OXYGENHUB
# c8df33a6829f oxygenhub / crates / hub / src / flow.rs
75.0 KiB · Rust · last changed Reports: the Reports tab, each report's page, its frame, and checks from thresholds 24 days ago source blame history raw
1
//! OxygenFlow on the hub (hub spec §13.2): runs started by pushes, the
2
//! runner API, job tokens and the checks that mirror jobs. Pipelines are
3
//! parsed by `flow-core`, the same code `flow check` runs.
4
5
use crate::api::ApiError;
6
use crate::db::{self, flow as fdb};
7
use crate::repos::RepoName;
8
use crate::Shared;
9
use axum::extract::{Path, State};
10
use axum::http::{HeaderMap, StatusCode};
11
use axum::response::{IntoResponse, Response};
12
use axum::Json;
13
use flow_core::api::{Claim, ClaimRequest, Finish, LogBatch, MAX_RECORDS, MAX_RECORD_TEXT};
14
use flow_core::{Pipeline, PIPELINE_PATH};
15
use oxy_core::hash::ChangeHash;
16
17
/// Prefix of job tokens, so the wire protocol can tell them from personal tokens.
18
pub const JOB_TOKEN_PREFIX: &str = "flowjob_";
19
/// A job token is refused this long after its job started, whatever its state.
20
pub const JOB_TOKEN_MAX_AGE_MS: i64 = 2 * 3600 * 1000;
21
22
/// Whether pushes to this repository start runs: its own setting when a
23
/// site administrator made one, else hub.toml's `[flow] repos`.
24
pub fn enabled(app: &Shared, row: &db::repos::RepoRow) -> bool {
25
    match row.flow.as_str() {
26
        "on" => true,
27
        "off" => false,
28
        _ => app.config.flow.allows(&row.owner, &row.name),
29
    }
30
}
31
32
/// A bookmark moved by a push: start a run when the repository is allowed
33
/// and the new state has a pipeline.
34
pub async fn on_push(app: &Shared, row: &db::repos::RepoRow, bookmark: &str, state: &str, heads: &[ChangeHash], actor: &str) {
35
    if heads.is_empty() || !enabled(app, row) {
36
        return;
37
    }
38
    // A pushed tag runs the jobs whose `on` names it (oxygenflow#24).
39
    let started = match bookmark.strip_prefix("tags/") {
40
        Some(tag) => start_run_with(app, row, bookmark, state, heads, actor, &format!("tag:{tag}"), None).await,
41
        None => {
42
            let touched = touched_since_last_run(app, row, bookmark, heads).await;
43
            let started = start_run_scoped(app, row, bookmark, state, heads, actor, "push", touched.as_deref()).await;
44
            if let Ok(Some(n)) = started {
45
                supersede(app, row, bookmark, n).await;
46
            }
47
            started
48
        }
49
    };
50
    if let Err(e) = started {
51
        tracing::warn!("flow: starting a run of {}/{}: {e}", row.owner, row.name);
52
    }
53
}
54
55
/// "Run pipeline" (hub spec §13.2): run a bookmark's current state without
56
/// moving it. Returns the new run's number.
57
pub async fn run_bookmark(app: &Shared, row: &db::repos::RepoRow, bookmark: &str, actor: &str) -> Result<i64, String> {
58
    if !enabled(app, row) {
59
        return Err("OxygenFlow is not enabled for this repository".into());
60
    }
61
    let repo = RepoName::new(&row.owner, &row.name).ok_or("bad repository name")?;
62
    let b = bookmark.to_string();
63
    let found = app
64
        .repos
65
        .read(&repo, move |r, _| {
66
            let Some(st) = r.view()?.bookmarks.get(&b).copied() else { return Ok(None) };
67
            Ok(Some((st, r.store().get_state(&st)?.heads)))
68
        })
69
        .await
70
        .map_err(|e| e.to_string())?;
71
    let Some((state, heads)) = found else { return Err(format!("no bookmark {bookmark}")) };
72
    if heads.is_empty() {
73
        return Err(format!("{bookmark} has no changes yet"));
74
    }
75
    start_run(app, row, bookmark, &state.to_hex(), &heads, actor, "manual")
76
        .await?
77
        .ok_or_else(|| format!("{bookmark} has no {PIPELINE_PATH}"))
78
}
79
80
/// "Run pipeline" on one committed change: the state whose only head is
81
/// `change`, whatever bookmarks hold it. Returns the new run's number.
82
pub async fn run_change(app: &Shared, row: &db::repos::RepoRow, change: &str, actor: &str) -> Result<i64, String> {
83
    if !enabled(app, row) {
84
        return Err("OxygenFlow is not enabled for this repository".into());
85
    }
86
    let h = ChangeHash::from_hex(change.trim().trim_start_matches('#')).ok_or("give the full hash of a change")?;
87
    let repo = RepoName::new(&row.owner, &row.name).ok_or("bad repository name")?;
88
    let known = app.repos.read(&repo, move |r, _| Ok(r.store().get_change_meta(&h).is_ok())).await.map_err(|e| e.to_string())?;
89
    if !known {
90
        return Err(format!("no change {} in this repository", &change.trim()[..12.min(change.trim().len())]));
91
    }
92
    let state = oxy_core::hash::hash_object(oxy_core::hash::ObjectType::State, &oxy_core::objects::StateObject::new(vec![h]).encode());
93
    start_run(app, row, "", &state.to_hex(), &[h], actor, "manual")
94
        .await?
95
        .ok_or_else(|| format!("that change's state has no {PIPELINE_PATH}"))
96
}
97
98
/// Re-run a finished run: the same state, bookmark and pipeline, as a new
99
/// run whose checks replace the old ones. Returns the new run's number.
100
pub async fn rerun(app: &Shared, row: &db::repos::RepoRow, run: &fdb::Run, actor: &str) -> Result<i64, String> {
101
    if !enabled(app, row) {
102
        return Err("OxygenFlow is not enabled for this repository".into());
103
    }
104
    if run.live() {
105
        return Err(format!("run #{} is still {}", run.number, run.status));
106
    }
107
    let heads: Vec<ChangeHash> = run.head_list().into_iter().filter_map(ChangeHash::from_hex).collect();
108
    if heads.is_empty() {
109
        return Err("the run names no changes".into());
110
    }
111
    // A rerun repeats its run's event: the jobs it had are the jobs it gets.
112
    let trigger = format!("rerun:{}:{}", run.number, original_event(run));
113
    start_run(app, row, &run.bookmark, &run.state, &heads, actor, &trigger)
114
        .await?
115
        .ok_or_else(|| format!("that state has no {PIPELINE_PATH}"))
116
}
117
118
/// Cancel a run still going (oxygenflow#32), its checks with it.
119
pub async fn cancel(app: &Shared, row: &db::repos::RepoRow, run: &fdb::Run, actor: &str) -> Result<(), String> {
120
    if !run.live() {
121
        return Err(format!("run #{} is already {}", run.number, run.status));
122
    }
123
    cancel_with(app, run, &format!("cancelled by {actor}")).await.map_err(|e| {
124
        tracing::warn!("flow: cancelling run #{} of {}/{}: {e}", run.number, row.owner, row.name);
125
        "the run could not be cancelled".to_string()
126
    })
127
}
128
129
async fn cancel_with(app: &Shared, run: &fdb::Run, why: &str) -> sqlx::Result<()> {
130
    let changed = fdb::cancel_run(&app.db, run.id, why).await?;
131
    report_changed(app, run.id, changed).await
132
}
133
134
/// A push makes the push runs still going on its bookmark moot: they are
135
/// cancelled, and the new run's scope reaches back past them.
136
async fn supersede(app: &Shared, row: &db::repos::RepoRow, bookmark: &str, number: i64) {
137
    let Ok(older) = fdb::live_push_runs_before(&app.db, row.id, bookmark, number).await else { return };
138
    for run in older {
139
        if let Err(e) = cancel_with(app, &run, &format!("superseded by run #{number}")).await {
140
            tracing::warn!("flow: superseding run #{} of {}/{}: {e}", run.number, row.owner, row.name);
141
        }
142
    }
143
}
144
145
/// The event a run's trigger stands for: `push`, `tag:NAME`, `manual`,
146
/// `schedule`, and `rerun:N:EVENT` repeating EVENT.
147
fn event_of<'a>(trigger: &'a str, bookmark: &'a str) -> flow_core::pipeline::Event<'a> {
148
    use flow_core::pipeline::Event;
149
    let t = match trigger.strip_prefix("rerun:") {
150
        Some(rest) => rest.split_once(':').map(|(_, e)| e).unwrap_or("manual"),
151
        None => trigger,
152
    };
153
    if let Some(tag) = t.strip_prefix("tag:") {
154
        Event::Tag(tag)
155
    } else if t == "push" {
156
        Event::Push(bookmark)
157
    } else if t == "schedule" {
158
        Event::Schedule
159
    } else {
160
        Event::Manual
161
    }
162
}
163
164
/// The event to repeat when rerunning `run`.
165
fn original_event(run: &fdb::Run) -> String {
166
    let t = run.trigger.as_str();
167
    match t.strip_prefix("rerun:") {
168
        Some(rest) => rest.split_once(':').map(|(_, e)| e.to_string()).unwrap_or_else(|| "manual".into()),
169
        None => t.to_string(),
170
    }
171
}
172
173
/// The paths the changes a push brought touch: the changes in `heads`'
174
/// state that the bookmark's previous run did not have, and the files their
175
/// edits and tree operations name, as the new state places them. `None`
176
/// (run every job) for the bookmark's first run, or when it cannot be told.
177
async fn touched_since_last_run(app: &Shared, row: &db::repos::RepoRow, bookmark: &str, heads: &[ChangeHash]) -> Option<Vec<String>> {
178
    let last = fdb::latest_on(&app.db, row.id, bookmark).await.ok()??;
179
    let old: Vec<ChangeHash> = last.head_list().into_iter().filter_map(ChangeHash::from_hex).collect();
180
    if old.is_empty() {
181
        return None;
182
    }
183
    let repo = RepoName::new(&row.owner, &row.name)?;
184
    let new = heads.to_vec();
185
    app.repos
186
        .read(&repo, move |r, cache| {
187
            let fresh = r.negotiate(&old, &new)?;
188
            let loaded = cache.load(r, &new)?;
189
            let by_node: std::collections::HashMap<oxy_core::ids::NodeId, &String> =
190
                loaded.snapshot.files.iter().map(|(p, f)| (f.node, p)).collect();
191
            let mut paths = std::collections::BTreeSet::new();
192
            for h in fresh {
193
                let (c, _) = r.store().get_change_lazy(&h)?;
194
                let nodes = c.edits.iter().map(|e| e.node()).chain(c.tree_ops.iter().map(|o| o.node()));
195
                for n in nodes {
196
                    if let Some(p) = by_node.get(&n) {
197
                        paths.insert((*p).clone());
198
                    }
199
                }
200
            }
201
            Ok(paths.into_iter().collect::<Vec<String>>())
202
        })
203
        .await
204
        .ok()
205
}
206
207
/// Whether a job runs for a push that touched `touched`: always without a
208
/// scope or without knowing what was touched, else when a path matches.
