OXYGENHUB
# 6f66e397539c oxygenflow / crates / flow / src / runner.rs
102.1 KiB · Rust · last changed Windows runner: helpers start from the running binary; microVM says Linux only 11 days ago source blame history raw
e0918b88 Andy McCall 1
//! `flow runner` (docs/m0.md §5): claim a job from OxygenHub, fetch the
2
//! repository into a cache with the job's token, write the state into a
3
//! workspace, run it in its tier, stream the log, report the receipt.
4
//! One job at a time.
5
47c1d64d Andy McCall ↶ 6
use crate::home::{Home, RunnerConfig, RunnerKey};
e0918b88 Andy McCall 7
use crate::{microvm, workspace};
8
use anyhow::{bail, Context, Result};
b11e30f5 Andy McCall ↶ 9
use flow_core::api::{Claim, ClaimRequest, Decline, Finish, LogBatch, Record, Register, Registered, RunnerInfo, MAX_RECORD_TEXT};
e0918b88 Andy McCall 10
use flow_core::{scope_tree_hash, Receipt};
11
use oxy_core::hash::ChangeHash;
12
use oxy_core::repo::Repo;
46fe2c81 Andy McCall ↶ 13
use std::path::{Path, PathBuf};
e0918b88 Andy McCall 14
use std::time::{Duration, Instant};
15
280ef951 Andy McCall ↶ 16
/// A request the hub refused (4xx). On a job's endpoints, 409 means the
17
/// job no longer runs: it was cancelled, or given up.
18
#[derive(Debug)]
19
pub struct Refused {
20
    pub code: u16,
21
    pub text: String,
22
}
23
24
impl std::fmt::Display for Refused {
25
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
26
        write!(f, "HTTP {}: {}", self.code, self.text)
27
    }
28
}
29
30
impl std::error::Error for Refused {}
31
32
/// Whether `e` is the hub saying the job no longer runs.
33
fn job_gone(e: &anyhow::Error) -> bool {
34
    e.downcast_ref::<Refused>().is_some_and(|r| r.code == 409)
35
}
36
37
/// What a job ends with when its run was cancelled: nothing is reported,
38
/// as the hub has ended it already.
39
const CANCELLED: &str = "the run was cancelled";
40
41
fn cancelled() -> Finish {
8f211e20 Andy McCall ↶ 42
    Finish {
43
        exit_code: None,
44
        duration_ms: 0,
45
        receipt: None,
46
        signed_receipt: None,
47
        error: Some(CANCELLED.into()),
48
        peak_disk_mib: None,
49
        peak_memory_mib: None,
50
    }
280ef951 Andy McCall ↶ 51
}
52
e0918b88 Andy McCall 53
pub struct Hub {
54
    base: String,
55
    token: String,
56
    agent: ureq::Agent,
57
}
58
59
impl Hub {
60
    pub fn new(cfg: &RunnerConfig) -> Result<Hub> {
61
        let tls = oxy_core::protocol::tls_client_config(cfg.ca.as_deref()).map_err(|e| anyhow::anyhow!("{e}"))?;
62
        let agent = ureq::AgentBuilder::new().tls_config(tls).timeout(Duration::from_secs(60)).build();
63
        Ok(Hub { base: cfg.hub.trim_end_matches('/').to_string(), token: cfg.token.clone(), agent })
64
    }
65
1b2229da Andy McCall ↶ 66
    /// POST to the hub. A server error (5xx) means nothing was committed,
67
    /// so it is tried again after 1, 2 and 4 seconds before giving up: a
68
    /// busy hub costs a job a few seconds, not its log lines or its claim.
e0918b88 Andy McCall 69
    fn post(&self, path: &str, body: &impl serde::Serialize) -> Result<Option<ureq::Response>> {
1b2229da Andy McCall ↶ 70
        self.post_with(path, body, &RETRY_DELAYS)
71
    }
72
73
    fn post_with(&self, path: &str, body: &impl serde::Serialize, delays: &[Duration]) -> Result<Option<ureq::Response>> {
e0918b88 Andy McCall 74
        let url = format!("{}/api/v1{path}", self.base);
89171fca Andy McCall ↶ 75
        let r = retrying(&url, delays, false, || {
76
            self.agent.post(&url).set("Authorization", &format!("Bearer {}", self.token)).send_json(body).map_err(Box::new)
77
        })?;
78
        Ok(if r.status() == 204 { None } else { Some(r) })
79
    }
80
81
    /// POST what a job leaves behind (oxygenflow#49): through a hub that is
82
    /// away for a few minutes, as while a deploy restarts it, so a job's
83
    /// declines and uploads outlast it.
84
    fn post_patiently(&self, path: &str, body: &impl serde::Serialize) -> Result<Option<ureq::Response>> {
85
        let url = format!("{}/api/v1{path}", self.base);
86
        let r = retrying(&url, &PATIENT_DELAYS, true, || {
87
            self.agent.post(&url).set("Authorization", &format!("Bearer {}", self.token)).send_json(body).map_err(Box::new)
88
        })?;
89
        Ok(if r.status() == 204 { None } else { Some(r) })
e0918b88 Andy McCall 90
    }
91
92
    pub fn claim(&self, req: &ClaimRequest) -> Result<Option<Claim>> {
93
        match self.post("/flow/claim", req)? {
94
            Some(r) => Ok(Some(r.into_json()?)),
95
            None => Ok(None),
96
        }
97
    }
98
b11e30f5 Andy McCall ↶ 99
    /// Hand job `job` back, saying why this runner will not run it.
100
    pub fn decline(&self, job: i64, reason: &str) -> Result<()> {
89171fca Andy McCall ↶ 101
        self.post_patiently(&format!("/flow/jobs/{job}/decline"), &Decline { reason: reason.into() })?;
b11e30f5 Andy McCall ↶ 102
        Ok(())
103
    }
104
06a4c472 Andy McCall ↶ 105
    /// Hand back a job this runner holds as it leaves (oxygenflow#76): the
106
    /// hub queues it again ahead of new work for another runner.
107
    pub fn release(&self, job: i64, reason: &str) -> Result<()> {
108
        self.post_patiently(&format!("/flow/jobs/{job}/release"), &Decline { reason: reason.into() })?;
109
        Ok(())
110
    }
111
a4b3f8cb Andy McCall ↶ 112
    /// Say this runner is there while it claims nothing: its slots are full,
113
    /// or it is draining (oxygenhub#243). Once, no retries: the next comes
114
    /// soon. A hub without the endpoint (404) is not an error.
35f7a849 Andy McCall ↶ 115
    /// A ten-minute certificate for `identity`, asked with a job's token
116
    /// (keyless, oxygenhub#241), and the identity of the hub's key that
117
    /// issued it, from `/.well-known/oxyhub-key`.
118
    pub fn certify(
119
        &self,
120
        job_token: &str,
121
        identity: &oxy_core::identity::Identity,
122
        proof: &oxy_core::identity::Signature,
123
    ) -> Result<(oxy_core::cert::SignedCertificate, oxy_core::identity::Identity)> {
124
        use oxy_core::dsse::{base64, base64_decode};
125
        let body = serde_json::json!({
126
            "identity": base64(&identity.encode()),
127
            "proof": base64(&oxy_core::cbor::encode(&proof.to_value())),
128
            "scope": "seal",
129
        });
130
        let url = format!("{}/api/v1/keyless/certify", self.base);
131
        let r = self
132
            .agent
133
            .post(&url)
134
            .set("Authorization", &format!("Bearer {job_token}"))
135
            .send_json(body)
136
            .map_err(|e| anyhow::anyhow!("asking the hub to certify the key: {e}"))?;
137
        let v: serde_json::Value = r.into_json()?;
138
        let cert = base64_decode(v["certificate"].as_str().unwrap_or(""))
139
            .context("the hub's answer has no certificate")
140
            .and_then(|b| oxy_core::cert::SignedCertificate::decode(&b).map_err(|e| anyhow::anyhow!("{e}")))?;
141
        let mut bytes = Vec::new();
142
        std::io::Read::read_to_end(
143
            &mut self.agent.get(&format!("{}/.well-known/oxyhub-key", self.base)).call()?.into_reader(),
144
            &mut bytes,
145
        )?;
146
        let issuer = oxy_core::identity::Identity::decode(&bytes)?;
147
        if issuer.hash() != cert.cert.issuer || cert.verify(&issuer).is_err() {
148
            bail!("the certificate is not signed by the hub's published key");
149
        }
150
        if cert.cert.key_id() != identity.hash() {
151
            bail!("the hub certified another key");
152
        }
153
        Ok((cert, issuer))
154
    }
155
a4b3f8cb Andy McCall ↶ 156
    pub fn heartbeat(&self) -> Result<()> {
157
        match self.post_with("/flow/runners/heartbeat", &serde_json::json!({}), &[]) {
158
            Ok(_) => Ok(()),
159
            Err(e) if format!("{e:#}").contains("404") => Ok(()),
160
            Err(e) => Err(e),
161
        }
162
    }
163
cedd1234 Andy McCall ↶ 164
    /// The update policy for this runner and the build it should run
165
    /// (oxygenhub#322); `None` from a hub without runner updates.
166
    pub fn update_answer(&self) -> Result<Option<flow_core::api::UpdateAnswer>> {
167
        let url = format!("{}/api/v1/flow/runner/update", self.base);
168
        match self.agent.get(&url).set("Authorization", &format!("Bearer {}", self.token)).call() {
169
            Ok(r) => Ok(Some(r.into_json()?)),
170
            Err(ureq::Error::Status(404, _)) => Ok(None),
171
            Err(e) => Err(anyhow::anyhow!("{url}: {e}")),
172
        }
173
    }
174
175
    /// Tell the hub what this runner did about its update.
176
    pub fn report_update(&self, note: &str) -> Result<()> {
177
        self.post_with("/flow/runner/update", &flow_core::api::UpdateReport { note: note.to_string() }, &[])?;
178
        Ok(())
179
    }
180
181
    /// Download an offered build to `to`, refusing it unless its BLAKE3
182
    /// is the one the hub attests and it is the size it said.
183
    pub fn download_build(&self, offer: &flow_core::api::UpdateOffer, to: &Path) -> Result<()> {
184
        use std::io::{Read, Write};
185
        let url = format!("{}{}", self.base, offer.url);
186
        let r = self.agent.get(&url).timeout(Duration::from_secs(1800)).call().map_err(|e| anyhow::anyhow!("{url}: {e}"))?;
187
        let part = to.with_extension("partial");
188
        let mut out = std::fs::File::create(&part).with_context(|| format!("writing {}", part.display()))?;
189
        let (mut hasher, mut n, mut buf) = (blake3::Hasher::new(), 0u64, vec![0u8; 1 << 16]);
190
        let mut body = r.into_reader().take(offer.size + 1);
191
        loop {
192
            let k = body.read(&mut buf)?;
193
            if k == 0 {
194
                break;
195
            }
196
            hasher.update(&buf[..k]);
197
            out.write_all(&buf[..k])?;
198
            n += k as u64;
199
        }
200
        drop(out);
201
        let got = hasher.finalize().to_hex().to_string();
202
        if n != offer.size || got != offer.blake3 {
203
            let _ = std::fs::remove_file(&part);
204
            bail!("{} does not match what the hub attests ({n} bytes, BLAKE3 {}…)", offer.asset, &got[..12]);
205
        }
206
        std::fs::rename(&part, to)?;
207
        Ok(())
208
    }
209
e0918b88 Andy McCall 210
    pub fn log(&self, job: i64, records: Vec<Record>) -> Result<()> {
211
        self.post(&format!("/flow/jobs/{job}/log"), &LogBatch { records })?;
212
        Ok(())
213
    }
214
a67b97e9 Andy McCall ↶ 215
    /// Upload artifact `name` of job `job` (spec §5.2); the hub keeps it as
216
    /// a blob of the job's repository and answers with its ObjectRef.
217
    pub fn artifact(&self, job: i64, name: &str, bytes: &[u8]) -> Result<String> {
218
        let url = format!("{}/api/v1/flow/jobs/{job}/artifacts/{name}", self.base);
89171fca Andy McCall ↶ 219
        let r = retrying(&url, &PATIENT_DELAYS, true, || {
220
            self.agent
a67b97e9 Andy McCall ↶ 221
                .post(&url)
222
                .timeout(Duration::from_secs(600))
223
                .set("Authorization", &format!("Bearer {}", self.token))
89171fca Andy McCall ↶ 224
                .set("Content-Type", "application/octet-stream")
225
                .send_bytes(bytes)
226
                .map_err(Box::new)
227
        })?;
228
        let v: serde_json::Value = r.into_json()?;
229
        v["blob"].as_str().map(String::from).context("the hub named no blob")
a67b97e9 Andy McCall ↶ 230
    }
231
f201e1bc Andy McCall ↶ 232
    /// Keep a checkpoint of job `job` (oxygenflow#77) as its runner hands
233
    /// it back: the hub answers with its blob, which the next attempt's
234
    /// claim names.
235
    pub fn save_checkpoint(&self, job: i64, bytes: &[u8]) -> Result<String> {
236
        let url = format!("{}/api/v1/flow/jobs/{job}/checkpoint", self.base);
237
        let r = retrying(&url, &PATIENT_DELAYS, true, || {
238
            self.agent
239
                .post(&url)
240
                .timeout(Duration::from_secs(600))
241
                .set("Authorization", &format!("Bearer {}", self.token))
242
                .set("Content-Type", "application/octet-stream")
243
                .send_bytes(bytes)
244
                .map_err(Box::new)
245
        })?;
246
        let v: serde_json::Value = r.into_json()?;
247
        v["blob"].as_str().map(String::from).context("the hub named no blob")
248
    }
249
250
    /// The checkpoint an earlier attempt of job `job` saved (oxygenflow#77),
251
    /// which the caller checks against the blob its claim names.
252
    pub fn checkpoint(&self, job: i64) -> Result<Vec<u8>> {
253
        let url = format!("{}/api/v1/flow/jobs/{job}/checkpoint", self.base);
254
        let r = retrying(&url, &PATIENT_DELAYS, false, || {
255
            self.agent
256
                .get(&url)
257
                .timeout(Duration::from_secs(600))
258
                .set("Authorization", &format!("Bearer {}", self.token))
259
                .call()
260
                .map_err(Box::new)
261
        })?;
262
        let mut bytes = Vec::new();
263
        std::io::Read::read_to_end(&mut std::io::Read::take(r.into_reader(), crate::checkpoint::MAX_BYTES as u64 + 1), &mut bytes)?;
264
        Ok(bytes)
265
    }
266
46fe2c81 Andy McCall ↶ 267
    /// Input `index` of job `job` (oxygenflow#58): the bytes of the
268
    /// producing job's artifact, which the caller checks against the blob the
269
    /// claim names.
270
    pub fn input(&self, job: i64, index: usize) -> Result<Vec<u8>> {
271
        let url = format!("{}/api/v1/flow/jobs/{job}/inputs/{index}", self.base);
272
        let r = retrying(&url, &PATIENT_DELAYS, false, || {
273
            self.agent
274
                .get(&url)
275
                .timeout(Duration::from_secs(600))
276
                .set("Authorization", &format!("Bearer {}", self.token))
277
                .call()
278
                .map_err(Box::new)
279
        })?;
280
        let mut bytes = Vec::new();
281
        let mut reader = r.into_reader();
282
        std::io::Read::read_to_end(&mut std::io::Read::take(&mut reader, MAX_INPUT_BYTES as u64 + 1), &mut bytes)?;
283
        if bytes.len() > MAX_INPUT_BYTES {
284
            bail!("larger than {} MiB", MAX_INPUT_BYTES >> 20);
285
        }
286
        Ok(bytes)
287
    }
288
6d53fb7d Andy McCall ↶ 289
    /// Upload file `path` of report `name` of job `job` (oxygenhub#127);
290
    /// the hub keeps it as a blob and answers with its ObjectRef. The
291
    /// report's manifest follows through [`Hub::artifact`].
