kanade-agent 0.43.98

Windows-side resident daemon for the kanade endpoint-management system. Subscribes to commands.* over NATS, runs scripts, publishes WMI inventory + heartbeats, watches for self-updates
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
//! Agent-side cache of operator-defined health-check results (#290).
//!
//! A job whose Manifest carries a `check:` hint prints a JSON object
//! on stdout; after a successful run the command path calls
//! [`CheckSink::record`], which maps the hint's `status_field` /
//! `detail_field` into a KLP [`Check`] and stores it keyed by check
//! name. The KLP state evaluator ([`crate::klp::state::eval_once`])
//! merges these cached checks into every `StateSnapshot.checks`, and
//! [`CheckSink::wait`] lets the evaluator re-publish immediately when
//! a new result lands instead of waiting for the 30 s tick.
//!
//! Cross-platform on purpose: the writer ([`CheckSink::record`] in
//! `commands.rs`) compiles everywhere, while the reader (`klp::state`)
//! is Windows-only. On non-Windows the cache is written but never
//! read — harmless.

use std::collections::{HashMap, HashSet};
use std::path::PathBuf;
use std::sync::{Arc, Mutex};

use kanade_shared::ipc::state::{Check, CheckStatus};
use kanade_shared::manifest::CheckHint;
use tokio::sync::Notify;

/// Shared, cheap-to-clone handle threaded into both the command
/// result path (writer) and the KLP state evaluator (reader).
#[derive(Clone)]
pub struct CheckSink {
    inner: Arc<Inner>,
}

struct Inner {
    results: Mutex<HashMap<String, Check>>,
    updated: Notify,
    /// Where the cache is persisted as JSON (`None` = in-memory only,
    /// used by tests / non-Windows). Operator-defined check results
    /// are written here on every `record` and re-loaded on boot so the
    /// Client App's Health tab survives an agent restart and shows the
    /// last-known status even while offline (#290). It's a bounded
    /// snapshot — one entry per check name, a few KB — so a single
    /// atomically-replaced JSON file is the right store (no SQLite;
    /// same pattern as `local_completions.json` / the outboxes).
    path: Option<PathBuf>,
    /// Check slugs whose defining job currently exists in `BUCKET_JOBS`,
    /// kept fresh by the local scheduler (see
    /// [`CheckSink::set_active_slugs`]). `checks()` filters its cached
    /// results to this set so a check whose job an operator **deleted**
    /// stops appearing on the client's Health tab — deleting a `check:`
    /// job otherwise leaves its last result stranded in the cache
    /// forever (no new result ever overwrites it).
    ///
    /// `None` (the initial value, before the scheduler's first resync)
    /// means "active set unknown — don't filter": a broker-down boot
    /// can't enumerate the live jobs, so we show the last-known checks
    /// rather than blanking the tab. Self-heals to `Some(..)` on the
    /// first successful resync.
    ///
    /// Wrapped in `Arc` so `checks()` (called every evaluator tick) reads
    /// it with a refcount bump instead of cloning the whole set on the
    /// read path.
    active_slugs: Mutex<Option<Arc<HashSet<String>>>>,
}

impl CheckSink {
    /// In-memory only (no persistence). For tests and the non-Windows
    /// build where the KLP reader isn't compiled.
    pub fn new() -> Self {
        Self::with_inner(HashMap::new(), None)
    }

    /// Load the cache from `path` (an empty cache if the file is
    /// missing or unreadable — a corrupt/old file must not stop the
    /// agent booting) and persist every subsequent `record` back to it.
    pub fn load(path: PathBuf) -> Self {
        let results = match std::fs::read(&path) {
            Ok(bytes) => serde_json::from_slice::<HashMap<String, Check>>(&bytes)
                .unwrap_or_else(|e| {
                    tracing::warn!(error = %e, path = %path.display(), "check_cache: ignoring unreadable persisted cache");
                    HashMap::new()
                }),
            Err(e) if e.kind() == std::io::ErrorKind::NotFound => HashMap::new(),
            Err(e) => {
                tracing::warn!(error = %e, path = %path.display(), "check_cache: failed to read persisted cache");
                HashMap::new()
            }
        };
        Self::with_inner(results, Some(path))
    }

    fn with_inner(results: HashMap<String, Check>, path: Option<PathBuf>) -> Self {
        Self {
            inner: Arc::new(Inner {
                results: Mutex::new(results),
                updated: Notify::new(),
                path,
                // Unknown until the scheduler's first resync — see the
                // field doc. Filtering stays off (show everything) until
                // then so a boot before the jobs KV is reachable doesn't
                // wipe the Health tab.
                active_slugs: Mutex::new(None),
            }),
        }
    }

