OXYGENHUB
# d94449e42718 oxygenhub / crates / hub / src / flow.rs
42.7 KiB · Rust · last changed A runner busy with its only job stays online 1 month 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
    let repo = RepoName::new(&row.owner, &row.name).ok_or("bad repository name")?;
273
    let h = heads.to_vec();
274
    let text = app
275
        .repos
276
        .read(&repo, move |r, cache| {
277
            let loaded = cache.load(r, &h)?;
278
            Ok(match loaded.snapshot.files.get(PIPELINE_PATH).map(|f| &f.content) {
279
                Some(oxy_core::state::FileContent::Text { bytes, .. }) => Some(String::from_utf8_lossy(bytes).into_owned()),
280
                _ => None,
281
            })
282
        })
283
        .await
284
        .map_err(|e| format!("reading the pipeline: {e}"))?;
285
    let Some(text) = text else { return Ok(None) };
286
    let mut hex: Vec<String> = heads.iter().map(|h| h.to_hex()).collect();
287
    hex.sort();
288
    let heads_s = hex.join("+");
289
    let base = db::flow::NewRun { repo_id: row.id, bookmark, state, heads: &heads_s, pipeline_hash: "", actor, trigger, error: None };
290
    let id = match Pipeline::parse(&text) {
291
        Err(problems) => {
292
            let msg = problems.to_string();
293
            db::flow::create_run(&app.db, db::flow::NewRun { error: Some(&msg), ..base }, &[]).await.map_err(|e| e.to_string())?
294
        }
295
        Ok(p) => {
296
            let depths = p.depths();
297
            // Jobs out of the push's scope stay out, unless a kept job
298
            // needs them.
299
            // The jobs this run's event starts (`on`), out of those its
300
            // push touched (`scope`); then what they depend on.
301
            let event = event_of(trigger, bookmark);
302
            let mut keep: std::collections::HashSet<String> = p
303
                .order()
304
                .into_iter()
305
                .filter(|j| p.on_of(j).matches(event) && in_scope(&j.scope, touched))
306
                .map(|j| j.name.clone())
307
                .collect();
308
            loop {
309
                let needed: Vec<String> = p
310
                    .order()
311
                    .into_iter()
312
                    .filter(|j| keep.contains(&j.name))
313
                    .flat_map(|j| j.depends_on.iter().cloned())
314
                    .filter(|d| !keep.contains(d))
315
                    .collect();
316
                if needed.is_empty() {
317
                    break;
318
                }
319
                keep.extend(needed);
320
            }
321
            let jobs: Vec<fdb::NewJob> = p
322
                .order()
323
                .into_iter()
324
                .filter(|j| keep.contains(&j.name))
325
                .map(|j| {
326
                    let sb = &p.sandboxes[&j.sandbox];
327
                    fdb::NewJob {
328
                        name: j.name.clone(),
329
                        depth: depths[j.name.as_str()] as i64,
330
                        depends: j.depends_on.clone(),
331
                        spec: serde_json::to_string(j).unwrap_or_default(),
332
                        sandbox: serde_json::to_string(sb).unwrap_or_default(),
333
                        runtime: sb.runtime.as_str().into(),
334
                    }
335
                })
336
                .collect();
337
            if jobs.is_empty() {
338
                // Nothing this event starts (or this push touched).
339
                return Ok(None);
340
            }
341
            let hash = hex::encode(p.hash());
342
            let id =
343
                db::flow::create_run(&app.db, db::flow::NewRun { pipeline_hash: &hash, ..base }, &jobs).await.map_err(|e| e.to_string())?;
344
            if let Ok(Some(run)) = fdb::run_by_id(&app.db, id).await {
345
                for j in &jobs {
346
                    report_check(app, row, state, run.number, &j.name, "queued", "waiting for a runner").await;
347
                }
348
            }
349
            id
350
        }
351
    };
352
    let run = fdb::run_by_id(&app.db, id).await.map_err(|e| e.to_string())?.ok_or("the run vanished")?;
353
    Ok(Some(run.number))
354
}
355
356
/// Record a job's check; it links to the run's page by the run's number.
357
async fn report_check(app: &Shared, row: &db::repos::RepoRow, state: &str, run_number: i64, job: &str, status: &str, summary: &str) {
358
    let url = format!("{}/{}/{}/flow/{run_number}", app.config.base_url(), row.owner, row.name);
359
    let name = format!("flow/{job}");
360
    let c = db::hooks::upsert_check(
361
        &app.db,
362
        row.id,
363
        db::hooks::CheckReport { state, name: &name, status, summary, url: &url, reporter: None, seal: "" },
364
    )
365
    .await;
366
    if let Ok(c) = c {
367
        let api = crate::api::check_of(&c);
368
        app.events.proposal(row.id, hub_api::ProposalEvent { number: 0, kind: "check".into(), check: Some(api.clone()) });
369
        crate::hooks::emit(app, row, "check", "flow", serde_json::json!({ "check": api })).await;
370
    }
371
}
372
373
/// The check status a job's status shows as.