209
fn in_scope(scope: &[String], touched: Option<&[String]>) -> bool {
210
    let Some(touched) = touched else { return true };
211
    if scope.is_empty() {
212
        return true;
213
    }
214
    let mut b = globset::GlobSetBuilder::new();
215
    for g in scope {
216
        match globset::Glob::new(g) {
217
            Ok(g) => {
218
                b.add(g);
219
            }
220
            Err(_) => return true,
221
        }
222
    }
223
    let Ok(set) = b.build() else { return true };
224
    touched.iter().any(|p| set.is_match(p))
225
}
226
227
/// [`start_run`] for a push: jobs whose `scope` the push did not touch are
228
/// left out of the run, unless a job that runs depends on them.
229
#[allow(clippy::too_many_arguments)]
230
pub async fn start_run_scoped(
231
    app: &Shared,
232
    row: &db::repos::RepoRow,
233
    bookmark: &str,
234
    state: &str,
235
    heads: &[ChangeHash],
236
    actor: &str,
237
    trigger: &str,
238
    touched: Option<&[String]>,
239
) -> Result<Option<i64>, String> {
240
    start_run_with(app, row, bookmark, state, heads, actor, trigger, touched).await
241
}
242
243
/// Start a run of the pipeline in the state `heads` make (whose state
244
/// hash is `state`), recorded on `bookmark`: for a push, "Run pipeline"
245
/// (`trigger` = `manual`) and re-runs (`rerun:N`). `actor` is the account
246
/// whose access the run's subrepo reads are checked against. Returns the
247
/// run's number, or `None` when the state has no pipeline. The caller
248
/// checks that Flow is enabled.
249
pub async fn start_run(
250
    app: &Shared,
251
    row: &db::repos::RepoRow,
252
    bookmark: &str,
253
    state: &str,
254
    heads: &[ChangeHash],
255
    actor: &str,
256
    trigger: &str,
257
) -> Result<Option<i64>, String> {
258
    start_run_with(app, row, bookmark, state, heads, actor, trigger, None).await
259
}
260
261
#[allow(clippy::too_many_arguments)]
262
async fn start_run_with(
263
    app: &Shared,
264
    row: &db::repos::RepoRow,
265
    bookmark: &str,
266
    state: &str,
267
    heads: &[ChangeHash],
268
    actor: &str,
269
    trigger: &str,
270
    touched: Option<&[String]>,
271
) -> Result<Option<i64>, String> {
272
    start_run_at(app, row, bookmark, state, heads, actor, trigger, touched, None).await
273
}
274
275
/// [`start_run_with`], for a scheduled run due at `at` on the hub's clock.
276
#[allow(clippy::too_many_arguments)]
277
async fn start_run_at(
278
    app: &Shared,
279
    row: &db::repos::RepoRow,
280
    bookmark: &str,
281
    state: &str,
282
    heads: &[ChangeHash],
283
    actor: &str,
284
    trigger: &str,
285
    touched: Option<&[String]>,
286
    at: Option<flow_core::pipeline::When>,
287
) -> Result<Option<i64>, String> {
288
    let repo = RepoName::new(&row.owner, &row.name).ok_or("bad repository name")?;
289
    let h = heads.to_vec();
290
    let text = app
291
        .repos
292
        .read(&repo, move |r, cache| {
293
            let loaded = cache.load(r, &h)?;
294
            Ok(match loaded.snapshot.files.get(PIPELINE_PATH).map(|f| &f.content) {
295
                Some(oxy_core::state::FileContent::Text { bytes, .. }) => Some(String::from_utf8_lossy(bytes).into_owned()),
296
                _ => None,
297
            })
298
        })
299
        .await
300
        .map_err(|e| format!("reading the pipeline: {e}"))?;
301
    let Some(text) = text else { return Ok(None) };
302
    let mut hex: Vec<String> = heads.iter().map(|h| h.to_hex()).collect();
303
    hex.sort();
304
    let heads_s = hex.join("+");
305
    let base = db::flow::NewRun { repo_id: row.id, bookmark, state, heads: &heads_s, pipeline_hash: "", actor, trigger, error: None };
306
    let id = match Pipeline::parse(&text) {
307
        Err(problems) => {
308
            let msg = problems.to_string();
309
            db::flow::create_run(&app.db, db::flow::NewRun { error: Some(&msg), ..base }, &[]).await.map_err(|e| e.to_string())?
310
        }
311
        Ok(p) => {
312
            let depths = p.depths();
313
            // Jobs out of the push's scope stay out, unless a kept job
314
            // needs them.
315
            // The jobs this run's event starts (`on`), out of those its
316
            // push touched (`scope`); then what they depend on.
317
            let event = match at {
318
                Some(w) => flow_core::pipeline::Event::ScheduleAt(w),
319
                None => event_of(trigger, bookmark),
320
            };
321
            let mut keep: std::collections::HashSet<String> = p
322
                .order()
323
                .into_iter()
324
                .filter(|j| p.on_of(j).matches(event) && in_scope(&j.scope, touched))
325
                .map(|j| j.name.clone())
326
                .collect();
327
            loop {
328
                let needed: Vec<String> = p
329
                    .order()
330
                    .into_iter()
331
                    .filter(|j| keep.contains(&j.name))
332
                    .flat_map(|j| j.depends_on.iter().cloned())
333
                    .filter(|d| !keep.contains(d))
334
                    .collect();
335
                if needed.is_empty() {
336
                    break;
337
                }
338
                keep.extend(needed);
339
            }
340
            let jobs: Vec<fdb::NewJob> = p
341
                .order()
342
                .into_iter()
343
                .filter(|j| keep.contains(&j.name))
344
                .map(|j| {
345
                    let sb = &p.sandboxes[&j.sandbox];
346
                    fdb::NewJob {
347
                        name: j.name.clone(),
348
                        depth: depths[j.name.as_str()] as i64,
349
                        depends: j.depends_on.clone(),
350
                        spec: serde_json::to_string(j).unwrap_or_default(),
351
                        sandbox: serde_json::to_string(sb).unwrap_or_default(),
352
                        runtime: sb.runtime.as_str().into(),
353
                    }
354
                })
355
                .collect();
356
            if jobs.is_empty() {
357
                // Nothing this event starts (or this push touched).
358
                return Ok(None);
359
            }
360
            let hash = hex::encode(p.hash());
361
            let id =
362
                db::flow::create_run(&app.db, db::flow::NewRun { pipeline_hash: &hash, ..base }, &jobs).await.map_err(|e| e.to_string())?;
363
            if let Ok(Some(run)) = fdb::run_by_id(&app.db, id).await {
364
                for j in &jobs {
365
                    report_check(app, row, state, run.number, &j.name, "queued", "waiting for a runner").await;
366
                }
367
            }
368
            id
369
        }
370
    };
371
    let run = fdb::run_by_id(&app.db, id).await.map_err(|e| e.to_string())?.ok_or("the run vanished")?;
372
    Ok(Some(run.number))
373
}
374
375
/// Record a job's check; it links to the run's page by the run's number.
376
async fn report_check(app: &Shared, row: &db::repos::RepoRow, state: &str, run_number: i64, job: &str, status: &str, summary: &str) {
377
    let url = format!("{}/{}/{}/flow/{run_number}", app.config.base_url(), row.owner, row.name);
378
    let name = format!("flow/{job}");
379
    let c = db::hooks::upsert_check(
380
        &app.db,
381
        row.id,
382
        db::hooks::CheckReport { state, name: &name, status, summary, url: &url, reporter: None, seal: "" },
383
    )
384
    .await;
385
    if let Ok(c) = c {
386
        let api = crate::api::check_of(&c);
387
        app.events.proposal(row.id, hub_api::ProposalEvent { number: 0, kind: "check".into(), check: Some(api.clone()) });
388
        crate::hooks::emit(app, row, "check", "flow", serde_json::json!({ "check": api })).await;
389
    }
390
}
391
392
/// The check status a job's status shows as.
393
fn check_status(job_status: &str) -> &'static str {
394
    match job_status {
395
        "queued" => "queued",
396
        "running" => "running",
397
        "passed" => "passed",
398
        "skipped" => "skipped",
399
        "cancelled" => "cancelled",
400
        _ => "failed",
401
    }
402
}
403
404
/// The runner a bearer token belongs to: a registered runner, or one
405
/// configured with a hub.toml token (tracked from its first request). A
406
/// disabled runner is refused.
407
async fn runner_auth(app: &Shared, headers: &HeaderMap) -> Result<db::runners::Runner, ApiError> {
408
    let Some(t) = oxy_server::bearer_token(headers) else {
409
        return Err(ApiError(StatusCode::UNAUTHORIZED, "a runner token is required".into()));
410
    };
411
    let hash = blake3::hash(t.as_bytes()).to_hex().to_string();
412
    let runner = match db::runners::by_token(&app.db, &hash).await? {
413
        Some(r) => r,
414
        None if app.config.flow.runner_tokens.iter().any(|x| crate::auth::mac_equal(x, t)) => {
415
            db::runners::configured(&app.db, &hash, "runner").await?
416
        }
417
        None => return Err(ApiError(StatusCode::UNAUTHORIZED, "not a runner token".into())),
418
    };
419
    if runner.state == "disabled" {
420
        return Err(ApiError(StatusCode::FORBIDDEN, format!("runner {} is disabled on this hub", runner.name)));
421
    }
422
    Ok(runner)
423
}
424
425
/// `POST /api/v1/flow/runners/register`: a one-time registration token
426
/// from Admin → Runners buys a runner credential, binding the runner's
427
/// signing identity (hub spec §13.2).
428
pub async fn register(
429
    State(app): State<Shared>,
430
    Json(req): Json<flow_core::api::Register>,
431
) -> Result<Json<flow_core::api::Registered>, ApiError> {
432
    let bytes = hex::decode(req.identity.trim()).map_err(|_| ApiError(StatusCode::BAD_REQUEST, "identity: not hex".into()))?;
433
    let identity = oxy_core::identity::Identity::decode(&bytes).map_err(|e| ApiError(StatusCode::BAD_REQUEST, format!("identity: {e}")))?;
434
    let name: String = req.name.chars().filter(|c| c.is_ascii_alphanumeric() || matches!(c, '-' | '_' | '.')).take(64).collect();
435
    if name.is_empty() {
436
        return Err(ApiError(StatusCode::BAD_REQUEST, "give the runner a name of letters, digits, - _ or .".into()));
437
    }
438
    let credential = format!("flowrunner_{}", crate::auth::random_hex(32));
439
    let hash = blake3::hash(credential.as_bytes()).to_hex().to_string();
440
    let reg = blake3::hash(req.token.trim().as_bytes()).to_hex().to_string();
441
    let runner =
442
        db::runners::register(&app.db, &reg, &name, &hash, &identity.hash().to_hex(), &identity.name, "").await?.ok_or_else(|| {
443
            ApiError(StatusCode::FORBIDDEN, "that registration token is unknown, used or expired; create one on Admin → Runners".into())
444
        })?;
445
    db::runners::seen(&app.db, runner.id, None, Some(&req.info)).await?;
446
    tracing::info!("flow: runner {} registered (key {})", runner.name, &runner.key[..12]);
447
    Ok(Json(flow_core::api::Registered { id: runner.id, name: runner.name, credential }))
448
}
449
450
/// `POST /api/v1/flow/claim`.
