OXYGENHUB
# c30444e35f2e oxygenhub / crates / hub / src / db / runners.rs
16.1 KiB · Rust · last changed Runners: Connectivity and Status apart, a heartbeat, a filterable queue 19 days ago source blame history raw
1
//! OxygenFlow runners (hub spec §13.2): who may claim jobs, what they are,
2
//! whether they are taking work, and what they have run.
3
4
use super::{now_ms, Pool};
5
use sqlx::FromRow;
6
7
#[derive(Clone, Debug, Default, FromRow)]
8
pub struct Runner {
9
    pub id: i64,
10
    pub name: String,
11
    pub token_hash: String,
12
    pub source: String,
13
    pub key: String,
14
    pub key_name: String,
15
    pub arch: String,
16
    pub os: String,
17
    pub version: String,
18
    pub labels: String,
19
    pub max_jobs: i64,
20
    pub memory_mib: Option<i64>,
21
    pub cpus: Option<i64>,
22
    pub state: String,
23
    pub created_ms: i64,
24
    pub created_by: String,
25
    pub last_seen_ms: Option<i64>,
26
    /// The owner whose repositories it serves; empty for the whole site.
27
    pub scope_owner: String,
28
    /// The one repository it serves, when it serves one.
29
    pub scope_repo_id: Option<i64>,
30
    /// Labels an administrator gave it, beside those it reports (#98).
31
    pub admin_labels: String,
32
    /// A time limit for its jobs in minutes, below the site's, when set.
33
    pub timeout_minutes: Option<i64>,
34
    /// It fetches images it lacks from the hub (oxygenflow#39).
35
    pub fetches_images: bool,
36
    /// What it measured at its last claim (oxygenflow#40): free disk (MiB),
37
    /// the 1-minute load ×100, its speed score and class, its kernel.
38
    pub free_disk_mib: Option<i64>,
39
    pub load_x100: Option<i64>,
40
    pub speed: Option<i64>,
41
    pub speed_class: Option<i64>,
42
    pub kernel: String,
43
    /// The tiers it offered at its last claim, comma-separated.
44
    pub tiers: String,
45
    /// The images it held then (a JSON array); empty when it did not say.
46
    pub images: String,
47
    /// BLAKE3 of what it offers; its declines hold until this changes.
48
    pub profile: String,
49
}
50
51
/// Which repositories' jobs a runner may take (oxygenhub#41).
52
#[derive(Clone, Debug, Default, PartialEq, Eq)]
53
pub struct Scope {
54
    /// Empty: any owner.
55
    pub owner: String,
56
    /// `None`: any of the owner's repositories.
57
    pub repo_id: Option<i64>,
58
}
59
60
impl Scope {
61
    pub fn is_site(&self) -> bool {
62
        self.owner.is_empty() && self.repo_id.is_none()
63
    }
64
}
65
66
/// A runner that has not polled for this long is shown offline.
67
pub const ONLINE_MS: i64 = 90_000;
68
69
impl Runner {
70
    pub fn online(&self) -> bool {
71
        self.last_seen_ms.is_some_and(|t| now_ms() - t < ONLINE_MS)
72
    }
73
74
    /// Whether the hub has heard from it lately (oxygenhub#243): any call it
75
    /// makes counts, a busy runner's heartbeat included.
76
    pub fn connectivity(&self) -> &'static str {
77
        if self.online() {
78
            "online"
79
        } else {
80
            "offline"
81
        }
82
    }
83
84
    /// `online`, `offline`, `draining` or `disabled`.
85
    pub fn status(&self) -> &'static str {
86
        match self.state.as_str() {
87
            "disabled" => "disabled",
88
            "draining" => "draining",
89
            _ if self.online() => "online",
90
            _ => "offline",
91
        }
92
    }
93
94
    /// Every label it carries: what it reports, and what an administrator
95
    /// gave it.
96
    pub fn label_list(&self) -> Vec<&str> {
97
        let mut out: Vec<&str> = self.labels.split(',').filter(|l| !l.is_empty()).collect();
98
        for l in self.admin_label_list() {
99
            if !out.contains(&l) {
100
                out.push(l);
101
            }
102
        }
103
        out
104
    }
105
106
    pub fn admin_label_list(&self) -> Vec<&str> {
107
        self.admin_labels.split(',').filter(|l| !l.is_empty()).collect()
108
    }
109
110
    /// The labels it reported itself, as the admin page distinguishes them.
