OXYGENHUB
# 76f93eaefe45 oxygenhub / crates / hub / src / db / flow.rs
45.3 KiB · Rust · last changed Flow priorities: an organisation's level, and a cap on expedited runs 19 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`, `manual` or `rerun:N` (migration 0020).
14 
    #[sqlx(default)]
15 
    pub trigger: String,
16 
    pub bookmark: String,
17 
    pub state: String,
18 
    pub heads: String,
19 
    pub pipeline_hash: String,
20 
    pub status: String,
21 
    pub error: String,
22 
    pub actor: String,
23 
    pub created_ms: i64,
24 
    pub finished_ms: Option<i64>,
25 
    /// `urgent` when expedited (oxygenhub#244), else empty: the run follows
26 
    /// its repository's, organisation's or the site's level.
27 
    #[sqlx(default)]
28 
    pub priority: String,
29 
    /// Who expedited it and why.
30 
    #[sqlx(default)]
31 
    pub priority_note: String,
32 
}
33
34 
/// The priority levels, most urgent first (oxygenhub#244).
35 
pub const PRIORITIES: [&str; 4] = ["urgent", "high", "normal", "low"];
36
37 
/// SQL for a queued job's level as a rank (0 urgent … 3 low): the run's
38 
/// own (expedited), else its repository's, its organisation's, the site's
39 
/// (`settings.flow_priority`), else normal. Needs `r` (the run) and `rep`
40 
/// (its repository) in the query.
41 
pub const PRIORITY_RANK: &str = "CASE COALESCE(NULLIF(r.priority, ''), NULLIF(rep.flow_priority, ''), \
42 
       NULLIF((SELECT o.flow_priority FROM organisations o WHERE o.name = rep.owner), ''), \
43 
       NULLIF((SELECT s.value FROM settings s WHERE s.key = 'flow_priority'), ''), 'normal') \
44 
     WHEN 'urgent' THEN 0 WHEN 'high' THEN 1 WHEN 'low' THEN 3 ELSE 2 END";
45
46 
/// How many runs of `owner`'s repositories are expedited and still live.
47 
pub async fn expedited_live(pool: &Pool, owner: &str) -> sqlx::Result<i64> {
48 
    sqlx::query_scalar(
49 
        "SELECT COUNT(*) FROM flow_runs r JOIN repositories rep ON rep.id = r.repo_id \
50 
         WHERE rep.owner = $1 AND r.priority = 'urgent' AND r.status IN ('queued', 'running')",
51 
    )
52 
    .bind(owner)
53 
    .fetch_one(pool)
54 
    .await
55 
}
56
57 
/// Raise run `run_id` to urgent, saying who and why; a run already ended
58 
/// is left alone. Whether it changed.
59 
pub async fn expedite(pool: &Pool, run_id: i64, note: &str) -> sqlx::Result<bool> {
60 
    let r = sqlx::query("UPDATE flow_runs SET priority = 'urgent', priority_note = $1 WHERE id = $2 AND status IN ('queued', 'running')")
61 
        .bind(note)
62 
        .bind(run_id)
63 
        .execute(pool)
64 
        .await?;
65 
    Ok(r.rows_affected() > 0)
66 
}
67
68 
impl Run {
69 
    /// How the run page says what started it.
70 
    pub fn started_by(&self) -> String {
71 
        match self.trigger.as_str() {
72 
            "manual" => format!("started by {}", self.actor),
73 
            "schedule" => "started on schedule".to_string(),
74 
            t if t.starts_with("tag:") => format!("tag {} pushed by {}", &t[4..], self.actor),
75 
            t if t.starts_with("rerun:") => format!("re-run of #{} by {}", self.rerun_of().unwrap_or(""), self.actor),
76 
            _ => format!("pushed by {}", self.actor),
77 
        }
78 
    }
79
80 
    /// The run a rerun repeats: `N` of `rerun:N` or `rerun:N:EVENT`.
81 
    pub fn rerun_of(&self) -> Option<&str> {
82 
        let rest = self.trigger.strip_prefix("rerun:")?;
83 
        Some(rest.split(':').next().unwrap_or(rest))
84 
    }
85
86 
    /// The trigger as people read it: `rerun:N`, without the event it repeats.
87 
    pub fn trigger_label(&self) -> String {
88 
        match self.rerun_of() {
89 
            Some(n) => format!("rerun:{n}"),
90 
            None => self.trigger.clone(),
91 
        }
92 
    }
93
94 
    pub fn head_list(&self) -> Vec<&str> {
95 
        self.heads.split('+').filter(|h| !h.is_empty()).collect()
96 
    }
97
98 
    pub fn live(&self) -> bool {
99 
        matches!(self.status.as_str(), "queued" | "running")
100 
    }
101 
}
102
103 
#[derive(Clone, Debug, Default, FromRow)]
104 
pub struct Job {
105 
    pub id: i64,
106 
    pub run_id: i64,
107 
    pub name: String,
108 
    pub depth: i64,
109 
    pub depends: String,
110 
    pub spec: String,
111 
    pub sandbox: String,
112 
    pub runtime: String,
113 
    pub status: String,
114 
    pub runner: String,
115 
    pub token_hash: String,
116 
    pub started_ms: Option<i64>,
117 
    pub finished_ms: Option<i64>,
118 
    pub exit_code: Option<i64>,
119 
    pub error: String,
120 
    pub receipt: String,
121 
    pub receipt_hash: String,
122 
    pub log_count: i64,
123 
    /// Comma-separated ids of other repositories the job token may read.
124 
    pub also_read: String,
125 
    /// The receipt attestation kept in the repository (Oxygen spec §14.6).
126 
    pub receipt_attest: String,
127 
    /// Who signed it: the runner's identity name.
128 
    pub receipt_signer: String,
129 
    /// Why a signed receipt was not kept.
130 
    pub receipt_note: String,
131 
    /// Why a queued job no runner can take waits (spec §11.4), from the
132 
    /// last placement pass; empty when it is only waiting its turn.
133 
    pub waiting: String,
134 
    /// What it took, as its runner reported (oxygenflow#41), in MiB.
135 
    pub peak_disk_mib: Option<i64>,
136 
    pub peak_memory_mib: Option<i64>,
137 
    /// The free disk it needs, learned from past runs of the same job.
138 
    pub need_disk_mib: Option<i64>,
139 
    /// Why it ran on its runner (oxygenflow#42).
140 
    pub placed: String,
141 
    /// `waiting` for a person, `approved`, or empty (oxygenflow#57).
142 
    pub approval: String,
143 
    /// Who approved or rejected it, and when.
144 
    pub approved_by: String,
145 
    pub approved_ms: Option<i64>,
146 
    /// The environment it deploys to, or empty (oxygenflow#59).
147 
    pub environment: String,
148 
    /// The approval attestation that released it, when sealed (#235).
149 
    pub approval_seal: String,
150 
}
151
152 
impl Job {
153 
    pub fn depends_on(&self) -> Vec<String> {
154 
        serde_json::from_str(&self.depends).unwrap_or_default()
155 
    }
156
157 
    pub fn done(&self) -> bool {
158 
        matches!(self.status.as_str(), "passed" | "failed" | "skipped" | "error" | "cancelled")
159 
    }
160 
}
161
162 
#[derive(Clone, Debug, FromRow)]
163 
pub struct LogLine {
164 
    pub seq: i64,
165 
    pub t: i64,
166 
    pub stream: String,
167 
    pub text: String,
168 
}
169
170 
/// A job as the trigger creates it.