374
fn check_status(job_status: &str) -> &'static str {
375
    match job_status {
376
        "queued" => "queued",
377
        "running" => "running",
378
        "passed" => "passed",
379
        "skipped" => "skipped",
380
        "cancelled" => "cancelled",
381
        _ => "failed",
382
    }
383
}
384
385
/// The runner a bearer token belongs to: a registered runner, or one
386
/// configured with a hub.toml token (tracked from its first request). A
387
/// disabled runner is refused.
388
async fn runner_auth(app: &Shared, headers: &HeaderMap) -> Result<db::runners::Runner, ApiError> {
389
    let Some(t) = oxy_server::bearer_token(headers) else {
390
        return Err(ApiError(StatusCode::UNAUTHORIZED, "a runner token is required".into()));
391
    };
392
    let hash = blake3::hash(t.as_bytes()).to_hex().to_string();
393
    let runner = match db::runners::by_token(&app.db, &hash).await? {
394
        Some(r) => r,
395
        None if app.config.flow.runner_tokens.iter().any(|x| crate::auth::mac_equal(x, t)) => {
396
            db::runners::configured(&app.db, &hash, "runner").await?
397
        }
398
        None => return Err(ApiError(StatusCode::UNAUTHORIZED, "not a runner token".into())),
399
    };
400
    if runner.state == "disabled" {
401
        return Err(ApiError(StatusCode::FORBIDDEN, format!("runner {} is disabled on this hub", runner.name)));
402
    }
403
    Ok(runner)
404
}
405
406
/// `POST /api/v1/flow/runners/register`: a one-time registration token
407
/// from Admin → Runners buys a runner credential, binding the runner's
408
/// signing identity (hub spec §13.2).
409
pub async fn register(
410
    State(app): State<Shared>,
411
    Json(req): Json<flow_core::api::Register>,
412
) -> Result<Json<flow_core::api::Registered>, ApiError> {
413
    let bytes = hex::decode(req.identity.trim()).map_err(|_| ApiError(StatusCode::BAD_REQUEST, "identity: not hex".into()))?;
414
    let identity = oxy_core::identity::Identity::decode(&bytes).map_err(|e| ApiError(StatusCode::BAD_REQUEST, format!("identity: {e}")))?;
415
    let name: String = req.name.chars().filter(|c| c.is_ascii_alphanumeric() || matches!(c, '-' | '_' | '.')).take(64).collect();
416
    if name.is_empty() {
417
        return Err(ApiError(StatusCode::BAD_REQUEST, "give the runner a name of letters, digits, - _ or .".into()));
418
    }
419
    let credential = format!("flowrunner_{}", crate::auth::random_hex(32));
420
    let hash = blake3::hash(credential.as_bytes()).to_hex().to_string();
421
    let reg = blake3::hash(req.token.trim().as_bytes()).to_hex().to_string();
422
    let runner =
423
        db::runners::register(&app.db, &reg, &name, &hash, &identity.hash().to_hex(), &identity.name, "").await?.ok_or_else(|| {
424
            ApiError(StatusCode::FORBIDDEN, "that registration token is unknown, used or expired; create one on Admin → Runners".into())
425
        })?;
426
    db::runners::seen(&app.db, runner.id, None, Some(&req.info)).await?;
427
    tracing::info!("flow: runner {} registered (key {})", runner.name, &runner.key[..12]);
428
    Ok(Json(flow_core::api::Registered { id: runner.id, name: runner.name, credential }))
429
}
430
431
/// `POST /api/v1/flow/claim`.
432
pub async fn claim(State(app): State<Shared>, headers: HeaderMap, Json(req): Json<ClaimRequest>) -> Result<Response, ApiError> {
433
    let me = runner_auth(&app, &headers).await?;
434
    let claimed: String = req.runner.chars().filter(|c| !c.is_control()).take(64).collect();
435
    db::runners::seen(&app.db, me.id, Some(&claimed), req.info.as_ref()).await?;
436
    if let Some(holding) = &req.holding {
437
        end_jobs_not_held(&app, &me, holding).await?;
438
    }
439
    // A draining runner finishes what it has and takes nothing new.