111
    pub fn reported_label_list(&self) -> Vec<&str> {
112
        self.labels.split(',').filter(|l| !l.is_empty()).collect()
113
    }
114
115
    pub fn scope(&self) -> Scope {
116
        Scope { owner: self.scope_owner.clone(), repo_id: self.scope_repo_id }
117
    }
118
}
119
120
pub async fn by_token(pool: &Pool, token_hash: &str) -> sqlx::Result<Option<Runner>> {
121
    sqlx::query_as("SELECT * FROM flow_runners WHERE token_hash = $1").bind(token_hash).fetch_optional(pool).await
122
}
123
124
pub async fn get(pool: &Pool, id: i64) -> sqlx::Result<Option<Runner>> {
125
    sqlx::query_as("SELECT * FROM flow_runners WHERE id = $1").bind(id).fetch_optional(pool).await
126
}
127
128
pub async fn list(pool: &Pool) -> sqlx::Result<Vec<Runner>> {
129
    sqlx::query_as("SELECT * FROM flow_runners ORDER BY name, id").fetch_all(pool).await
130
}
131
132
/// The row of a runner configured with a hub.toml token, made on first sight.
133
pub async fn configured(pool: &Pool, token_hash: &str, name: &str) -> sqlx::Result<Runner> {
134
    sqlx::query("INSERT OR IGNORE INTO flow_runners (name, token_hash, source, created_ms) VALUES ($1, $2, 'config', $3)")
135
        .bind(name)
136
        .bind(token_hash)
137
        .bind(now_ms())
138
        .execute(pool)
139
        .await?;
140
    Ok(by_token(pool, token_hash).await?.expect("just made"))
141
}
142
143
/// What a claim says about the runner, and that it was seen now. A
144
/// configured runner takes the name it claims with.
145
/// The administrator's settings for a runner (#98).
146
pub async fn set_limits(pool: &Pool, id: i64, admin_labels: &str, timeout_minutes: Option<i64>) -> sqlx::Result<()> {
147
    sqlx::query("UPDATE flow_runners SET admin_labels = $1, timeout_minutes = $2 WHERE id = $3")
148
        .bind(admin_labels)
149
        .bind(timeout_minutes)
150
        .bind(id)
151
        .execute(pool)
152
        .await?;
153
    Ok(())
154
}
155
156
pub async fn seen(pool: &Pool, id: i64, name: Option<&str>, info: Option<&flow_core::api::RunnerInfo>) -> sqlx::Result<()> {
157
    let now = now_ms();
158
    match info {
159
        Some(i) => {
160
            sqlx::query(
161
                "UPDATE flow_runners SET arch = $1, os = $2, version = $3, labels = $4, max_jobs = $5, memory_mib = $6, cpus = $7, last_seen_ms = $8 WHERE id = $9",
162
            )
163
            .bind(&i.arch)
164
            .bind(&i.os)
165
            .bind(&i.version)
166
            .bind(i.labels.join(","))
167
            .bind(i.max_jobs as i64)
168
            .bind(i.memory_mib.map(i64::from))
169
            .bind(i.cpus.map(i64::from))
170
            .bind(now)
171
            .bind(id)
172
            .execute(pool)
173
            .await?;
174
        }
175
        None => {
176
            sqlx::query("UPDATE flow_runners SET last_seen_ms = $1 WHERE id = $2").bind(now).bind(id).execute(pool).await?;
177
        }
178
    }
179
    if let Some(n) = name {
180
        sqlx::query("UPDATE flow_runners SET name = $1 WHERE id = $2 AND source = 'config'").bind(n).bind(id).execute(pool).await?;
181
    }
182
    Ok(())
183
}
184
185
/// Keep what a runner measured about itself at a claim (oxygenflow#40).