292
    pub fn report_file(&self, job: i64, name: &str, path: &str, bytes: &[u8]) -> Result<String> {
293
        let url = format!("{}/api/v1/flow/jobs/{job}/reports/{name}/files", self.base);
89171fca Andy McCall ↶ 294
        let r = retrying(&url, &PATIENT_DELAYS, true, || {
295
            self.agent
6d53fb7d Andy McCall ↶ 296
                .post(&url)
297
                .timeout(Duration::from_secs(600))
298
                .set("Authorization", &format!("Bearer {}", self.token))
299
                .set("Content-Type", "application/octet-stream")
89171fca Andy McCall ↶ 300
                .set("X-Flow-Report-Path", path)
301
                .send_bytes(bytes)
302
                .map_err(Box::new)
303
        })?;
304
        let v: serde_json::Value = r.into_json()?;
305
        v["blob"].as_str().map(String::from).context("the hub named no blob")
6d53fb7d Andy McCall ↶ 306
    }
307
3007d8f0 Andy McCall ↶ 308
    /// Report a job's end until the hub has it (oxygenhub#71): while the hub
309
    /// cannot be reached or fails (5xx), try again after `first_wait`,
310
    /// doubling to a minute, for up to `patience`. A refusal (4xx: the job was ended
311
    /// already, say) is final. A result lost here leaves the job running
312
    /// on the hub until its time limit.
313
    pub fn finish_until_accepted(&self, job: i64, f: &Finish, patience: Duration, first_wait: Duration) -> Result<()> {
314
        let url = format!("{}/api/v1/flow/jobs/{job}/finish", self.base);
315
        let deadline = std::time::Instant::now() + patience;
316
        let mut wait = first_wait;
317
        loop {
318
            let why = match self.agent.post(&url).set("Authorization", &format!("Bearer {}", self.token)).send_json(f) {
319
                Ok(_) => return Ok(()),
320
                Err(ureq::Error::Status(code, r)) if code < 500 => {
321
                    bail!("{url}: HTTP {code}: {}", r.into_string().unwrap_or_default())
322
                }
323
                Err(ureq::Error::Status(code, _)) => format!("HTTP {code}"),
324
                Err(e) => e.to_string(),
325
            };
326
            if std::time::Instant::now() + wait > deadline {
327
                bail!("{url}: gave up after {} minutes: {why}", patience.as_secs() / 60);
328
            }
329
            eprintln!("job {job}: the hub did not take the result ({why}); trying again in {} s", wait.as_secs());
330
            std::thread::sleep(wait);
331
            wait = (wait * 2).min(Duration::from_secs(60));
332
        }
333
    }
334
}
335
336
/// How long a runner keeps trying to report a job's end.
337
const FINISH_PATIENCE: Duration = Duration::from_secs(2 * 3600);
338
339
/// Where a job's result waits until the hub has it, so a runner that
340
/// restarts still reports it.
341
fn pending_path(home: &Home, job: i64) -> PathBuf {
342
    home.root.join("pending").join(format!("{job}.json"))
343
}
344
345
/// Report `f`, keeping it on disk until the hub has it.
346
fn report_finish(home: &Home, hub: &Hub, job: i64, f: &Finish) {
347
    let path = pending_path(home, job);
348
    let kept = std::fs::create_dir_all(path.parent().unwrap())
349
        .and_then(|_| std::fs::write(&path, serde_json::to_vec(f).unwrap_or_default()))
350
        .is_ok();
351
    match hub.finish_until_accepted(job, f, FINISH_PATIENCE, Duration::from_secs(5)) {
352
        Ok(()) => {}
353
        Err(e) => eprintln!("job {job}: reporting failed: {e:#}"),
354
    }
355
    if kept {
356
        let _ = std::fs::remove_file(&path);
f98a8f78 Andy McCall ↶ 357
    }
3007d8f0 Andy McCall ↶ 358
}
359
360
/// Results a previous run of the runner could not report, to send again.
361
fn report_pending(home: &Home) -> Vec<(i64, Finish)> {
362
    let Ok(dir) = std::fs::read_dir(home.root.join("pending")) else {
363
        return Vec::new();
364
    };
365
    dir.filter_map(|e| {
366
        let p = e.ok()?.path();
367
        let job: i64 = p.file_stem()?.to_str()?.parse().ok()?;
368
        let f: Finish = serde_json::from_slice(&std::fs::read(&p).ok()?).ok()?;
369
        Some((job, f))
370
    })
371
    .collect()
f98a8f78 Andy McCall ↶ 372
}
373
374
/// `flow runner register`: exchange a one-time registration token from
375
/// the hub for this runner's credential, binding the runner's signing key,
376
/// and keep hub, credential and name in runner.toml.
377
pub fn register(home: &Home, hub: &str, token: &str, name: &str, ca: Option<PathBuf>, labels: Vec<String>) -> Result<Registered> {
378
    if home.config.runner.is_some() && !home.root.join(crate::home::RUNNER_TOML).is_file() {
379
        bail!("flow.toml already has a [runner] section; remove it to register this runner instead");
380
    }
381
    let key = home.runner_key(name)?;
382
    let hub = hub.trim_end_matches('/').to_string();
383
    let tls = oxy_core::protocol::tls_client_config(ca.as_deref()).map_err(|e| anyhow::anyhow!("{e}"))?;
384
    let agent = ureq::AgentBuilder::new().tls_config(tls).timeout(Duration::from_secs(60)).build();
385
    let req = Register {
386
        token: token.trim().to_string(),
387
        name: name.to_string(),
388
        identity: hex::encode(key.identity.encode()),
2d025bbf Andy McCall ↶ 389
        info: RunnerInfo { version: env!("OXY_VERSION_LONG").into(), ..RunnerInfo::here(labels.clone(), 1, None, None) },
f98a8f78 Andy McCall ↶ 390
    };
391
    let url = format!("{hub}/api/v1/flow/runners/register");
392
    let got: Registered = match agent.post(&url).send_json(&req) {
393
        Ok(r) => r.into_json()?,
394
        Err(ureq::Error::Status(code, r)) => bail!("{url}: HTTP {code}: {}", r.into_string().unwrap_or_default()),
395
        Err(e) => bail!("{url}: {e}"),
396
    };
397
    let mut text = format!(
398
        "# Written by `flow runner register`; the credential is secret.\nhub = {hub:?}\ntoken = {:?}\nname = {:?}\n",
399
        got.credential, got.name
400
    );
401
    if let Some(ca) = &ca {
402
        text.push_str(&format!("ca = {:?}\n", ca.display().to_string()));
403
    }
404
    if !labels.is_empty() {
405
        text.push_str(&format!("labels = {labels:?}\n"));
406
    }
407
    let path = home.root.join(crate::home::RUNNER_TOML);
408
    {
409
        use std::io::Write;
410
        let tmp = path.with_extension("toml.tmp");
3554afef Andy McCall ↶ 411
        let mut f = crate::portable::private(std::fs::OpenOptions::new().write(true).create(true).truncate(true)).open(&tmp)?;
f98a8f78 Andy McCall ↶ 412
        f.write_all(text.as_bytes())?;
413
        f.sync_all()?;
414
        std::fs::rename(&tmp, &path)?;
e0918b88 Andy McCall 415
    }
f98a8f78 Andy McCall ↶ 416
    Ok(got)
e0918b88 Andy McCall 417
}
1b2229da Andy McCall ↶ 418
419
/// Waits before trying a hub request again after a server error.
420
const RETRY_DELAYS: [Duration; 3] = [Duration::from_secs(1), Duration::from_secs(2), Duration::from_secs(4)];
89171fca Andy McCall ↶ 421
422
/// Waits for what a job leaves behind (oxygenflow#49): three minutes in
423
/// all, longer than a hub takes to restart.
46fe2c81 Andy McCall ↶ 424
/// Check an input's bytes against the blob the producing job's receipt
425
/// names, then write them at its path under `tree` (the workspace's `/src`).
426
fn place_input(tree: &Path, i: &flow_core::api::ArtifactInput, bytes: &[u8]) -> Result<[u8; 32]> {
427
    let Some(rel) = i.path.strip_prefix("/src/").filter(|r| r.split('/').all(|p| !p.is_empty() && p != "." && p != ".." && p != ".oxy"))
428
    else {
429
        bail!("path {:?} is not a clean path under /src", i.path);
430
    };
431
    let blob = Repo::blob_ref(bytes);
432
    if blob.to_hex() != i.blob {
433
        bail!("it is not what {}'s receipt names (blob {}, not {})", i.from, &blob.to_hex()[..12], &i.blob[..i.blob.len().min(12)]);
434
    }
435
    let dest = tree.join(rel);
436
    if let Some(dir) = dest.parent() {
437
        std::fs::create_dir_all(dir)?;
438
    }
439
    std::fs::write(&dest, bytes)?;
440
    Ok(blob.0)
441
}
442
443
/// The largest input a job takes: the hub's artifact limit.
444
pub const MAX_INPUT_BYTES: usize = 512 << 20;
445
89171fca Andy McCall ↶ 446
const PATIENT_DELAYS: [Duration; 15] = {
447
    let mut d = [Duration::from_secs(15); 15];
448
    d[0] = Duration::from_secs(1);
449
    d[1] = Duration::from_secs(2);
450
    d[2] = Duration::from_secs(4);
451
    d[3] = Duration::from_secs(8);
452
    d
453
};
454
455
/// Send a request until the hub answers, waiting `delays` in turn after a
456
/// server error (5xx: nothing was committed) and, when `connection`, after
457
/// the hub could not be reached at all. A refusal (4xx) is final, as a
458
/// [`Refused`] the caller can look at.
459
fn retrying(
460
    url: &str,
461
    delays: &[Duration],
462
    connection: bool,
463
    mut send: impl FnMut() -> Result<ureq::Response, Box<ureq::Error>>,
464
) -> Result<ureq::Response> {
465
    let mut delays = delays.iter();
466
    loop {
467
        let why = match send().map_err(|e| *e) {
468
            Ok(r) => return Ok(r),
469
            Err(ureq::Error::Status(code, r)) if code < 500 => {
470
                let text = r.into_string().unwrap_or_default();
471
                return Err(anyhow::Error::new(Refused { code, text }).context(url.to_string()));
472
            }
473
            Err(ureq::Error::Status(code, r)) => format!("HTTP {code}: {}", r.into_string().unwrap_or_default()),
474
            Err(e) if connection => e.to_string(),
475
            Err(e) => bail!("{url}: {e}"),
476
        };
477
        match delays.next() {
478
            Some(d) => std::thread::sleep(*d),
479
            None => bail!("{url}: {why}"),
480
        }
481
    }
482
}
e0918b88 Andy McCall 483
bb643c25 Andy McCall ↶ 484
/// What the runner may run at once (flow.toml `[runner]`): at most
485
/// `max_jobs` jobs, whose sandboxes together fit `memory_mib` and `cpus`.
486
#[derive(Debug, Default, PartialEq, Eq)]
487
pub struct Budget {
488
    pub max_jobs: usize,
489
    pub memory_mib: Option<u32>,
490
    pub cpus: Option<u32>,
491
}
492
493
/// What is in use now.
494
#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
495
pub struct Used {
496
    pub jobs: usize,
497
    pub memory_mib: u32,
498
    pub cpus: u32,
499
}
500
501
impl Budget {
502
    pub fn of(cfg: &RunnerConfig) -> Budget {
503
        Budget { max_jobs: cfg.max_jobs.unwrap_or(1).max(1), memory_mib: cfg.memory_mib, cpus: cfg.cpus }
504
    }
505
506
    /// The room to ask the hub for, or `None` when no job more can start.
507
    /// A lone job may use the whole budget even if it asks for more, so a
508
    /// large sandbox is never starved.
509
    pub fn room(&self, used: Used) -> Option<(Option<u32>, Option<u32>)> {
510
        if used.jobs >= self.max_jobs {
511
            return None;
512
        }
513
        if used.jobs == 0 {
514
            return Some((None, None));
515
        }
516
        let mem = self.memory_mib.map(|m| m.saturating_sub(used.memory_mib));
517
        let cpus = self.cpus.map(|c| c.saturating_sub(used.cpus));
518
        if mem == Some(0) || cpus == Some(0) {
519
            return None;
520
        }
521
        Some((mem, cpus))
522
    }
523
}
524
c5b702e5 Andy McCall ↶ 525
/// The isolation tiers this machine can honour now (spec §11.1): the
ea56c56e Andy McCall ↶ 526
/// MicroVM tier when `/dev/kvm` opens for reading and writing and
527
/// flow.toml pins Firecracker; the syscall tier when this binary may use
528
/// user namespaces, Landlock and seccomp and manage a delegated cgroup.
529
/// `Err` says why there are none; the reasons for missing tiers come back
530
/// with the ones there are.
531
pub fn tiers(home: &Home) -> std::result::Result<(Vec<String>, Vec<String>), String> {
532
    let mut have = Vec::new();
533
    let mut missing = Vec::new();
c5b702e5 Andy McCall ↶ 534
    let kvm = std::fs::OpenOptions::new().read(true).write(true).open("/dev/kvm");
535
    match (kvm, home.config.microvm.is_some()) {
e6949d87 Andy McCall ↶ 536
        _ if !cfg!(target_os = "linux") => missing.push("microvm: the tier runs only on Linux (Firecracker and KVM)".to_string()),
ea56c56e Andy McCall ↶ 537
        (Ok(_), true) => have.push("microvm".to_string()),
538
        (Ok(_), false) => missing.push("microvm: flow.toml has no [microvm] section".to_string()),
539
        (Err(e), _) => missing.push(format!(
540
            "microvm: /dev/kvm cannot be opened ({e}); the tier needs hardware virtualisation and this user in the kvm group"
c5b702e5 Andy McCall ↶ 541
        )),
542
    }
ea56c56e Andy McCall ↶ 543
    match crate::syscall::available() {
544
        Ok(()) => have.push("syscall".to_string()),
545
        Err(e) => missing.push(format!("syscall: {e}")),
546
    }
28cc603f Andy McCall ↶ 547
    // wasmtime runs on every machine this binary does (oxygenflow#75).
548
    have.push("wasm".to_string());
ea56c56e Andy McCall ↶ 549
    if have.is_empty() {
550
        return Err(format!("no isolation tier can run here: {}", missing.join("; ")));
551
    }
552
    Ok((have, missing))
b11e30f5 Andy McCall ↶ 553
}
554
4cf0c99a Andy McCall ↶ 555
/// Fetch `claim`'s image from the hub when this home lacks it
556
/// (oxygenflow#39), saying so in the job's log. A hub that does not hold it
557
/// leaves it missing, and the preflight declines the job as before.
b14aeaae Andy McCall ↶ 558
fn fetch_image(home: &Home, hub: &Hub, cfg: &RunnerConfig, swarm: Option<&crate::swarm::Swarm>, claim: &Claim) {
4cf0c99a Andy McCall ↶ 559
    let hash = &claim.sandbox.rootfs;
560
    if hash.is_empty() || home.rootfs(hash).is_file() || !jet::images::is_hash(hash) {
561
        return;
562
    }
563
    let note = |text: String| {
564
        eprintln!("job {}: {text}", claim.job_id);
565
        hub.log(claim.job_id, vec![Record { t: 0, stream: "flow".into(), text }]).ok();
566
    };
567
    // One fetch of an image at a time: two jobs wanting it wait for one.
568
    static FETCHING: std::sync::Mutex<()> = std::sync::Mutex::new(());
569
    let _one = FETCHING.lock().unwrap_or_else(|p| p.into_inner());
570
    if home.rootfs(hash).is_file() {
571
        return;
572
    }
573
    let client = match crate::images::Client::new(&cfg.hub, &cfg.token, cfg.ca.as_deref()) {
574
        Ok(c) => c,
575
        Err(e) => return note(format!("image {}: cannot reach the hub: {e:#}", &hash[..12])),
576
    };
b14aeaae Andy McCall ↶ 577
    let fetched = match swarm {
578
        Some(sw) => {
579
            note(format!("image {} is not on this runner: fetching it from its peers and the hub", &hash[..12]));
580
            sw.fetch(home, hash)
581
        }
582
        None => {
583
            note(format!("image {} is not on this runner: fetching it from the hub", &hash[..12]));
584
            client.fetch(home, hash)
585
        }
586
    };
587
    match fetched {
4cf0c99a Andy McCall ↶ 588
        Ok(Some(m)) => note(format!(
b14aeaae Andy McCall ↶ 589
            "image {}: fetched {} chunks, {}{} in {:.1} s",
4cf0c99a Andy McCall ↶ 590
            &hash[..12],
591
            m.chunks,
592
            crate::images::human(m.bytes),
b14aeaae Andy McCall ↶ 593
            if swarm.is_some() { format!(" ({} from peers)", crate::images::human(m.peer_bytes)) } else { String::new() },
4cf0c99a Andy McCall ↶ 594
            m.elapsed.as_secs_f64()
595
        )),
596
        Ok(None) => note(format!("image {}: the hub does not hold it either", &hash[..12])),
597
        Err(e) => note(format!("image {}: fetch failed: {e:#}", &hash[..12])),
598
    }
599
}
600
b11e30f5 Andy McCall ↶ 601
/// Why this runner cannot run `claim` now (spec §11.5), checked before
602
/// anything is fetched: the tier, the architecture, the image and the
603
/// memory. The hub chose it from what the runner last said, which may be
604
/// out of date.