440
    if me.state == "draining" {
441
        return Ok(StatusCode::NO_CONTENT.into_response());
442
    }
443
    let runner = if me.source == "registered" { me.name.clone() } else { claimed };
444
    let secret = format!("{JOB_TOKEN_PREFIX}{}", crate::auth::random_hex(32));
445
    let room = fdb::Room {
446
        memory_mib: req.free_memory_mib,
447
        cpus: req.free_cpus,
448
        images: req.images.as_ref().map(|i| i.iter().map(|h| h.to_ascii_lowercase()).collect()),
449
    };
450
    // The labels it reports now, or those it registered with.
451
    let labels: Vec<String> = match &req.info {
452
        Some(i) => i.labels.clone(),
453
        None => me.label_list().into_iter().map(String::from).collect(),
454
    };
455
    let Some(job) =
456
        fdb::claim_within(&app.db, &req.runtimes, &runner, &crate::auth::token_hash(&secret), room, &me.scope(), &labels).await?
457
    else {
458
        return Ok(StatusCode::NO_CONTENT.into_response());
459
    };
460
    db::runners::set_job_runner(&app.db, job.id, me.id).await?;
461
    let run = fdb::run_by_id(&app.db, job.run_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such run".into()))?;
462
    let row =
463
        db::repos::get_by_id(&app.db, run.repo_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such repository".into()))?;
464
    report_check(&app, &row, &run.state, run.number, &job.name, "running", &format!("running on {runner}")).await;
465
    let bad = |e: serde_json::Error| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("stored job: {e}"));
466
    // The job's limit: its own timeout, else its sandbox's, else the site
467
    // default, never past the site maximum; recorded for the reaper.
468
    let limits = limits(&app).await;
469
    let spec: flow_core::pipeline::Job = serde_json::from_str(&job.spec).map_err(bad)?;
470
    let sandbox: flow_core::pipeline::Sandbox = serde_json::from_str(&job.sandbox).map_err(bad)?;
471
    let timeout_s = spec.timeout_s.or(sandbox.timeout_s).unwrap_or(limits.default_s).min(limits.max_s).max(60);
472
    fdb::set_job_timeout(&app.db, job.id, timeout_s).await?;
473
    let claim = Claim {
474
        job_id: job.id,
475
        run_id: run.id,
476
        repo: format!("{}/{}", row.owner, row.name),
477
        clone_url: format!("{}/{}/{}", app.config.base_url(), row.owner, row.name),
478
        token: secret,
479
        heads: run.head_list().into_iter().map(String::from).collect(),
480
        pipeline_hash: run.pipeline_hash.clone(),
481
        job: spec,
482
        sandbox,
483
        timeout_s,
484
        subrepos: vec![],
485
        state: run.state.clone(),
486
        bookmark: run.bookmark.clone(),
487
        run_number: run.number,
488
    };
489
    let mut claim = claim;
490
    let (inputs, ids) = subrepo_inputs(&app, &row, &run).await;
491
    if !ids.is_empty() {
492
        fdb::set_also_read(&app.db, job.id, &ids).await?;
493
    }
494
    claim.subrepos = inputs;
495
    Ok(Json(claim).into_response())
496
}
497
498
/// The Oxygen subrepo pins in `name`'s state `heads`: (path, repository id, pinned heads).
499
async fn pins_in(
500
    app: &Shared,
501
    name: crate::repos::RepoName,
502
    heads: Vec<oxy_core::hash::ChangeHash>,
503
) -> Vec<(String, [u8; 16], Vec<oxy_core::hash::ChangeHash>)> {
504
    app.repos
505
        .read(&name, move |repo, cache| {
506
            let loaded = cache.load(repo, &heads)?;
507
            let mut out = Vec::new();
508
            for (path, f) in &loaded.snapshot.files {
509
                if f.kind != oxy_core::objects::NodeKind::Subrepo {
510
                    continue;
511
                }
512
                if let oxy_core::state::FileContent::Opaque { blobs } = &f.content {
513
                    if let Some(b) = blobs.first() {
514
                        if let Ok(oxy_core::subrepo::Pin::Oxy { repo: id, heads, .. }) = repo
515
                            .read_blob(b)
516
                            .map_err(crate::repos::RepoError::Oxy)
517
                            .and_then(|bytes| oxy_core::subrepo::Pin::decode(&bytes).map_err(|e| crate::repos::RepoError::Oxy(e.into())))
518
                        {
519
                            out.push((path.clone(), id, heads));
520
                        }
521
                    }
522
                }
523
            }
524
            Ok(out)
525
        })
526
        .await
527
        .unwrap_or_default()
528
}
529
530
/// How deep subrepos of subrepos are followed.