186
pub async fn set_facts(pool: &Pool, id: i64, f: &flow_core::api::Facts) -> sqlx::Result<()> {
187
    sqlx::query("UPDATE flow_runners SET free_disk_mib = $1, load_x100 = $2, speed = $3, speed_class = $4, kernel = $5 WHERE id = $6")
188
        .bind(f.free_disk_mib.map(|v| v.min(i64::MAX as u64) as i64))
189
        .bind(f.load_x100.map(i64::from))
190
        .bind(f.speed.map(i64::from))
191
        .bind(f.speed_class.map(i64::from))
192
        .bind(f.kernel.clone().unwrap_or_default())
193
        .bind(id)
194
        .execute(pool)
195
        .await?;
196
    Ok(())
197
}
198
199
impl Runner {
200
    /// Its facts as words for the Runners page: "speed 100 (class 3) · load
201
    /// 1.12 · 41 GiB free"; empty when it has sent none.
202
    pub fn facts_line(&self) -> String {
203
        let mut out = Vec::new();
204
        if let Some(s) = self.speed {
205
            out.push(format!("speed {s} (class {})", self.speed_class.unwrap_or(0)));
206
        }
207
        if let Some(l) = self.load_x100 {
208
            out.push(format!("load {:.2}", l as f64 / 100.0));
209
        }
210
        if let Some(d) = self.free_disk_mib {
211
            out.push(if d >= 1024 { format!("{} GiB free", d / 1024) } else { format!("{d} MiB free") });
212
        }
213
        out.join(" · ")
214
    }
215
}
216
217
/// What a runner offers at a claim (OxygenFlow spec §11.1): its tiers and
218
/// images, and the profile hash over those, its architecture, labels and
219
/// capacity.
220
pub async fn set_offer(
221
    pool: &Pool,
222
    id: i64,
223
    tiers: &[String],
224
    images: Option<&[String]>,
225
    fetches_images: bool,
226
    profile: &str,
227
) -> sqlx::Result<()> {
228
    let images = images.map(|i| serde_json::to_string(i).unwrap_or_default()).unwrap_or_default();
229
    sqlx::query("UPDATE flow_runners SET tiers = $1, images = $2, fetches_images = $3, profile = $4 WHERE id = $5")
230
        .bind(tiers.join(","))
231
        .bind(images)
232
        .bind(fetches_images)
233
        .bind(profile)
234
        .bind(id)
235
        .execute(pool)
236
        .await?;
237
    Ok(())
238
}
239
240
impl Runner {
241
    pub fn tier_list(&self) -> Vec<&str> {
242
        self.tiers.split(',').filter(|t| !t.is_empty()).collect()
243
    }
244
245
    /// The images it holds, `None` when it does not say.
246
    pub fn image_list(&self) -> Option<Vec<String>> {
247
        if self.images.is_empty() {
248
            None
249
        } else {
250
            serde_json::from_str(&self.images).ok()
251
        }
252
    }
253
}
254
255
pub async fn set_state(pool: &Pool, id: i64, state: &str) -> sqlx::Result<()> {
256
    sqlx::query("UPDATE flow_runners SET state = $1, state_ms = $2 WHERE id = $3")
257
        .bind(state)
258
        .bind(super::now_ms())
259
        .bind(id)
260
        .execute(pool)
261
        .await?;
262
    Ok(())
263
}
264
265
/// Whether runner `id`'s last `n` jobs since its state last changed all
266
/// ended in an infrastructure error (oxygenflow#30), with the last one's.
267
pub async fn erroring(pool: &Pool, id: i64, n: u32) -> sqlx::Result<Option<String>> {
268
    if n == 0 {
269
        return Ok(None);
270
    }
271
    let last: Vec<(String, String)> = sqlx::query_as(
272
        "SELECT j.status, j.error FROM flow_jobs j JOIN flow_runners r ON r.id = j.runner_id
273
         WHERE j.runner_id = $1 AND j.finished_ms IS NOT NULL AND j.finished_ms >= r.state_ms
274
         ORDER BY j.finished_ms DESC, j.id DESC LIMIT $2",
275
    )
276
    .bind(id)
277
    .bind(n as i64)
278
    .fetch_all(pool)
279
    .await?;
280
    Ok((last.len() == n as usize && last.iter().all(|(s, _)| s == "error")).then(|| last[0].1.clone()))
281
}
282
283
/// Forget a runner: its credential stops working. Its jobs keep its name.
284
pub async fn remove(pool: &Pool, id: i64) -> sqlx::Result<()> {
285
    sqlx::query("DELETE FROM flow_runners WHERE id = $1").bind(id).execute(pool).await?;
286
    Ok(())
287
}
288
289
/// The key a runner signs receipts with: the one bound, or `key` (named
290
/// `name`) when none is yet.