605
fn preflight(home: &Home, claim: &Claim) -> Option<String> {
606
    let sb = &claim.sandbox;
607
    match sb.runtime {
608
        flow_core::Runtime::Microvm => {
609
            if let Err(e) = std::fs::OpenOptions::new().read(true).write(true).open("/dev/kvm") {
610
                return Some(format!("the MicroVM tier needs /dev/kvm, which cannot be opened here ({e})"));
611
            }
612
            if home.config.microvm.is_none() {
613
                return Some("the MicroVM tier is not set up here (flow.toml has no [microvm] section)".into());
614
            }
615
        }
616
        flow_core::Runtime::Syscall => {
617
            if let Err(e) = crate::syscall::available() {
618
                return Some(format!("the syscall tier cannot run here: {e}"));
619
            }
620
        }
28cc603f Andy McCall ↶ 621
        // No image and no architecture: only the memory matters.
622
        flow_core::Runtime::Wasm => {
623
            return available_memory_mib()
624
                .filter(|free| *free < u64::from(sb.memory_mib))
625
                .map(|free| format!("only {free} MiB of memory is free here; the sandbox needs {} MiB", sb.memory_mib));
626
        }
b11e30f5 Andy McCall ↶ 627
    }
628
    sandbox_problem(home, sb, std::env::consts::ARCH, available_memory_mib())
629
}
630
631
/// What stops `sb` running on this home's images, on a machine of
632
/// architecture `here` with `free_mib` of memory free.
633
fn sandbox_problem(home: &Home, sb: &flow_core::Sandbox, here: &str, free_mib: Option<u64>) -> Option<String> {
634
    if let Some(want) = sb.arch.as_deref().filter(|a| *a != here) {
635
        return Some(format!("the sandbox needs {want}; this runner is {here}"));
636
    }
637
    let short = &sb.rootfs[..12.min(sb.rootfs.len())];
638
    let image = home.rootfs(&sb.rootfs);
639
    if !image.is_file() {
640
        return Some(format!("rootfs {short} is not on this runner"));
641
    }
642
    if let Some(arch) = home.image_arch(&sb.rootfs).filter(|a| a != here) {
643
        return Some(format!("rootfs {short} is for {arch}; this runner is {here}"));
644
    }
645
    if let Err(e) = home.verified(&image, &sb.rootfs) {
646
        return Some(format!("rootfs {short} does not verify here: {e:#}"));
647
    }
648
    if let Some(free) = free_mib.filter(|free| *free < u64::from(sb.memory_mib)) {
649
        return Some(format!("only {free} MiB of memory is free here; the sandbox needs {} MiB", sb.memory_mib));
650
    }
651
    None
652
}
653
654
/// `MemAvailable` from /proc/meminfo, in MiB.
655
fn available_memory_mib() -> Option<u64> {
656
    let info = std::fs::read_to_string("/proc/meminfo").ok()?;
657
    let kib: u64 = info.lines().find_map(|l| l.strip_prefix("MemAvailable:"))?.trim().trim_end_matches("kB").trim().parse().ok()?;
658
    Some(kib / 1024)
c5b702e5 Andy McCall ↶ 659
}
660
e0918b88 Andy McCall 661
pub fn serve(home: &Home) -> Result<()> {
5cf1e8ae Andy McCall ↶ 662
    drain_on_stop();
85275e1e Andy McCall ↶ 663
    crate::update::note_runner(&home.root);
664
    // `flow runner update` swapped a new build in and asks this one to
665
    // leave with the notice it gave (#85).
3554afef Andy McCall ↶ 666
    #[cfg(unix)]
667
    {
668
        let handler = on_update as extern "C" fn(libc::c_int) as libc::sighandler_t;
669
        unsafe { libc::signal(libc::SIGUSR1, handler) };
670
    }
85275e1e Andy McCall ↶ 671
    if let Some(p) = crate::update::pending(&home.root) {
672
        // On trial: a build that does not reach the hub in time goes back.
673
        let root = home.root.clone();
674
        std::thread::spawn(move || {
675
            std::thread::sleep(crate::update::PROVE_WITHIN);
676
            if crate::update::pending(&root).is_some_and(|now| now.to == p.to && now.at_ms == p.at_ms) {
677
                let why = format!("did not reach the hub within {} minutes", crate::update::PROVE_WITHIN.as_secs() / 60);
678
                if crate::update::roll_back(&root, &p, &why).is_ok() {
679
                    std::process::exit(1);
680
                }
681
            }
682
        });
683
    }
684
    let served = serve_until_drained(home);
685
    crate::update::forget_runner(&home.root);
686
    served
5cf1e8ae Andy McCall ↶ 687
}
688
85275e1e Andy McCall ↶ 689
/// Set by SIGUSR1: a new build is in place; leave for it.
690
static UPDATE: std::sync::atomic::AtomicBool = std::sync::atomic::AtomicBool::new(false);
691
3554afef Andy McCall ↶ 692
#[cfg(unix)]
85275e1e Andy McCall ↶ 693
extern "C" fn on_update(_: libc::c_int) {
694
    UPDATE.store(true, std::sync::atomic::Ordering::SeqCst);
695
}
696
5cf1e8ae Andy McCall ↶ 697
/// The runner as a library, as the Android app runs it (oxygenhub-android#8):
698
/// no signal handlers; [`stop`] drains it, and this returns once the jobs
699
/// it holds are done and reported. It may be started again afterwards.
700
pub fn serve_embedded(home: &Home) -> Result<()> {
701
    STOPS.store(0, std::sync::atomic::Ordering::SeqCst);
06a4c472 Andy McCall ↶ 702
    *LEAVING.lock().unwrap_or_else(|p| p.into_inner()) = None;
5cf1e8ae Andy McCall ↶ 703
    serve_until_drained(home)
06a4c472 Andy McCall ↶ 704
}
705
1b160d37 Andy McCall ↶ 706
/// A departure ([`leave_within`]): why, and when the runner must be gone.
707
#[derive(Clone, Debug)]
708
struct Leave {
709
    reason: String,
710
    by: Instant,
711
}
712
713
/// Set once the runner is leaving: a job whose expected rest fits before
714
/// the deadline finishes; the others are stopped and handed back
715
/// (oxygenflow#76).
716
static LEAVING: std::sync::Mutex<Option<Leave>> = std::sync::Mutex::new(None);
06a4c472 Andy McCall ↶ 717
718
fn leaving() -> Option<String> {
1b160d37 Andy McCall ↶ 719
    LEAVING.lock().unwrap_or_else(|p| p.into_inner()).as_ref().map(|l| l.reason.clone())
06a4c472 Andy McCall ↶ 720
}
721
722
/// Ask a runner started with [`serve_embedded`] to leave now: claim nothing
723
/// more, stop the jobs it holds and hand them back to the hub, which queues
724
/// them ahead of new work for another runner (oxygenflow#76), then return.
725
/// For a departure with no time to finish: a phone off the allowed network.
726
pub fn leave(reason: &str) {
1b160d37 Andy McCall ↶ 727
    leave_within(reason, Duration::ZERO);
728
}
729
730
/// [`leave`], with `notice`: a job the hub expects to finish within it
731
/// (the claim's `expected_ms`, less the time it has run) finishes and is
732
/// reported; the others, and any still running when the notice is up, are
733
/// handed back. For a phone taken off charge, a spot reclaim.
734
pub fn leave_within(reason: &str, notice: Duration) {
735
    *LEAVING.lock().unwrap_or_else(|p| p.into_inner()) = Some(Leave { reason: reason.to_string(), by: Instant::now() + notice });
06a4c472 Andy McCall ↶ 736
    stop();
5cf1e8ae Andy McCall ↶ 737
}
738
739
/// Ask a runner started with [`serve_embedded`] to drain: claim nothing
740
/// more, finish and report what it holds, then return (oxygenflow#76's
741
/// drain, for a phone taken off charge or off the allowed network).
742
pub fn stop() {
743
    STOPS.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
744
}
745
746
fn serve_until_drained(home: &Home) -> Result<()> {
e0918b88 Andy McCall 747
    let cfg = home.config.runner.clone().context("flow.toml has no [runner] section")?;
c5b702e5 Andy McCall ↶ 748
    // A runner offers only what it can run: a machine without KVM claims
749
    // nothing rather than taking MicroVM jobs and failing them (#36).
ea56c56e Andy McCall ↶ 750
    let (runtimes, missing) = tiers(home).map_err(|why| anyhow::anyhow!("flow runner: {why}"))?;
751
    for why in &missing {
752
        eprintln!("flow runner: not offering {why}");
753
    }
e0918b88 Andy McCall 754
    let hub = Hub::new(&cfg)?;
755
    let name = cfg.name.clone().unwrap_or_else(|| "flow".into());
bb643c25 Andy McCall ↶ 756
    let budget = Budget::of(&cfg);
47c1d64d Andy McCall ↶ 757
    let key = home.runner_key(&name)?;
b14aeaae Andy McCall ↶ 758
    // The swarm (docs/jetstream.md §8.1): a runner that cannot serve its
759
    // peers still runs jobs, fetching images from the hub alone.
760
    let swarm = crate::swarm::Swarm::start(home, &cfg).unwrap_or_else(|e| {
761
        eprintln!("flow runner: not joining the swarm: {e:#}");
762
        None
763
    });
47c1d64d Andy McCall ↶ 764
    eprintln!(
c5b702e5 Andy McCall ↶ 765
        "flow runner {name}: polling {} (up to {} jobs at once, tiers {}), signing receipts as {}",
47c1d64d Andy McCall ↶ 766
        cfg.hub,
767
        budget.max_jobs,
c5b702e5 Andy McCall ↶ 768
        runtimes.join(", "),
47c1d64d Andy McCall ↶ 769
        key.identity.hash().to_hex()
770
    );
4be5e4cd Andy McCall ↶ 771
    if let Some(addr) = &cfg.metrics {
772
        crate::metrics::serve(addr, cfg.metrics_token.clone(), name.clone());
773
    }
bb643c25 Andy McCall ↶ 774
    let used = std::sync::Mutex::new(Used::default());
e0918b88 Andy McCall 775
    let mut idle = Duration::from_secs(2);
3007d8f0 Andy McCall ↶ 776
    let pending = report_pending(home);
1daca0b6 Andy McCall ↶ 777
    // The jobs held now, told to the hub at each claim so it can end the
778
    // ones this runner lost: a job is held from its claim until its result
779
    // is reported.
780
    let held = std::sync::Mutex::new(std::collections::BTreeSet::<i64>::new());
781
    held.lock().unwrap().extend(pending.iter().map(|(job, _)| *job));
3007d8f0 Andy McCall ↶ 782
    std::thread::scope(|scope| {
783
        for (job, f) in pending {
1daca0b6 Andy McCall ↶ 784
            let (hub, held) = (&hub, &held);
3007d8f0 Andy McCall ↶ 785
            scope.spawn(move || {
786
                eprintln!("job {job}: reporting the result kept from before the restart");
787
                report_finish(home, hub, job, &f);
1daca0b6 Andy McCall ↶ 788
                held.lock().unwrap().remove(&job);
3007d8f0 Andy McCall ↶ 789
            });
790
        }
62829d15 Andy McCall ↶ 791
        // The speed (spec §11.1): measured now, and again every 15 minutes
792
        // while no job runs, so a job never shares the CPU with it.
793
        let measure = || {
794
            let (speed, ms) = crate::facts::bench();
795
            eprintln!("flow runner {name}: speed {speed} (class {}), benchmark {ms} ms", flow_core::api::Facts::class_of(speed));
796
            speed
797
        };
798
        let mut speed = measure();
799
        let mut benched = Instant::now();
89171fca Andy McCall ↶ 800
        let mut said_draining = false;
a4b3f8cb Andy McCall ↶ 801
        // When the hub last heard from us (oxygenhub#243): a claim, or a
802
        // heartbeat while there is no room to claim.
803
        let mut heard = Instant::now();
804
        let beat = |heard: &mut Instant| {
805
            if heard.elapsed() >= HUB_HEARTBEAT {
806
                if let Err(e) = hub.heartbeat() {
807
                    eprintln!("flow runner {name}: heartbeat: {e:#}");
808
                }
809
                *heard = Instant::now();
810
            }
811
        };
85275e1e Andy McCall ↶ 812
        let mut on_trial = crate::update::pending(&home.root).is_some();
cedd1234 Andy McCall ↶ 813
        // Runner updates (oxygenhub#322): asked of the hub once it has
814
        // answered a claim, then every 15 minutes.
815
        let mut update_asked: Option<Instant> = None;
816
        let mut update_said = String::new();
3007d8f0 Andy McCall ↶ 817
        loop {
85275e1e Andy McCall ↶ 818
            if UPDATE.swap(false, std::sync::atomic::Ordering::SeqCst) {
819
                let p = crate::update::pending(&home.root);
820
                let notice = Duration::from_secs(p.as_ref().map_or(0, |p| p.notice_secs));
821
                let to = p.map_or_else(|| "a new build".to_string(), |p| format!("flow {}", p.to));
822
                eprintln!("flow runner {name}: updating to {to}: leaving within {} s", notice.as_secs());
823
                leave_within(&format!("updating to {to}"), notice);
824
            }
89171fca Andy McCall ↶ 825
            // Draining (oxygenflow#50): told to stop, claim nothing more and
826
            // stop once every job held is done and reported.
827
            if stopping() {
828
                let (jobs, holding) = (used.lock().unwrap().jobs, held.lock().unwrap().len());
829
                if jobs == 0 && holding == 0 {
830
                    eprintln!("flow runner {name}: drained, stopping");
831
                    break;
832
                }
833
                if !said_draining {
834
                    eprintln!("flow runner {name}: draining: claiming nothing more, stopping when the {jobs} job(s) held are done (stop again to stop now)");
835
                    said_draining = true;
836
                }
a4b3f8cb Andy McCall ↶ 837
                beat(&mut heard);
89171fca Andy McCall ↶ 838
                std::thread::sleep(Duration::from_millis(500));
839
                continue;
840
            }
62829d15 Andy McCall ↶ 841
            if used.lock().unwrap().jobs == 0 && benched.elapsed() >= Duration::from_secs(15 * 60) {
842
                speed = measure();
843
                benched = Instant::now();
844
            }
3007d8f0 Andy McCall ↶ 845
            let room = budget.room(*used.lock().unwrap());
846
            let Some((free_memory_mib, free_cpus)) = room else {
a4b3f8cb Andy McCall ↶ 847
                beat(&mut heard);
3007d8f0 Andy McCall ↶ 848
                std::thread::sleep(Duration::from_millis(500));
849
                continue;
850
            };
a4b3f8cb Andy McCall ↶ 851
            heard = Instant::now();
1e6b9fb1 Andy McCall ↶ 852
            let cache_times = crate::cache::Pool::new(&home.root, cfg.cache_gb).slot_times();
3007d8f0 Andy McCall ↶ 853
            let req = ClaimRequest {
854
                runner: name.clone(),
c5b702e5 Andy McCall ↶ 855
                runtimes: runtimes.clone(),
3007d8f0 Andy McCall ↶ 856
                free_memory_mib,
857
                free_cpus,
858
                info: Some(RunnerInfo {
859
                    version: env!("OXY_VERSION_LONG").into(),
ea81796b Andy McCall ↶ 860
                    ..RunnerInfo::here(cfg.offered_labels(), budget.max_jobs as u32, cfg.memory_mib, cfg.cpus)
3007d8f0 Andy McCall ↶ 861
                }),
8c0a4e3a Andy McCall ↶ 862
                // Read each time: an image imported while the runner runs counts.