    /// Store a freshly-built [`Check`] under its `name`, persist the
    /// cache, and wake the state evaluator. The caller builds the
    /// `Check` — [`build_check`] on a clean exit, [`build_check_failed`]
    /// when the script crashed — so this stays a dumb sink.
    pub fn record(&self, check: Check) {
        self.guarded(|map| {
            map.insert(check.name.clone(), check);
            // Persist UNDER the lock: two concurrent records can't
            // interleave and write an older snapshot over a newer one
            // (the previous out-of-lock version had that race). The
            // file is a few KB — sub-ms write — so holding the lock
            // briefly doesn't meaningfully delay a reader's `checks()`.
            self.persist(map);
        });
        // Re-publish the snapshot now instead of on the next 30 s tick.
        self.inner.updated.notify_one();
    }

    /// Atomically write the cache to its JSON file (temp + rename), if
    /// persistence is configured. Best-effort: a write failure is
    /// logged, not fatal — the in-memory cache is still authoritative
    /// for this run. Call while holding the results lock so writes are
    /// serialised.
    fn persist(&self, results: &HashMap<String, Check>) {
        let Some(path) = &self.inner.path else { return };
        let json = match serde_json::to_vec_pretty(results) {
            Ok(j) => j,
            Err(e) => {
                tracing::warn!(error = %e, "check_cache: serialize failed");
                return;
            }
        };
        // The data dir may not exist yet on a fresh install.
        if let Some(parent) = path.parent() {
            if let Err(e) = std::fs::create_dir_all(parent) {
                tracing::warn!(error = %e, path = %parent.display(), "check_cache: create data dir failed");
                return;
            }
        }
        let tmp = path.with_extension("json.tmp");
        if let Err(e) = std::fs::write(&tmp, &json) {
            tracing::warn!(error = %e, path = %tmp.display(), "check_cache: temp write failed");
            return;
        }
        // rename replaces the destination atomically (MoveFileEx with
        // REPLACE_EXISTING on Windows, rename(2) on Unix).
        if let Err(e) = std::fs::rename(&tmp, path) {
            tracing::warn!(error = %e, path = %path.display(), "check_cache: atomic rename failed");
        }
    }

    /// Replace the set of check slugs whose defining job currently lives
    /// in `BUCKET_JOBS`. Called by the local scheduler whenever its jobs
    /// snapshot changes (resync + per-key watch events) so a **deleted**
    /// `check:` job's stale result drops off the Health tab instead of
    /// lingering forever (nothing ever overwrites a cached check once
    /// its job stops running). Wakes the evaluator so the prune lands on
    /// the next push rather than waiting up to the 30 s tick.
    ///
    /// Note this filters at read time only — `record`/`persist` still
    /// keep every result on disk, so a transient empty/stale active set
    /// (or a job that comes back) can never destroy the last-known data;
    /// the row simply reappears once its slug is active again.
    ///
    /// Not `#[cfg]`-gated: `local_scheduler` calls this on every platform
    /// (it has no Windows guard), so the function is live everywhere —
    /// unlike `checks()`, whose only caller is the Windows-only KLP
    /// evaluator.
    pub fn set_active_slugs(&self, slugs: HashSet<String>) {
        {
            let mut active = self
                .inner
                .active_slugs
                .lock()
                .unwrap_or_else(|poisoned| poisoned.into_inner());
            *active = Some(Arc::new(slugs));
        }
        self.inner.updated.notify_one();
    }

    /// Current cached checks, filtered to those whose job is still live
    /// (see [`set_active_slugs`](Self::set_active_slugs)). Cheap clone of
    /// the small map's values; order is unspecified (the snapshot builder
    /// positions them). When the active set is still unknown (`None`,
    /// pre-first-resync) nothing is filtered.
    #[cfg_attr(not(target_os = "windows"), allow(dead_code))]
    pub fn checks(&self) -> Vec<Check> {
        // Snapshot the active set first (and release its lock) so we
        // never hold both mutexes at once — `set_active_slugs` only ever
        // takes the active-set lock, so this ordering can't deadlock.
        let active = self
            .inner
            .active_slugs
            .lock()
            .unwrap_or_else(|poisoned| poisoned.into_inner())
            .clone();
        self.guarded(|map| {
            map.values()
                .filter(|c| match &active {
                    Some(set) => set.contains(c.name.as_str()),
                    None => true,
                })
                .cloned()
                .collect()
        })
    }