171 
pub struct NewJob {
172 
    pub name: String,
173 
    pub depth: i64,
174 
    pub depends: Vec<String>,
175 
    pub spec: String,
176 
    pub sandbox: String,
177 
    pub runtime: String,
178 
    /// The free disk it needs, from [`learned_disk_need`].
179 
    pub need_disk_mib: Option<i64>,
180 
}
181
182 
/// Runs of a job whose peaks set its next need.
183 
const NEED_FROM_RUNS: i64 = 5;
184
185 
/// The free disk a new run of job `name` in repository `repo_id` needs
186 
/// (oxygenflow#41): the most its last runs took, plus a quarter; `None`
187 
/// for a job that has not run here yet.
188 
pub async fn learned_disk_need(pool: &Pool, repo_id: i64, name: &str) -> sqlx::Result<Option<i64>> {
189 
    let peak: Option<i64> = sqlx::query_scalar(
190 
        "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 \
191 
         WHERE r.repo_id = $1 AND j.name = $2 AND j.peak_disk_mib IS NOT NULL ORDER BY j.id DESC LIMIT $3)",
192 
    )
193 
    .bind(repo_id)
194 
    .bind(name)
195 
    .bind(NEED_FROM_RUNS)
196 
    .fetch_one(pool)
197 
    .await?;
198 
    Ok(peak.map(|p| p + p / 4 + 64))
199 
}
200
201 
/// Record why job `job_id` ran where it did.
202 
pub async fn set_placed(pool: &Pool, job_id: i64, placed: &str) -> sqlx::Result<()> {
203 
    sqlx::query("UPDATE flow_jobs SET placed = $1 WHERE id = $2").bind(placed).bind(job_id).execute(pool).await?;
204 
    Ok(())
205 
}
206
207 
/// Keep what a job took, as its runner reported.
208 
pub async fn set_peaks(pool: &Pool, job_id: i64, disk_mib: Option<u64>, memory_mib: Option<u64>) -> sqlx::Result<()> {
209 
    sqlx::query("UPDATE flow_jobs SET peak_disk_mib = $1, peak_memory_mib = $2 WHERE id = $3")
210 
        .bind(disk_mib.map(|v| v.min(i64::MAX as u64) as i64))
211 
        .bind(memory_mib.map(|v| v.min(i64::MAX as u64) as i64))
212 
        .bind(job_id)
213 
        .execute(pool)
214 
        .await?;
215 
    Ok(())
216 
}
217
218 
pub struct NewRun<'a> {
219 
    pub repo_id: i64,
220 
    pub bookmark: &'a str,
221 
    pub state: &'a str,
222 
    pub heads: &'a str,
223 
    pub pipeline_hash: &'a str,
224 
    pub actor: &'a str,
225 
    /// `push`, `manual` or `rerun:N`.
226 
    pub trigger: &'a str,
227 
    /// Problems when the pipeline is invalid: the run is recorded in error.
228 
    pub error: Option<&'a str>,
229 
}
230
231 
pub async fn create_run(pool: &Pool, r: NewRun<'_>, jobs: &[NewJob]) -> sqlx::Result<i64> {
232 
    let mut tx = super::begin_write(pool).await?;
233 
    let now = now_ms();
234 
    let status = if r.error.is_some() { "error" } else { "queued" };
235 
    let id: i64 = sqlx::query_scalar(
236 
        "INSERT INTO flow_runs (repo_id, number, bookmark, state, heads, pipeline_hash, status, error, actor, trigger, created_ms, finished_ms) \
237 
         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",
238 
    )
239 
    .bind(r.repo_id)
240 
    .bind(r.repo_id)
241 
    .bind(r.bookmark)
242 
    .bind(r.state)
243 
    .bind(r.heads)
244 
    .bind(r.pipeline_hash)
245 
    .bind(status)
246 
    .bind(r.error.unwrap_or(""))
247 
    .bind(r.actor)
248 
    .bind(r.trigger)
249 
    .bind(now)
250 
    .bind(r.error.map(|_| now))
251 
    .fetch_one(&mut *tx)
252 
    .await?;
253 
    for j in jobs {
254 
        sqlx::query(
255 
            "INSERT INTO flow_jobs (run_id, name, depth, depends, spec, sandbox, runtime, status, need_disk_mib, approval, environment) \
256 
             VALUES ($1, $2, $3, $4, $5, $6, $7, 'queued', $8, $9, $10)",
257 
        )
258 
        .bind(id)
259 
        .bind(&j.name)
260 
        .bind(j.depth)
261 
        .bind(serde_json::to_string(&j.depends).unwrap_or_else(|_| "[]".into()))
262 
        .bind(&j.spec)
263 
        .bind(&j.sandbox)
264 
        .bind(&j.runtime)
265 
        .bind(j.need_disk_mib)
266 
        .bind(if needs_approval(&j.spec) { "waiting" } else { "" })
267 
        .bind(environment_of(&j.spec))
268 
        .execute(&mut *tx)
269 
        .await?;
270 
    }
271 
    tx.commit().await?;
272 
    Ok(id)
273 
}
274
275 
/// The environment a compiled job deploys to, or empty (oxygenflow#59).
276 
fn environment_of(spec: &str) -> String {
277 
    serde_json::from_str::<serde_json::Value>(spec).ok().and_then(|v| v["environment"].as_str().map(String::from)).unwrap_or_default()
278 
}
279
280 
/// Whether a compiled job asks for approval (oxygenflow#57).
281 
fn needs_approval(spec: &str) -> bool {
282 
    serde_json::from_str::<serde_json::Value>(spec).is_ok_and(|v| !v["approval"].is_null())
283 
}
284
285 
/// The labels a compiled job and its sandbox ask a runner for (`runs-on`).
286 
fn labels_in(spec: &str, sandbox: &str) -> Vec<String> {
287 
    let mut out: Vec<String> = [spec, sandbox]
288 
        .iter()
289 
        .filter_map(|j| serde_json::from_str::<serde_json::Value>(j).ok())
290 
        .flat_map(|v| v["runs_on"].as_array().cloned().unwrap_or_default())
291 
        .filter_map(|l| l.as_str().map(String::from))
292 
        .collect();
293 
    out.sort();
294 
    out.dedup();
295 
    out
296 
}
297
298 
impl Job {
299 
    /// The runner labels this job needs.
300 
    pub fn needs(&self) -> Vec<String> {
301 
        labels_in(&self.spec, &self.sandbox)
302 
    }
303
304 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
305 
    pub fn arch(&self) -> Option<String> {
306 
        arch_in(&self.sandbox)
307 
    }
308 
}
309
310 
/// The architecture a sandbox's JSON asks for, if any.
311 
fn arch_in(sandbox: &str) -> Option<String> {
312 
    serde_json::from_str::<serde_json::Value>(sandbox).ok()?["arch"].as_str().map(String::from)
313 
}
314
315 
/// A job waiting for or held by a runner, with where it comes from.