863
                images: Some(home.images()),
1abadd4a Andy McCall ↶ 864
                image_archs: Some(home.image_archs()),
1daca0b6 Andy McCall ↶ 865
                holding: Some(held.lock().unwrap().iter().copied().collect()),
4cf0c99a Andy McCall ↶ 866
                fetches_images: true,
62829d15 Andy McCall ↶ 867
                facts: Some(crate::facts::now(home, Some(speed))),
b14aeaae Andy McCall ↶ 868
                swarm: swarm.as_ref().map(|s| s.addr.clone()),
1e6b9fb1 Andy McCall ↶ 869
                caches: Some(cache_times.keys().cloned().collect()),
870
                cache_times: Some(cache_times),
3007d8f0 Andy McCall ↶ 871
            };
85275e1e Andy McCall ↶ 872
            let answer = hub.claim(&req);
873
            if on_trial && answer.is_ok() {
874
                crate::update::proven(&home.root);
875
                on_trial = false;
cedd1234 Andy McCall ↶ 876
            }
877
            if answer.is_ok() && update_asked.is_none_or(|t| t.elapsed() >= UPDATE_EVERY) {
878
                update_asked = Some(Instant::now());
879
                if let Some(note) = crate::update::take_note(&home.root) {
880
                    let _ = hub.report_update(&note);
881
                }
882
                consider_update(home, &hub, &mut update_said);
85275e1e Andy McCall ↶ 883
            }
884
            match answer {
3007d8f0 Andy McCall ↶ 885
                Ok(Some(claim)) => {
886
                    idle = Duration::from_secs(2);
4be5e4cd Andy McCall ↶ 887
                    crate::metrics::CLAIMS.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
3007d8f0 Andy McCall ↶ 888
                    let (mem, cpus) = (claim.sandbox.memory_mib, claim.sandbox.cpus);
889
                    {
890
                        let mut u = used.lock().unwrap();
891
                        u.jobs += 1;
892
                        u.memory_mib += mem;
893
                        u.cpus += cpus;
bb643c25 Andy McCall ↶ 894
                    }
1daca0b6 Andy McCall ↶ 895
                    held.lock().unwrap().insert(claim.job_id);
b14aeaae Andy McCall ↶ 896
                    let (home, hub, cfg, used, key, held, swarm) = (home, &hub, &cfg, &used, &key, &held, swarm.as_ref());
3007d8f0 Andy McCall ↶ 897
                    scope.spawn(move || {
898
                        let id = claim.job_id;
b14aeaae Andy McCall ↶ 899
                        fetch_image(home, hub, cfg, swarm, &claim);
b11e30f5 Andy McCall ↶ 900
                        if let Some(why) = preflight(home, &claim) {
901
                            eprintln!("job {id}: declined: {why}");
4be5e4cd Andy McCall ↶ 902
                            crate::metrics::DECLINED.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
b11e30f5 Andy McCall ↶ 903
                            if let Err(e) = hub.decline(id, &why) {
904
                                eprintln!("job {id}: decline: {e:#}");
905
                            }
906
                            held.lock().unwrap().remove(&id);
907
                            let mut u = used.lock().unwrap();
908
                            u.jobs -= 1;
909
                            u.memory_mib -= mem;
910
                            u.cpus -= cpus;
911
                            return;
912
                        }
3007d8f0 Andy McCall ↶ 913
                        eprintln!(
914
                            "job {id}: {} {} on {}",
915
                            claim.repo,
916
                            claim.job.name,
917
                            claim.heads.first().map(|h| &h[..12]).unwrap_or("-")
918
                        );
4be5e4cd Andy McCall ↶ 919
                        crate::metrics::RUNNING.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
3007d8f0 Andy McCall ↶ 920
                        let finish = run_claim(home, hub, cfg, key, &claim).unwrap_or_else(|e| Finish {
921
                            exit_code: None,
922
                            duration_ms: 0,
923
                            receipt: None,
924
                            signed_receipt: None,
925
                            error: Some(format!("{e:#}")),
8f211e20 Andy McCall ↶ 926
                            peak_disk_mib: None,
927
                            peak_memory_mib: None,
3007d8f0 Andy McCall ↶ 928
                        });
929
                        eprintln!(
930
                            "job {id}: exit {:?}{}",
931
                            finish.exit_code,
932
                            finish.error.as_deref().map(|e| format!(" ({e})")).unwrap_or_default()
933
                        );
4be5e4cd Andy McCall ↶ 934
                        crate::metrics::RUNNING.fetch_sub(1, std::sync::atomic::Ordering::Relaxed);
935
                        let cancelled = finish.error.as_deref() == Some(CANCELLED);
936
                        crate::metrics::finished(finish.exit_code, finish.error.as_deref(), finish.duration_ms, cancelled);
06a4c472 Andy McCall ↶ 937
                        match leaving() {
938
                            // Stopped as the runner leaves: another runner takes it.
939
                            Some(why) if cancelled => {
940
                                eprintln!("job {id}: handed back: {why}");
941
                                if let Err(e) = hub.release(id, &why) {
942
                                    eprintln!("job {id}: handing back: {e:#}; the hub requeues it once this runner's lease lapses");
943
                                }
944
                            }
945
                            _ if !cancelled => report_finish(home, hub, id, &finish),
946
                            _ => {}
280ef951 Andy McCall ↶ 947
                        }
1daca0b6 Andy McCall ↶ 948
                        held.lock().unwrap().remove(&id);
3007d8f0 Andy McCall ↶ 949
                        let mut u = used.lock().unwrap();
950
                        u.jobs -= 1;
951
                        u.memory_mib -= mem;
952
                        u.cpus -= cpus;
953
                    });
954
                }
955
                Ok(None) => std::thread::sleep(idle),
956
                Err(e) => {
957
                    eprintln!("claim: {e:#}");
958
                    std::thread::sleep(idle);
959
                    idle = (idle * 2).min(Duration::from_secs(60));
960
                }
e0918b88 Andy McCall 961
            }
962
        }
89171fca Andy McCall ↶ 963
        Ok(())
bb643c25 Andy McCall ↶ 964
    })
f201e1bc Andy McCall ↶ 965
}
966
967
/// Pack and keep a handed-back job's checkpoints (oxygenflow#77); what
968
/// happened goes to its log, and a failure only means the next attempt
969
/// starts again.
970
#[allow(clippy::too_many_arguments)]
971
fn save_checkpoint(hub: &Hub, id: i64, job: &flow_core::Job, scratch: &Path, dir_tier: bool, wasm: bool, mut say: impl FnMut(String)) {
972
    if !dir_tier {
973
        say("checkpoint: only the WASM and syscall tiers save one yet; the next attempt starts again".into());
974
        return;
975
    }
976
    let dirs: Vec<(String, PathBuf)> = job
977
        .checkpoints
978
        .iter()
979
        .filter_map(|c| crate::checkpoint::host_dir(c, job, &scratch.join("tree"), scratch, wasm).map(|d| (c.clone(), d)))
980
        .collect();
981
    match crate::checkpoint::pack(&dirs).and_then(|(bytes, n)| Ok((hub.save_checkpoint(id, &bytes)?, n))) {
982
        Ok((blob, n)) => {
983
            say(format!("checkpoint {}: {} files, {} bytes, kept for the next attempt", &blob[..12.min(blob.len())], n.files, n.bytes))
984
        }
985
        Err(e) => say(format!("checkpoint: not kept ({e:#}); the next attempt starts again")),
986
    }
89171fca Andy McCall ↶ 987
}
988
989
/// How many times the runner was told to stop (SIGTERM or SIGINT).
990
static STOPS: std::sync::atomic::AtomicUsize = std::sync::atomic::AtomicUsize::new(0);
991
992
extern "C" fn on_stop(_: libc::c_int) {
993
    // A second stop means now: only async-signal-safe calls here.
994
    if STOPS.fetch_add(1, std::sync::atomic::Ordering::SeqCst) >= 1 {
995
        unsafe { libc::_exit(143) };
996
    }
997
}
998
999
/// Drain instead of dying on the first SIGTERM or SIGINT (oxygenflow#50),
1000
/// so restarting a runner never kills the job it runs: the unit's
1001
/// `KillMode=mixed` sends the signal to the runner alone, not its microVMs.
1002
fn drain_on_stop() {
1003
    let handler = on_stop as extern "C" fn(libc::c_int) as libc::sighandler_t;
1004
    unsafe {
1005
        libc::signal(libc::SIGTERM, handler);
1006
        libc::signal(libc::SIGINT, handler);
1007
    }
e0918b88 Andy McCall 1008
}
1009
89171fca Andy McCall ↶ 1010
fn stopping() -> bool {
1011
    STOPS.load(std::sync::atomic::Ordering::SeqCst) > 0
1012
}
1013
280ef951 Andy McCall ↶ 1014
/// Batches records to the hub every second (or every 200 lines). A quiet
1015
/// job's sink still asks the hub every few seconds whether the job runs,
1016
/// with an empty batch; once the hub answers 409 the job is not wanted.
e0918b88 Andy McCall 1017
struct LogSink<'a> {
1018
    hub: &'a Hub,
1019
    job: i64,
1020
    buf: Vec<Record>,
1021
    last: Instant,
280ef951 Andy McCall ↶ 1022
    gone: bool,
06a4c472 Andy McCall ↶ 1023
    /// Set when the runner leaves ([`LEAVING`]): the job stops, to be
1b160d37 Andy McCall ↶ 1024
    /// handed back, unless its expected rest fits the notice.
1025
    leaving: &'a std::sync::Mutex<Option<Leave>>,
1026
    /// When the job started, and how long the hub expects it to take.
1027
    started: Instant,
1028
    expected: Option<Duration>,
280ef951 Andy McCall ↶ 1029
}
89171fca Andy McCall ↶ 1030
1031
/// The most log lines sent at once, as a busy job's sink flushes them.
1032
const BATCH: usize = 200;
280ef951 Andy McCall ↶ 1033
89171fca Andy McCall ↶ 1034
/// The most log lines kept while the hub cannot be reached.
1035
const MAX_UNSENT: usize = 50_000;
1036
280ef951 Andy McCall ↶ 1037
/// How often a quiet job asks whether it was cancelled.
1038
const HEARTBEAT: Duration = Duration::from_secs(5);
1039
1040
impl microvm::Sink for LogSink<'_> {
1041
    fn record(&mut self, r: Record) {
1042
        self.push(r)
1043
    }
1044
1045
    fn cancelled(&mut self) -> bool {
1b160d37 Andy McCall ↶ 1046
        let leave = self.leaving.lock().unwrap_or_else(|p| p.into_inner()).clone();
1047
        if let Some(l) = leave {
1048
            // Finished if what is left of it fits the notice; else handed
1049
            // back as the runner leaves, what the job said so far first.
1050
            let now = Instant::now();
1051
            let fits = self.expected.is_some_and(|e| now + e.saturating_sub(self.started.elapsed()) <= l.by);
1052
            if !fits || now >= l.by {
1053
                self.flush();
1054
                return true;
1055
            }
06a4c472 Andy McCall ↶ 1056
        }
280ef951 Andy McCall ↶ 1057
        if !self.gone && self.last.elapsed() >= HEARTBEAT {
1058
            self.send();
1059
        }
1060
        self.gone
1061
    }
e0918b88 Andy McCall 1062
}
1063
1064
impl LogSink<'_> {
1065
    fn push(&mut self, mut r: Record) {
280ef951 Andy McCall ↶ 1066
        if self.gone {
1067
            return;
1068
        }
e0918b88 Andy McCall 1069
        if r.text.len() > MAX_RECORD_TEXT {
1070
            let mut cut = MAX_RECORD_TEXT;
1071
            while !r.text.is_char_boundary(cut) {
1072
                cut -= 1;
1073
            }
1074
            r.text.truncate(cut);
1075
        }
1076
        self.buf.push(r);
89171fca Andy McCall ↶ 1077
        if self.buf.len() >= BATCH || self.last.elapsed() >= Duration::from_secs(1) {
e0918b88 Andy McCall 1078
            self.flush();
1079
        }
1080
    }
1081
1082
    fn flush(&mut self) {
1083
        if !self.buf.is_empty() {
280ef951 Andy McCall ↶ 1084
            self.send();
1085
        }
1086
        self.last = Instant::now();
1087
    }
1088
89171fca Andy McCall ↶ 1089
    /// Send what is buffered, even nothing, in batches of at most
1090
    /// `BATCH` lines.
280ef951 Andy McCall ↶ 1091
    fn send(&mut self) {
89171fca Andy McCall ↶ 1092
        loop {
1093
            if self.gone {
1094
                return;
1095
            }
1096
            let n = self.buf.len().min(BATCH);
1097
            let batch: Vec<Record> = self.buf.drain(..n).collect();
1098
            match self.hub.log(self.job, batch.clone()) {
1099
                Ok(()) => {}
1100
                Err(e) if job_gone(&e) => self.gone = true,
1101
                Err(e) => {
1102
                    // The hub is away (a restart, say): keep the lines for
1103
                    // the next batch, up to a bound, so the log loses none
1104
                    // (oxygenflow#49).
1105
                    eprintln!("job {}: log: {e:#}; kept for the next batch", self.job);
1106
                    let mut kept = batch;
1107
                    kept.append(&mut self.buf);
1108
                    let over = kept.len().saturating_sub(MAX_UNSENT);
1109
                    if over > 0 {
1110
                        kept.drain(..over);
1111
                        kept.insert(
1112
                            0,
1113
                            Record {
1114
                                t: 0,
1115
                                stream: "flow".into(),
1116
                                text: format!("({over} log lines lost while the hub could not be reached)"),
1117
                            },
1118
                        );
1119
                    }
1120
                    self.buf = kept;
1121
                    break;
1122
                }
1123
            }
1124
            if self.buf.is_empty() {
1125
                break;
1126
            }
e0918b88 Andy McCall 1127
        }
1128
        self.last = Instant::now();
1129
    }
1130
}
a4b3f8cb Andy McCall ↶ 1131
1132
/// How often a runner with no room to claim tells the hub it is there:
1133
/// well inside the hub's 90 s before it shows a runner offline.
1134
const HUB_HEARTBEAT: Duration = Duration::from_secs(30);
cedd1234 Andy McCall ↶ 1135
1136
/// How often a runner asks the hub whether it should run another build.
1137
const UPDATE_EVERY: Duration = Duration::from_secs(15 * 60);
1138
1139
/// Notice a runner gives itself to leave for an update the hub offered.
1140
const UPDATE_NOTICE: Duration = Duration::from_secs(600);
bb643c25 Andy McCall ↶ 1141
cedd1234 Andy McCall ↶ 1142
/// Ask the hub for this runner's update, and act on it: download the build
1143
/// it offers, check it against the hub's BLAKE3 and as a runner build, and
1144
/// install it, or for a dry run say what would happen. What it did goes to
1145
/// the hub once per outcome (`said`).
1146
fn consider_update(home: &Home, hub: &Hub, said: &mut String) {
1147
    let answer = match hub.update_answer() {
1148
        Ok(Some(a)) => a,
1149
        Ok(None) => return,
1150
        Err(e) => {
1151
            eprintln!("flow runner: asking for an update: {e:#}");
1152
            return;
1153
        }
1154
    };
1155
    let Some(offer) = answer.offer else { return };
1156
    let mut say = |note: String| {
1157
        if *said != note {
1158
            eprintln!("flow runner: {note}");
1159
            if let Err(e) = hub.report_update(&note) {
1160
                eprintln!("flow runner: telling the hub: {e:#}");
1161
            }
1162
            *said = note;
1163
        }
1164
    };
1165
    if home.config.runner.as_ref().and_then(|r| r.update.as_ref()).is_some_and(|u| !u.allow) {
1166
        return say(format!("refuses flow {}: flow.toml says [runner.update] allow = false", offer.version));
1167
    }
1168
    let dir = home.root.join("updates");
1169
    let file = dir.join(format!("flow-{}", offer.version));
1170
    let fresh = std::fs::read(&file).is_ok_and(|b| blake3::hash(&b).to_hex().as_str() == offer.blake3);
1171
    if !fresh {
1172
        if let Err(e) = std::fs::create_dir_all(&dir).map_err(anyhow::Error::from).and_then(|_| hub.download_build(&offer, &file)) {
1173
            return say(format!("flow {}: not fetched: {e:#}", offer.version));
1174
        }
1175
    }
1176
    let exe = match std::env::current_exe() {
1177
        Ok(e) => e,
1178
        Err(e) => return say(format!("flow {}: cannot find its own binary: {e}", offer.version)),
1179
    };
1180
    match crate::update::install(home, &exe, &file, UPDATE_NOTICE, answer.policy.dry_run, offer.older) {
1181
        Ok(done) if answer.policy.dry_run => say(format!("dry run: flow {} checks out here and would replace flow {}", done.to, done.from)),
1182
        Ok(done) => {
1183
            let _ = std::fs::remove_file(&file);
1184
            say(format!("installed flow {} (was {}); leaving within {} s to start it", done.to, done.from, UPDATE_NOTICE.as_secs()))
1185
        }
1186
        Err(e) => say(format!("flow {}: not installed: {e:#}", offer.version)),
1187
    }
1188
}
1189
ed4aef8b Andy McCall ↶ 1190
/// Held while a job takes its cache disks and images its workspace (see
1191
/// `run_claim`): one job at a time on the runner's disk and cache pool.
bb643c25 Andy McCall ↶ 1192
static PREPARE: std::sync::Mutex<()> = std::sync::Mutex::new(());
e0918b88 Andy McCall 1193
ed4aef8b Andy McCall ↶ 1194
/// Held while the first clone of any repository passes its token through
1195
/// the process environment.