451
pub async fn claim(State(app): State<Shared>, headers: HeaderMap, Json(req): Json<ClaimRequest>) -> Result<Response, ApiError> {
452
    let me = runner_auth(&app, &headers).await?;
453
    let claimed: String = req.runner.chars().filter(|c| !c.is_control()).take(64).collect();
454
    db::runners::seen(&app.db, me.id, Some(&claimed), req.info.as_ref()).await?;
455
    if let Some(holding) = &req.holding {
456
        end_jobs_not_held(&app, &me, holding).await?;
457
    }
458
    // A draining runner finishes what it has and takes nothing new.
459
    if me.state == "draining" {
460
        return Ok(StatusCode::NO_CONTENT.into_response());
461
    }
462
    let runner = if me.source == "registered" { me.name.clone() } else { claimed };
463
    let secret = format!("{JOB_TOKEN_PREFIX}{}", crate::auth::random_hex(32));
464
    // What it holds and what it is (OxygenFlow spec §11): the images'
465
    // architectures it reports are remembered for every runner's claims.
466
    if let Some(archs) = &req.image_archs {
467
        fdb::record_image_archs(&app.db, archs).await?;
468
    }
469
    let arch = req.info.as_ref().map(|i| i.arch.clone()).filter(|a| !a.is_empty()).unwrap_or_else(|| me.arch.clone());
470
    // What it offers, kept for the placement pass; a job it declined is
471
    // not offered to it again while this stays the same (spec §11.5).
472
    let profile = profile_of(&req, &arch);
473
    db::runners::set_offer(&app.db, me.id, &req.runtimes, req.images.as_deref(), &profile).await?;
474
    let room = fdb::Room {
475
        declined: fdb::declined_by(&app.db, me.id, &profile).await?,
476
        memory_mib: req.free_memory_mib,
477
        cpus: req.free_cpus,
478
        images: req.images.as_ref().map(|i| i.iter().map(|h| h.to_ascii_lowercase()).collect()),
479
        arch,
480
        image_archs: fdb::image_archs(&app.db).await?,
481
    };
482
    // The labels it reports now, or those it registered with, and those
483
    // an administrator gave it (#98).
484
    let mut labels: Vec<String> = match &req.info {
485
        Some(i) => i.labels.clone(),
486
        None => me.reported_label_list().into_iter().map(String::from).collect(),
487
    };
488
    for l in me.admin_label_list() {
489
        if !labels.iter().any(|x| x == l) {
490
            labels.push(l.to_string());
491
        }
492
    }
493
    let Some(job) =
494
        fdb::claim_within(&app.db, &req.runtimes, &runner, &crate::auth::token_hash(&secret), room, &me.scope(), &labels).await?
495
    else {
496
        return Ok(StatusCode::NO_CONTENT.into_response());
497
    };
498
    db::runners::set_job_runner(&app.db, job.id, me.id).await?;
499
    let run = fdb::run_by_id(&app.db, job.run_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such run".into()))?;
500
    let row =
501
        db::repos::get_by_id(&app.db, run.repo_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such repository".into()))?;
502
    report_check(&app, &row, &run.state, run.number, &job.name, "running", &format!("running on {runner}")).await;
503
    let bad = |e: serde_json::Error| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("stored job: {e}"));
504
    // The job's limit: its own timeout, else its sandbox's, else the site
505
    // default, never past the site maximum; recorded for the reaper.
506
    let limits = limits(&app).await;
507
    let spec: flow_core::pipeline::Job = serde_json::from_str(&job.spec).map_err(bad)?;
508
    let sandbox: flow_core::pipeline::Sandbox = serde_json::from_str(&job.sandbox).map_err(bad)?;
509
    let mut timeout_s = spec.timeout_s.or(sandbox.timeout_s).unwrap_or(limits.default_s).min(limits.max_s).max(60);
510
    // ...and never past the runner's own limit, when it has one (#98).
511
    if let Some(m) = me.timeout_minutes.filter(|m| *m > 0) {
512
        timeout_s = timeout_s.min(m as u64 * 60).max(60);
513
    }
514
    fdb::set_job_timeout(&app.db, job.id, timeout_s).await?;
515
    let claim = Claim {
516
        job_id: job.id,
517
        run_id: run.id,
518
        repo: format!("{}/{}", row.owner, row.name),
519
        clone_url: format!("{}/{}/{}", app.config.base_url(), row.owner, row.name),
520
        token: secret,
521
        heads: run.head_list().into_iter().map(String::from).collect(),
522
        pipeline_hash: run.pipeline_hash.clone(),
523
        job: spec,
524
        sandbox,
525
        timeout_s,
526
        subrepos: vec![],
527
        state: run.state.clone(),
528
        bookmark: run.bookmark.clone(),
529
        run_number: run.number,
530
        // The site's, the owner's and the repository's variables (#102).
531
        env: crate::variables::env_for_repo(&app.db, &row).await?,
532
        secrets: vec![],
533
    };
534
    let mut claim = claim;
535
    let (inputs, ids, mut left_out) = subrepo_inputs(&app, &row, &run).await;
536
    // The secrets the job declares, as files (#107); one not set is said
537
    // at the top of the log and left out, so the job fails on the file.
538
    let names: Vec<String> = claim.job.secrets.iter().map(|s| s.name.clone()).collect();
539
    let found = crate::secrets::for_job(&app, &row, &names).await?;
540
    // The identities it asks for (#103): a token each, as a file.
541
    if !claim.job.identities.is_empty() {
542
        let issuer = app.oidc().await.map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("oidc: {e}")))?;
543
        for i in &claim.job.identities {
544
            let token = issuer.mint(&crate::oidc::JobClaims {
545
                issuer: app.config.base_url(),
546
                audience: &i.audience,
547
                owner: &row.owner,
548
                name: &row.name,
549
                visibility: &row.visibility,
550
                bookmark: &run.bookmark,
551
                run_number: run.number,
552
                job: &job.name,
553
                state: &run.state,
554
                runner: &runner,
555
                actor: &run.actor,
556
                now_ms: db::now_ms(),
557
                ttl_s: timeout_s,
558
            });
559
            claim.secrets.push(flow_core::api::SecretInput {
560
                name: format!("identity:{}", i.audience),
561
                path: i.mount_path.clone(),
562
                value: token,
563
            });
564
        }
565
    }
566
    for s in &claim.job.secrets {
567
        match found.iter().find(|(n, _)| *n == s.name) {
568
            Some((name, value)) => {
569
                claim.secrets.push(flow_core::api::SecretInput { name: name.clone(), path: s.mount_path.clone(), value: value.clone() })
570
            }
571
            None => left_out.push(format!(
572
                "secret {:?} is not set for this repository, its owner or the site; {} will be missing",
573
                s.name, s.mount_path
574
            )),
575
        }
576
    }
577
    if !ids.is_empty() {
578
        fdb::set_also_read(&app.db, job.id, &ids).await?;
579
    }
580
    // Say at the top of the job's log what it will not have (oxygenhub#56),
581
    // rather than let it fail later on missing files.
582
    if !left_out.is_empty() {
583
        let records: Vec<flow_core::api::Record> =
584
            left_out.into_iter().map(|text| flow_core::api::Record { t: 0, stream: "flow".into(), text }).collect();
585
        fdb::append_log(&app.db, job.id, &records, MAX_RECORDS as i64).await?;
586
    }
587
    claim.subrepos = inputs;
588
    Ok(Json(claim).into_response())
589
}
590
591
/// BLAKE3 over what a claiming runner offers: tiers, images, architecture,
592
/// labels and capacity, each list sorted.
593
fn profile_of(req: &ClaimRequest, arch: &str) -> String {
594
    let sorted = |v: &[String]| {
595
        let mut v = v.to_vec();
596
        v.sort();
597
        v
598
    };
599
    let info = req.info.as_ref();
600
    let v = serde_json::json!({
601
        "arch": arch,
602
        "tiers": sorted(&req.runtimes),
603
        "images": req.images.as_deref().map(sorted),
604
        "labels": info.map(|i| sorted(&i.labels)),
605
        "memory_mib": info.and_then(|i| i.memory_mib),
606
        "cpus": info.and_then(|i| i.cpus),
607
    });
608
    blake3::hash(v.to_string().as_bytes()).to_hex().to_string()
609
}
610
611
/// A job handed back by this many different runners is unplaceable.
612
pub const UNPLACEABLE_AFTER: usize = 3;
613
614
/// `POST /api/v1/flow/jobs/{id}/decline` (OxygenFlow spec §11.5): the
615
/// runner will not run a job it was handed. The job is queued again and
616
/// not offered to that runner while its profile stays the same (and for
617
/// [`fdb::DECLINE_HOLDS_MS`] at most); once [`UNPLACEABLE_AFTER`]
618
/// different runners have declined it, it ends as unplaceable with their
619
/// reasons. A decline is never a failed check.
620
pub async fn decline(
621
    State(app): State<Shared>,
622
    headers: HeaderMap,
623
    Path(id): Path<i64>,
624
    Json(d): Json<flow_core::api::Decline>,
625
) -> Result<StatusCode, ApiError> {
626
    let me = runner_auth(&app, &headers).await?;
627
    db::runners::seen(&app.db, me.id, None, None).await?;
628
    let reason: String = d.reason.chars().map(|c| if c.is_control() { ' ' } else { c }).take(500).collect();
629
    let reason = match reason.trim() {
630
        "" => "no reason given".to_string(),
631
        r => r.to_string(),
632
    };
633
    let Some(all) = fdb::decline(&app.db, id, me.id, &me.name, &me.profile, &reason).await? else {
634
        return Err(ApiError(StatusCode::CONFLICT, "that job is not running on this runner".into()));
635
    };
636
    tracing::info!("flow: runner {} declined job {id}: {reason}", me.name);
637
    let job = fdb::job(&app.db, id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such job".into()))?;
638
    let runners: std::collections::BTreeSet<i64> = all.iter().map(|d| d.runner_id).collect();
639
    if runners.len() >= UNPLACEABLE_AFTER {
640
        let error = format!(
641
            "unplaceable: declined by {} runners: {}",
642
            runners.len(),
643
            all.iter().map(|d| format!("{}: {}", d.runner, d.reason)).collect::<Vec<_>>().join("; ")
644
        );
645
        let changed =
646
            fdb::finish(&app.db, id, fdb::Outcome { status: "error", exit_code: None, error: &error, receipt: "", receipt_hash: "" })
647
                .await?;
648
        report_changed(&app, job.run_id, changed).await?;
649
    } else if let Some(run) = fdb::run_by_id(&app.db, job.run_id).await? {
650
        if let Some(row) = db::repos::get_by_id(&app.db, run.repo_id).await? {
651
            let summary = format!("declined by {}: {reason}; waiting for another runner", me.name);
652
            report_check(&app, &row, &run.state, run.number, &job.name, "queued", &summary).await;
653
            fdb::set_waiting(&app.db, id, &summary).await?;
654
        }
655
    }
656
    Ok(StatusCode::NO_CONTENT)
657
}
658
659
/// Why no runner can take `job` of a repository of `owner` (id `repo_id`)
660
/// now (spec §11.4), from what each runner offered at its last claim:
661
/// empty when one could (it waits its turn or a free slot), else what
662
/// each one lacks. `declines` are the job's.