316 
#[derive(Clone, Debug, FromRow)]
317 
pub struct QueueEntry {
318 
    pub job_id: i64,
319 
    pub run_id: i64,
320 
    pub repo_id: i64,
321 
    pub owner: String,
322 
    pub name: String,
323 
    pub visibility: String,
324 
    pub run_number: i64,
325 
    pub job: String,
326 
    pub status: String,
327 
    /// The runner holding it, when running.
328 
    pub runner: String,
329 
    pub spec: String,
330 
    pub sandbox: String,
331 
    /// When it started running.
332 
    pub started_ms: Option<i64>,
333 
    /// Its level (0 urgent … 3 low, oxygenhub#244) and, when expedited, who
334 
    /// did it and why.
335 
    pub priority_rank: i64,
336 
    pub priority_note: String,
337 
}
338
339 
impl QueueEntry {
340 
    pub fn priority(&self) -> &'static str {
341 
        PRIORITIES.get(self.priority_rank as usize).copied().unwrap_or("normal")
342 
    }
343 
}
344
345 
impl QueueEntry {
346 
    /// The runner labels the job needs.
347 
    pub fn needs(&self) -> Vec<String> {
348 
        labels_in(&self.spec, &self.sandbox)
349 
    }
350
351 
    /// The architecture the job's sandbox asks for (OxygenFlow spec §11.3).
352 
    pub fn arch(&self) -> Option<String> {
353 
        arch_in(&self.sandbox)
354 
    }
355 
}
356
357 
/// Every job of a live run that is running or queued: running ones first,
358 
/// then queued ones in the order runners claim them (hub spec §13.2).
359 
pub async fn queue(pool: &Pool) -> sqlx::Result<Vec<QueueEntry>> {
360 
    sqlx::query_as(&format!(
361 
        "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, \
362 
           CAST({PRIORITY_RANK} AS BIGINT) AS priority_rank, r.priority_note \
363 
         FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories rep ON rep.id = r.repo_id \
364 
         WHERE j.status IN ('queued', 'running') AND j.approval != 'waiting' AND r.status IN ('queued', 'running') \
365 
         ORDER BY CASE j.status WHEN 'running' THEN 0 ELSE 1 END, \
366 
           CASE j.status WHEN 'running' THEN 0 ELSE {PRIORITY_RANK} END, \
367 
           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) \
368 
           + (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, \
369 
           j.run_id, j.depth, j.id"
370 
    ))
371 
    .fetch_all(pool)
372 
    .await
373 
}
374
375 
/// A repository's runs, newest first, numbered below `before`.
376 
pub async fn runs_by_number(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
377 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND number < $2 ORDER BY number DESC LIMIT $3")
378 
        .bind(repo_id)
379 
        .bind(before.unwrap_or(i64::MAX))
380 
        .bind(limit)
381 
        .fetch_all(pool)
382 
        .await
383 
}
384
385 
/// The latest run on this bookmark of this repository, for what a push
386 
/// changed since (a job's scope).
387 
/// The run a push to `bookmark` is scoped against: the newest that was not
388 
/// cancelled, nor a push run still going (a new push supersedes it), so
389 
/// that what a cancelled run would have checked is checked by the next.
390 
pub async fn latest_on(pool: &Pool, repo_id: i64, bookmark: &str) -> sqlx::Result<Option<Run>> {
391 
    sqlx::query_as(
392 
        "SELECT * FROM flow_runs WHERE repo_id = $1 AND bookmark = $2 AND heads != '' AND status != 'cancelled'
393 
         AND NOT (trigger = 'push' AND status IN ('queued', 'running')) ORDER BY id DESC LIMIT 1",
394 
    )
395 
    .bind(repo_id)
396 
    .bind(bookmark)
397 
    .fetch_optional(pool)
398 
    .await
399 
}
400
401 
pub async fn runs(pool: &Pool, repo_id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<Run>> {
402 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND id < $2 ORDER BY id DESC LIMIT $3")
403 
        .bind(repo_id)
404 
        .bind(before.unwrap_or(i64::MAX))
405 
        .bind(limit)
406 
        .fetch_all(pool)
407 
        .await
408 
}
409
410 
/// A repository's run by its number there.
411 
pub async fn run(pool: &Pool, repo_id: i64, number: i64) -> sqlx::Result<Option<Run>> {
412 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND number = $2").bind(repo_id).bind(number).fetch_optional(pool).await
413 
}
414
415 
pub async fn run_by_id(pool: &Pool, id: i64) -> sqlx::Result<Option<Run>> {
416 
    sqlx::query_as("SELECT * FROM flow_runs WHERE id = $1").bind(id).fetch_optional(pool).await
417 
}
418
419 
pub async fn jobs(pool: &Pool, run_id: i64) -> sqlx::Result<Vec<Job>> {
420 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1 ORDER BY depth, name").bind(run_id).fetch_all(pool).await
421 
}
422
423 
pub async fn job(pool: &Pool, id: i64) -> sqlx::Result<Option<Job>> {
424 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE id = $1").bind(id).fetch_optional(pool).await
425 
}
426
427 
/// The running job a job token belongs to, while it may still be used.
428 
pub async fn job_by_token(pool: &Pool, token_hash: &str, max_age_ms: i64) -> sqlx::Result<Option<Job>> {
429 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE token_hash = $1 AND status = 'running' AND started_ms >= $2")
430 
        .bind(token_hash)
431 
        .bind(now_ms() - max_age_ms)
432 
        .fetch_optional(pool)
433 
        .await
434 
}
435
436 
/// Take the oldest queued job whose dependencies all passed and whose tier
437 
/// the runner offers, and mark it running for `runner` with a token.
438 
/// Let a job's token also read these repositories (its pinned subrepos).
439 
pub async fn set_also_read(pool: &Pool, job_id: i64, repo_ids: &[i64]) -> sqlx::Result<()> {
440 
    let list: Vec<String> = repo_ids.iter().map(|i| i.to_string()).collect();
441 
    sqlx::query("UPDATE flow_jobs SET also_read = $1 WHERE id = $2").bind(list.join(",")).bind(job_id).execute(pool).await?;
442 
    Ok(())
443 
}
444
445 
impl Job {
446 
    pub fn also_read_ids(&self) -> Vec<i64> {
447 
        self.also_read.split(',').filter_map(|x| x.trim().parse().ok()).collect()
448 
    }
449 
}
450
451 
/// What a runner can take on: free memory (MiB) and vCPUs, `None` for no
452 
/// limit, and the rootfs images it holds, `None` when it does not say.
453 
#[derive(Clone, Debug, Default)]
454 
pub struct Room {
455 
    pub memory_mib: Option<u32>,
456 
    pub cpus: Option<u32>,
457 
    pub images: Option<std::collections::HashSet<String>>,
458 
    /// The runner's architecture (`x86_64`, `riscv64`, …), empty when unknown.
459 
    pub arch: String,
460 
    /// Each image's architecture as runners have reported it (`flow_images`).
461 
    pub image_archs: std::collections::HashMap<String, String>,
462 
    /// Jobs this runner declined, not to be offered to it again (spec §11.5).
463 
    pub declined: std::collections::HashSet<i64>,
464 
    /// Images the hub holds that this runner would fetch when it lacks
465 
    /// them (oxygenflow#39); empty for a runner that does not fetch.