291
pub async fn bind_key(pool: &Pool, id: i64, key: &str, name: &str) -> sqlx::Result<String> {
292
    sqlx::query("UPDATE flow_runners SET key = $1, key_name = $2 WHERE id = $3 AND key = ''")
293
        .bind(key)
294
        .bind(name)
295
        .bind(id)
296
        .execute(pool)
297
        .await?;
298
    sqlx::query_scalar("SELECT key FROM flow_runners WHERE id = $1").bind(id).fetch_one(pool).await
299
}
300
301
/// Keys that runners of this hub sign with.
302
pub async fn keys(pool: &Pool) -> sqlx::Result<Vec<String>> {
303
    sqlx::query_scalar("SELECT key FROM flow_runners WHERE key != ''").fetch_all(pool).await
304
}
305
306
/// A one-time registration token, valid for a day, for a runner of `scope`.
307
pub async fn add_registration(pool: &Pool, token_hash: &str, by: &str, scope: &Scope) -> sqlx::Result<()> {
308
    let now = now_ms();
309
    sqlx::query(
310
        "INSERT INTO flow_runner_registrations (token_hash, created_by, created_ms, expires_ms, scope_owner, scope_repo_id) VALUES ($1, $2, $3, $4, $5, $6)",
311
    )
312
    .bind(token_hash)
313
    .bind(by)
314
    .bind(now)
315
    .bind(now + 24 * 3_600_000)
316
    .bind(&scope.owner)
317
    .bind(scope.repo_id)
318
    .execute(pool)
319
    .await?;
320
    Ok(())
321
}
322
323
/// The runners that may take a repository's jobs: the site's, its owner's
324
/// and its own (oxygenhub#41).
325
pub async fn serving(pool: &Pool, owner: &str, repo_id: i64) -> sqlx::Result<Vec<Runner>> {
326
    sqlx::query_as(
327
        "SELECT * FROM flow_runners WHERE (scope_owner = '' AND scope_repo_id IS NULL) \
328
         OR (scope_owner = $1 AND scope_repo_id IS NULL) OR scope_repo_id = $2 ORDER BY name",
329
    )
330
    .bind(owner)
331
    .bind(repo_id)
332
    .fetch_all(pool)
333
    .await
334
}
335
336
/// The runners that serve an owner: the site's, and those scoped to the
337
/// owner or one of its repositories.
338
pub async fn serving_owner(pool: &Pool, owner: &str) -> sqlx::Result<Vec<Runner>> {
339
    sqlx::query_as("SELECT * FROM flow_runners WHERE (scope_owner = '' AND scope_repo_id IS NULL) OR scope_owner = $1 ORDER BY name")
340
        .bind(owner)
341
        .fetch_all(pool)
342
        .await
343
}
344
345
/// Change the repositories a runner serves.
346
pub async fn set_scope(pool: &Pool, id: i64, scope: &Scope) -> sqlx::Result<()> {
347
    sqlx::query("UPDATE flow_runners SET scope_owner = $1, scope_repo_id = $2 WHERE id = $3")
348
        .bind(&scope.owner)
349
        .bind(scope.repo_id)
350
        .bind(id)
351
        .execute(pool)
352
        .await?;
353
    Ok(())
354
}
355
356
/// Registration tokens not used yet and not expired: (created by, created, expires).
357
pub async fn pending_registrations(pool: &Pool) -> sqlx::Result<Vec<(String, i64, i64)>> {
358
    sqlx::query_as("SELECT created_by, created_ms, expires_ms FROM flow_runner_registrations WHERE used_ms IS NULL AND expires_ms > $1 ORDER BY created_ms DESC")
359
        .bind(now_ms())
360
        .fetch_all(pool)
361
        .await
362
}
363
364
/// Use registration token `token_hash` to register a runner: `None` when
365
/// the token is unknown, used or expired.