1196
static CLONE: std::sync::Mutex<()> = std::sync::Mutex::new(());
1197
1198
/// One lock per cached repository: fetches into it, and exports from it,
1199
/// never overlap, while jobs of other repositories go on. A slow repository
1200
/// (one kept in a bucket, say) once held every job on the runner waiting
1201
/// behind a single global lock.
1202
fn repo_lock(name: &str) -> std::sync::Arc<std::sync::Mutex<()>> {
1203
    static LOCKS: std::sync::OnceLock<std::sync::Mutex<std::collections::HashMap<String, std::sync::Arc<std::sync::Mutex<()>>>>> =
1204
        std::sync::OnceLock::new();
1205
    let mut locks = LOCKS.get_or_init(Default::default).lock().unwrap_or_else(|p| p.into_inner());
1206
    locks.entry(name.to_string()).or_default().clone()
1207
}
1208
1209
/// Seconds, to one decimal place, for the job's log.
1210
fn secs(d: std::time::Duration) -> String {
1211
    format!("{:.1} s", d.as_secs_f64())
1212
}
1213
47c1d64d Andy McCall ↶ 1214
fn run_claim(home: &Home, hub: &Hub, cfg: &RunnerConfig, key: &RunnerKey, claim: &Claim) -> Result<Finish> {
e0918b88 Andy McCall 1215
    let started = Instant::now();
1b160d37 Andy McCall ↶ 1216
    let mut sink = LogSink {
1217
        hub,
1218
        job: claim.job_id,
1219
        buf: Vec::new(),
1220
        last: Instant::now(),
1221
        gone: false,
1222
        leaving: &LEAVING,
1223
        started: Instant::now(),
1224
        expected: claim.expected_ms.map(Duration::from_millis),
1225
    };
e0918b88 Andy McCall 1226
    let note =
1227
        |sink: &mut LogSink, text: String| sink.push(Record { t: started.elapsed().as_millis() as u64, stream: "flow".into(), text });
1228
    let heads: Vec<ChangeHash> = claim.heads.iter().map(|h| ChangeHash::from_hex(h).context("bad head")).collect::<Result<_>>()?;
1229
    note(&mut sink, format!("fetching {} for state {}", claim.repo, claim.heads.iter().map(|h| &h[..12]).collect::<Vec<_>>().join("+")));
ed4aef8b Andy McCall ↶ 1230
    // Preparing a workspace: fetches into a cached repository, and exports
1231
    // from it, are one job at a time per repository (`repo_lock`); taking
1232
    // cache disks and imaging the workspace are one job at a time on the
1233
    // runner (`PREPARE`); a first clone, whose token goes through the process
1234
    // environment, one at a time (`CLONE`).
1235
    let main_lock = repo_lock(&claim.repo);
1236
    let asked = Instant::now();
1237
    let mut main_guard = Some(main_lock.lock().unwrap_or_else(|p| p.into_inner()));
1238
    if asked.elapsed().as_secs() >= 2 {
1239
        note(&mut sink, format!("waited {} for another job fetching {}", secs(asked.elapsed()), claim.repo));
1240
    }
1241
    let fetching = Instant::now();
e0918b88 Andy McCall 1242
    let repo = cached_repo(home, cfg, claim)?;
ed4aef8b Andy McCall ↶ 1243
    let fetched = fetching.elapsed();
1244
    let mut preparing = None;
e0918b88 Andy McCall 1245
    let scratch = home.root.join("work").join(format!("job-{}", claim.job_id));
1246
    let _ = std::fs::remove_dir_all(&scratch);
1247
    std::fs::create_dir_all(scratch.join("tree"))?;
1248
    let result = (|| -> Result<Finish> {
ed4aef8b Andy McCall ↶ 1249
        let exporting = Instant::now();
c23d5191 Andy McCall ↶ 1250
        let mut files = workspace::export_state(&repo, &heads, &scratch.join("tree"))?;
ed4aef8b Andy McCall ↶ 1251
        main_guard.take();
1252
        note(&mut sink, format!("fetched in {}, {} files exported in {}", secs(fetched), files.len(), secs(exporting.elapsed())));
c23d5191 Andy McCall ↶ 1253
        // Pinned subrepos (Oxygen spec §7.8), at their paths, as part of the
1254
        // workspace and of the scope hash.
1255
        for sub in &claim.subrepos {
1256
            if sub.path.split('/').any(|p| p.is_empty() || p == "." || p == ".." || p == ".oxy") {
1257
                bail!("subrepo path {:?} is not a clean relative path", sub.path);
1258
            }
1259
            note(
1260
                &mut sink,
1261
                format!(
1262
                    "fetching subrepo {} at {} for {}",
1263
                    sub.repo,
1264
                    sub.heads.iter().map(|h| &h[..12]).collect::<Vec<_>>().join("+"),
1265
                    sub.path
1266
                ),
1267
            );
ed4aef8b Andy McCall ↶ 1268
            let sub_lock = repo_lock(&sub.repo);
1269
            let asked = Instant::now();
1270
            let sub_guard = sub_lock.lock().unwrap_or_else(|p| p.into_inner());
1271
            if asked.elapsed().as_secs() >= 2 {
1272
                note(&mut sink, format!("waited {} for another job fetching {}", secs(asked.elapsed()), sub.repo));
1273
            }
c23d5191 Andy McCall ↶ 1274
            let sub_repo = cached_repo_at(home, cfg, &sub.repo, &sub.clone_url, &claim.token)?;
1275
            let sub_heads: Vec<ChangeHash> =
1276
                sub.heads.iter().map(|h| ChangeHash::from_hex(h).context("bad subrepo head")).collect::<Result<_>>()?;
1277
            let dir = scratch.join("tree").join(&sub.path);
1278
            std::fs::create_dir_all(&dir)?;
1279
            let sub_files = workspace::export_state(&sub_repo, &sub_heads, &dir)?;
ed4aef8b Andy McCall ↶ 1280
            drop(sub_guard);
c23d5191 Andy McCall ↶ 1281
            files.extend(sub_files.into_iter().map(|(p, h)| (format!("{}/{p}", sub.path), h)));
1282
        }
e0918b88 Andy McCall 1283
        let scope_hash = scope_tree_hash(&claim.job.scope, files.iter().map(|(p, h)| (p.as_str(), *h)));
46fe2c81 Andy McCall ↶ 1284
        // Inputs (oxygenflow#58): each producing job's artifact, checked
1285
        // against the blob its receipt names, then put at its path. Not part
1286
        // of the scope: they are the run's, not the repository's.
1287
        let mut input_blobs = Vec::new();
1288
        for (index, i) in claim.inputs.iter().enumerate() {
1289
            let what = format!("input {} from {}", i.artifact, i.from);
1290
            let bytes = hub.input(claim.job_id, index).with_context(|| format!("{what}: fetching it"))?;
1291
            let blob = place_input(&scratch.join("tree"), i, &bytes).with_context(|| what.clone())?;
1292
            note(&mut sink, format!("{what}: {} bytes at {}, blob {} as its receipt names", bytes.len(), i.path, &i.blob[..12]));
1293
            input_blobs.push(blob);
1294
        }
ea56c56e Andy McCall ↶ 1295
        // The syscall tier runs on the tree itself; a microVM gets an image.
1296
        let syscall_tier = claim.sandbox.runtime == flow_core::Runtime::Syscall;
28cc603f Andy McCall ↶ 1297
        let wasm_tier = claim.sandbox.runtime == flow_core::Runtime::Wasm;
1298
        // The syscall and WASM tiers see the job's files as a directory;
1299
        // the MicroVM tier as an ext4 image.
1300
        let dir_tier = syscall_tier || wasm_tier;
82cf6e91 Andy McCall ↶ 1301
        let job = claim.job.in_sandbox(&claim.sandbox).with_flow_env(&claim.flow_env());
2594278f Andy McCall ↶ 1302
        // Disk mounts that name a cache start from a kept disk (#28): a
1303
        // microVM takes an image already in place as it is, the syscall tier
1304
        // a directory (#70). Taken before the workspace is imaged: the
1305
        // sources the disk was built from give unchanged files their old
1306
        // times back (#52).
ed4aef8b Andy McCall ↶ 1307
        let asked = Instant::now();
1308
        preparing = Some(PREPARE.lock().unwrap_or_else(|p| p.into_inner()));
1309
        if asked.elapsed().as_secs() >= 2 {
1310
            note(&mut sink, format!("waited {} for another job imaging its workspace", secs(asked.elapsed())));
1311
        }
82cf6e91 Andy McCall ↶ 1312
        let pool = crate::cache::Pool::new(&home.root, cfg.cache_gb);
28cc603f Andy McCall ↶ 1313
        let form = if dir_tier { crate::cache::Form::Dir } else { crate::cache::Form::Image };
82cf6e91 Andy McCall ↶ 1314
        let cache_ctx =
2594278f Andy McCall ↶ 1315
            crate::cache::Ctx { repo: &claim.repo, bookmark: &claim.bookmark, job: &claim.job.name, image: &claim.sandbox.rootfs, form };
82cf6e91 Andy McCall ↶ 1316
        let mut cached: Vec<(usize, String, String)> = Vec::new();
1317
        let mut built_from: Option<crate::cache::Sources> = None;
2594278f Andy McCall ↶ 1318
        for (i, m) in job.mounts.iter().filter(|m| m.source == flow_core::pipeline::MountSource::Disk).enumerate() {
1319
            let Some(name) = &m.cache else { continue };
1320
            let key = crate::cache::Pool::key(&cache_ctx, m, |p| files.iter().find(|(f, _)| f == p).map(|(_, h)| hex::encode(h)));
1321
            let how = match pool.take(&cache_ctx, m, &key, &crate::cache::disk_path(form, &scratch, i)) {
1322
                Some(t) => {
1323
                    if built_from.is_none() {
1324
                        built_from = t.sources;
82cf6e91 Andy McCall ↶ 1325
                    }
2594278f Andy McCall ↶ 1326
                    t.how
1327
                }
1328
                None => format!("cache {name}: none kept yet, an empty disk"),
1329
            };
1330
            note(&mut sink, format!("{} ({})", how, m.target));
077a6524 Andy McCall ↶ 1331
            // What the job reads is seen when the disk is pruned (#72).
1332
            if let Err(e) = crate::cache::forget_reads(form, &crate::cache::disk_path(form, &scratch, i)) {
1333
                note(&mut sink, format!("cache {name}: {e:#}"));
1334
            }
2594278f Andy McCall ↶ 1335
            cached.push((i, key, how));
82cf6e91 Andy McCall ↶ 1336
        }
077a6524 Andy McCall ↶ 1337
        // What the job reads or writes on its cache from now on is what it
1338
        // used; the rest may be pruned before the cache is kept (#72).
1339
        let cache_started = if cached.is_empty() { std::time::SystemTime::now() } else { flow_init::prune::start_clock() };
82cf6e91 Andy McCall ↶ 1340
        let sources = if cached.is_empty() {
1341
            crate::cache::Sources::new()
1342
        } else {
1343
            let (sources, restored) = workspace::restore_times(&scratch.join("tree"), &files, built_from.as_ref());
1344
            if built_from.is_some() {
1345
                note(&mut sink, format!("cache: {restored} of {} files unchanged since the cached build, their times kept", sources.len()));
1346
            }
1347
            sources
1348
        };
f201e1bc Andy McCall ↶ 1349
        // An earlier attempt's checkpoint (oxygenflow#77): back in place,
1350
        // over the cache, before the same exec runs again; the receipt
1351
        // lists it among the job's inputs.
1352
        if let Some(blob) = &claim.checkpoint {
1353
            if dir_tier {
1354
                let bytes = hub.checkpoint(claim.job_id).context("fetching the checkpoint")?;
1355
                if Repo::blob_ref(&bytes).to_hex() != *blob {
1356
                    bail!("the checkpoint fetched is not blob {}, as the claim names", &blob[..12.min(blob.len())]);
1357
                }
1358
                let place = |root: &str| crate::checkpoint::host_dir(root, &job, &scratch.join("tree"), &scratch, wasm_tier);
1359
                let n = crate::checkpoint::unpack(&bytes, &job.checkpoints, place)?;
1360
                note(
1361
                    &mut sink,
1362
                    format!("restored checkpoint {}: {} files, {} bytes, from the attempt handed back", &blob[..12], n.files, n.bytes),
1363
                );
1364
                input_blobs.push(Repo::blob_ref(&bytes).0);
1365
            } else {
1366
                note(&mut sink, "a checkpoint was saved, but only the WASM and syscall tiers restore one yet: starting again".into());
1367
            }
1368
        }
28cc603f Andy McCall ↶ 1369
        if !dir_tier {
ea56c56e Andy McCall ↶ 1370
            workspace::image(&scratch.join("tree"), &scratch.join("workspace.ext4"))?;
1371
            std::fs::remove_dir_all(scratch.join("tree"))?;
1372
        }
bb643c25 Andy McCall ↶ 1373
        preparing.take();
ea56c56e Andy McCall ↶ 1374
        // Cancelled while it was being prepared: no sandbox.
280ef951 Andy McCall ↶ 1375
        if microvm::Sink::cancelled(&mut sink) {
1376
            return Ok(cancelled());
1377
        }
28cc603f Andy McCall ↶ 1378
        if wasm_tier {
1379
            note(&mut sink, format!("starting WASM (wasmtime {}): {} MiB", crate::wasm::WASMTIME, claim.sandbox.memory_mib));
1380
        } else {
1381
            note(
1382
                &mut sink,
1383
                format!(
1384
                    "{}: {} {}, {} MiB, rootfs {}",
1385
                    if syscall_tier { "starting a syscall sandbox" } else { "booting a microVM" },
1386
                    claim.sandbox.cpus,
1387
                    if syscall_tier { "CPU" } else { "vCPU" },
1388
                    claim.sandbox.memory_mib,
1389
                    &claim.sandbox.rootfs[..12]
1390
                ),
1391
            );
1392
        }
ea56c56e Andy McCall ↶ 1393
        let timeout = Duration::from_secs(claim.timeout_s);
28cc603f Andy McCall ↶ 1394
        let mut outcome = if wasm_tier {
1395
            crate::wasm::run(&claim.sandbox, &job, &scratch.join("tree"), &scratch, timeout, &mut sink)?
1396
        } else if syscall_tier {
ea56c56e Andy McCall ↶ 1397
            let group = format!("job-{}", claim.job_id);
1398
            crate::syscall::run(home, &claim.sandbox, &job, &claim.secrets, &scratch.join("tree"), &scratch, timeout, &mut sink, &group)?
1399
        } else {
1400
            microvm::run(home, &claim.sandbox, &job, &claim.secrets, &scratch.join("workspace.ext4"), &scratch, timeout, &mut sink)?
1401
        };
280ef951 Andy McCall ↶ 1402
        if outcome.cancelled {
f201e1bc Andy McCall ↶ 1403
            // Handed back as the runner leaves: its checkpoints first, for
1404
            // the next attempt (oxygenflow#77).
1405
            if leaving().is_some() && !job.checkpoints.is_empty() {
1406
                save_checkpoint(hub, claim.job_id, &job, &scratch, dir_tier, wasm_tier, |t| note(&mut sink, t));
1407
            }
280ef951 Andy McCall ↶ 1408
            return Ok(cancelled());
d60aac6c Andy McCall ↶ 1409
        }
1410
        // The receipt says what a cached disk started from.