466 
    pub fetchable: std::collections::HashSet<String>,
467 
    /// Its free disk now, from its claim's facts (oxygenflow#40); `None`
468 
    /// when it does not say.
469 
    pub free_disk_mib: Option<u64>,
470 
}
471
472 
impl Room {
473 
    /// Whether a job needing `need` MiB of disk fits in its free disk.
474 
    pub fn disk_fits(&self, need: Option<i64>) -> bool {
475 
        match (need, self.free_disk_mib) {
476 
            (Some(n), Some(free)) => n.max(0) as u64 <= free,
477 
            _ => true,
478 
        }
479 
    }
480 
}
481
482 
impl Room {
483 
    /// Whether a job's sandbox (its stored JSON) fits: its memory and vCPUs,
484 
    /// and its rootfs image among the runner's (oxygenhub#73).
485 
    pub fn fits(&self, sandbox_json: &str) -> bool {
486 
        let v: serde_json::Value = serde_json::from_str(sandbox_json).unwrap_or_default();
487 
        let need = |k: &str| v[k].as_u64().unwrap_or(0);
488 
        let rootfs = v["rootfs"].as_str().unwrap_or("");
489 
        // The architecture: the sandbox's own ask, else the image's as read
490 
        // or learned from its passes (oxygenhub#173); one nobody has read
491 
        // or run yet matches (OxygenFlow spec §11.3).
492 
        let wanted = v["arch"].as_str().map(String::from).or_else(|| self.image_archs.get(rootfs).cloned());
493 
        let arch_ok = self.arch.is_empty() || wanted.as_deref().is_none_or(|w| w == self.arch);
494 
        arch_ok
495 
            && self.memory_mib.is_none_or(|m| need("memory_mib") <= m as u64)
496 
            && self.cpus.is_none_or(|c| need("cpus") <= c as u64)
497 
            && (rootfs.is_empty() || self.images.as_ref().is_none_or(|have| have.contains(rootfs)) || self.fetchable.contains(rootfs))
498 
    }
499 
}
500
501 
/// Remember which architecture each image is for, as a runner reports it.
502 
pub async fn record_image_archs(pool: &Pool, archs: &std::collections::BTreeMap<String, String>) -> sqlx::Result<()> {
503 
    for (hash, arch) in archs.iter().take(500) {
504 
        if hash.len() != 64 || !flow_core::arch_ok(arch) {
505 
            continue;
506 
        }
507 
        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")
508 
        .bind(hash.to_ascii_lowercase())
509 
        .bind(arch)
510 
        .bind(now_ms())
511 
        .execute(pool)
512 
        .await?;
513 
    }
514 
    Ok(())
515 
}
516
517 
/// Give every image nobody has read the architecture of the runners its
518 
/// jobs passed on (oxygenhub#173), for the jobs before the hub learned at
519 
/// each pass. Never overrides a known architecture.
520 
pub async fn learn_archs_from_runs(pool: &Pool) -> sqlx::Result<u64> {
521 
    let rootfs = crate::db::json_text("j.sandbox", "rootfs");
522 
    let r = sqlx::query(&format!(
523 
        "INSERT INTO flow_images (hash, arch, seen_ms)
524 
         SELECT lower({rootfs}), MIN(r.arch), $1 FROM flow_jobs j JOIN flow_runners r ON r.id = j.runner_id
525 
         WHERE j.status = 'passed' AND r.arch != '' AND length({rootfs}) = 64
526 
         GROUP BY lower({rootfs})
527 
         ON CONFLICT (hash) DO NOTHING"
528 
    ))
529 
    .bind(now_ms())
530 
    .execute(pool)
531 
    .await?;
532 
    Ok(r.rows_affected())
533 
}
534
535 
/// Every image's architecture the hub knows.
536 
pub async fn image_archs(pool: &Pool) -> sqlx::Result<std::collections::HashMap<String, String>> {
537 
    let rows: Vec<(String, String)> = sqlx::query_as("SELECT hash, arch FROM flow_images").fetch_all(pool).await?;
538 
    Ok(rows.into_iter().collect())
539 
}
540
541 
pub async fn claim(pool: &Pool, runtimes: &[String], runner: &str, token_hash: &str) -> sqlx::Result<Option<Job>> {
542 
    claim_within(pool, runtimes, runner, token_hash, Room::default(), &Default::default(), &[]).await
543 
}
544
545 
/// The first queued job a runner with these tiers, this room and this
546 
/// scope may take, now running under `runner`. Repositories take turns
547 
/// (fair share, oxygenhub#244): a job's turn is its place in its own
548 
/// repository's queue plus that repository's running jobs, lowest first,
549 
/// oldest first among equals, so one repository's burst can't hold up the
550 
/// rest. A scoped runner never sees
551 
/// another scope's jobs (oxygenhub#41).
552 
pub async fn claim_within(
553 
    pool: &Pool,
554 
    runtimes: &[String],
555 
    runner: &str,
556 
    token_hash: &str,
557 
    room: Room,
558 
    scope: &super::runners::Scope,
559 
    labels: &[String],
560 
) -> sqlx::Result<Option<Job>> {
561 
    claim_ranked(pool, runtimes, runner, token_hash, room, scope, labels, &|_| false).await
562 
}
563
564 
/// [`claim_within`], passing over a job `elsewhere` says a better runner
565 
/// should take (oxygenflow#42).
566 
#[allow(clippy::too_many_arguments)]
567 
pub async fn claim_ranked(
568 
    pool: &Pool,
569 
    runtimes: &[String],
570 
    runner: &str,
571 
    token_hash: &str,
572 
    room: Room,
573 
    scope: &super::runners::Scope,
574 
    labels: &[String],
575 
    elsewhere: &(dyn Fn(&Job) -> bool + Sync),
576 
) -> sqlx::Result<Option<Job>> {
577 
    let mut tx = super::begin_write(pool).await?;
578 
    let queued: Vec<Job> = sqlx::query_as(&format!(
579 
        "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 \
580 
         WHERE j.status = 'queued' AND j.approval != 'waiting' AND r.status IN ('queued', 'running') \
581 
         AND ($1 = '' OR rep.owner = $2) AND ($3 IS NULL OR r.repo_id = $4) \
582 
         AND NOT ($5 = '' AND $6 IS NULL AND rep.flow = 'own') \
583 
         ORDER BY {PRIORITY_RANK}, ROW_NUMBER() OVER (PARTITION BY r.repo_id ORDER BY j.run_id, j.depth, j.id) \
584 
           + (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), \
585 
           j.run_id, j.depth, j.id LIMIT 200"
586 
    ))
587 
    .bind(&scope.owner)
588 
    .bind(&scope.owner)
589 
    .bind(scope.repo_id)
590 
    .bind(scope.repo_id)
591 
    // A repository on its own runners only (#225): the site's runners skip it.
592 
    .bind(&scope.owner)
593 
    .bind(scope.repo_id)
594 
    .fetch_all(&mut *tx)
595 
    .await?;
596 
    for j in queued {
597 
        if !runtimes.iter().any(|r| r == &j.runtime)
598 
            || !room.fits(&j.sandbox)
599 
            || !room.disk_fits(j.need_disk_mib)
600 
            || room.declined.contains(&j.id)
601 
        {
602 
            continue;
603 
        }
604 
        // Every label the job and its sandbox ask for (`runs-on`).