663
pub fn why_waiting(
664
    job: &fdb::Job,
665
    owner: &str,
666
    repo_id: i64,
667
    runners: &[db::runners::Runner],
668
    image_archs: &std::collections::HashMap<String, String>,
669
    declines: &[fdb::DeclineRow],
670
    now: i64,
671
) -> String {
672
    let sandbox: serde_json::Value = serde_json::from_str(&job.sandbox).unwrap_or_default();
673
    let rootfs = sandbox["rootfs"].as_str().unwrap_or("");
674
    let wanted_arch = job.arch().or_else(|| image_archs.get(rootfs).cloned());
675
    let need = |k: &str| sandbox[k].as_u64().unwrap_or(0);
676
    let needs = job.needs();
677
    let lacks = |r: &db::runners::Runner| -> Option<String> {
678
        if r.state == "draining" {
679
            return Some("is draining".into());
680
        }
681
        let scope = r.scope();
682
        if (!scope.owner.is_empty() && scope.owner != owner) || scope.repo_id.is_some_and(|id| id != repo_id) {
683
            return Some("serves other repositories".into());
684
        }
685
        let tiers = r.tier_list();
686
        if !tiers.is_empty() && !tiers.contains(&job.runtime.as_str()) {
687
            return Some(format!("has no {} tier (it offers {})", job.runtime, tiers.join(", ")));
688
        }
689
        if let Some(w) = wanted_arch.as_deref().filter(|w| !r.arch.is_empty() && *w != r.arch) {
690
            return Some(format!("is {}; the job needs {w}", r.arch));
691
        }
692
        if r.image_list().is_some_and(|have| !have.iter().any(|h| h == rootfs)) {
693
            return Some(format!("does not hold image {}", &rootfs[..12.min(rootfs.len())]));
694
        }
695
        let labels = r.label_list();
696
        if let Some(l) = needs.iter().find(|l| !labels.contains(&l.as_str())) {
697
            return Some(format!("lacks label {l}"));
698
        }
699
        if let Some(m) = r.memory_mib.filter(|m| (*m as u64) < need("memory_mib")) {
700
            return Some(format!("has {m} MiB; the job needs {} MiB", need("memory_mib")));
701
        }
702
        if let Some(c) = r.cpus.filter(|c| (*c as u64) < need("cpus")) {
703
            return Some(format!("has {c} CPUs; the job needs {}", need("cpus")));
704
        }
705
        if let Some(d) = declines.iter().find(|d| d.runner_id == r.id && now - d.ms < fdb::DECLINE_HOLDS_MS) {
706
            return Some(format!("declined it: {}", d.reason));
707
        }
708
        None
709
    };
710
    let online: Vec<&db::runners::Runner> = runners.iter().filter(|r| r.state != "disabled" && r.online()).collect();
711
    if online.is_empty() {
712
        return "no runner is online".into();
713
    }
714
    let mut why = Vec::new();
715
    for r in &online {
716
        match lacks(r) {
717
            None => return String::new(),
718
            Some(w) => why.push(format!("{} {w}", r.name)),
719
        }
720
    }
721
    let more = why.len().saturating_sub(4);
722
    why.truncate(4);
723
    let mut text = format!("no runner can take it: {}", why.join("; "));
724
    if more > 0 {
725
        text.push_str(&format!("; and {more} more"));
726
    }
727
    text
728
}
729
730
/// One placement pass: say why each ready queued job waits.
731
pub async fn explain_waits(app: &Shared) -> sqlx::Result<()> {
732
    let ready = fdb::ready_queued(&app.db).await?;
733
    if ready.is_empty() {
734
        return Ok(());
735
    }
736
    let runners = db::runners::list(&app.db).await?;
737
    let archs = fdb::image_archs(&app.db).await?;
738
    let now = db::now_ms();
739
    for (job, owner, repo_id) in ready {
740
        let declines = fdb::declines_of_job(&app.db, job.id).await?;
741
        let why = why_waiting(&job, &owner, repo_id, &runners, &archs, &declines, now);
742
        fdb::set_waiting(&app.db, job.id, &why).await?;
743
    }
744
    Ok(())
745
}
746
747
/// Run a placement pass every 15 seconds.
748
pub async fn placement(app: Shared) {
749
    loop {
750
        tokio::time::sleep(std::time::Duration::from_secs(15)).await;
751
        if let Err(e) = explain_waits(&app).await {
752
            tracing::warn!("flow: placement pass: {e}");
753
        }
754
    }
755
}
756
757
/// The Oxygen subrepo pins in `name`'s state `heads`: (path, repository id, pinned heads).
758
async fn pins_in(
759
    app: &Shared,
760
    name: crate::repos::RepoName,
761
    heads: Vec<oxy_core::hash::ChangeHash>,
762
) -> Vec<(String, [u8; 16], Vec<oxy_core::hash::ChangeHash>)> {
763
    app.repos
764
        .read(&name, move |repo, cache| {
765
            let loaded = cache.load(repo, &heads)?;
766
            let mut out = Vec::new();
767
            for (path, f) in &loaded.snapshot.files {
768
                if f.kind != oxy_core::objects::NodeKind::Subrepo {
769
                    continue;
770
                }
771
                if let oxy_core::state::FileContent::Opaque { blobs } = &f.content {
772
                    if let Some(b) = blobs.first() {
773
                        if let Ok(oxy_core::subrepo::Pin::Oxy { repo: id, heads, .. }) = repo
774
                            .read_blob(b)
775
                            .map_err(crate::repos::RepoError::Oxy)
776
                            .and_then(|bytes| oxy_core::subrepo::Pin::decode(&bytes).map_err(|e| crate::repos::RepoError::Oxy(e.into())))
777
                        {
778
                            out.push((path.clone(), id, heads));
779
                        }
780
                    }
781
                }
782
            }
783
            Ok(out)
784
        })
785
        .await
786
        .unwrap_or_default()
787
}
788
789
/// How deep subrepos of subrepos are followed.
790
const MAX_SUBREPO_DEPTH: usize = 3;
791
792
/// The Oxygen subrepos the run's state pins, and those their states pin in
793
/// turn (parents first, paths joined), that live on this hub and that the
794
/// run's pusher may read: what the job's workspace needs beside its own
795
/// tree. A pin the hub cannot serve is left out, with what it pins, and
796
/// the third list says why, for the job's log.
797
async fn subrepo_inputs(
798
    app: &Shared,
799
    row: &db::repos::RepoRow,
800
    run: &fdb::Run,
801
) -> (Vec<flow_core::api::SubrepoInput>, Vec<i64>, Vec<String>) {
802
    let Some(name) = crate::repos::RepoName::new(&row.owner, &row.name) else { return (vec![], vec![], vec![]) };
803
    let heads: Vec<oxy_core::hash::ChangeHash> = run.head_list().into_iter().filter_map(oxy_core::hash::ChangeHash::from_hex).collect();
804
    let pusher = db::accounts::account_by_name(&app.db, &run.actor).await.ok().flatten();
805
    let rows = db::repos::list(&app.db).await.unwrap_or_default();
806
    let mut inputs = Vec::new();
807
    let mut ids = Vec::new();
808
    let mut left_out = Vec::new();
809
    let actor = if run.actor.is_empty() { "the run's starter".to_string() } else { run.actor.clone() };
810
    // (path prefix, repository, heads, depth), parents before children.
811
    let mut queue = std::collections::VecDeque::from([(String::new(), name, heads, 0usize)]);
812
    while let Some((prefix, name, heads, depth)) = queue.pop_front() {
813
        if depth == MAX_SUBREPO_DEPTH {
814
            continue;
815
        }
816
        for (path, id, pinned) in pins_in(app, name, heads).await {
817
            let path = if prefix.is_empty() { path } else { format!("{prefix}/{path}") };
818
            let mut hosted = false;
819
            for other in &rows {
820
                let Some(other_name) = crate::repos::RepoName::new(&other.owner, &other.name) else { continue };
821
                let matches = app.repos.read(&other_name, move |r, _| Ok(r.id().ok() == Some(id))).await.unwrap_or(false);
822
                if !matches {
823
                    continue;
824
                }
825
                hosted = true;
826
                let role = crate::auth::roles::role_for(&app.db, pusher.as_ref(), other).await.ok().flatten();
827
                if crate::auth::roles::allows(role, crate::auth::Access::Read) {
828
                    inputs.push(flow_core::api::SubrepoInput {
829
                        path: path.clone(),
830
                        repo: format!("{}/{}", other.owner, other.name),
831
                        clone_url: format!("{}/{}/{}", app.config.base_url(), other.owner, other.name),
832
                        heads: pinned.iter().map(|h| h.to_hex()).collect(),
833
                    });
834
                    if !ids.contains(&other.id) {
835
                        ids.push(other.id);
836
                    }
837
                    queue.push_back((path.clone(), other_name, pinned.clone(), depth + 1));
838
                } else {
839
                    left_out.push(format!(
840
                        "subrepo {path} ({}/{}) is left out: {actor} cannot read it, so the job has no files there",
841
                        other.owner, other.name
842
                    ));
843
                }
844
                break;
845
            }
846
            if !hosted {
847
                left_out.push(format!("subrepo {path} is left out: the repository it pins is not on this hub"));
848
            }
849
        }
850
    }
851
    (inputs, ids, left_out)
852
}
853
854
async fn running_job(app: &Shared, id: i64) -> Result<fdb::Job, ApiError> {
855
    match fdb::job(&app.db, id).await? {
856
        Some(j) if j.status == "running" => Ok(j),
857
        Some(_) => Err(ApiError(StatusCode::CONFLICT, "this job is not running".into())),
858
        None => Err(ApiError(StatusCode::NOT_FOUND, "no such job".into())),
859
    }
860
}
861
862
/// `POST /api/v1/flow/jobs/{id}/log`.
863
pub async fn log(
864
    State(app): State<Shared>,
865
    headers: HeaderMap,
866
    Path(id): Path<i64>,
867
    Json(mut batch): Json<LogBatch>,
868
) -> Result<StatusCode, ApiError> {
869
    runner_auth(&app, &headers).await?;
870
    let job = running_job(&app, id).await?;
871
    // An empty batch asks only whether the job still runs: a cancelled
872
    // one answers 409 and its runner stops it (oxygenflow#32).