366
#[allow(clippy::too_many_arguments)]
367
pub async fn register(
368
    pool: &Pool,
369
    registration: &str,
370
    name: &str,
371
    token_hash: &str,
372
    key: &str,
373
    key_name: &str,
374
    by_default: &str,
375
) -> sqlx::Result<Option<Runner>> {
376
    let mut tx = super::begin_write(pool).await?;
377
    let now = now_ms();
378
    let found: Option<(String, String, Option<i64>)> = sqlx::query_as(
379
        "SELECT created_by, scope_owner, scope_repo_id FROM flow_runner_registrations WHERE token_hash = $1 AND used_ms IS NULL AND expires_ms > $2",
380
    )
381
    .bind(registration)
382
    .bind(now)
383
    .fetch_optional(&mut *tx)
384
    .await?;
385
    let Some((who, scope_owner, scope_repo_id)) = found else { return Ok(None) };
386
    let id: i64 = sqlx::query_scalar(
387
        "INSERT INTO flow_runners (name, token_hash, source, key, key_name, created_ms, created_by, scope_owner, scope_repo_id) VALUES ($1, $2, 'registered', $3, $4, $5, $6, $7, $8) RETURNING id",
388
    )
389
    .bind(name)
390
    .bind(token_hash)
391
    .bind(key)
392
    .bind(key_name)
393
    .bind(now)
394
    .bind(if who.is_empty() { by_default } else { &who })
395
    .bind(scope_owner)
396
    .bind(scope_repo_id)
397
    .fetch_one(&mut *tx)
398
    .await?;
399
    sqlx::query("UPDATE flow_runner_registrations SET used_ms = $1, runner_id = $2 WHERE token_hash = $3")
400
        .bind(now)
401
        .bind(id)
402
        .bind(registration)
403
        .execute(&mut *tx)
404
        .await?;
405
    tx.commit().await?;
406
    get(pool, id).await
407
}
408
409
/// A job a runner ran, for its history.
410
#[derive(Clone, Debug, FromRow)]
411
pub struct RunnerJob {
412
    pub owner: String,
413
    pub repo: String,
414
    pub run_number: i64,
415
    pub job: String,
416
    pub status: String,
417
    pub started_ms: Option<i64>,
418
    pub finished_ms: Option<i64>,
419
    pub receipt_attest: String,
420
}
421
422
/// Jobs runner `id` ran or is running, newest first.
423
pub async fn history(pool: &Pool, id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<RunnerJob>> {
424
    sqlx::query_as(
425
        "SELECT p.owner AS owner, p.name AS repo, r.number AS run_number, j.name AS job, j.status AS status, j.started_ms AS started_ms, \
426
         j.finished_ms AS finished_ms, j.receipt_attest AS receipt_attest \
427
         FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories p ON p.id = r.repo_id \
428
         WHERE j.runner_id = $1 AND ($2 IS NULL OR j.started_ms < $3) ORDER BY j.started_ms DESC LIMIT $4",
429
    )
430
    .bind(id)
431
    .bind(before)
432
    .bind(before)
433
    .bind(limit)
434
    .fetch_all(pool)
435
    .await
436
}
437
438
/// (jobs run, passed, busy milliseconds) of runner `id`.
439
pub async fn totals(pool: &Pool, id: i64) -> sqlx::Result<(i64, i64, i64)> {
440
    sqlx::query_as(
441
        "SELECT COUNT(*), COALESCE(SUM(CASE WHEN status = 'passed' THEN 1 ELSE 0 END), 0), CAST(COALESCE(SUM(finished_ms - started_ms), 0) AS BIGINT) \
442
         FROM flow_jobs WHERE runner_id = $1 AND finished_ms IS NOT NULL",
443
    )
444
    .bind(id)
445
    .fetch_one(pool)
446
    .await
447
}
448
449
/// Jobs running on each runner now, by runner id.
450
pub async fn running(pool: &Pool) -> sqlx::Result<Vec<(i64, i64)>> {
451
    sqlx::query_as("SELECT runner_id, COUNT(*) FROM flow_jobs WHERE status = 'running' AND runner_id IS NOT NULL GROUP BY runner_id")
452
        .fetch_all(pool)
453
        .await
454
}
455
456
pub async fn set_job_runner(pool: &Pool, job_id: i64, runner_id: i64) -> sqlx::Result<()> {
457
    sqlx::query("UPDATE flow_jobs SET runner_id = $1 WHERE id = $2").bind(runner_id).bind(job_id).execute(pool).await?;
458
    Ok(())
459
}