605 
        if !j.needs().iter().all(|l| labels.contains(l)) {
606 
            continue;
607 
        }
608 
        let deps = j.depends_on();
609 
        if !deps.is_empty() {
610 
            let mut passed = 0i64;
611 
            for d in &deps {
612 
                let st: Option<String> = sqlx::query_scalar("SELECT status FROM flow_jobs WHERE run_id = $1 AND name = $2")
613 
                    .bind(j.run_id)
614 
                    .bind(d)
615 
                    .fetch_optional(&mut *tx)
616 
                    .await?;
617 
                if st.as_deref() == Some("passed") {
618 
                    passed += 1;
619 
                }
620 
            }
621 
            if passed != deps.len() as i64 {
622 
                continue;
623 
            }
624 
        }
625 
        if elsewhere(&j) {
626 
            continue;
627 
        }
628 
        let now = now_ms();
629 
        let done = sqlx::query(
630 
            "UPDATE flow_jobs SET status = 'running', runner = $1, token_hash = $2, started_ms = $3, waiting = '' WHERE id = $4 AND status = 'queued'",
631 
        )
632 
        .bind(runner)
633 
        .bind(token_hash)
634 
        .bind(now)
635 
        .bind(j.id)
636 
        .execute(&mut *tx)
637 
        .await?;
638 
        if done.rows_affected() == 1 {
639 
            sqlx::query("UPDATE flow_runs SET status = 'running' WHERE id = $1 AND status = 'queued'")
640 
                .bind(j.run_id)
641 
                .execute(&mut *tx)
642 
                .await?;
643 
            tx.commit().await?;
644 
            return job(pool, j.id).await;
645 
        }
646 
    }
647 
    tx.commit().await?;
648 
    Ok(None)
649 
}
650
651 
/// How long a decline keeps a job from its runner when the runner's
652 
/// profile stays the same: a reason such as memory in use passes.
653 
pub const DECLINE_HOLDS_MS: i64 = 10 * 60_000;
654
655 
/// The jobs runner `runner_id` declined with this profile, recently.
656 
pub async fn declined_by(pool: &Pool, runner_id: i64, profile: &str) -> sqlx::Result<std::collections::HashSet<i64>> {
657 
    let rows: Vec<(i64,)> = sqlx::query_as("SELECT job_id FROM flow_declines WHERE runner_id = $1 AND profile = $2 AND ms > $3")
658 
        .bind(runner_id)
659 
        .bind(profile)
660 
        .bind(now_ms() - DECLINE_HOLDS_MS)
661 
        .fetch_all(pool)
662 
        .await?;
663 
    Ok(rows.into_iter().map(|(j,)| j).collect())
664 
}
665
666 
/// A decline: one runner's reason for handing a job back.
667 
#[derive(Clone, Debug, sqlx::FromRow)]
668 
pub struct DeclineRow {
669 
    pub job_id: i64,
670 
    pub runner_id: i64,
671 
    pub runner: String,
672 
    pub reason: String,
673 
    pub ms: i64,
674 
}
675
676 
/// Runner `runner_id` hands back job `job_id`, running on it: the decline
677 
/// is kept (one per job and runner, the latest), and the job is queued
678 
/// again. Returns every decline the job has had, or `None` when the job
679 
/// was not running on that runner.
680 
pub async fn decline(
681 
    pool: &Pool,
682 
    job_id: i64,
683 
    runner_id: i64,
684 
    runner: &str,
685 
    profile: &str,
686 
    reason: &str,
687 
) -> sqlx::Result<Option<Vec<DeclineRow>>> {
688 
    let mut tx = super::begin_write(pool).await?;
689 
    let back = sqlx::query(
690 
        "UPDATE flow_jobs SET status = 'queued', runner = '', token_hash = '', started_ms = NULL, runner_id = NULL \
691 
         WHERE id = $1 AND status = 'running' AND runner_id = $2",
692 
    )
693 
    .bind(job_id)
694 
    .bind(runner_id)
695 
    .execute(&mut *tx)
696 
    .await?;
697 
    if back.rows_affected() != 1 {
698 
        tx.rollback().await?;
699 
        return Ok(None);
700 
    }
701 
    sqlx::query(
702 
        "INSERT INTO flow_declines (job_id, runner_id, runner, profile, reason, ms) VALUES ($1, $2, $3, $4, $5, $6) \
703 
         ON CONFLICT (job_id, runner_id) DO UPDATE SET runner = excluded.runner, profile = excluded.profile, reason = excluded.reason, ms = excluded.ms",
704 
    )
705 
    .bind(job_id)
706 
    .bind(runner_id)
707 
    .bind(runner)
708 
    .bind(profile)
709 
    .bind(reason)
710 
    .bind(now_ms())
711 
    .execute(&mut *tx)
712 
    .await?;
713 
    let all: Vec<DeclineRow> =
714 
        sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = $1 ORDER BY ms")
715 
            .bind(job_id)
716 
            .fetch_all(&mut *tx)
717 
            .await?;
718 
    tx.commit().await?;
719 
    Ok(Some(all))
720 
}
721
722 
/// The declines of job `job_id`, oldest first.
723 
pub async fn declines_of_job(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<DeclineRow>> {
724 
    sqlx::query_as("SELECT job_id, runner_id, runner, reason, ms FROM flow_declines WHERE job_id = $1 ORDER BY ms")
725 
        .bind(job_id)
726 
        .fetch_all(pool)
727 
        .await
728 
}
729
730 
/// A runner's latest declines, newest first, with where the job is.
731 
#[derive(Clone, Debug, sqlx::FromRow)]
732 
pub struct RunnerDecline {
733 
    pub job_id: i64,
734 
    pub job: String,
735 
    pub run_number: i64,
736 
    pub owner: String,
737 
    pub name: String,
738 
    pub reason: String,
739 
    pub ms: i64,
740 
}
741
742 
pub async fn declines_of_runner(pool: &Pool, runner_id: i64, limit: i64) -> sqlx::Result<Vec<RunnerDecline>> {
743 
    sqlx::query_as(
744 
        "SELECT d.job_id, j.name AS job, r.number AS run_number, rep.owner, rep.name, d.reason, d.ms \
745 
         FROM flow_declines d JOIN flow_jobs j ON j.id = d.job_id JOIN flow_runs r ON r.id = j.run_id \
746 
         JOIN repositories rep ON rep.id = r.repo_id WHERE d.runner_id = $1 ORDER BY d.ms DESC LIMIT $2",
747 
    )
748 
    .bind(runner_id)
749 
    .bind(limit)
750 
    .fetch_all(pool)
751 
    .await
752 
}
753
754 
/// Queued jobs whose dependencies have passed, in their live runs, with
755 
/// their repositories (owner, id): the ones a runner could take now.