873
    if batch.records.is_empty() {
874
        return Ok(StatusCode::NO_CONTENT);
875
    }
876
    // A secret's value never reaches the stored log (#107).
877
    let values: Vec<String> = match serde_json::from_str::<flow_core::Job>(&job.spec) {
878
        Ok(spec) if !spec.secrets.is_empty() => {
879
            let run = fdb::run_by_id(&app.db, job.run_id).await?;
880
            let row = match run {
881
                Some(run) => db::repos::get_by_id(&app.db, run.repo_id).await?,
882
                None => None,
883
            };
884
            match row {
885
                Some(row) => {
886
                    let names: Vec<String> = spec.secrets.iter().map(|s| s.name.clone()).collect();
887
                    crate::secrets::for_job(&app, &row, &names).await?.into_iter().map(|(_, v)| v).collect()
888
                }
889
                None => vec![],
890
            }
891
        }
892
        _ => vec![],
893
    };
894
    let kid = app.oidc.get().map(|i| i.kid.clone());
895
    for r in &mut batch.records {
896
        if !values.is_empty() {
897
            r.text = crate::secrets::mask(&r.text, &values);
898
        }
899
        if let Some(kid) = &kid {
900
            r.text = crate::oidc::mask(&r.text, kid);
901
        }
902
        if r.text.len() > MAX_RECORD_TEXT {
903
            let mut cut = MAX_RECORD_TEXT;
904
            while !r.text.is_char_boundary(cut) {
905
                cut -= 1;
906
            }
907
            r.text.truncate(cut);
908
        }
909
        if !matches!(r.stream.as_str(), "out" | "err" | "console" | "flow") {
910
            r.stream = "console".into();
911
        }
912
    }
913
    let kept = fdb::append_log(&app.db, job.id, &batch.records, MAX_RECORDS as i64).await?;
914
    if kept < batch.records.len() && job.log_count < MAX_RECORDS as i64 {
915
        tracing::info!("flow: job {} reached the log limit", job.id);
916
    }
917
    Ok(StatusCode::NO_CONTENT)
918
}
919
920
/// Largest artifact the hub keeps; the runner refuses larger ones first.
921
pub const MAX_ARTIFACT_BYTES: usize = 512 << 20;
922
923
/// `POST /api/v1/flow/jobs/{id}/artifacts/{name}`: the bytes of an artifact
924
/// the job declares (OxygenFlow spec §5.2), kept as a blob of the run's
925
/// repository. Answers `{"blob"}`.
926
pub async fn artifact(
927
    State(app): State<Shared>,
928
    headers: HeaderMap,
929
    Path((id, name)): Path<(i64, String)>,
930
    body: axum::body::Bytes,
931
) -> Result<Response, ApiError> {
932
    runner_auth(&app, &headers).await?;
933
    let job = running_job(&app, id).await?;
934
    let spec: flow_core::Job =
935
        serde_json::from_str(&job.spec).map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("stored job: {e}")))?;
936
    let Some(out) = spec.outputs.iter().find(|o| o.name == name) else {
937
        return Err(ApiError(StatusCode::BAD_REQUEST, format!("job {} declares no artifact {name:?}", job.name)));
938
    };
939
    let run = fdb::run_by_id(&app.db, job.run_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such run".into()))?;
940
    let row =
941
        db::repos::get_by_id(&app.db, run.repo_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such repository".into()))?;
942
    let repo = RepoName::new(&row.owner, &row.name).ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "bad repository name".into()))?;
943
    // A report's artifact is its manifest (#127): checked, and its files
944
    // found in the store, before it is kept.
945
    let report = match out.kind {
946
        flow_core::OutputKind::Report => Some(report_manifest(&app, &repo, &name, &body).await?),
947
        flow_core::OutputKind::Artifact => None,
948
    };
949
    let content = blake3::hash(&body).to_hex().to_string();
950
    let size = body.len() as i64;
951
    let artifact = report.is_none();
952
    let report_bytes = (artifact && name == crate::coverage::ARTIFACT).then(|| body.clone());
953
    let junit_bytes = (artifact && name == crate::junit::ARTIFACT).then(|| body.clone());
954
    let blob = app
955
        .repos
956
        .write(&repo, move |r, _| Ok(r.store_blob(&body)?))
957
        .await
958
        .map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("keep the artifact: {e}")))?;
959
    let a = fdb::Artifact { job_id: job.id, name, path: out.path.clone(), blob: blob.to_hex(), size, content, created_ms: db::now_ms() };
960
    fdb::add_artifact(&app.db, &a).await?;
961
    if let Some((manifest, declaration)) = report {
962
        let origin = crate::reports::Origin {
963
            repo_id: row.id,
964
            run_id: run.id,
965
            run_number: run.number,
966
            job_id: job.id,
967
            job: &job.name,
968
            bookmark: &run.bookmark,
969
            state: &run.state,
970
        };
971
        let said = match declaration {
972
            Ok(d) => match crate::reports::keep(&app, &origin, &a.name, &d, &manifest, &a.blob).await {
973
                Ok(id) => {
974
                    report_threshold_checks(&app, &row, &run, &a.name, id, &d.metrics).await;
975
                    let s = if d.metrics.len() == 1 { "" } else { "s" };
976
                    format!("report {}: {} \"{}\", {} metric{s}", a.name, d.kind, d.title, d.metrics.len())
977
                }
978
                Err(e) => format!("report {}: not indexed: {e}", a.name),
979
            },
980
            Err(why) => {
981
                let d = crate::reports::Declaration::unreadable(&a.name, &why);
982
                crate::reports::keep(&app, &origin, &a.name, &d, &manifest, &a.blob).await.ok();
983
                format!("report {}: report.toml: {why}", a.name)
984
            }
985
        };
986
        let record = flow_core::api::Record { t: 0, stream: "flow".into(), text: said };
987
        fdb::append_log(&app.db, job.id, &[record], MAX_RECORDS as i64).await.ok();
988
    }
989
    // A coverage report (#113) is read now; what it says, or why it could
990
    // not be read, goes in the job's log. It never fails the upload.
991
    if let Some(bytes) = report_bytes {
992
        let said = match read_coverage(&app, &repo, &row, &run, &job, &a.blob, bytes).await {
993
            Ok(text) | Err(text) => text,
994
        };
995
        let record = flow_core::api::Record { t: 0, stream: "flow".into(), text: said };
996
        fdb::append_log(&app.db, job.id, &[record], MAX_RECORDS as i64).await.ok();
997
    }
998
    // A test report (#114) likewise.
999
    if let Some(bytes) = junit_bytes {
1000
        let said = match read_junit(&app, &row, &run, &job, &a.blob, bytes).await {
1001
            Ok(text) | Err(text) => text,
1002
        };
1003
        let record = flow_core::api::Record { t: 0, stream: "flow".into(), text: said };
1004
        fdb::append_log(&app.db, job.id, &[record], MAX_RECORDS as i64).await.ok();
1005
    }
1006
    Ok(Json(serde_json::json!({ "blob": a.blob })).into_response())
1007
}
1008
1009
/// A check per metric with a threshold (#127, docs/reports.md §5), named
1010
/// `report/NAME/METRIC` on the run's state: failed when the value crosses
1011
/// `fail_above` or `fail_below`, which a proposal may require.
1012
async fn report_threshold_checks(
1013
    app: &Shared,
1014
    row: &db::repos::RepoRow,
1015
    run: &fdb::Run,
1016
    name: &str,
1017
    id: i64,
1018
    metrics: &[crate::reports::Metric],
1019
) {
1020
    let url = format!("{}/{}/{}/reports/{name}?id={id}", app.config.base_url(), row.owner, row.name);
1021
    for m in metrics.iter().filter(|m| m.fail_above.is_some() || m.fail_below.is_some()) {
1022
        let unit = if m.unit.is_empty() { String::new() } else { format!(" {}", m.unit) };
1023
        let above = m.fail_above.filter(|a| m.value > *a);
1024
        let below = m.fail_below.filter(|b| m.value < *b);
1025
        let (status, summary) = match (above, below) {
1026
            (Some(a), _) => ("failed", format!("{}{unit}, above {a}", m.value)),
1027
            (_, Some(b)) => ("failed", format!("{}{unit}, below {b}", m.value)),
1028
            _ => {
1029
                let limits: Vec<String> = [m.fail_above.map(|a| format!("at most {a}")), m.fail_below.map(|b| format!("at least {b}"))]
1030
                    .into_iter()
1031
                    .flatten()
1032
                    .collect();
1033
                ("passed", format!("{}{unit} ({})", m.value, limits.join(", ")))
1034
            }
1035
        };
1036
        let check = format!("report/{name}/{}", m.key);
1037
        if let Ok(c) = db::hooks::upsert_check(
1038
            &app.db,
1039
            row.id,
1040
            db::hooks::CheckReport { state: &run.state, name: &check, status, summary: &summary, url: &url, reporter: None, seal: "" },
1041
        )
1042
        .await
1043
        {
1044
            let api = crate::api::check_of(&c);
1045
            app.events.proposal(row.id, hub_api::ProposalEvent { number: 0, kind: "check".into(), check: Some(api.clone()) });
1046
            crate::hooks::emit(app, row, "check", "flow", serde_json::json!({ "check": api })).await;
1047
        }
1048
    }
1049
}
1050
1051
/// Check report manifest `body` (#127) and that the repository holds every
1052
/// file it names; answers it with the report's declaration, or why that
1053
/// could not be read.
1054
async fn report_manifest(
1055
    app: &Shared,
1056
    repo: &RepoName,
1057
    name: &str,
1058
    body: &[u8],
1059
) -> Result<(crate::reports::Manifest, Result<crate::reports::Declaration, String>), ApiError> {
1060
    let manifest = crate::reports::parse_manifest(body).map_err(|e| ApiError(StatusCode::BAD_REQUEST, e))?;
1061
    let refs: Vec<(String, oxy_core::hash::ObjectRef)> =
1062
        manifest.files.iter().filter_map(|f| Some((f.path.clone(), oxy_core::hash::ObjectRef::from_hex(&f.blob)?))).collect();
1063
    let declaration = manifest.blob("report.toml").and_then(oxy_core::hash::ObjectRef::from_hex);
1064
    let found = app
1065
        .repos
1066
        .read(repo, move |r, _| {
1067
            if let Some((path, _)) = refs.iter().find(|(_, b)| !r.store().has(b)) {
1068
                return Ok(Err(path.clone()));
1069
            }
1070
            Ok(Ok(declaration.map(|b| r.read_blob(&b)).transpose()?))
