        "INSERT OR REPLACE INTO flow_artifacts (job_id, name, path, blob, size, content, created_ms) VALUES ($1, $2, $3, $4, $5, $6, $7)",
1147 
    )
1148 
    .bind(a.job_id)
1149 
    .bind(&a.name)
1150 
    .bind(&a.path)
1151 
    .bind(&a.blob)
1152 
    .bind(a.size)
1153 
    .bind(&a.content)
1154 
    .bind(a.created_ms)
1155 
    .execute(pool)
1156 
    .await?;
1157 
    Ok(())
1158 
}
1159
1160 
/// Every artifact blob a repository's runs keep, for garbage collection to
1161 
/// keep too (#126): the database is the only thing that names them.
1162 
pub async fn artifact_blobs(pool: &Pool, repo_id: i64) -> sqlx::Result<Vec<String>> {
1163 
    sqlx::query_scalar(
1164 
        "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",
1165 
    )
1166 
    .bind(repo_id)
1167 
    .fetch_all(pool)
1168 
    .await
1169 
}
1170
1171 
/// Job `job_id`'s artifacts, by name.
1172 
pub async fn artifacts(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<Artifact>> {
1173 
    sqlx::query_as("SELECT * FROM flow_artifacts WHERE job_id = $1 ORDER BY name").bind(job_id).fetch_all(pool).await
1174 
}
1175
1176 
/// An artifact with the run and job that made it, for a repository's list
1177 
/// of every run's outputs (oxygenhub#254).
1178 
#[derive(Clone, Debug, FromRow)]
1179 
pub struct RepoArtifact {
1180 
    pub run_number: i64,
1181 
    pub bookmark: String,
1182 
    pub job: String,
1183 
    pub name: String,
1184 
    pub path: String,
1185 
    pub size: i64,
1186 
    pub content: String,
1187 
    pub created_ms: i64,
1188 
}
1189
1190 
/// A repository's newest artifacts, at most `limit`.
1191 
pub async fn repo_artifacts(pool: &Pool, repo_id: i64, limit: i64) -> sqlx::Result<Vec<RepoArtifact>> {
1192 
    sqlx::query_as(
1193 
        "SELECT r.number AS run_number, r.bookmark, j.name AS job, a.name, a.path, a.size, a.content, a.created_ms
1194 
         FROM flow_artifacts a JOIN flow_jobs j ON j.id = a.job_id JOIN flow_runs r ON r.id = j.run_id
1195 
         WHERE r.repo_id = $1 ORDER BY a.created_ms DESC, r.number DESC, j.name, a.name LIMIT $2",
1196 
    )
1197 
    .bind(repo_id)
1198 
    .bind(limit)
1199 
    .fetch_all(pool)
1200 
    .await
1201 
}
1202
1203 
/// Where an artifact with given content came from: the run and job that
1204 
/// built it, and the job's signed receipt when one was kept.
1205 
#[derive(Clone, Debug, FromRow)]
1206 
pub struct Provenance {
1207 
    pub run_number: i64,
1208 
    pub job: String,
1209 
    pub name: String,
1210 
    pub receipt_attest: String,
1211 
    pub receipt_signer: String,
1212 
}
1213
1214 
/// The build of the bytes whose BLAKE3 is `content` in repository
1215 
/// `repo_id`: a passed job, one with a kept signed receipt first, newest
1216 
/// run first.
1217 
pub async fn provenance(pool: &Pool, repo_id: i64, content: &str) -> sqlx::Result<Option<Provenance>> {
1218 
    sqlx::query_as(
1219 
        "SELECT r.number AS run_number, j.name AS job, a.name AS name, j.receipt_attest, j.receipt_signer \
1220 
         FROM flow_artifacts a JOIN flow_jobs j ON j.id = a.job_id JOIN flow_runs r ON r.id = j.run_id \
1221 
         WHERE r.repo_id = $1 AND a.content = $2 AND j.status = 'passed' \
1222 
         ORDER BY (j.receipt_attest != '') DESC, r.number DESC LIMIT 1",
1223 
    )
1224 
    .bind(repo_id)
1225 
    .bind(content)
1226 
    .fetch_optional(pool)
1227 
    .await
1228 
}
1229
1230 
/// Approve job `id` if it is still waiting for a person (oxygenflow#57);
1231 
/// whether it was.
1232 
pub async fn approve_job(pool: &Pool, id: i64, who: &str, seal: &str, via: &str) -> sqlx::Result<bool> {
1233 
    let r = sqlx::query(
1234 
        "UPDATE flow_jobs SET approval = 'approved', approved_by = $1, approved_ms = $2, approval_seal = $3, approved_via = $5 \
1235 
         WHERE id = $4 AND status = 'queued' AND approval = 'waiting'",
1236 
    )
1237 
    .bind(who)
1238 
    .bind(now_ms())
1239 
    .bind(seal)
1240 
    .bind(id)
1241 
    .bind(via)
1242 
    .execute(pool)
1243 
    .await?;
1244 
    Ok(r.rows_affected() > 0)
1245 
}
1246
1247 
/// Note who rejected job `id` (it is then finished as cancelled).
1248 
pub async fn note_rejection(pool: &Pool, id: i64, who: &str, via: &str) -> sqlx::Result<()> {
1249 
    sqlx::query("UPDATE flow_jobs SET approved_by = $1, approved_ms = $2, approved_via = $4 WHERE id = $3")
1250 
        .bind(who)
1251 
        .bind(now_ms())
1252 
        .bind(id)
1253 
        .bind(via)
1254 
        .execute(pool)
1255 
        .await?;
1256 
    Ok(())
1257 
}
1258
1259 
/// Jobs waiting for a person, in runs still going.
1260 
pub async fn waiting_for_approval(pool: &Pool) -> sqlx::Result<Vec<Job>> {
1261 
    sqlx::query_as(
1262 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
1263 
         WHERE j.status = 'queued' AND j.approval = 'waiting' AND r.status IN ('queued', 'running')",
1264 
    )
1265 
    .fetch_all(pool)
1266 
    .await
1267 
}
1268
1269 
/// One deployment: a job for an environment that passed.
1270 
#[derive(Clone, Debug, FromRow)]
1271 
pub struct Deployment {
1272 
    pub environment: String,
1273 
    pub job: String,
1274 
    pub run_number: i64,
1275 
    pub state: String,
1276 
    pub bookmark: String,
1277 
    pub actor: String,
1278 
    pub approved_by: String,
1279 
    pub finished_ms: i64,
1280 
}
1281
1282 
/// A repository's deployments, newest first (oxygenflow#59).
1283 
pub async fn deployments(pool: &Pool, repo_id: i64, limit: i64) -> sqlx::Result<Vec<Deployment>> {
1284 
    sqlx::query_as(
1285 
        "SELECT j.environment, j.name AS job, r.number AS run_number, r.state, r.bookmark, r.actor, j.approved_by, \
1286 
         COALESCE(j.finished_ms, 0) AS finished_ms FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
1287 
         WHERE r.repo_id = $1 AND j.environment != '' AND j.status = 'passed' ORDER BY j.finished_ms DESC LIMIT $2",
1288 
    )
1289 
    .bind(repo_id)
1290 
    .bind(limit)
1291 
    .fetch_all(pool)
1292 
    .await
1293 
}