756 
pub async fn ready_queued(pool: &Pool) -> sqlx::Result<Vec<(Job, String, i64)>> {
757 
    let queued: Vec<Job> = sqlx::query_as(
758 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
759 
         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",
760 
    )
761 
    .fetch_all(pool)
762 
    .await?;
763 
    let mut out = Vec::new();
764 
    for j in queued {
765 
        let mut ready = true;
766 
        for d in j.depends_on() {
767 
            let st: Option<String> = sqlx::query_scalar("SELECT status FROM flow_jobs WHERE run_id = $1 AND name = $2")
768 
                .bind(j.run_id)
769 
                .bind(&d)
770 
                .fetch_optional(pool)
771 
                .await?;
772 
            if st.as_deref() != Some("passed") {
773 
                ready = false;
774 
                break;
775 
            }
776 
        }
777 
        if !ready {
778 
            continue;
779 
        }
780 
        let repo: Option<(String, i64)> =
781 
            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")
782 
                .bind(j.run_id)
783 
                .fetch_optional(pool)
784 
                .await?;
785 
        if let Some((owner, id)) = repo {
786 
            out.push((j, owner, id));
787 
        }
788 
    }
789 
    Ok(out)
790 
}
791
792 
pub async fn set_waiting(pool: &Pool, job_id: i64, waiting: &str) -> sqlx::Result<()> {
793 
    sqlx::query("UPDATE flow_jobs SET waiting = $1 WHERE id = $2 AND waiting <> $3")
794 
        .bind(waiting)
795 
        .bind(job_id)
796 
        .bind(waiting)
797 
        .execute(pool)
798 
        .await?;
799 
    Ok(())
800 
}
801
802 
/// Append records; returns how many were kept (the rest exceed the cap).
803 
pub async fn append_log(pool: &Pool, job_id: i64, records: &[flow_core::api::Record], cap: i64) -> sqlx::Result<usize> {
804 
    let mut tx = super::begin_write(pool).await?;
805 
    let count: i64 = sqlx::query_scalar("SELECT log_count FROM flow_jobs WHERE id = $1").bind(job_id).fetch_one(&mut *tx).await?;
806 
    let room = (cap - count).max(0) as usize;
807 
    let keep = &records[..records.len().min(room)];
808 
    for (i, r) in keep.iter().enumerate() {
809 
        sqlx::query("INSERT INTO flow_logs (job_id, seq, t, stream, text) VALUES ($1, $2, $3, $4, $5)")
810 
            .bind(job_id)
811 
            .bind(count + i as i64 + 1)
812 
            .bind(r.t as i64)
813 
            .bind(&r.stream)
814 
            .bind(&r.text)
815 
            .execute(&mut *tx)
816 
            .await?;
817 
    }
818 
    sqlx::query("UPDATE flow_jobs SET log_count = log_count + $1 WHERE id = $2")
819 
        .bind(keep.len() as i64)
820 
        .bind(job_id)
821 
        .execute(&mut *tx)
822 
        .await?;
823 
    tx.commit().await?;
824 
    Ok(keep.len())
825 
}
826
827 
pub async fn log(pool: &Pool, job_id: i64, after: i64, limit: i64) -> sqlx::Result<Vec<LogLine>> {
828 
    sqlx::query_as("SELECT seq, t, stream, text FROM flow_logs WHERE job_id = $1 AND seq > $2 ORDER BY seq LIMIT $3")
829 
        .bind(job_id)
830 
        .bind(after)
831 
        .bind(limit)
832 
        .fetch_all(pool)
833 
        .await
834 
}
835
836 
/// Jobs running since before `ms` (the oldest first).
837 
/// Jobs still running at `now` past their own time limit (`default_s` for
838 
/// one claimed without) plus `grace_ms`.
839 
pub async fn running_past_limit(pool: &Pool, now: i64, default_s: u64, grace_ms: i64) -> sqlx::Result<Vec<Job>> {
840 
    sqlx::query_as(
841 
        "SELECT * FROM flow_jobs WHERE status = 'running' \
842 
         AND COALESCE(started_ms, 0) + (CASE WHEN timeout_s > 0 THEN timeout_s ELSE $1 END) * 1000 + $2 < $3 ORDER BY started_ms",
843 
    )
844 
    .bind(default_s as i64)
845 
    .bind(grace_ms)
846 
    .bind(now)
847 
    .fetch_all(pool)
848 
    .await
849 
}
850
851 
/// Record the limit a claimed job runs under.
852 
pub async fn set_job_timeout(pool: &Pool, job_id: i64, timeout_s: u64) -> sqlx::Result<()> {
853 
    sqlx::query("UPDATE flow_jobs SET timeout_s = $1 WHERE id = $2").bind(timeout_s as i64).bind(job_id).execute(pool).await?;
854 
    Ok(())
855 
}
856
857 
pub async fn running_since_before(pool: &Pool, ms: i64) -> sqlx::Result<Vec<Job>> {
858 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND COALESCE(started_ms, 0) < $1 ORDER BY started_ms")
859 
        .bind(ms)
860 
        .fetch_all(pool)
861 
        .await
862 
}
863
864 
/// The jobs running on the runner with this id.
865 
pub async fn running_on(pool: &Pool, runner_id: i64) -> sqlx::Result<Vec<Job>> {
866 
    sqlx::query_as("SELECT * FROM flow_jobs WHERE status = 'running' AND runner_id = $1 ORDER BY id").bind(runner_id).fetch_all(pool).await
867 
}
868
869 
pub struct Outcome<'a> {
870 
    pub status: &'a str,
871 
    pub exit_code: Option<i64>,
872 
    pub error: &'a str,
873 
    pub receipt: &'a str,
874 
    pub receipt_hash: &'a str,
875 
}
876
877 
/// Record a job's end, skip the jobs that can no longer run, and settle the
878 
/// run. Returns the jobs whose status changed (for their checks).
879 
pub async fn finish(pool: &Pool, job_id: i64, o: Outcome<'_>) -> sqlx::Result<Vec<Job>> {
880 
    let mut tx = super::begin_write(pool).await?;
881 
    let now = now_ms();
882 
    let j: Job = sqlx::query_as("SELECT * FROM flow_jobs WHERE id = $1").bind(job_id).fetch_one(&mut *tx).await?;
883 
    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")
884 
        .bind(o.status)
885 
        .bind(o.exit_code)
886 
        .bind(o.error)
887 
        .bind(o.receipt)
888 
        .bind(o.receipt_hash)
889 
        .bind(now)
890 
        .bind(job_id)
891 
        .execute(&mut *tx)
892 
        .await?;
893 
    // A pass teaches what its image is for, when nobody has read it
894 
    // (oxygenhub#173): the architecture of the runner it passed on.
895 
    if o.status == "passed" {
896 
        let rootfs = serde_json::from_str::<serde_json::Value>(&j.sandbox)
897 
            .ok()
898 
            .and_then(|v| v["rootfs"].as_str().map(str::to_ascii_lowercase))
899 
            .unwrap_or_default();
900 
        if rootfs.len() == 64 {
901 
            sqlx::query(
902 
                "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 \
903 
                 WHERE j.id = $3 AND r.arch != '' ON CONFLICT (hash) DO NOTHING",
904 
            )
905 
            .bind(&rootfs)
906 
            .bind(now)
907 
            .bind(job_id)
908 
            .execute(&mut *tx)
909 
            .await?;
910 
        }
911 
    }
912 
    let mut changed = vec![job_id];
913 
    // Dependents of anything that did not pass are skipped, transitively.