1071
        })
1072
        .await
1073
        .map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("read the report: {e}")))?;
1074
    let toml = found.map_err(|path| ApiError(StatusCode::BAD_REQUEST, format!("report file {path} was not uploaded; upload it first")))?;
1075
    let declaration = crate::reports::parse_declaration(name, &toml.unwrap_or_default());
1076
    Ok((manifest, declaration))
1077
}
1078
1079
/// `POST /api/v1/flow/jobs/{id}/reports/{name}/files`: one file of a report
1080
/// the job declares (#127), its path in `X-Flow-Report-Path`, kept as a
1081
/// blob of the run's repository. Answers `{"blob"}`; the manifest naming
1082
/// the files follows as the report's artifact.
1083
pub async fn report_file(
1084
    State(app): State<Shared>,
1085
    headers: HeaderMap,
1086
    Path((id, name)): Path<(i64, String)>,
1087
    body: axum::body::Bytes,
1088
) -> Result<Response, ApiError> {
1089
    runner_auth(&app, &headers).await?;
1090
    let job = running_job(&app, id).await?;
1091
    let spec: flow_core::Job =
1092
        serde_json::from_str(&job.spec).map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("stored job: {e}")))?;
1093
    if !spec.outputs.iter().any(|o| o.name == name && o.kind == flow_core::OutputKind::Report) {
1094
        return Err(ApiError(StatusCode::BAD_REQUEST, format!("job {} declares no report {name:?}", job.name)));
1095
    }
1096
    let path = headers.get("x-flow-report-path").and_then(|v| v.to_str().ok()).unwrap_or_default();
1097
    if !crate::reports::path_ok(path) {
1098
        return Err(ApiError(StatusCode::BAD_REQUEST, format!("report file {path:?}: give its path in X-Flow-Report-Path")));
1099
    }
1100
    let run = fdb::run_by_id(&app.db, job.run_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such run".into()))?;
1101
    let row =
1102
        db::repos::get_by_id(&app.db, run.repo_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such repository".into()))?;
1103
    let repo = RepoName::new(&row.owner, &row.name).ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "bad repository name".into()))?;
1104
    let blob = app
1105
        .repos
1106
        .write(&repo, move |r, _| Ok(r.store_blob(&body)?))
1107
        .await
1108
        .map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("keep the report file: {e}")))?;
1109
    Ok(Json(serde_json::json!({ "blob": blob.to_hex() })).into_response())
1110
}
1111
1112
/// Parse and keep a job's JUnit report: what it says, or why not.
1113
async fn read_junit(
1114
    app: &Shared,
1115
    row: &db::repos::RepoRow,
1116
    run: &fdb::Run,
1117
    job: &fdb::Job,
1118
    blob: &str,
1119
    bytes: axum::body::Bytes,
1120
) -> Result<String, String> {
1121
    if bytes.len() > crate::junit::MAX_REPORT {
1122
        return Err(format!("tests: the report is {} MiB; at most {} MiB is read", bytes.len() >> 20, crate::junit::MAX_REPORT >> 20));
1123
    }
1124
    let text = String::from_utf8(bytes.to_vec()).map_err(|_| "tests: the report is not UTF-8 text (JUnit XML is expected)".to_string())?;
1125
    let cases = tokio::task::spawn_blocking(move || crate::junit::parse(&text))
1126
        .await
1127
        .map_err(|e| format!("tests: {e}"))?
1128
        .map_err(|e| format!("tests: {e}"))?;
1129
    let origin = crate::junit::Origin {
1130
        repo_id: row.id,
1131
        run_id: run.id,
1132
        run_number: run.number,
1133
        job_id: job.id,
1134
        job: &job.name,
1135
        bookmark: &run.bookmark,
1136
        state: &run.state,
1137
        blob,
1138
    };
1139
    crate::junit::keep(app, &origin, &cases).await.map_err(|e| format!("tests: {e}"))?;
1140
    let r = crate::junit::for_run(&app.db, run.id)
1141
        .await
1142
        .map_err(|e| format!("tests: {e}"))?
1143
        .into_iter()
1144
        .find(|r| r.job_id == job.id)
1145
        .ok_or_else(|| "tests: the report was not kept".to_string())?;
1146
    // The `tests` check on the run's state: what fails, and which of those
1147
    // passed in the same job's latest report on the default bookmark, so a
1148
    // proposal names what it broke.
1149
    let base = crate::junit::baseline(&app.db, &r, &row.default_bookmark).await.map_err(|e| format!("tests: {e}"))?;
1150
    let newly = match &base {
1151
        Some(b) if r.broken() > 0 => crate::junit::newly_failing(&app.db, r.id, b.id).await.map_err(|e| format!("tests: {e}"))?,
1152
        _ => vec![],
1153
    };
1154
    let (status, summary) = if r.broken() == 0 {
1155
        ("passed", format!("{} passed", r.passed))
1156
    } else if let Some(b) = &base {
1157
        let names: Vec<&str> = newly.iter().take(3).map(String::as_str).collect();
1158
        let more = if newly.len() > 3 { format!(", +{}", newly.len() - 3) } else { String::new() };
1159
        let new = if newly.is_empty() {
1160
            format!("none new against {} run #{}", row.default_bookmark, b.run_number)
1161
        } else {
1162
            format!("{} new against {}: {}{more}", newly.len(), row.default_bookmark, names.join(", "))
1163
        };
1164
        ("failed", format!("{} failing; {new}", r.broken()))
1165
    } else {
1166
        ("failed", format!("{} failing", r.broken()))
1167
    };
1168
    let url = format!("{}/{}/{}/tests?report={}", app.config.base_url(), row.owner, row.name, r.id);
1169
    if let Ok(c) = db::hooks::upsert_check(
1170
        &app.db,
1171
        row.id,
1172
        db::hooks::CheckReport { state: &run.state, name: "tests", status, summary: &summary, url: &url, reporter: None, seal: "" },
1173
    )
1174
    .await
1175
    {
1176
        let api = crate::api::check_of(&c);
1177
        app.events.proposal(row.id, hub_api::ProposalEvent { number: 0, kind: "check".into(), check: Some(api.clone()) });
1178
        crate::hooks::emit(app, row, "check", "flow", serde_json::json!({ "check": api })).await;
1179
    }
1180
    Ok(format!("tests: {} in {}", r.summary(), human_ms(r.time_ms)))
1181
}
1182
1183
/// Parse and keep a job's coverage report: what it says, or why not.
1184
async fn read_coverage(
1185
    app: &Shared,
1186
    repo: &RepoName,
1187
    row: &db::repos::RepoRow,
1188
    run: &fdb::Run,
1189
    job: &fdb::Job,
1190
    blob: &str,
1191
    bytes: axum::body::Bytes,
1192
) -> Result<String, String> {
1193
    if bytes.len() > crate::coverage::MAX_REPORT {
1194
        return Err(format!(
1195
            "coverage: the report is {} MiB; at most {} MiB is read",
1196
            bytes.len() >> 20,
1197
            crate::coverage::MAX_REPORT >> 20
1198
        ));
1199
    }
1200
    let text = String::from_utf8(bytes.to_vec())
1201
        .map_err(|_| "coverage: the report is not text (LCOV or Cobertura XML is expected)".to_string())?;
1202
    let (files, outside, format) = tokio::task::spawn_blocking(move || crate::coverage::parse(&text))
1203
        .await
1204
        .map_err(|e| format!("coverage: {e}"))?
1205
        .map_err(|e| format!("coverage: {e}"))?;
1206
    let heads: Vec<ChangeHash> = run.head_list().into_iter().filter_map(ChangeHash::from_hex).collect();
1207
    let in_state: std::collections::HashSet<String> = app
1208
        .repos
1209
        .read(repo, move |r, cache| Ok(cache.load(r, &heads)?.snapshot.files.keys().cloned().collect()))
1210
        .await
1211
        .map_err(|e| format!("coverage: the run's files could not be read: {e}"))?;
1212
    let heads: Vec<String> = run.head_list().into_iter().map(String::from).collect();
1213
    let origin = crate::coverage::Origin {
1214
        repo_id: row.id,
1215
        run_id: run.id,
1216
        run_number: run.number,
1217
        job_id: job.id,
1218
        job: &job.name,
1219
        bookmark: &run.bookmark,
1220
        state: &run.state,
1221
        heads: &heads,
1222
        blob,
1223
        format,
1224
    };
1225
    let id = crate::coverage::keep(app, &origin, files, outside, &in_state).await.map_err(|e| format!("coverage: {e}"))?;
1226
    let Some(r) = crate::coverage::report(&app.db, row.id, id).await.map_err(|e| format!("coverage: {e}"))? else {
1227
        return Err("coverage: the report was not kept".into());
1228
    };
1229
    // The `coverage` check on the run's state (#113): against the default
1230
    // bookmark, under the repository's rules.
1231
    let base = crate::coverage::baseline(&app.db, &r, &row.default_bookmark).await.map_err(|e| format!("coverage: {e}"))?;
1232
    let rules = crate::coverage::rules(&app.db, row.id).await.map_err(|e| format!("coverage: {e}"))?;
1233
    let (status, summary) = crate::coverage::verdict(&r, base.as_ref(), &row.default_bookmark, rules);
1234
    let url = format!("{}/{}/{}/coverage?report={}", app.config.base_url(), row.owner, row.name, r.id);
1235
    if let Ok(c) = db::hooks::upsert_check(
1236
        &app.db,
1237
        row.id,
1238
        db::hooks::CheckReport { state: &run.state, name: "coverage", status, summary: &summary, url: &url, reporter: None, seal: "" },
1239
    )
1240
    .await
1241
    {
1242
        let api = crate::api::check_of(&c);
1243
        app.events.proposal(row.id, hub_api::ProposalEvent { number: 0, kind: "check".into(), check: Some(api.clone()) });
1244
        crate::hooks::emit(app, row, "check", "flow", serde_json::json!({ "check": api })).await;
1245
    }
1246
    let pct = r.percent().map(|p| format!("{p:.1}%")).unwrap_or_else(|| "no lines".into());
1247
    let dropped =
1248
        if r.dropped > 0 { format!(" ({} named files are not in this state and were left out)", r.dropped) } else { String::new() };
1249
    Ok(format!("coverage: {pct} of lines ({} of {}) in {} files{dropped}", r.lines_hit, r.lines_found, r.files))
1250
}
1251
1252
/// `POST /api/v1/flow/jobs/{id}/finish`.
1253
pub async fn finish(
1254
    State(app): State<Shared>,
1255
    headers: HeaderMap,
1256
    Path(id): Path<i64>,
1257
    Json(f): Json<Finish>,
1258
) -> Result<StatusCode, ApiError> {
1259
    let me = runner_auth(&app, &headers).await?;
1260
    let job = running_job(&app, id).await?;
1261
    let status = match (f.exit_code, &f.error) {
1262
        (Some(0), None) => "passed",
1263
        (Some(_), _) => "failed",
1264
        (None, _) => "error",
1265
    };
1266
    let (receipt, receipt_hash) = match &f.receipt {
1267
        Some(r) if r.job_name == job.name && Some(r.exit_code as i32) == f.exit_code => {
1268
            // A passing job's receipt names exactly the artifacts uploaded,
1269
            // in name order; a failing one's names none, though it may have
1270
            // kept `when "always"` artifacts (oxygenflow#37).