1411
        let disk_mounts: Vec<&flow_core::pipeline::Mount> =
1412
            job.mounts.iter().filter(|m| m.source == flow_core::pipeline::MountSource::Disk).collect();
1413
        for (i, key, how) in &cached {
1414
            outcome.environment.parts.push((format!("cache {}", disk_mounts[*i].target), format!("{key}: {how}")));
280ef951 Andy McCall ↶ 1415
        }
a67b97e9 Andy McCall ↶ 1416
        // A passing job's artifacts, read from its disks and uploaded; a
9a6fac2b Andy McCall ↶ 1417
        // declared artifact that cannot be kept fails the job. A failing
1418
        // job keeps only its `when "always"` ones (oxygenflow#37), outside
1419
        // its receipt, and one it cannot keep is only noted.
a67b97e9 Andy McCall ↶ 1420
        let mut outputs = vec![];
1421
        let mut output_error = None;
9a6fac2b Andy McCall ↶ 1422
        let passed = outcome.exit_code == Some(0);
1423
        let wanted = claim.job.outputs.iter().filter(|o| passed || o.always);
1424
        for o in wanted {
6d53fb7d Andy McCall ↶ 1425
            if o.kind == flow_core::OutputKind::Report {
28cc603f Andy McCall ↶ 1426
                let read = if dir_tier {
6d53fb7d Andy McCall ↶ 1427
                    let (dir, inner) = crate::outputs::locate_dir(&claim.job, &scratch.join("tree"), &scratch, &o.path);
1428
                    crate::outputs::read_report_dir(&dir, &inner)
1429
                } else {
1430
                    let (image, inner) = crate::outputs::locate(&claim.job, &scratch.join("workspace.ext4"), &scratch, &o.path);
1431
                    crate::outputs::read_report(&image, &inner, &scratch)
1432
                };
1433
                let kept = read.and_then(|files| upload_report(hub, claim.job_id, &o.name, &files));
1434
                match kept {
1435
                    Ok((blob, files, bytes)) => {
1436
                        let unattested = if passed { "" } else { " (the job failed: not in its receipt)" };
1437
                        let hex = blob.to_hex();
1438
                        note(&mut sink, format!("report {}: {files} files, {bytes} bytes, manifest {}{unattested}", o.name, &hex[..12]));
1439
                        if passed {
1440
                            outputs.push(blob.0);
1441
                        }
1442
                    }
1443
                    Err(e) if passed => {
1444
                        output_error = Some(format!("report {}: {e:#}", o.name));
1445
                        break;
1446
                    }
1447
                    Err(e) => note(&mut sink, format!("report {}: not kept: {e:#}", o.name)),
1448
                }
1449
                continue;
1450
            }
28cc603f Andy McCall ↶ 1451
            let read = if dir_tier {
9a6fac2b Andy McCall ↶ 1452
                let (dir, inner) = crate::outputs::locate_dir(&claim.job, &scratch.join("tree"), &scratch, &o.path);
1453
                crate::outputs::read_dir_file(&dir, &inner)
1454
            } else {
1455
                let (image, inner) = crate::outputs::locate(&claim.job, &scratch.join("workspace.ext4"), &scratch, &o.path);
1456
                crate::outputs::read(&image, &inner, &scratch)
1457
            };
1458
            let kept = read.and_then(|bytes| {
1459
                let blob = Repo::blob_ref(&bytes);
1460
                let named = hub.artifact(claim.job_id, &o.name, &bytes)?;
1461
                if named != blob.to_hex() {
1462
                    bail!("the hub kept it as {named}, not {}", blob.to_hex());
1463
                }
1464
                let unattested = if passed { "" } else { " (the job failed: not in its receipt)" };
1465
                note(&mut sink, format!("artifact {}: {} bytes, blob {}{unattested}", o.name, bytes.len(), &named[..12]));
1466
                Ok(blob.0)
1467
            });
1468
            match kept {
1469
                Ok(b) if passed => outputs.push(b),
1470
                Ok(_) => {}
1471
                Err(e) if passed => {
1472
                    output_error = Some(format!("artifact {}: {e:#}", o.name));
1473
                    break;
a67b97e9 Andy McCall ↶ 1474
                }
9a6fac2b Andy McCall ↶ 1475
                Err(e) => note(&mut sink, format!("artifact {}: not kept: {e:#}", o.name)),
a67b97e9 Andy McCall ↶ 1476
            }
1477
        }
1478
        if let Some(e) = &output_error {
1479
            note(&mut sink, e.clone());
1480
        }
1481
        let receipt = outcome.exit_code.filter(|_| output_error.is_none()).map(|code| Receipt {
e0918b88 Andy McCall 1482
            job_name: claim.job.name.clone(),
1483
            pipeline_hash: hex32(&claim.pipeline_hash),
1484
            scope_tree_hash: scope_hash,
46fe2c81 Andy McCall ↶ 1485
            inputs: input_blobs.clone(),
e0918b88 Andy McCall 1486
            environment: outcome.environment.hash(),
a67b97e9 Andy McCall ↶ 1487
            outputs,
e0918b88 Andy McCall 1488
            exit_code: code as u32,
1489
            duration_ms: outcome.duration_ms,
1490
        });
47c1d64d Andy McCall ↶ 1491
        let signed_receipt = match &receipt {
35f7a849 Andy McCall ↶ 1492
            Some(r) => {
1493
                let mut signed = sign_receipt(&repo, &heads, cfg, key, claim, r)?;
1494
                if cfg.keyless {
1495
                    match sign_keyless(&repo, &heads, cfg, hub, claim, r) {
1496
                        Ok(k) => signed = k,
1497
                        Err(e) => {
1498
                            note(&mut sink, format!("receipt: signed with the runner's key; a short-lived key was not certified: {e:#}"))
1499
                        }
1500
                    }
1501
                }
1502
                Some(hex::encode(signed.encode()))
1503
            }
47c1d64d Andy McCall ↶ 1504
            None => None,
1505
        };
d60aac6c Andy McCall ↶ 1506
        // Measured before a cached disk leaves the scratch directory.
1507
        let peak_disk_mib = crate::facts::allocated_mib(&scratch);
077a6524 Andy McCall ↶ 1508
        // A passing job's cached disks are pruned and kept (#72); a
1509
        // failing job's are thrown away, so nothing it wrote reaches a
1510
        // later job.
2594278f Andy McCall ↶ 1511
        let kept = passed && output_error.is_none();
1512
        for (i, key, _) in &cached {
1513
            let disk = crate::cache::disk_path(form, &scratch, *i);
077a6524 Andy McCall ↶ 1514
            let m = disk_mounts[*i];
1515
            let prune = crate::cache::Prune {
1516
                started: cache_started,
1517
                percent: cfg.cache_prune_percent.unwrap_or(flow_init::prune::DEFAULT_PERCENT),
1518
                guest: outcome.pruned.iter().find(|(t, _)| *t == m.target).map(|(_, p)| p),
1519
            };
1520
            match pool.settle(&cache_ctx, m, key, &disk, &sources, kept, &prune) {
1521
                Ok(Some(line)) => note(&mut sink, line),
1522
                Ok(None) => {}
1523
                Err(e) => note(&mut sink, format!("cache: not {}: {e:#}", if kept { "kept" } else { "discarded" })),
d60aac6c Andy McCall ↶ 1524
            }
1525
        }
e0918b88 Andy McCall 1526
        Ok(Finish {
1527
            exit_code: outcome.exit_code,
1528
            duration_ms: started.elapsed().as_millis() as u64,
1529
            receipt,
47c1d64d Andy McCall ↶ 1530
            signed_receipt,
a67b97e9 Andy McCall ↶ 1531
            error: output_error.or_else(|| outcome.timed_out.then(|| "stopped at the time limit".to_string())),
8f211e20 Andy McCall ↶ 1532
            // What the job took (oxygenflow#41): its disks are sparse files
1533
            // in the scratch directory, so their blocks now are their peak.
d60aac6c Andy McCall ↶ 1534
            peak_disk_mib,
8f211e20 Andy McCall ↶ 1535
            peak_memory_mib: outcome.peak_memory_mib,
e0918b88 Andy McCall 1536
        })
1537
    })();
ea56c56e Andy McCall ↶ 1538
    let _ = workspace::remove_tree(&scratch);
e0918b88 Andy McCall 1539
    sink.flush();
1540
    result
47c1d64d Andy McCall ↶ 1541
}
1542
1543
/// Bind `receipt` to the state of `heads` in `repo` and sign it as the
1544
/// runner (Oxygen spec §14.6). The state is computed here, not taken from
1545
/// the claim, so the runner vouches only for what it fetched.
6d53fb7d Andy McCall ↶ 1546
/// Upload a report's files, then its manifest under the report's name, and
1547
/// check the hub kept each as the blob it should (oxygenhub#127). Answers
1548
/// the manifest's blob, the file count and their total size.
1549
fn upload_report(hub: &Hub, job: i64, name: &str, files: &[(String, Vec<u8>)]) -> Result<(oxy_core::hash::ObjectRef, usize, u64)> {
1550
    let mut listed = Vec::with_capacity(files.len());
1551
    let mut total = 0u64;
1552
    for (path, bytes) in files {
1553
        let blob = Repo::blob_ref(bytes).to_hex();
1554
        let named = hub.report_file(job, name, path, bytes).with_context(|| format!("uploading {path}"))?;
1555
        if named != blob {
1556
            bail!("the hub kept {path} as {named}, not {blob}");
1557
        }
1558
        total += bytes.len() as u64;
1559
        listed.push(serde_json::json!({ "path": path, "blob": blob, "size": bytes.len() }));
1560
    }
1561
    let manifest = serde_json::to_vec(&serde_json::json!({ "version": 1, "files": listed }))?;
1562
    let blob = Repo::blob_ref(&manifest);
1563
    let named = hub.artifact(job, name, &manifest)?;
1564
    if named != blob.to_hex() {
1565
        bail!("the hub kept the manifest as {named}, not {}", blob.to_hex());
1566
    }
1567
    Ok((blob, files.len(), total))
1568
}
1569
47c1d64d Andy McCall ↶ 1570
fn sign_receipt(
1571
    repo: &Repo,
1572
    heads: &[ChangeHash],
1573
    cfg: &RunnerConfig,
1574
    key: &RunnerKey,
1575
    claim: &Claim,
1576
    receipt: &Receipt,
1577
) -> Result<oxy_core::receipt::SignedReceipt> {
1578
    let state = state_of(heads);
1579
    let run = format!("{}/{}/flow/{}", cfg.hub.trim_end_matches('/'), claim.repo, claim.run_number);
1580
    Ok(oxy_core::receipt::SignedReceipt::sign(receipt.to_oxy(), repo.id()?, state, &run, &key.identity, &key.secret))
1581
}
1582
35f7a849 Andy McCall ↶ 1583
/// Sign `receipt` with a fresh key the hub certifies for this job now,
1584
/// just before signing, so the certificate covers the moment the hub
1585
/// accepts it (keyless, oxygenhub#241). The key is dropped with the
1586
/// receipt signed: nothing long-lived signs.
1587
fn sign_keyless(
1588
    repo: &Repo,
1589
    heads: &[ChangeHash],
1590
    cfg: &RunnerConfig,
1591
    hub: &Hub,
1592
    claim: &Claim,
1593
    receipt: &Receipt,
1594
) -> Result<oxy_core::receipt::SignedReceipt> {
1595
    let sk = oxy_core::identity::SecretKey::generate();
1596
    let now = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).map(|d| d.as_millis() as u64).unwrap_or(0);
1597
    let name = format!("{} job {}", cfg.name.as_deref().unwrap_or("runner"), claim.job_id);
1598
    let id = sk.identity(&name, "", now);
1599
    let proof = sk.sign_challenge(&id.hash(), &oxy_core::cert::certify_challenge(&claim.token)).map_err(|e| anyhow::anyhow!("{e}"))?;
1600
    let (cert, issuer) = hub.certify(&claim.token, &id, &proof)?;
1601
    let state = state_of(heads);
1602
    let run = format!("{}/{}/flow/{}", cfg.hub.trim_end_matches('/'), claim.repo, claim.run_number);
1603
    Ok(oxy_core::receipt::SignedReceipt::sign(receipt.to_oxy(), repo.id()?, state, &run, &id, &sk).with_certificate(cert, issuer))
1604
}
1605
47c1d64d Andy McCall ↶ 1606
/// The ObjectRef of the state whose heads are `heads` (Oxygen spec §10.1).
1607
pub fn state_of(heads: &[ChangeHash]) -> oxy_core::hash::ObjectRef {
1608
    use oxy_core::hash::{hash_object, ObjectType};
1609
    hash_object(ObjectType::State, &oxy_core::objects::StateObject::new(heads.to_vec()).encode())
e0918b88 Andy McCall 1610
}
1611
1612
fn hex32(s: &str) -> [u8; 32] {
1613
    let mut out = [0; 32];
1614
    if let Ok(v) = hex::decode(s) {
1615
        if v.len() == 32 {
1616
            out.copy_from_slice(&v);
1617
        }
1618
    }
1619
    out
1620
}
1621
f53e6d47 Andy McCall ↶ 1622
/// Whether a network error is worth another try: DNS, a refused or reset
1623
/// connection, a timeout, or a gateway answering 502, 503 or 504 while the
1624
/// hub restarts (oxygenhub#111).
1625
fn transient(msg: &str) -> bool {
1626
    let m = msg.to_lowercase();
1627
    [
1628
        "dns failed",
1629
        "name resolution",
1630
        "connection refused",
1631
        "connection reset",
1632
        "timed out",
1633
        "timeout",
1634
        "http 502",
1635
        "http 503",
1636
        "http 504",
1637
        "broken pipe",
1638
        "unexpected eof",
1639
    ]
1640
    .iter()
1641
    .any(|k| m.contains(k))
1642
}
1643
1644
/// Run `f` again after a transient network error, waiting `waits` between
1645
/// tries (about a minute in all), and give up on any other error at once.
1646
fn with_retries<T>(waits: &[Duration], mut f: impl FnMut() -> Result<T>) -> Result<T> {
1647
    let mut tries = waits.iter();
1648
    loop {
1649
        match f() {
1650
            Ok(v) => return Ok(v),
1651
            Err(e) => match tries.next() {
1652
                Some(w) if transient(&format!("{e:#}")) => std::thread::sleep(*w),
1653
                _ => return Err(e),
1654
            },
1655
        }
1656
    }
1657
}
1658
1659
/// The waits between tries of a fetch: 2, 4, 8, 16 and 30 seconds.
1660
const FETCH_WAITS: [Duration; 5] =
1661
    [Duration::from_secs(2), Duration::from_secs(4), Duration::from_secs(8), Duration::from_secs(16), Duration::from_secs(30)];
1662
e0918b88 Andy McCall 1663
/// The runner's copy of a repository, fetched up to date with the job's token.
1664
fn cached_repo(home: &Home, cfg: &RunnerConfig, claim: &Claim) -> Result<Repo> {
c23d5191 Andy McCall ↶ 1665
    cached_repo_at(home, cfg, &claim.repo, &claim.clone_url, &claim.token)
1666
}
1667
1668
/// The runner's copy of `repo`, cloned or fetched with `token`.
1669
fn cached_repo_at(home: &Home, cfg: &RunnerConfig, repo_name: &str, clone_url: &str, token: &str) -> Result<Repo> {
1670
    let dir: PathBuf = home.root.join("repos").join(repo_name.replace('/', "__"));
e0918b88 Andy McCall 1671
    let set_remote = |repo: &mut Repo| -> Result<()> {
c23d5191 Andy McCall ↶ 1672
        repo.config.set("remote.origin.url", clone_url).map_err(|e| anyhow::anyhow!("{e}"))?;
1673
        repo.config.set("remote.origin.token", token).map_err(|e| anyhow::anyhow!("{e}"))?;
e0918b88 Andy McCall 1674
        if let Some(ca) = &cfg.ca {
1675
            repo.config.set("remote.origin.ca", &ca.to_string_lossy()).map_err(|e| anyhow::anyhow!("{e}"))?;
1676
        }
1677
        repo.save_config().map_err(|e| anyhow::anyhow!("{e}"))
1678
    };
1679
    if dir.join(".oxy").is_dir() {
1680
        let mut repo = Repo::open(&dir).map_err(|e| anyhow::anyhow!("{e}"))?;
1681
        set_remote(&mut repo)?;
f53e6d47 Andy McCall ↶ 1682
        with_retries(&FETCH_WAITS, || repo.fetch("origin").map(|_| ()).map_err(|e| anyhow::anyhow!("fetch {clone_url}: {e}")))?;
e0918b88 Andy McCall 1683
        return Ok(repo);
1684
    }
ed4aef8b Andy McCall ↶ 1685
    // First job for this repository: clone with the job's token, which goes
1686
    // through the process environment, so one clone at a time.