914 
    loop {
915 
        let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(j.run_id).fetch_all(&mut *tx).await?;
916 
        let blocked: Vec<i64> = all
917 
            .iter()
918 
            .filter(|x| x.status == "queued")
919 
            .filter(|x| {
920 
                x.depends_on()
921 
                    .iter()
922 
                    .any(|d| all.iter().any(|y| &y.name == d && matches!(y.status.as_str(), "failed" | "skipped" | "error" | "cancelled")))
923 
            })
924 
            .map(|x| x.id)
925 
            .collect();
926 
        if blocked.is_empty() {
927 
            break;
928 
        }
929 
        for id in blocked {
930 
            sqlx::query("UPDATE flow_jobs SET status = 'skipped', finished_ms = $1 WHERE id = $2")
931 
                .bind(now)
932 
                .bind(id)
933 
                .execute(&mut *tx)
934 
                .await?;
935 
            changed.push(id);
936 
        }
937 
    }
938 
    let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(j.run_id).fetch_all(&mut *tx).await?;
939 
    if all.iter().all(Job::done) {
940 
        let status = if all.iter().all(|x| x.status == "passed") { "passed" } else { "failed" };
941 
        sqlx::query("UPDATE flow_runs SET status = $1, finished_ms = $2 WHERE id = $3")
942 
            .bind(status)
943 
            .bind(now)
944 
            .bind(j.run_id)
945 
            .execute(&mut *tx)
946 
            .await?;
947 
    }
948 
    tx.commit().await?;
949 
    let mut out = Vec::new();
950 
    for id in changed {
951 
        if let Some(x) = job(pool, id).await? {
952 
            out.push(x);
953 
        }
954 
    }
955 
    Ok(out)
956 
}
957
958 
/// Cancel a run (oxygenflow#32): its jobs not yet done end as cancelled,
959 
/// and so does the run. A runner learns of it when the hub next refuses
960 
/// its log lines, and stops the job. Returns the jobs whose status changed.
961 
pub async fn cancel_run(pool: &Pool, run_id: i64, why: &str) -> sqlx::Result<Vec<Job>> {
962 
    let now = now_ms();
963 
    let mut tx = pool.begin().await?;
964 
    let ids: Vec<(i64,)> = sqlx::query_as("SELECT id FROM flow_jobs WHERE run_id = $1 AND status IN ('queued', 'running')")
965 
        .bind(run_id)
966 
        .fetch_all(&mut *tx)
967 
        .await?;
968 
    sqlx::query(
969 
        "UPDATE flow_jobs SET status = 'cancelled', error = $1, finished_ms = $2, token_hash = ''
970 
         WHERE run_id = $3 AND status IN ('queued', 'running')",
971 
    )
972 
    .bind(why)
973 
    .bind(now)
974 
    .bind(run_id)
975 
    .execute(&mut *tx)
976 
    .await?;
977 
    sqlx::query("UPDATE flow_runs SET status = 'cancelled', finished_ms = $1 WHERE id = $2 AND status IN ('queued', 'running')")
978 
        .bind(now)
979 
        .bind(run_id)
980 
        .execute(&mut *tx)
981 
        .await?;
982 
    tx.commit().await?;
983 
    let mut out = Vec::new();
984 
    for (id,) in ids {
985 
        if let Some(x) = job(pool, id).await? {
986 
            out.push(x);
987 
        }
988 
    }
989 
    Ok(out)
990 
}
991
992 
/// Run job `name` of finished run `run_id` again, with every job that
993 
/// depends on it (oxygenflow#31): back to the queue as if new, their logs,
994 
/// artifacts and reports gone, and the run running again. The jobs they
995 
/// depend on keep their results. The jobs reset; none when there is no
996 
/// such job.
997 
pub async fn rerun_job(pool: &Pool, run_id: i64, name: &str) -> sqlx::Result<Vec<Job>> {
998 
    let all: Vec<Job> = sqlx::query_as("SELECT * FROM flow_jobs WHERE run_id = $1").bind(run_id).fetch_all(pool).await?;
999 
    if !all.iter().any(|j| j.name == name) {
1000 
        return Ok(Vec::new());
1001 
    }
1002 
    let mut again = vec![name.to_string()];
1003 
    loop {
1004 
        let more: Vec<String> = all
1005 
            .iter()
1006 
            .filter(|j| !again.contains(&j.name) && j.depends_on().iter().any(|d| again.contains(d)))
1007 
            .map(|j| j.name.clone())
1008 
            .collect();
1009 
        if more.is_empty() {
1010 
            break;
1011 
        }
1012 
        again.extend(more);
1013 
    }
1014 
    let ids: Vec<i64> = all.iter().filter(|j| again.contains(&j.name)).map(|j| j.id).collect();
1015 
    let mut tx = pool.begin().await?;
1016 
    for id in &ids {
1017 
        for table in ["flow_logs", "flow_artifacts", "flow_declines", "test_reports", "coverage_reports", "reports"] {
1018 
            sqlx::query(&format!("DELETE FROM {table} WHERE job_id = $1")).bind(id).execute(&mut *tx).await?;
1019 
        }
1020 
        sqlx::query(
1021 
            "UPDATE flow_jobs SET status = 'queued', runner = '', token_hash = '', started_ms = NULL, finished_ms = NULL,
1022 
             exit_code = NULL, error = '', receipt = '', receipt_hash = '', log_count = 0, also_read = '', receipt_attest = '',
1023 
             receipt_signer = '', receipt_note = '', runner_id = NULL, waiting = '', placed = '', peak_disk_mib = NULL,
1024 
             peak_memory_mib = NULL, approval = CASE WHEN approval = '' THEN '' ELSE 'waiting' END, approved_by = '',
1025 
             approved_ms = NULL, approval_seal = '' WHERE id = $1",
1026 
        )
1027 
        .bind(id)
1028 
        .execute(&mut *tx)
1029 
        .await?;
1030 
    }
1031 
    sqlx::query("UPDATE flow_runs SET status = 'running', finished_ms = NULL WHERE id = $1").bind(run_id).execute(&mut *tx).await?;
1032 
    tx.commit().await?;
1033 
    let mut out = Vec::new();
1034 
    for id in ids {
1035 
        if let Some(x) = job(pool, id).await? {
1036 
            out.push(x);
1037 
        }
1038 
    }
1039 
    Ok(out)
1040 
}
1041
1042 
/// The push runs of `bookmark` still going, older than run `number`.
1043 
pub async fn live_push_runs_before(pool: &Pool, repo_id: i64, bookmark: &str, number: i64) -> sqlx::Result<Vec<Run>> {
1044 
    sqlx::query_as(
1045 
        "SELECT * FROM flow_runs WHERE repo_id = $1 AND bookmark = $2 AND number < $3 AND trigger = 'push'
1046 
         AND status IN ('queued', 'running') ORDER BY id",
1047 
    )
1048 
    .bind(repo_id)
1049 
    .bind(bookmark)
1050 
    .bind(number)
1051 
    .fetch_all(pool)
1052 
    .await
1053 
}
1054
1055 
/// The newest run of exactly this state (heads joined with `+`, sorted).
1056 
pub async fn latest_for_heads(pool: &Pool, repo_id: i64, heads: &str) -> sqlx::Result<Option<Run>> {
1057 
    sqlx::query_as("SELECT * FROM flow_runs WHERE repo_id = $1 AND heads = $2 ORDER BY id DESC LIMIT 1")
1058 
        .bind(repo_id)
1059 
        .bind(heads)
1060 
        .fetch_optional(pool)
1061 
        .await
1062 
}
1063
1064 
/// Record what became of job `job_id`'s signed receipt.