1271
            let kept: Vec<String> =
1272
                if r.exit_code == 0 { fdb::artifacts(&app.db, job.id).await?.into_iter().map(|a| a.blob).collect() } else { vec![] };
1273
            if r.outputs.iter().map(hex::encode).collect::<Vec<_>>() != kept {
1274
                return Err(ApiError(StatusCode::BAD_REQUEST, "the receipt names other outputs than the artifacts uploaded".into()));
1275
            }
1276
            (serde_json::to_string(r).unwrap_or_default(), hex::encode(r.hash()))
1277
        }
1278
        Some(_) => return Err(ApiError(StatusCode::BAD_REQUEST, "the receipt does not match the job".into())),
1279
        None => (String::new(), String::new()),
1280
    };
1281
    let error = f.error.clone().unwrap_or_default();
1282
    let changed = fdb::finish(
1283
        &app.db,
1284
        job.id,
1285
        fdb::Outcome { status, exit_code: f.exit_code.map(i64::from), error: &error, receipt: &receipt, receipt_hash: &receipt_hash },
1286
    )
1287
    .await?;
1288
    if let Some(run) = fdb::run_by_id(&app.db, job.run_id).await? {
1289
        if let Some(row) = db::repos::get_by_id(&app.db, run.repo_id).await? {
1290
            if let (Some(r), Some(signed)) = (&f.receipt, &f.signed_receipt) {
1291
                let (attest, signer, note) = match keep_receipt(&app, &me, &row, &run, r, signed).await {
1292
                    Ok((attest, signer)) => (attest, signer, String::new()),
1293
                    Err(e) => {
1294
                        tracing::warn!("flow: job {} receipt not kept: {e}", job.id);
1295
                        (String::new(), String::new(), e)
1296
                    }
1297
                };
1298
                fdb::set_receipt(&app.db, job.id, &attest, &signer, &note).await?;
1299
            }
1300
        }
1301
    }
1302
    report_changed(&app, job.run_id, changed).await?;
1303
    Ok(StatusCode::NO_CONTENT)
1304
}
1305
1306
/// Report the checks of jobs whose status changed.
1307
async fn report_changed(app: &Shared, run_id: i64, changed: Vec<fdb::Job>) -> sqlx::Result<()> {
1308
    let Some(run) = fdb::run_by_id(&app.db, run_id).await? else {
1309
        return Ok(());
1310
    };
1311
    let Some(row) = db::repos::get_by_id(&app.db, run.repo_id).await? else {
1312
        return Ok(());
1313
    };
1314
    for j in changed {
1315
        let summary = match j.status.as_str() {
1316
            "passed" => format!("passed in {}", duration(&j)),
1317
            "failed" => format!("exit {} after {}", j.exit_code.unwrap_or(-1), duration(&j)),
1318
            "skipped" => "skipped: a dependency did not pass".into(),
1319
            "cancelled" => j.error.clone(),
1320
            _ => format!("could not run: {}", j.error),
1321
        };
1322
        report_check(app, &row, &run.state, run.number, &j.name, check_status(&j.status), &summary).await;
1323
    }
1324
    Ok(())
1325
}
1326
1327
/// The site's job time limits (oxygenhub#89): set on Admin → Runners →
1328
/// Settings, else from `hub.toml`.
1329
pub struct Limits {
1330
    /// For a job whose pipeline names none.
1331
    pub default_s: u64,
1332
    /// The most a pipeline may ask for.
1333
    pub max_s: u64,
1334
    /// Grace past a job's limit before the hub gives it up: a runner stops
1335
    /// a job at its limit and reports it within moments, so one still
1336
    /// running this long after is one the runner no longer reports (it was
1337
    /// restarted, or lost the hub while the job ended, oxygenhub#71).
1338
    pub grace_ms: i64,
1339
}
1340
1341
pub const SETTING_DEFAULT: &str = "flow.job_timeout_minutes";
1342
pub const SETTING_MAX: &str = "flow.job_timeout_max_minutes";
1343
pub const SETTING_GRACE: &str = "flow.lost_grace_minutes";
1344
1345
pub async fn limits(app: &Shared) -> Limits {
1346
    let minutes = |key: &'static str| async move {
1347
        db::accounts::setting(&app.db, key).await.ok().flatten().and_then(|v| v.parse::<u64>().ok()).filter(|m| *m > 0)
1348
    };
1349
    let default_m = minutes(SETTING_DEFAULT).await.unwrap_or(app.config.flow.job_timeout_minutes.max(1));
1350
    let max_m = minutes(SETTING_MAX).await.unwrap_or(default_m.max(6 * 60)).max(default_m);
1351
    let grace_m = minutes(SETTING_GRACE).await.unwrap_or(5);
1352
    Limits { default_s: default_m * 60, max_s: max_m * 60, grace_ms: grace_m as i64 * 60_000 }
1353
}
1354
1355
/// End every job still running past its time limit plus the grace:
1356
/// it is marked an error with the reason, the jobs that depended on it are
1357
/// skipped and its run settled, as when a runner reports an error.
1358
pub async fn reap_lost_jobs(app: &Shared) -> sqlx::Result<usize> {
1359
    let limits = limits(app).await;
1360
    let lost = fdb::running_past_limit(&app.db, db::now_ms(), limits.default_s, limits.grace_ms).await?;
1361
    let n = lost.len();
1362
    for job in lost {
1363
        let error = format!(
1364
            "lost: still running {} minutes after it started, past the time limit; the runner{} stopped reporting it",
1365
            (db::now_ms() - job.started_ms.unwrap_or(0)) / 60_000,
1366
            if job.runner.is_empty() { String::new() } else { format!(" {}", job.runner) },
1367
        );
1368
        let changed =
1369
            fdb::finish(&app.db, job.id, fdb::Outcome { status: "error", exit_code: None, error: &error, receipt: "", receipt_hash: "" })
1370
                .await?;
1371
        tracing::warn!("flow: job {} {error}", job.id);
1372
        report_changed(app, job.run_id, changed).await?;
1373
    }
1374
    Ok(n)
1375
}
1376
1377
/// End the jobs running on `runner` that it does not hold (oxygenhub#71):
1378
/// it restarted since claiming them, or never received the claim. A
1379
/// runner claims one job at a time, so a job it holds is always listed.
1380
async fn end_jobs_not_held(app: &Shared, runner: &db::runners::Runner, holding: &[i64]) -> sqlx::Result<()> {
1381
    for job in fdb::running_on(&app.db, runner.id).await? {
1382
        if holding.contains(&job.id) {
1383
            continue;
1384
        }
1385
        let error = format!("lost: the runner {} no longer holds this job (it restarted, or never received it)", runner.name);
1386
        let changed =
1387
            fdb::finish(&app.db, job.id, fdb::Outcome { status: "error", exit_code: None, error: &error, receipt: "", receipt_hash: "" })
1388
                .await?;
1389
        tracing::warn!("flow: job {} {error}", job.id);
1390
        report_changed(app, job.run_id, changed).await?;
1391
    }
1392
    Ok(())
1393
}
1394
1395
/// Start the scheduled runs due at `when` (oxygenflow#24): on each
1396
/// repository that runs Flow, the jobs of its default bookmark's pipeline
1397
/// with a `schedule` line due then. Returns how many runs started.
1398
pub async fn run_due(app: &Shared, when: flow_core::pipeline::When) -> usize {
1399
    let Ok(rows) = db::repos::list(&app.db).await else { return 0 };
1400
    let mut started = 0;
1401
    for row in rows.into_iter().filter(|r| enabled(app, r)) {
1402
        let Some(repo) = RepoName::new(&row.owner, &row.name) else { continue };
1403
        let b = row.default_bookmark.clone();
1404
        let found = app
1405
            .repos
1406
            .read(&repo, move |r, _| {
1407
                let Some(st) = r.view()?.bookmarks.get(&b).copied() else { return Ok(None) };
1408
                Ok(Some((st, r.store().get_state(&st)?.heads)))
1409
            })
1410
            .await;
1411
        let Ok(Some((state, heads))) = found else { continue };
1412
        if heads.is_empty() {
1413
            continue;
1414
        }
1415
        match start_run_at(app, &row, &row.default_bookmark, &state.to_hex(), &heads, &row.owner, "schedule", None, Some(when)).await {
1416
            Ok(Some(_)) => started += 1,
1417
            Ok(None) => {}
1418
            Err(e) => tracing::warn!("flow: scheduled run of {}/{}: {e}", row.owner, row.name),
1419
        }
1420
    }
1421
    started
1422
}
1423
1424
/// The hub's clock as a [`When`](flow_core::pipeline::When).
1425
pub fn when_now() -> flow_core::pipeline::When {
1426
    let now = time::OffsetDateTime::now_utc();
1427
    flow_core::pipeline::When {
1428
        minute: now.minute() as u32,
1429
        hour: now.hour() as u32,
1430
        day: now.day() as u32,
1431
        month: u8::from(now.month()) as u32,
1432
        weekday: now.weekday().number_days_from_sunday() as u32,
1433
    }
1434
}
1435
1436
/// Start the scheduled runs due each minute, once a minute.
1437
pub async fn scheduler(app: Shared) {
1438
    let mut last = None;
1439
    loop {
1440
        tokio::time::sleep(std::time::Duration::from_secs(15)).await;
1441
        let minute = time::OffsetDateTime::now_utc().unix_timestamp() / 60;
1442
        if last == Some(minute) {
1443
            continue;
1444
        }
1445
        last = Some(minute);
1446
        let n = run_due(&app, when_now()).await;
1447
        if n > 0 {
1448
            tracing::info!("flow: {n} scheduled run(s) started");
1449
        }
1450
    }
1451
}
1452
1453
/// Look for lost jobs once a minute.
1454
pub async fn reaper(app: Shared) {
1455
    loop {
1456
        tokio::time::sleep(std::time::Duration::from_secs(60)).await;
1457
        if let Err(e) = reap_lost_jobs(&app).await {
1458
            tracing::warn!("flow: looking for lost jobs: {e}");
1459
        }
1460
    }
1461
}
1462
1463
/// Check a runner's signed receipt for a job of `run` and keep it in the
1464
/// repository (Oxygen spec §14.6): the attestation's hex and the runner's
1465
/// name, or why it was not kept. It must be the job's receipt, for the
1466
/// run's state and repository, signed by the key the runner's token first
1467
/// signed with.