531
const MAX_SUBREPO_DEPTH: usize = 3;
532
533
/// The Oxygen subrepos the run's state pins, and those their states pin in
534
/// turn (parents first, paths joined), that live on this hub and that the
535
/// run's pusher may read: what the job's workspace needs beside its own
536
/// tree. A pin the hub cannot serve is left out, with what it pins; the job
537
/// then fails on the missing files with an ordinary error.
538
async fn subrepo_inputs(app: &Shared, row: &db::repos::RepoRow, run: &fdb::Run) -> (Vec<flow_core::api::SubrepoInput>, Vec<i64>) {
539
    let Some(name) = crate::repos::RepoName::new(&row.owner, &row.name) else { return (vec![], vec![]) };
540
    let heads: Vec<oxy_core::hash::ChangeHash> = run.head_list().into_iter().filter_map(oxy_core::hash::ChangeHash::from_hex).collect();
541
    let pusher = db::accounts::account_by_name(&app.db, &run.actor).await.ok().flatten();
542
    let rows = db::repos::list(&app.db).await.unwrap_or_default();
543
    let mut inputs = Vec::new();
544
    let mut ids = Vec::new();
545
    // (path prefix, repository, heads, depth), parents before children.
546
    let mut queue = std::collections::VecDeque::from([(String::new(), name, heads, 0usize)]);
547
    while let Some((prefix, name, heads, depth)) = queue.pop_front() {
548
        if depth == MAX_SUBREPO_DEPTH {
549
            continue;
550
        }
551
        for (path, id, pinned) in pins_in(app, name, heads).await {
552
            let path = if prefix.is_empty() { path } else { format!("{prefix}/{path}") };
553
            for other in &rows {
554
                let Some(other_name) = crate::repos::RepoName::new(&other.owner, &other.name) else { continue };
555
                let matches = app.repos.read(&other_name, move |r, _| Ok(r.id().ok() == Some(id))).await.unwrap_or(false);
556
                if !matches {
557
                    continue;
558
                }
559
                let role = crate::auth::roles::role_for(&app.db, pusher.as_ref(), other).await.ok().flatten();
560
                if crate::auth::roles::allows(role, crate::auth::Access::Read) {
561
                    inputs.push(flow_core::api::SubrepoInput {
562
                        path: path.clone(),
563
                        repo: format!("{}/{}", other.owner, other.name),
564
                        clone_url: format!("{}/{}/{}", app.config.base_url(), other.owner, other.name),
565
                        heads: pinned.iter().map(|h| h.to_hex()).collect(),
566
                    });
567
                    if !ids.contains(&other.id) {
568
                        ids.push(other.id);
569
                    }
570
                    queue.push_back((path.clone(), other_name, pinned.clone(), depth + 1));
571
                }
572
                break;
573
            }
574
        }
575
    }
576
    (inputs, ids)
577
}
578
579
async fn running_job(app: &Shared, id: i64) -> Result<fdb::Job, ApiError> {
580
    match fdb::job(&app.db, id).await? {
581
        Some(j) if j.status == "running" => Ok(j),
582
        Some(_) => Err(ApiError(StatusCode::CONFLICT, "this job is not running".into())),
583
        None => Err(ApiError(StatusCode::NOT_FOUND, "no such job".into())),
584
    }
585
}
586
587
/// `POST /api/v1/flow/jobs/{id}/log`.
588
pub async fn log(
589
    State(app): State<Shared>,
590
    headers: HeaderMap,
591
    Path(id): Path<i64>,
592
    Json(mut batch): Json<LogBatch>,
593
) -> Result<StatusCode, ApiError> {
594
    let me = runner_auth(&app, &headers).await?;
595
    let job = running_job(&app, id).await?;
596
    // A runner busy with its only job claims nothing, so its log and
597
    // finish calls are what keep it online (oxygenhub#78).
598
    db::runners::seen(&app.db, me.id, None, None).await?;
599
    // An empty batch asks only whether the job still runs: a cancelled
600
    // one answers 409 and its runner stops it (oxygenflow#32).
601
    if batch.records.is_empty() {
602
        return Ok(StatusCode::NO_CONTENT);
603
    }
604
    for r in &mut batch.records {
605
        if r.text.len() > MAX_RECORD_TEXT {
606
            let mut cut = MAX_RECORD_TEXT;
607
            while !r.text.is_char_boundary(cut) {
608
                cut -= 1;
609
            }
610
            r.text.truncate(cut);
611
        }
612
        if !matches!(r.stream.as_str(), "out" | "err" | "console" | "flow") {
613
            r.stream = "console".into();
614
        }
615
    }
616
    let kept = fdb::append_log(&app.db, job.id, &batch.records, MAX_RECORDS as i64).await?;
617
    if kept < batch.records.len() && job.log_count < MAX_RECORDS as i64 {
618
        tracing::info!("flow: job {} reached the log limit", job.id);
619
    }
620
    Ok(StatusCode::NO_CONTENT)
621
}
622
623
/// Largest artifact the hub keeps; the runner refuses larger ones first.