1687
    let _clone = CLONE.lock().unwrap_or_else(|p| p.into_inner());
c23d5191 Andy McCall ↶ 1688
    std::env::set_var("OXY_TOKEN", token);
e0918b88 Andy McCall 1689
    if let Some(ca) = &cfg.ca {
1690
        std::env::set_var("OXY_CA", ca);
1691
    }
f53e6d47 Andy McCall ↶ 1692
    let cloned = with_retries(&FETCH_WAITS, || {
1693
        let _ = std::fs::remove_dir_all(&dir);
1694
        Repo::clone(clone_url, &dir).map_err(|e| anyhow::anyhow!("{e}"))
1695
    });
e0918b88 Andy McCall 1696
    std::env::remove_var("OXY_TOKEN");
1697
    std::env::remove_var("OXY_CA");
1698
    let mut repo = cloned.map_err(|e| {
1699
        let _ = std::fs::remove_dir_all(&dir);
c23d5191 Andy McCall ↶ 1700
        anyhow::anyhow!("clone {}: {e}", clone_url)
e0918b88 Andy McCall 1701
    })?;
1702
    set_remote(&mut repo)?;
1703
    Ok(repo)
bb643c25 Andy McCall ↶ 1704
}
1705
1706
#[cfg(test)]
1707
mod tests {
46fe2c81 Andy McCall ↶ 1708
1709
    #[test]
1710
    fn an_input_is_placed_only_when_it_is_what_the_receipt_names() {
1711
        let dir = tempfile::tempdir().unwrap();
1712
        let bytes = b"the reviewed plan";
1713
        let blob = Repo::blob_ref(bytes).to_hex();
1714
        let input = |path: &str, blob: &str| flow_core::api::ArtifactInput {
1715
            artifact: "tfplan".into(),
1716
            from: "plan".into(),
1717
            path: path.into(),
1718
            blob: blob.into(),
1719
        };
1720
        let got = super::place_input(dir.path(), &input("/src/plans/tfplan", &blob), bytes).unwrap();
1721
        assert_eq!(hex::encode(got), blob);
1722
        assert_eq!(std::fs::read(dir.path().join("plans/tfplan")).unwrap(), bytes);
1723
        let wrong = super::place_input(dir.path(), &input("/src/other", &"0".repeat(64)), bytes).unwrap_err();
1724
        assert!(format!("{wrong:#}").contains("not what plan's receipt names"), "{wrong:#}");
1725
        assert!(!dir.path().join("other").exists(), "nothing written when it does not match");
1726
        for bad in ["/etc/passwd", "/src/../x", "/src/.oxy/x", "/src/"] {
1727
            assert!(super::place_input(dir.path(), &input(bad, &blob), bytes).is_err(), "{bad}");
1728
        }
1729
    }
1730
bb643c25 Andy McCall ↶ 1731
    use super::*;
b11e30f5 Andy McCall ↶ 1732
1733
    #[test]
1734
    fn a_runner_names_what_stops_a_sandbox() {
1735
        let dir = tempfile::tempdir().unwrap();
1736
        let home = Home::locate(Some(dir.path())).unwrap();
1737
        let bytes = b"not really ext4";
1738
        let hash = blake3::hash(bytes).to_hex().to_string();
1739
        let mut sb: flow_core::Sandbox = serde_json::from_value(serde_json::json!({
1740
            "name": "s", "runtime": "syscall", "rootfs": "ab".repeat(32), "memory_mib": 1024, "cpus": 1, "network": "isolated"
1741
        }))
1742
        .unwrap();
1743
        let problem = |sb: &flow_core::Sandbox, free| sandbox_problem(&home, sb, "x86_64", free);
1744
        assert_eq!(problem(&sb, None).unwrap(), "rootfs abababababab is not on this runner");
1745
        std::fs::write(home.rootfs(&hash), bytes).unwrap();
1746
        sb.rootfs = hash.clone();
1747
        assert_eq!(problem(&sb, Some(4096)), None);
1748
        assert_eq!(problem(&sb, Some(512)).unwrap(), "only 512 MiB of memory is free here; the sandbox needs 1024 MiB");
1749
        std::fs::write(home.image_manifest(&hash), r#"{"arch":"riscv64"}"#).unwrap();
1750
        assert_eq!(problem(&sb, None).unwrap(), format!("rootfs {} is for riscv64; this runner is x86_64", &hash[..12]));
1751
        sb.arch = Some("aarch64".into());
1752
        assert_eq!(problem(&sb, None).unwrap(), "the sandbox needs aarch64; this runner is x86_64");
1753
        sb.arch = None;
1754
        std::fs::remove_file(home.image_manifest(&hash)).unwrap();
1755
        std::fs::write(home.rootfs(&hash), b"changed").unwrap();
1756
        assert!(problem(&sb, None).unwrap().contains("does not verify"));
1757
    }
f53e6d47 Andy McCall ↶ 1758
1759
    #[test]
1760
    fn transient_network_errors_are_tried_again_and_others_are_not() {
1761
        assert!(transient("fetch x: Dns Failed: resolve dns name 'oxygenhub.io:443': Temporary failure in name resolution"));
1762
        assert!(transient("HTTP 503: the hub is restarting"));
1763
        assert!(!transient("HTTP 404: no such repository"));
1764
        let mut n = 0;
1765
        let got: Result<u32> = with_retries(&[Duration::ZERO, Duration::ZERO], || {
1766
            n += 1;
1767
            if n < 3 {
1768
                Err(anyhow::anyhow!("connection refused"))
1769
            } else {
1770
                Ok(7)
1771
            }
1772
        });
1773
        assert_eq!((got.unwrap(), n), (7, 3), "two retries, then success");
1774
        let mut m = 0;
1775
        let got: Result<u32> = with_retries(&[Duration::ZERO; 4], || {
1776
            m += 1;
1777
            Err(anyhow::anyhow!("HTTP 403: forbidden"))
1778
        });
1779
        assert!(got.is_err());
1780
        assert_eq!(m, 1, "a refusal is not tried again");
1781
        let mut k = 0;
1782
        let got: Result<u32> = with_retries(&[Duration::ZERO; 2], || {
1783
            k += 1;
1784
            Err(anyhow::anyhow!("timed out"))
1785
        });
1786
        assert!(got.is_err());
1787
        assert_eq!(k, 3, "gives up after the last wait");
1788
    }
1b2229da Andy McCall ↶ 1789
1790
    /// A hub that answers each request with the next of `codes`, counting them.
1791
    fn fake_hub(codes: Vec<u16>) -> (String, std::sync::Arc<std::sync::atomic::AtomicUsize>) {
89171fca Andy McCall ↶ 1792
        fake_hub_on(std::net::TcpListener::bind("127.0.0.1:0").unwrap(), codes)
1793
    }
1794
1795
    /// A hub at `listener` answering each request with the next of `codes`.
1796
    fn fake_hub_on(listener: std::net::TcpListener, codes: Vec<u16>) -> (String, std::sync::Arc<std::sync::atomic::AtomicUsize>) {
1b2229da Andy McCall ↶ 1797
        use std::io::{Read, Write};
1798
        let addr = listener.local_addr().unwrap();
1799
        let seen = std::sync::Arc::new(std::sync::atomic::AtomicUsize::new(0));
1800
        let count = seen.clone();
1801
        std::thread::spawn(move || {
1802
            for code in codes {
1803
                let (mut s, _) = listener.accept().unwrap();
1804
                let mut buf = [0u8; 8192];
1805
                let _ = s.read(&mut buf);
1806
                count.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
1807
                let reason = if code == 204 { "No Content" } else { "Error" };
1808
                let _ = write!(s, "HTTP/1.1 {code} {reason}\r\nContent-Length: 0\r\nConnection: close\r\n\r\n");
1809
            }
1810
        });
1811
        (format!("http://{addr}"), seen)
f98a8f78 Andy McCall ↶ 1812
    }
1813
1814
    /// A hub that answers one request with `body` (status 200) and hands
1815
    /// back what it was sent.
1816
    fn fake_json_hub(body: &'static str) -> (String, std::sync::mpsc::Receiver<String>) {
1817
        use std::io::{Read, Write};
1818
        let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
1819
        let addr = listener.local_addr().unwrap();
1820
        let (tx, rx) = std::sync::mpsc::channel();
1821
        std::thread::spawn(move || {
1822
            let (mut s, _) = listener.accept().unwrap();
1823
            let mut req = Vec::new();
1824
            let mut buf = [0u8; 65536];
1825
            loop {
1826
                let n = s.read(&mut buf).unwrap_or(0);
1827
                if n == 0 {
1828
                    break;
1829
                }
1830
                req.extend_from_slice(&buf[..n]);
1831
                let text = String::from_utf8_lossy(&req);
1832
                if let Some(end) = text.find("\r\n\r\n") {
1833
                    let len = text[..end]
1834
                        .lines()
1835
                        .find_map(|l| {
1836
                            l.to_ascii_lowercase().strip_prefix("content-length:").map(|v| v.trim().parse::<usize>().unwrap_or(0))
1837
                        })
1838
                        .unwrap_or(0);
1839
                    if req.len() >= end + 4 + len {
1840
                        break;
1841
                    }
1842
                }
1843
            }
1844
            tx.send(String::from_utf8_lossy(&req).into_owned()).unwrap();
1845
            let _ = write!(
1846
                s,
1847
                "HTTP/1.1 200 OK\r\nContent-Type: application/json\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{body}",
1848
                body.len()
1849
            );
1850
        });
1851
        (format!("http://{addr}"), rx)
1852
    }
1853
1854
    #[test]
1855
    fn registering_binds_the_key_and_keeps_the_credential_private() {
1856
        use std::os::unix::fs::PermissionsExt;
1857
        let dir = tempfile::tempdir().unwrap();
1858
        let home = Home::locate(Some(dir.path())).unwrap();
1859
        let (hub, sent) = fake_json_hub(r#"{"id":7,"name":"plex","credential":"flowrunner_secret"}"#);
1860
        let got = register(&home, &hub, " reg-token ", "plex", None, vec!["gpu".into()]).unwrap();
1861
        assert_eq!((got.id, got.name.as_str()), (7, "plex"));
1862
        let req = sent.recv().unwrap();
1863
        assert!(req.starts_with("POST /api/v1/flow/runners/register"), "{req}");
1864
        let body: Register = serde_json::from_str(&req[req.find("\r\n\r\n").unwrap() + 4..]).unwrap();
1865
        assert_eq!((body.token.as_str(), body.info.labels.clone()), ("reg-token", vec!["gpu".to_string()]));
1866
        let key = home.runner_key("plex").unwrap();
1867
        assert_eq!(body.identity, hex::encode(key.identity.encode()), "the runner's signing identity");
1868
        let path = dir.path().join(crate::home::RUNNER_TOML);
1869
        assert_eq!(std::fs::metadata(&path).unwrap().permissions().mode() & 0o777, 0o600);
1870
        // The runner then starts from runner.toml.
1871
        let again = Home::locate(Some(dir.path())).unwrap();
1872
        let cfg = again.config.runner.unwrap();
1873
        assert_eq!((cfg.hub.as_str(), cfg.token.as_str(), cfg.name.as_deref()), (hub.as_str(), "flowrunner_secret", Some("plex")));
1874
        assert_eq!(cfg.labels, ["gpu"]);
35f7a849 Andy McCall ↶ 1875
    }
1876
1877
    /// A hub that answers certify with a certificate `sign_as` signs for
1878
    /// `cert_key`, and publishes `published` as its key: two requests.
1879
    fn fake_certify_hub(
1880
        sign_as: (oxy_core::identity::Identity, oxy_core::identity::SecretKey),
1881
        cert_key: oxy_core::identity::Identity,
1882
        published: oxy_core::identity::Identity,
1883
    ) -> String {
1884
        use std::io::{Read, Write};
1885
        let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
1886
        let addr = listener.local_addr().unwrap();
1887
        std::thread::spawn(move || {
1888
            for _ in 0..2 {
1889
                let Ok((mut s, _)) = listener.accept() else { return };
1890
                let mut req = Vec::new();
1891
                let mut buf = [0u8; 65536];
1892
                loop {
1893
                    let n = s.read(&mut buf).unwrap_or(0);
1894
                    if n == 0 {
1895
                        break;
1896
                    }
1897
                    req.extend_from_slice(&buf[..n]);
1898
                    let text = String::from_utf8_lossy(&req);
1899
                    if let Some(end) = text.find("\r\n\r\n") {
1900
                        let len = text[..end]
1901
                            .lines()
1902
                            .find_map(|l| {
1903
                                l.to_ascii_lowercase().strip_prefix("content-length:").map(|v| v.trim().parse::<usize>().unwrap_or(0))
1904
                            })
1905
                            .unwrap_or(0);
1906
                        if req.len() >= end + 4 + len {
1907
                            break;
1908
                        }
1909
                    }
1910
                }
1911
                let line = String::from_utf8_lossy(&req).lines().next().unwrap_or("").to_string();
1912
                let body: Vec<u8> = if line.contains("/keyless/certify") {
1913
                    let cert = oxy_core::cert::SignedCertificate::sign(
1914
                        oxy_core::cert::Certificate {
1915
                            key: cert_key.clone(),
1916
                            issuer: sign_as.0.hash(),
1917
                            subject: "repo:andy/demo:bookmark:main".into(),
1918
                            claims: vec![],
1919
                            scope: oxy_core::identity::Scope::SEAL,
1920
                            not_before: 1,
1921
                            expires: 600_001,
fc2a6d18 Andy McCall ↶ 1922
                            grant: None,
35f7a849 Andy McCall ↶ 1923
                        },
1924
                        &sign_as.0,
1925
                        &sign_as.1,
1926
                    );
1927
                    serde_json::json!({ "certificate": oxy_core::dsse::base64(&cert.encode()) }).to_string().into_bytes()
1928
                } else {
1929
                    published.encode()
1930
                };
1931
                let _ = write!(s, "HTTP/1.1 200 OK\r\nContent-Length: {}\r\nConnection: close\r\n\r\n", body.len());
1932
                let _ = s.write_all(&body);
1933
            }
1934
        });
1935
        format!("http://{addr}")
1936
    }
1937
1938
    /// keyless (oxygenhub#241): the runner takes a certificate only when it
1939
    /// is for its fresh key and signed by the key the hub publishes.
1940
    #[test]
1941
    fn a_certificate_is_taken_only_for_this_key_from_the_hubs_key() {
1942
        use oxy_core::identity::SecretKey;
1943
        let hub_sk = SecretKey::generate();
1944
        let hub_id = hub_sk.identity("hub", "", 1);
1945
        let sk = SecretKey::generate();
1946
        let id = sk.identity("earth-1 job 7", "", 1);
1947
        let proof = sk.sign_challenge(&id.hash(), &oxy_core::cert::certify_challenge("flowjob_x")).unwrap();
1948
1949
        let base = fake_certify_hub((hub_id.clone(), hub_sk.clone()), id.clone(), hub_id.clone());
1950
        let (cert, issuer) = hub_at(base).certify("flowjob_x", &id, &proof).unwrap();
1951
        assert_eq!((cert.cert.key_id(), issuer.hash()), (id.hash(), hub_id.hash()));
1952
1953
        let mallory_sk = SecretKey::generate();
1954
        let mallory = mallory_sk.identity("mallory", "", 1);
1955
        let base = fake_certify_hub((mallory.clone(), mallory_sk), id.clone(), hub_id.clone());
1956
        let e = hub_at(base).certify("flowjob_x", &id, &proof).unwrap_err().to_string();
1957
        assert!(e.contains("not signed by the hub's published key"), "{e}");
1958
1959
        let other = SecretKey::generate().identity("other", "", 1);
1960
        let base = fake_certify_hub((hub_id.clone(), hub_sk), other, hub_id);
1961
        let e = hub_at(base).certify("flowjob_x", &id, &proof).unwrap_err().to_string();
1962
        assert!(e.contains("another key"), "{e}");
1b2229da Andy McCall ↶ 1963
    }
1964
1965
    fn hub_at(base: String) -> Hub {
1966
        Hub { base, token: "t".into(), agent: ureq::AgentBuilder::new().timeout(Duration::from_secs(5)).build() }
cedd1234 Andy McCall ↶ 1967
    }
1968
1969
    /// oxygenhub#322: an offered build is kept only if it is the bytes the
1970
    /// hub attests; a peer or a proxy cannot change what is installed.