1468
pub async fn keep_receipt(
1469
    app: &Shared,
1470
    runner: &db::runners::Runner,
1471
    row: &db::repos::RepoRow,
1472
    run: &fdb::Run,
1473
    receipt: &flow_core::Receipt,
1474
    signed_hex: &str,
1475
) -> Result<(String, String), String> {
1476
    use oxy_core::receipt::SignedReceipt;
1477
    let bytes = hex::decode(signed_hex).map_err(|_| "the signed receipt is not hex".to_string())?;
1478
    let sr = SignedReceipt::decode(&bytes).map_err(|e| format!("not a signed receipt: {e}"))?;
1479
    sr.verify().map_err(|e| format!("it does not verify: {e}"))?;
1480
    if sr.receipt.hash().0 != receipt.hash() {
1481
        return Err("it signs another receipt than the job's".into());
1482
    }
1483
    let heads: Vec<oxy_core::hash::ChangeHash> = run.head_list().into_iter().filter_map(oxy_core::hash::ChangeHash::from_hex).collect();
1484
    let state = oxy_core::hash::hash_object(oxy_core::hash::ObjectType::State, &oxy_core::objects::StateObject::new(heads).encode());
1485
    if sr.state() != state {
1486
        return Err("it is for another state than the run's".into());
1487
    }
1488
    let key = sr.runner.hash().to_hex();
1489
    let bound = db::runners::bind_key(&app.db, runner.id, &key, &sr.runner.name).await.map_err(|e| e.to_string())?;
1490
    if bound != key {
1491
        return Err(format!("signed by {}, not the key this runner signs with ({})", &key[..12], &bound[..12.min(bound.len())]));
1492
    }
1493
    let name = RepoName::new(&row.owner, &row.name).ok_or("bad repository name")?;
1494
    let signer = sr.runner.name.clone();
1495
    let attest = app.repos.write(&name, move |repo, _| Ok(repo.add_receipt(&sr)?)).await.map_err(|e| e.to_string())?;
1496
    Ok((attest.to_hex(), signer))
1497
}
1498
1499
/// A job's wall time, as the pages show it.
1500
pub fn duration(j: &fdb::Job) -> String {
1501
    match (j.started_ms, j.finished_ms) {
1502
        (Some(a), Some(b)) => human_ms(b - a),
1503
        (Some(a), None) => human_ms(db::now_ms() - a),
1504
        _ => "–".into(),
1505
    }
1506
}
1507
1508
pub fn human_ms(ms: i64) -> String {
1509
    let ms = ms.max(0);
1510
    if ms < 1000 {
1511
        format!("{ms} ms")
1512
    } else if ms < 60_000 {
1513
        format!("{:.1} s", ms as f64 / 1000.0)
1514
    } else {
1515
        format!("{} min {} s", ms / 60_000, (ms % 60_000) / 1000)
1516
    }
1517
}
1518
1519
/// Where a live run stands among the runners' work: what its viewers may
1520
/// know of it. Jobs of other repositories are named only when those are
1521
/// public; the rest are counted.
1522
#[derive(Clone, Debug, Default, PartialEq, Eq)]
1523
pub struct QueueInfo {
1524
    /// 1-based place of the run's first queued job among all queued jobs.
1525
    pub position: Option<usize>,
1526
    /// Jobs running now, and queued jobs ahead of the run's first one.
1527
    pub running: Vec<String>,
1528
    pub ahead: Vec<String>,
1529
    /// Of those, jobs in repositories the page may not name.
1530
    pub hidden: usize,
1531
}
1532
1533
impl QueueInfo {
1534
    /// "3rd in line" and the like.
1535
    pub fn place(&self) -> Option<String> {
1536
        self.position.map(|n| {
1537
            let suffix = match (n % 10, n % 100) {
1538
                (1, 11) | (2, 12) | (3, 13) => "th",
1539
                (1, _) => "st",
1540
                (2, _) => "nd",
1541
                (3, _) => "rd",
1542
                _ => "th",
1543
            };
1544
            format!("{n}{suffix} in line")
1545
        })
1546
    }
1547
}
1548
1549
/// The queue as `run` (of repository `repo_id`) sees it, from `queue`.
1550
pub fn queue_info(queue: &[fdb::QueueEntry], repo_id: i64, run_id: i64) -> QueueInfo {
1551
    let mut info = QueueInfo::default();
1552
    let first = queue.iter().filter(|e| e.status == "queued").position(|e| e.run_id == run_id);
1553
    info.position = first.map(|i| i + 1);
1554
    let label = |e: &fdb::QueueEntry| format!("{}/{} #{} {}", e.owner, e.name, e.run_number, e.job);
1555
    let mut queued_seen = 0;
1556
    for e in queue {
1557
        if e.run_id == run_id {
1558
            continue;
1559
        }
1560
        let ahead = match e.status.as_str() {
1561
            "running" => true,
1562
            _ => {
1563
                queued_seen += 1;
1564
                first.is_some_and(|f| queued_seen <= f)
1565
            }
1566
        };
1567
        if !ahead {
1568
            continue;
1569
        }
1570
        if e.repo_id != repo_id && e.visibility != "public" {
1571
            info.hidden += 1;
1572
        } else if e.status == "running" {
1573
            info.running.push(label(e));
1574
        } else {
1575
            info.ahead.push(label(e));
1576
        }
1577
    }
1578
    info
1579
}
1580
1581
#[cfg(test)]
1582
mod queue_tests {
1583
    use super::*;
1584
1585
    fn e(job_id: i64, run_id: i64, repo_id: i64, vis: &str, job: &str, status: &str) -> fdb::QueueEntry {
1586
        fdb::QueueEntry {
1587
            job_id,
1588
            run_id,
1589
            repo_id,
1590
            owner: "o".into(),
1591
            name: format!("r{repo_id}"),
1592
            visibility: vis.into(),
1593
            run_number: run_id,
1594
            job: job.into(),
1595
            status: status.into(),
1596
            runner: String::new(),
1597
            spec: String::new(),
1598
            sandbox: String::new(),
1599
        }
1600
    }
1601
1602
    #[test]
1603
    fn a_run_learns_its_place_and_what_is_ahead_of_it() {
1604
        // Running first, then queued in claim order (as `fdb::queue` returns them).
1605
        let q = vec![
1606
            e(1, 10, 1, "public", "build", "running"),
1607
            e(2, 11, 2, "private", "test", "running"),
1608
            e(3, 10, 1, "public", "test", "queued"),
1609
            e(4, 12, 3, "private", "lint", "queued"),
1610
            e(5, 13, 4, "public", "build", "queued"),
1611
            e(6, 13, 4, "public", "test", "queued"),
1612
            e(7, 14, 5, "public", "late", "queued"),
1613
        ];
1614
        let info = queue_info(&q, 4, 13);
1615
        assert_eq!(info.position, Some(3));
1616
        assert_eq!(info.place().as_deref(), Some("3rd in line"));
1617
        assert_eq!(info.running, ["o/r1 #10 build"]);
1618
        assert_eq!(info.ahead, ["o/r1 #10 test"], "queued ahead, not behind");
1619
        assert_eq!(info.hidden, 2, "private repositories are counted, not named");
1620
        // A private repository's own jobs are named to its viewers.
1621
        assert_eq!(queue_info(&q, 3, 12).ahead, ["o/r1 #10 test"]);
1622
        assert_eq!(queue_info(&q, 3, 12).position, Some(2));
1623
        // A run with nothing queued has no place.
1624
        assert_eq!(queue_info(&q, 1, 99).position, None);
1625
        let ordinal = |n| QueueInfo { position: Some(n), ..Default::default() }.place().unwrap();
1626
        assert_eq!(
1627
            [ordinal(1), ordinal(2), ordinal(11), ordinal(22), ordinal(113)],
1628
            ["1st in line", "2nd in line", "11th in line", "22nd in line", "113th in line"]
1629
        );
1630
    }
1631
}
1632
1633
#[cfg(test)]
1634
mod placement_tests {
1635
    use super::*;
1636
1637
    #[test]
1638
    fn a_waiting_job_says_what_each_runner_lacks() {
1639
        use crate::db::runners::Runner;
1640
        let rootfs = "ab".repeat(32);
1641
        let job = fdb::Job {
1642
            id: 7,
1643
            runtime: "syscall".into(),
1644
            sandbox: serde_json::json!({ "rootfs": rootfs, "memory_mib": 4096, "cpus": 2 }).to_string(),
1645
            spec: serde_json::json!({ "name": "build" }).to_string(),
1646
            ..Default::default()
1647
        };
1648
        let now = db::now_ms();
1649
        let runner = |id: i64, name: &str| Runner {
1650
            id,
1651
            name: name.into(),
1652
            state: "active".into(),
1653
            last_seen_ms: Some(now),
1654
            arch: "x86_64".into(),
1655
            tiers: "microvm,syscall".into(),
1656
            images: serde_json::to_string(&[&rootfs]).unwrap(),
1657
            memory_mib: Some(16384),
1658
            cpus: Some(8),
1659
            ..Default::default()
1660
        };
1661
        let archs = std::collections::HashMap::new();
1662
        let why = |runners: &[Runner], declines: &[fdb::DeclineRow]| why_waiting(&job, "andy", 1, runners, &archs, declines, now);
1663
        assert_eq!(why(&[], &[]), "no runner is online");
1664
        assert_eq!(why(&[runner(1, "earth-1")], &[]), "", "one can take it: it only waits its turn");
1665
        let microvm_only = Runner { tiers: "microvm".into(), ..runner(1, "earth-1") };
1666
        let riscv = Runner { arch: "riscv64".into(), ..runner(2, "narvi") };
1667
        let small = Runner { memory_mib: Some(2048), ..runner(3, "pi") };
1668
        let other = Runner { scope_owner: "bob".into(), ..runner(4, "bobs") };
1669
        let offline = Runner { last_seen_ms: Some(now - 3_600_000), ..runner(5, "gone") };
1670
        assert_eq!(
1671
            why(&[microvm_only, small, other, offline], &[]),
1672
            "no runner can take it: earth-1 has no syscall tier (it offers microvm); pi has 2048 MiB; the job needs 4096 MiB; bobs serves other repositories"
1673
        );
1674
        let archs_known: std::collections::HashMap<String, String> = [(rootfs.clone(), "x86_64".to_string())].into();
1675
        assert_eq!(
1676
            why_waiting(&job, "andy", 1, &[riscv], &archs_known, &[], now),
1677
            "no runner can take it: narvi is riscv64; the job needs x86_64"
1678
        );
1679
        let without = Runner { images: "[]".into(), ..runner(6, "new") };
1680
        assert_eq!(why(&[without], &[]), "no runner can take it: new does not hold image abababababab");
1681
        let d = fdb::DeclineRow {
1682
            job_id: 7,
1683
            runner_id: 1,
1684
            runner: "earth-1".into(),
1685
            reason: "only 512 MiB of memory is free here".into(),
1686
            ms: now,
1687
        };
1688
        assert_eq!(
1689
            why(&[runner(1, "earth-1")], std::slice::from_ref(&d)),
1690
            "no runner can take it: earth-1 declined it: only 512 MiB of memory is free here"
1691
        );
1692
        let old = fdb::DeclineRow { ms: now - fdb::DECLINE_HOLDS_MS - 1, ..d };
1693
        assert_eq!(why(&[runner(1, "earth-1")], &[old]), "", "an old decline no longer holds");
1694
    }
1695
}