624
pub const MAX_ARTIFACT_BYTES: usize = 512 << 20;
625
626
/// `POST /api/v1/flow/jobs/{id}/artifacts/{name}`: the bytes of an artifact
627
/// the job declares (OxygenFlow spec §5.2), kept as a blob of the run's
628
/// repository. Answers `{"blob"}`.
629
pub async fn artifact(
630
    State(app): State<Shared>,
631
    headers: HeaderMap,
632
    Path((id, name)): Path<(i64, String)>,
633
    body: axum::body::Bytes,
634
) -> Result<Response, ApiError> {
635
    runner_auth(&app, &headers).await?;
636
    let job = running_job(&app, id).await?;
637
    let spec: flow_core::Job =
638
        serde_json::from_str(&job.spec).map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("stored job: {e}")))?;
639
    let Some(out) = spec.outputs.iter().find(|o| o.name == name) else {
640
        return Err(ApiError(StatusCode::BAD_REQUEST, format!("job {} declares no artifact {name:?}", job.name)));
641
    };
642
    let run = fdb::run_by_id(&app.db, job.run_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such run".into()))?;
643
    let row =
644
        db::repos::get_by_id(&app.db, run.repo_id).await?.ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "no such repository".into()))?;
645
    let repo = RepoName::new(&row.owner, &row.name).ok_or_else(|| ApiError(StatusCode::NOT_FOUND, "bad repository name".into()))?;
646
    let content = blake3::hash(&body).to_hex().to_string();
647
    let size = body.len() as i64;
648
    let blob = app
649
        .repos
650
        .write(&repo, move |r, _| Ok(r.store_blob(&body)?))
651
        .await
652
        .map_err(|e| ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("keep the artifact: {e}")))?;
653
    let a = fdb::Artifact { job_id: job.id, name, path: out.path.clone(), blob: blob.to_hex(), size, content, created_ms: db::now_ms() };
654
    fdb::add_artifact(&app.db, &a).await?;
655
    Ok(Json(serde_json::json!({ "blob": a.blob })).into_response())
656
}
657
658
/// `POST /api/v1/flow/jobs/{id}/finish`.
659
pub async fn finish(
660
    State(app): State<Shared>,
661
    headers: HeaderMap,
662
    Path(id): Path<i64>,
663
    Json(f): Json<Finish>,
664
) -> Result<StatusCode, ApiError> {
665
    let me = runner_auth(&app, &headers).await?;
666
    let job = running_job(&app, id).await?;
667
    db::runners::seen(&app.db, me.id, None, None).await?;
668
    let status = match (f.exit_code, &f.error) {
669
        (Some(0), None) => "passed",
670
        (Some(_), _) => "failed",
671
        (None, _) => "error",
672
    };
673
    let (receipt, receipt_hash) = match &f.receipt {
674
        Some(r) if r.job_name == job.name && Some(r.exit_code as i32) == f.exit_code => {
675
            // The receipt names exactly the artifacts uploaded, in name order.
676
            let kept: Vec<String> = fdb::artifacts(&app.db, job.id).await?.into_iter().map(|a| a.blob).collect();
677
            if r.outputs.iter().map(hex::encode).collect::<Vec<_>>() != kept {
678
                return Err(ApiError(StatusCode::BAD_REQUEST, "the receipt names other outputs than the artifacts uploaded".into()));
679
            }
680
            (serde_json::to_string(r).unwrap_or_default(), hex::encode(r.hash()))
681
        }
682
        Some(_) => return Err(ApiError(StatusCode::BAD_REQUEST, "the receipt does not match the job".into())),
683
        None => (String::new(), String::new()),
684
    };
685
    let error = f.error.clone().unwrap_or_default();
686
    let changed = fdb::finish(
687
        &app.db,
688
        job.id,
689
        fdb::Outcome { status, exit_code: f.exit_code.map(i64::from), error: &error, receipt: &receipt, receipt_hash: &receipt_hash },
690
    )
691
    .await?;
692
    if let Some(run) = fdb::run_by_id(&app.db, job.run_id).await? {
693
        if let Some(row) = db::repos::get_by_id(&app.db, run.repo_id).await? {
694
            if let (Some(r), Some(signed)) = (&f.receipt, &f.signed_receipt) {
695
                let (attest, signer, note) = match keep_receipt(&app, &me, &row, &run, r, signed).await {
696
                    Ok((attest, signer)) => (attest, signer, String::new()),
697
                    Err(e) => {
698
                        tracing::warn!("flow: job {} receipt not kept: {e}", job.id);
699
                        (String::new(), String::new(), e)
700
                    }
701
                };
702
                fdb::set_receipt(&app.db, job.id, &attest, &signer, &note).await?;
703
            }
704
        }
705
    }
706
    report_changed(&app, job.run_id, changed).await?;
707
    Ok(StatusCode::NO_CONTENT)
708
}
709
710
/// Report the checks of jobs whose status changed.