1971
    #[test]
1972
    fn a_build_is_kept_only_if_it_matches_the_hubs_blake3() {
1973
        let (body, other) = (b"#!flow build 0.3.2".to_vec(), b"#!something else!!".to_vec());
1974
        let serve = |bytes: Vec<u8>| {
1975
            let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
1976
            let addr = listener.local_addr().unwrap();
1977
            std::thread::spawn(move || {
1978
                use std::io::{Read, Write};
1979
                let (mut s, _) = listener.accept().unwrap();
1980
                let mut buf = [0u8; 4096];
1981
                let _ = s.read(&mut buf);
1982
                let _ = write!(s, "HTTP/1.1 200 OK\r\nContent-Length: {}\r\nConnection: close\r\n\r\n", bytes.len());
1983
                let _ = s.write_all(&bytes);
1984
            });
1985
            format!("http://{addr}")
1986
        };
1987
        let offer = flow_core::api::UpdateOffer {
1988
            version: "0.3.2".into(),
1989
            asset: "flow-linux-x86_64".into(),
1990
            url: "/andy/oxygenflow/asset/1/flow-linux-x86_64".into(),
1991
            blake3: blake3::hash(&body).to_hex().to_string(),
1992
            size: body.len() as u64,
1993
            ..Default::default()
1994
        };
1995
        let dir = tempfile::tempdir().unwrap();
1996
        let to = dir.path().join("flow-0.3.2");
1997
        hub_at(serve(body.clone())).download_build(&offer, &to).unwrap();
1998
        assert_eq!(std::fs::read(&to).unwrap(), body);
1999
        std::fs::remove_file(&to).unwrap();
2000
        let e = hub_at(serve(other)).download_build(&offer, &to).unwrap_err();
2001
        assert!(e.to_string().contains("does not match what the hub attests"), "{e:#}");
2002
        assert!(!to.exists() && !to.with_extension("partial").exists(), "nothing of it is left");
3007d8f0 Andy McCall ↶ 2003
    }
2004
2005
    /// oxygenhub#71: a job's result is offered until the hub takes it,
2006
    /// through server errors, and a refusal ends the trying at once.
2007
    #[test]
2008
    fn a_result_is_reported_until_the_hub_takes_it() {
8f211e20 Andy McCall ↶ 2009
        let f = Finish {
2010
            exit_code: Some(0),
2011
            duration_ms: 1,
2012
            receipt: None,
2013
            signed_receipt: None,
2014
            error: None,
2015
            peak_disk_mib: None,
2016
            peak_memory_mib: None,
2017
        };
3007d8f0 Andy McCall ↶ 2018
        let quick = Duration::from_millis(10);
2019
        let (base, seen) = fake_hub(vec![502, 503, 500, 500, 204]);
2020
        hub_at(base).finish_until_accepted(7, &f, Duration::from_secs(30), quick).unwrap();
2021
        assert_eq!(seen.load(std::sync::atomic::Ordering::SeqCst), 5, "kept trying through four server errors");
2022
        let (base, seen) = fake_hub(vec![409]);
2023
        assert!(hub_at(base).finish_until_accepted(7, &f, Duration::from_secs(30), quick).is_err());
2024
        assert_eq!(seen.load(std::sync::atomic::Ordering::SeqCst), 1, "a refusal is final");
1b2229da Andy McCall ↶ 2025
    }
2026
2027
    #[test]
2028
    fn server_errors_are_tried_again_and_client_errors_are_not() {
2029
        let quick = [Duration::from_millis(10); 3];
2030
        let (base, seen) = fake_hub(vec![500, 503, 204]);
2031
        let r = hub_at(base).post_with("/flow/jobs/1/log", &serde_json::json!({}), &quick).unwrap();
2032
        assert!(r.is_none());
2033
        assert_eq!(seen.load(std::sync::atomic::Ordering::SeqCst), 3, "two retries, then accepted");
2034
2035
        let (base, seen) = fake_hub(vec![500, 500, 500, 500]);
2036
        let e = hub_at(base).post_with("/flow/claim", &serde_json::json!({}), &quick).unwrap_err().to_string();
2037
        assert!(e.contains("HTTP 500"), "{e}");
2038
        assert_eq!(seen.load(std::sync::atomic::Ordering::SeqCst), 4, "gives up after the last delay");
2039
2040
        let (base, seen) = fake_hub(vec![401]);
2041
        assert!(hub_at(base).post_with("/flow/claim", &serde_json::json!({}), &quick).is_err());
2042
        assert_eq!(seen.load(std::sync::atomic::Ordering::SeqCst), 1, "a refusal is final");
06a4c472 Andy McCall ↶ 2043
    }
2044
1b160d37 Andy McCall ↶ 2045
    static NOT_LEAVING: std::sync::Mutex<Option<Leave>> = std::sync::Mutex::new(None);
06a4c472 Andy McCall ↶ 2046
2047
    /// oxygenflow#76: once the runner leaves, a job's sink says it is
2048
    /// cancelled, having sent what the job said so far, and the job is
2049
    /// handed back to the hub.
2050
    #[test]
2051
    fn a_leaving_runner_stops_its_job_and_hands_it_back() {
2052
        use microvm::Sink;
2053
        let leaving = std::sync::Mutex::new(None);
2054
        let (base, seen) = fake_hub(vec![204, 204]);
2055
        let hub = hub_at(base);
1b160d37 Andy McCall ↶ 2056
        let mut sink = LogSink {
2057
            hub: &hub,
2058
            job: 7,
2059
            buf: Vec::new(),
2060
            last: Instant::now(),
2061
            gone: false,
2062
            leaving: &leaving,
2063
            started: Instant::now(),
2064
            expected: None,
2065
        };
06a4c472 Andy McCall ↶ 2066
        sink.buf.push(Record { t: 1, stream: "out".into(), text: "step 1".into() });
2067
        assert!(!sink.cancelled());
1b160d37 Andy McCall ↶ 2068
        *leaving.lock().unwrap() = Some(Leave { reason: "off Wi-Fi".into(), by: Instant::now() });
06a4c472 Andy McCall ↶ 2069
        assert!(sink.cancelled());
2070
        assert!(sink.buf.is_empty(), "flushed before the job stops");
2071
        assert_eq!(seen.load(std::sync::atomic::Ordering::SeqCst), 1);
2072
        hub.release(7, "off Wi-Fi").unwrap();
2073
        assert_eq!(seen.load(std::sync::atomic::Ordering::SeqCst), 2);
1b160d37 Andy McCall ↶ 2074
    }
2075
2076
    /// oxygenflow#76: with notice, a job the hub expects to finish within
2077
    /// it runs on; one that will not, or whose time is unknown, is handed
2078
    /// back at once, and any job still running when the notice is up.
2079
    #[test]
2080
    fn a_job_that_fits_the_notice_finishes_and_the_others_go_back() {
2081
        use microvm::Sink;
2082
        let (base, _) = fake_hub(vec![204; 8]);
2083
        let hub = hub_at(base);
2084
        let leaving = std::sync::Mutex::new(Some(Leave { reason: "off charge".into(), by: Instant::now() + Duration::from_secs(60) }));
2085
        let sink = |expected: Option<u64>, ran: u64| LogSink {
2086
            hub: &hub,
2087
            job: 1,
2088
            buf: Vec::new(),
2089
            last: Instant::now(),
2090
            gone: false,
2091
            leaving: &leaving,
2092
            started: Instant::now() - Duration::from_secs(ran),
2093
            expected: expected.map(Duration::from_secs),
2094
        };
2095
        assert!(!sink(Some(70), 30).cancelled(), "40 s left of it, within a minute: it finishes");
2096
        assert!(sink(Some(600), 30).cancelled(), "nearly ten minutes left: handed back");
2097
        assert!(sink(None, 0).cancelled(), "no idea how long: handed back");
2098
        *leaving.lock().unwrap() = Some(Leave { reason: "off charge".into(), by: Instant::now() - Duration::from_secs(1) });
2099
        assert!(sink(Some(70), 30).cancelled(), "the notice is up: handed back whatever was expected");
280ef951 Andy McCall ↶ 2100
    }
2101
2102
    /// oxygenflow#32: a quiet job asks the hub every few seconds whether
2103
    /// it still runs; a 409 marks it cancelled, and nothing more is sent.
2104
    #[test]
2105
    fn a_quiet_job_learns_that_its_run_was_cancelled() {
2106
        use microvm::Sink;
2107
        let n = |seen: &std::sync::Arc<std::sync::atomic::AtomicUsize>| seen.load(std::sync::atomic::Ordering::SeqCst);
2108
        let quiet = Instant::now() - HEARTBEAT - Duration::from_secs(1);
2109
        let (base, seen) = fake_hub(vec![204]);
2110
        let hub = hub_at(base);
1b160d37 Andy McCall ↶ 2111
        let mut sink = LogSink {
2112
            hub: &hub,
2113
            job: 1,
2114
            buf: Vec::new(),
2115
            last: quiet,
2116
            gone: false,
2117
            leaving: &NOT_LEAVING,
2118
            started: Instant::now(),
2119
            expected: None,
2120
        };
280ef951 Andy McCall ↶ 2121
        assert!(!sink.cancelled());
2122
        assert!(!sink.cancelled(), "asked again only after the heartbeat");
2123
        assert_eq!(n(&seen), 1);
2124
2125
        let (base, seen) = fake_hub(vec![409]);
2126
        let hub = hub_at(base);
1b160d37 Andy McCall ↶ 2127
        let mut sink = LogSink {
2128
            hub: &hub,
2129
            job: 1,
2130
            buf: Vec::new(),
2131
            last: quiet,
2132
            gone: false,
2133
            leaving: &NOT_LEAVING,
2134
            started: Instant::now(),
2135
            expected: None,
2136
        };
280ef951 Andy McCall ↶ 2137
        assert!(sink.cancelled());
2138
        sink.record(Record { t: 0, stream: "out".into(), text: "late".into() });
2139
        sink.flush();
2140
        assert!(sink.cancelled());
2141
        assert_eq!(n(&seen), 1, "nothing is sent for a cancelled job");
89171fca Andy McCall ↶ 2142
    }
2143
2144
    /// oxygenflow#49: a hub that is away for a while (a deploy restarting
2145
    /// it) costs a job's uploads a wait, not the upload, and costs its log
2146
    /// no lines.
2147
    #[test]
2148
    fn a_job_outlasts_a_hub_restart() {
2149
        let n = |seen: &std::sync::Arc<std::sync::atomic::AtomicUsize>| seen.load(std::sync::atomic::Ordering::SeqCst);
2150
        // A port nobody listens on yet: the hub is restarting.
2151
        let port = std::net::TcpListener::bind("127.0.0.1:0").unwrap().local_addr().unwrap().port();
2152
        let base = format!("http://127.0.0.1:{port}");
2153
        let hub = hub_at(base.clone());
2154
2155
        // Log lines sent while it is away are kept, and go when it is back.
1b160d37 Andy McCall ↶ 2156
        let mut sink = LogSink {
2157
            hub: &hub,
2158
            job: 1,
2159
            buf: Vec::new(),
2160
            last: Instant::now(),
2161
            gone: false,
2162
            leaving: &NOT_LEAVING,
2163
            started: Instant::now(),
2164
            expected: None,
2165
        };
89171fca Andy McCall ↶ 2166
        for i in 0..450 {
2167
            sink.buf.push(Record { t: i, stream: "out".into(), text: format!("line {i}") });
2168
        }
2169
        sink.send();
2170
        assert_eq!(sink.buf.len(), 450, "nothing lost while the hub is away");
2171
        let (_, seen) = fake_hub_on(std::net::TcpListener::bind(("127.0.0.1", port)).unwrap(), vec![204, 204, 204]);
2172
        sink.send();
2173
        assert!(sink.buf.is_empty() && !sink.gone);
2174
        assert_eq!(n(&seen), 3, "in batches of 200: 200, 200 and 50");
2175
2176
        // An upload waits for the hub to come back, through refused
2177
        // connections and server errors alike.
2178
        let port = std::net::TcpListener::bind("127.0.0.1:0").unwrap().local_addr().unwrap().port();
2179
        let url = format!("http://127.0.0.1:{port}/x");
2180
        let agent = ureq::AgentBuilder::new().timeout(Duration::from_secs(5)).build();
2181
        let later = std::thread::spawn(move || {
2182
            std::thread::sleep(Duration::from_millis(150));
2183
            fake_hub_on(std::net::TcpListener::bind(("127.0.0.1", port)).unwrap(), vec![502, 204])
2184
        });
2185
        let quick = [Duration::from_millis(50); 20];
2186
        let r = retrying(&url, &quick, true, || agent.post(&url).send_string("{}").map_err(Box::new)).unwrap();
2187
        assert_eq!(r.status(), 204);
2188
        assert_eq!(n(&later.join().unwrap().1), 2, "a 502 from the proxy, then the hub");
2189
        // Without `connection`, a hub that cannot be reached ends it at once.
2190
        let port = std::net::TcpListener::bind("127.0.0.1:0").unwrap().local_addr().unwrap().port();
2191
        let url = format!("http://127.0.0.1:{port}/x");
2192
        assert!(retrying(&url, &quick, false, || agent.post(&url).send_string("{}").map_err(Box::new)).is_err());
2193
        assert!(PATIENT_DELAYS.iter().sum::<Duration>() >= Duration::from_secs(180), "longer than a hub takes to restart");
2194
    }
2195
2196
    /// oxygenflow#50: the first stop drains (the second, not called here,
2197
    /// exits at once).
2198
    #[test]
2199
    fn a_stop_drains_the_runner() {
2200
        assert!(!stopping());
2201
        on_stop(libc::SIGTERM);
2202
        assert!(stopping(), "told to stop, it claims nothing more");
5cf1e8ae Andy McCall ↶ 2203
        // An embedded runner starts afresh, and stop() drains it the same way.
2204
        STOPS.store(0, std::sync::atomic::Ordering::SeqCst);
2205
        stop();
2206
        assert!(stopping());
2207
        STOPS.store(0, std::sync::atomic::Ordering::SeqCst);
89171fca Andy McCall ↶ 2208
    }
2209
2210
    /// Lines kept for a hub that stays away are bounded, saying how many went.
2211
    #[test]
2212
    fn unsent_log_lines_are_bounded() {
2213
        let port = std::net::TcpListener::bind("127.0.0.1:0").unwrap().local_addr().unwrap().port();
2214
        let hub = hub_at(format!("http://127.0.0.1:{port}"));
1b160d37 Andy McCall ↶ 2215
        let mut sink = LogSink {
2216
            hub: &hub,
2217
            job: 1,
2218
            buf: Vec::new(),
2219
            last: Instant::now(),
2220
            gone: false,
2221
            leaving: &NOT_LEAVING,
2222
            started: Instant::now(),
2223
            expected: None,
2224
        };
89171fca Andy McCall ↶ 2225
        sink.buf = (0..MAX_UNSENT as u64 + 10).map(|i| Record { t: i, stream: "out".into(), text: String::new() }).collect();
2226
        sink.send();
2227
        assert_eq!(sink.buf.len(), MAX_UNSENT + 1);
2228
        assert_eq!(sink.buf[0].text, "(10 log lines lost while the hub could not be reached)");
2229
        assert_eq!(sink.buf[1].t, 10, "the oldest go first");
1b2229da Andy McCall ↶ 2230
    }
bb643c25 Andy McCall ↶ 2231
2232
    #[test]
2233
    fn the_budget_admits_jobs_that_fit() {
2234
        let b = Budget { max_jobs: 3, memory_mib: Some(24_576), cpus: Some(8) };
2235
        let used = |jobs, memory_mib, cpus| Used { jobs, memory_mib, cpus };
2236
        assert_eq!(b.room(used(0, 0, 0)), Some((None, None)), "a lone job may take everything");
2237
        assert_eq!(b.room(used(1, 8192, 4)), Some((Some(16_384), Some(4))));
2238
        assert_eq!(b.room(used(2, 16_384, 8)), None, "no vCPU left");
2239
        assert_eq!(b.room(used(3, 1024, 1)), None, "at max_jobs");
2240
        let one = Budget { max_jobs: 1, memory_mib: None, cpus: None };
2241
        assert_eq!(one.room(used(0, 0, 0)), Some((None, None)));
2242
        assert_eq!(one.room(used(1, 0, 0)), None, "the default runs one job at a time");
2243
        let open = Budget { max_jobs: 2, memory_mib: None, cpus: None };
2244
        assert_eq!(open.room(used(1, 8192, 4)), Some((None, None)), "no limits configured");
2245
    }
e0918b88 Andy McCall 2246
}