1065 
pub async fn set_receipt(pool: &Pool, job_id: i64, attest: &str, signer: &str, note: &str) -> sqlx::Result<()> {
1066 
    sqlx::query("UPDATE flow_jobs SET receipt_attest = $1, receipt_signer = $2, receipt_note = $3 WHERE id = $4")
1067 
        .bind(attest)
1068 
        .bind(signer)
1069 
        .bind(note)
1070 
        .bind(job_id)
1071 
        .execute(pool)
1072 
        .await?;
1073 
    Ok(())
1074 
}
1075
1076 
/// A job's artifact, kept as a blob of the run's repository.
1077 
#[derive(Clone, Debug, FromRow)]
1078 
pub struct Artifact {
1079 
    pub job_id: i64,
1080 
    pub name: String,
1081 
    pub path: String,
1082 
    pub blob: String,
1083 
    pub size: i64,
1084 
    /// BLAKE3 of the bytes.
1085 
    pub content: String,
1086 
    pub created_ms: i64,
1087 
}
1088
1089 
pub async fn add_artifact(pool: &Pool, a: &Artifact) -> sqlx::Result<()> {
1090 
    sqlx::query(
1091 
        "INSERT OR REPLACE INTO flow_artifacts (job_id, name, path, blob, size, content, created_ms) VALUES ($1, $2, $3, $4, $5, $6, $7)",
1092 
    )
1093 
    .bind(a.job_id)
1094 
    .bind(&a.name)
1095 
    .bind(&a.path)
1096 
    .bind(&a.blob)
1097 
    .bind(a.size)
1098 
    .bind(&a.content)
1099 
    .bind(a.created_ms)
1100 
    .execute(pool)
1101 
    .await?;
1102 
    Ok(())
1103 
}
1104
1105 
/// Every artifact blob a repository's runs keep, for garbage collection to
1106 
/// keep too (#126): the database is the only thing that names them.
1107 
pub async fn artifact_blobs(pool: &Pool, repo_id: i64) -> sqlx::Result<Vec<String>> {
1108 
    sqlx::query_scalar(
1109 
        "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",
1110 
    )
1111 
    .bind(repo_id)
1112 
    .fetch_all(pool)
1113 
    .await
1114 
}
1115
1116 
/// Job `job_id`'s artifacts, by name.
1117 
pub async fn artifacts(pool: &Pool, job_id: i64) -> sqlx::Result<Vec<Artifact>> {
1118 
    sqlx::query_as("SELECT * FROM flow_artifacts WHERE job_id = $1 ORDER BY name").bind(job_id).fetch_all(pool).await
1119 
}
1120
1121 
/// Where an artifact with given content came from: the run and job that
1122 
/// built it, and the job's signed receipt when one was kept.
1123 
#[derive(Clone, Debug, FromRow)]
1124 
pub struct Provenance {
1125 
    pub run_number: i64,
1126 
    pub job: String,
1127 
    pub name: String,
1128 
    pub receipt_attest: String,
1129 
    pub receipt_signer: String,
1130 
}
1131
1132 
/// The build of the bytes whose BLAKE3 is `content` in repository
1133 
/// `repo_id`: a passed job, one with a kept signed receipt first, newest
1134 
/// run first.
1135 
pub async fn provenance(pool: &Pool, repo_id: i64, content: &str) -> sqlx::Result<Option<Provenance>> {
1136 
    sqlx::query_as(
1137 
        "SELECT r.number AS run_number, j.name AS job, a.name AS name, j.receipt_attest, j.receipt_signer \
1138 
         FROM flow_artifacts a JOIN flow_jobs j ON j.id = a.job_id JOIN flow_runs r ON r.id = j.run_id \
1139 
         WHERE r.repo_id = $1 AND a.content = $2 AND j.status = 'passed' \
1140 
         ORDER BY (j.receipt_attest != '') DESC, r.number DESC LIMIT 1",
1141 
    )
1142 
    .bind(repo_id)
1143 
    .bind(content)
1144 
    .fetch_optional(pool)
1145 
    .await
1146 
}
1147
1148 
/// Approve job `id` if it is still waiting for a person (oxygenflow#57);
1149 
/// whether it was.
1150 
pub async fn approve_job(pool: &Pool, id: i64, who: &str, seal: &str) -> sqlx::Result<bool> {
1151 
    let r = sqlx::query(
1152 
        "UPDATE flow_jobs SET approval = 'approved', approved_by = $1, approved_ms = $2, approval_seal = $3 \
1153 
         WHERE id = $4 AND status = 'queued' AND approval = 'waiting'",
1154 
    )
1155 
    .bind(who)
1156 
    .bind(now_ms())
1157 
    .bind(seal)
1158 
    .bind(id)
1159 
    .execute(pool)
1160 
    .await?;
1161 
    Ok(r.rows_affected() > 0)
1162 
}
1163
1164 
/// Note who rejected job `id` (it is then finished as cancelled).
1165 
pub async fn note_rejection(pool: &Pool, id: i64, who: &str) -> sqlx::Result<()> {
1166 
    sqlx::query("UPDATE flow_jobs SET approved_by = $1, approved_ms = $2 WHERE id = $3")
1167 
        .bind(who)
1168 
        .bind(now_ms())
1169 
        .bind(id)
1170 
        .execute(pool)
1171 
        .await?;
1172 
    Ok(())
1173 
}
1174
1175 
/// Jobs waiting for a person, in runs still going.
1176 
pub async fn waiting_for_approval(pool: &Pool) -> sqlx::Result<Vec<Job>> {
1177 
    sqlx::query_as(
1178 
        "SELECT j.* FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
1179 
         WHERE j.status = 'queued' AND j.approval = 'waiting' AND r.status IN ('queued', 'running')",
1180 
    )
1181 
    .fetch_all(pool)
1182 
    .await
1183 
}
1184
1185 
/// One deployment: a job for an environment that passed.
1186 
#[derive(Clone, Debug, FromRow)]
1187 
pub struct Deployment {
1188 
    pub environment: String,
1189 
    pub job: String,
1190 
    pub run_number: i64,
1191 
    pub state: String,
1192 
    pub bookmark: String,
1193 
    pub actor: String,
1194 
    pub approved_by: String,
1195 
    pub finished_ms: i64,
1196 
}
1197
1198 
/// A repository's deployments, newest first (oxygenflow#59).
1199 
pub async fn deployments(pool: &Pool, repo_id: i64, limit: i64) -> sqlx::Result<Vec<Deployment>> {
1200 
    sqlx::query_as(
1201 
        "SELECT j.environment, j.name AS job, r.number AS run_number, r.state, r.bookmark, r.actor, j.approved_by, \
1202 
         COALESCE(j.finished_ms, 0) AS finished_ms FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id \
1203 
         WHERE r.repo_id = $1 AND j.environment != '' AND j.status = 'passed' ORDER BY j.finished_ms DESC LIMIT $2",
1204 
    )
1205 
    .bind(repo_id)
1206 
    .bind(limit)
1207 
    .fetch_all(pool)
1208 
    .await
1209 
}