711
async fn report_changed(app: &Shared, run_id: i64, changed: Vec<fdb::Job>) -> sqlx::Result<()> {
712
    let Some(run) = fdb::run_by_id(&app.db, run_id).await? else {
713
        return Ok(());
714
    };
715
    let Some(row) = db::repos::get_by_id(&app.db, run.repo_id).await? else {
716
        return Ok(());
717
    };
718
    for j in changed {
719
        let summary = match j.status.as_str() {
720
            "passed" => format!("passed in {}", duration(&j)),
721
            "failed" => format!("exit {} after {}", j.exit_code.unwrap_or(-1), duration(&j)),
722
            "skipped" => "skipped: a dependency did not pass".into(),
723
            "cancelled" => j.error.clone(),
724
            _ => format!("could not run: {}", j.error),
725
        };
726
        report_check(app, &row, &run.state, run.number, &j.name, check_status(&j.status), &summary).await;
727
    }
728
    Ok(())
729
}
730
731
/// The site's job time limits (oxygenhub#89): set on Admin → Runners →
732
/// Settings, else from `hub.toml`.
733
pub struct Limits {
734
    /// For a job whose pipeline names none.
735
    pub default_s: u64,
736
    /// The most a pipeline may ask for.
737
    pub max_s: u64,
738
    /// Grace past a job's limit before the hub gives it up: a runner stops
739
    /// a job at its limit and reports it within moments, so one still
740
    /// running this long after is one the runner no longer reports (it was
741
    /// restarted, or lost the hub while the job ended, oxygenhub#71).
742
    pub grace_ms: i64,
743
}
744
745
pub const SETTING_DEFAULT: &str = "flow.job_timeout_minutes";
746
pub const SETTING_MAX: &str = "flow.job_timeout_max_minutes";
747
pub const SETTING_GRACE: &str = "flow.lost_grace_minutes";
748
749
pub async fn limits(app: &Shared) -> Limits {
750
    let minutes = |key: &'static str| async move {
751
        db::accounts::setting(&app.db, key).await.ok().flatten().and_then(|v| v.parse::<u64>().ok()).filter(|m| *m > 0)
752
    };
753
    let default_m = minutes(SETTING_DEFAULT).await.unwrap_or(app.config.flow.job_timeout_minutes.max(1));
754
    let max_m = minutes(SETTING_MAX).await.unwrap_or(default_m.max(6 * 60)).max(default_m);
755
    let grace_m = minutes(SETTING_GRACE).await.unwrap_or(5);
756
    Limits { default_s: default_m * 60, max_s: max_m * 60, grace_ms: grace_m as i64 * 60_000 }
757
}
758
759
/// End every job still running past its time limit plus the grace:
760
/// it is marked an error with the reason, the jobs that depended on it are
761
/// skipped and its run settled, as when a runner reports an error.
762
pub async fn reap_lost_jobs(app: &Shared) -> sqlx::Result<usize> {
763
    let limits = limits(app).await;
764
    let lost = fdb::running_past_limit(&app.db, db::now_ms(), limits.default_s, limits.grace_ms).await?;
765
    let n = lost.len();
766
    for job in lost {
767
        let error = format!(
768
            "lost: still running {} minutes after it started, past the time limit; the runner{} stopped reporting it",
769
            (db::now_ms() - job.started_ms.unwrap_or(0)) / 60_000,
770
            if job.runner.is_empty() { String::new() } else { format!(" {}", job.runner) },
771
        );
772
        let changed =
773
            fdb::finish(&app.db, job.id, fdb::Outcome { status: "error", exit_code: None, error: &error, receipt: "", receipt_hash: "" })
774
                .await?;
775
        tracing::warn!("flow: job {} {error}", job.id);
776
        report_changed(app, job.run_id, changed).await?;
777
    }
778
    Ok(n)
779
}
780
781
/// End the jobs running on `runner` that it does not hold (oxygenhub#71):
782
/// it restarted since claiming them, or never received the claim. A
783
/// runner claims one job at a time, so a job it holds is always listed.
784
async fn end_jobs_not_held(app: &Shared, runner: &db::runners::Runner, holding: &[i64]) -> sqlx::Result<()> {
785
    for job in fdb::running_on(&app.db, runner.id).await? {
786
        if holding.contains(&job.id) {
787
            continue;
788
        }
789
        let error = format!("lost: the runner {} no longer holds this job (it restarted, or never received it)", runner.name);
790
        let changed =
791
            fdb::finish(&app.db, job.id, fdb::Outcome { status: "error", exit_code: None, error: &error, receipt: "", receipt_hash: "" })
792
                .await?;
793
        tracing::warn!("flow: job {} {error}", job.id);
794
        report_changed(app, job.run_id, changed).await?;
795
    }
796
    Ok(())
797
}
798
799
/// Look for lost jobs once a minute.
800
pub async fn reaper(app: Shared) {
801
    loop {
802
        tokio::time::sleep(std::time::Duration::from_secs(60)).await;
803
        if let Err(e) = reap_lost_jobs(&app).await {
804
            tracing::warn!("flow: looking for lost jobs: {e}");
805
        }
806
    }
807
}
808
809
/// Check a runner's signed receipt for a job of `run` and keep it in the
810
/// repository (Oxygen spec §14.6): the attestation's hex and the runner's
811
/// name, or why it was not kept. It must be the job's receipt, for the
812
/// run's state and repository, signed by the key the runner's token first
813
/// signed with.
814
pub async fn keep_receipt(
815
    app: &Shared,
816
    runner: &db::runners::Runner,
817
    row: &db::repos::RepoRow,
818
    run: &fdb::Run,
819
    receipt: &flow_core::Receipt,
820
    signed_hex: &str,
821
) -> Result<(String, String), String> {
822
    use oxy_core::receipt::SignedReceipt;
823
    let bytes = hex::decode(signed_hex).map_err(|_| "the signed receipt is not hex".to_string())?;
824
    let sr = SignedReceipt::decode(&bytes).map_err(|e| format!("not a signed receipt: {e}"))?;
825
    sr.verify().map_err(|e| format!("it does not verify: {e}"))?;
826
    if sr.receipt.hash().0 != receipt.hash() {
827
        return Err("it signs another receipt than the job's".into());
828
    }
829
    let heads: Vec<oxy_core::hash::ChangeHash> = run.head_list().into_iter().filter_map(oxy_core::hash::ChangeHash::from_hex).collect();
830
    let state = oxy_core::hash::hash_object(oxy_core::hash::ObjectType::State, &oxy_core::objects::StateObject::new(heads).encode());
831
    if sr.state() != state {
832
        return Err("it is for another state than the run's".into());
833
    }
834
    let key = sr.runner.hash().to_hex();
835
    let bound = db::runners::bind_key(&app.db, runner.id, &key, &sr.runner.name).await.map_err(|e| e.to_string())?;
836
    if bound != key {
837
        return Err(format!("signed by {}, not the key this runner signs with ({})", &key[..12], &bound[..12.min(bound.len())]));
838
    }
839
    let name = RepoName::new(&row.owner, &row.name).ok_or("bad repository name")?;
840
    let signer = sr.runner.name.clone();
841
    let attest = app.repos.write(&name, move |repo, _| Ok(repo.add_receipt(&sr)?)).await.map_err(|e| e.to_string())?;
842
    Ok((attest.to_hex(), signer))
843
}
844
845
/// A job's wall time, as the pages show it.
846
pub fn duration(j: &fdb::Job) -> String {
847
    match (j.started_ms, j.finished_ms) {
848
        (Some(a), Some(b)) => human_ms(b - a),
849
        (Some(a), None) => human_ms(db::now_ms() - a),
850
        _ => "–".into(),
851
    }
852
}
853
854
pub fn human_ms(ms: i64) -> String {
855
    let ms = ms.max(0);
856
    if ms < 1000 {
857
        format!("{ms} ms")
858
    } else if ms < 60_000 {
859
        format!("{:.1} s", ms as f64 / 1000.0)
860
    } else {
861
        format!("{} min {} s", ms / 60_000, (ms % 60_000) / 1000)
862
    }
863
}
864
865
/// Where a live run stands among the runners' work: what its viewers may
866
/// know of it. Jobs of other repositories are named only when those are
867
/// public; the rest are counted.
868
#[derive(Clone, Debug, Default, PartialEq, Eq)]
869
pub struct QueueInfo {
870
    /// 1-based place of the run's first queued job among all queued jobs.
871
    pub position: Option<usize>,
872
    /// Jobs running now, and queued jobs ahead of the run's first one.
873
    pub running: Vec<String>,
874
    pub ahead: Vec<String>,
875
    /// Of those, jobs in repositories the page may not name.
876
    pub hidden: usize,
877
}
878
879
impl QueueInfo {
880
    /// "3rd in line" and the like.
881
    pub fn place(&self) -> Option<String> {
882
        self.position.map(|n| {
883
            let suffix = match (n % 10, n % 100) {
884
                (1, 11) | (2, 12) | (3, 13) => "th",
885
                (1, _) => "st",
886
                (2, _) => "nd",
887
                (3, _) => "rd",
888
                _ => "th",
889
            };
890
            format!("{n}{suffix} in line")
891
        })
892
    }
893
}
894
895
/// The queue as `run` (of repository `repo_id`) sees it, from `queue`.
896
pub fn queue_info(queue: &[fdb::QueueEntry], repo_id: i64, run_id: i64) -> QueueInfo {
897
    let mut info = QueueInfo::default();
898
    let first = queue.iter().filter(|e| e.status == "queued").position(|e| e.run_id == run_id);
899
    info.position = first.map(|i| i + 1);
900
    let label = |e: &fdb::QueueEntry| format!("{}/{} #{} {}", e.owner, e.name, e.run_number, e.job);
901
    let mut queued_seen = 0;
902
    for e in queue {
903
        if e.run_id == run_id {
904
            continue;
905
        }
906
        let ahead = match e.status.as_str() {
907
            "running" => true,
908
            _ => {
909
                queued_seen += 1;
910
                first.is_some_and(|f| queued_seen <= f)
911
            }
912
        };
913
        if !ahead {
914
            continue;
915
        }
916
        if e.repo_id != repo_id && e.visibility != "public" {
917
            info.hidden += 1;
918
        } else if e.status == "running" {
919
            info.running.push(label(e));
920
        } else {
921
            info.ahead.push(label(e));
922
        }
923
    }
924
    info
925
}
926
927
#[cfg(test)]
928
mod queue_tests {
929
    use super::*;
930
931
    fn e(job_id: i64, run_id: i64, repo_id: i64, vis: &str, job: &str, status: &str) -> fdb::QueueEntry {
932
        fdb::QueueEntry {
933
            job_id,
934
            run_id,
935
            repo_id,
936
            owner: "o".into(),
937
            name: format!("r{repo_id}"),
938
            visibility: vis.into(),
939
            run_number: run_id,
940
            job: job.into(),
941
            status: status.into(),
942
            runner: String::new(),
943
            spec: String::new(),
944
            sandbox: String::new(),
945
        }
946
    }
947
948
    #[test]
949
    fn a_run_learns_its_place_and_what_is_ahead_of_it() {
950
        // Running first, then queued in claim order (as `fdb::queue` returns them).
951
        let q = vec![
952
            e(1, 10, 1, "public", "build", "running"),
953
            e(2, 11, 2, "private", "test", "running"),
954
            e(3, 10, 1, "public", "test", "queued"),
955
            e(4, 12, 3, "private", "lint", "queued"),
956
            e(5, 13, 4, "public", "build", "queued"),
957
            e(6, 13, 4, "public", "test", "queued"),
958
            e(7, 14, 5, "public", "late", "queued"),
959
        ];
960
        let info = queue_info(&q, 4, 13);
961
        assert_eq!(info.position, Some(3));
962
        assert_eq!(info.place().as_deref(), Some("3rd in line"));
963
        assert_eq!(info.running, ["o/r1 #10 build"]);
964
        assert_eq!(info.ahead, ["o/r1 #10 test"], "queued ahead, not behind");
965
        assert_eq!(info.hidden, 2, "private repositories are counted, not named");
966
        // A private repository's own jobs are named to its viewers.
967
        assert_eq!(queue_info(&q, 3, 12).ahead, ["o/r1 #10 test"]);
968
        assert_eq!(queue_info(&q, 3, 12).position, Some(2));
969
        // A run with nothing queued has no place.
970
        assert_eq!(queue_info(&q, 1, 99).position, None);
971
        let ordinal = |n| QueueInfo { position: Some(n), ..Default::default() }.place().unwrap();
972
        assert_eq!(
973
            [ordinal(1), ordinal(2), ordinal(11), ordinal(22), ordinal(113)],
974
            ["1st in line", "2nd in line", "11th in line", "22nd in line", "113th in line"]
975
        );
976
    }
977
}