    /// Resolves the next time a check result is recorded.
    ///
    /// Relies on `tokio::sync::Notify` permit semantics: a
    /// `record()` → `notify_one()` that fires while no waiter is
    /// registered stores **one** permit, which the next `wait()`
    /// consumes immediately — so a result recorded between
    /// `eval_loop` iterations is never missed (multiple records
    /// coalesce into one wake, which is fine since the evaluator
    /// re-reads the whole map). Do NOT swap this for a condvar-style
    /// API that lacks the stored permit.
    #[cfg_attr(not(target_os = "windows"), allow(dead_code))]
    pub async fn wait(&self) {
        self.inner.updated.notified().await;
    }

    /// Run `f` under the results lock, recovering the guard (with a
    /// warning) if a previous holder panicked — a poisoned lock here
    /// must not silently freeze or wipe the Health tab. The map's
    /// data is still structurally valid after a panic elsewhere.
    fn guarded<R>(&self, f: impl FnOnce(&mut HashMap<String, Check>) -> R) -> R {
        let mut map = self.inner.results.lock().unwrap_or_else(|poisoned| {
            tracing::warn!("check_cache: results mutex poisoned — recovering");
            poisoned.into_inner()
        });
        f(&mut map)
    }
}

impl Default for CheckSink {
    fn default() -> Self {
        Self::new()
    }
}

/// Merge operator-defined `extra` checks onto the agent's intrinsic
/// `base` checks: an operator check overrides a built-in of the same
/// name (so a fleet could shadow e.g. `agent_self_update` with its
/// own), and any remaining extras are appended. Pure; shared by
/// `eval_once`.
#[cfg_attr(not(target_os = "windows"), allow(dead_code))]
pub fn merge_checks(base: Vec<Check>, extra: &[Check]) -> Vec<Check> {
    let overridden: HashSet<&str> = extra.iter().map(|c| c.name.as_str()).collect();
    let mut merged: Vec<Check> = base
        .into_iter()
        .filter(|c| !overridden.contains(c.name.as_str()))
        .collect();
    merged.extend(extra.iter().cloned());
    merged
}

/// Map a check job's stdout JSON object + its [`CheckHint`] into a KLP
/// [`Check`]. Pure (no I/O) so it's unit-testable. Non-JSON / non-object
/// stdout, or a missing / unrecognised `status_field`, degrades to
/// [`CheckStatus::Unknown`] with a diagnostic detail — everything else
/// is up to the operator's PowerShell.
pub fn build_check(hint: &CheckHint, stdout: &str) -> Check {
    // #821: read check's own fenced block so a job can compose check with
    // inventory / collect (and/or a user message) on one stdout. No fence
    // ⇒ the whole stdout (back-compat for a check-only job). `fenced_
    // payload` already trims.
    let payload = kanade_shared::manifest::fenced_payload(
        stdout,
        kanade_shared::manifest::CHECK_BLOCK_BEGIN,
        kanade_shared::manifest::CHECK_BLOCK_END,
    );
    let value: serde_json::Value = match serde_json::from_str(payload) {
        Ok(v) => v,
        Err(e) => return unknown(hint, format!("check stdout was not JSON: {e}")),
    };
    let Some(obj) = value.as_object() else {
        return unknown(hint, "check stdout was not a JSON object".to_string());
    };
    let status = match obj.get(&hint.status_field) {
        Some(v) => match serde_json::from_value::<CheckStatus>(v.clone()) {
            Ok(s) => s,
            Err(_) => {
                return unknown(
                    hint,
                    format!(
                        "`{}` = {v} is not one of ok/warn/fail/unknown",
                        hint.status_field
                    ),
                );
            }
        },
        None => {
            return unknown(hint, format!("stdout has no `{}` field", hint.status_field));
        }
    };
    let detail = obj.get(&hint.detail_field).and_then(json_to_detail);
    Check {
        name: hint.name.clone(),
        label: hint.label.clone(),
        status,
        detail,
        troubleshoot: hint.troubleshoot.clone(),
    }
}

/// Build the [`Check`] for a `check:` job that exited non-zero. The
/// script crashed before it could report a status, so we surface
/// `Unknown` with the exit code + a stderr snippet rather than
/// leaving a stale `Ok` on the Health tab (a persistently-crashing
/// check must not read as healthy).
pub fn build_check_failed(hint: &CheckHint, exit_code: i32, stderr: &str) -> Check {
    let snippet: String = stderr.trim().chars().take(200).collect();
    let detail = if snippet.is_empty() {
        format!("check script exited {exit_code}")
    } else {
        format!("check script exited {exit_code}: {snippet}")
    };
    unknown(hint, detail)
}

fn unknown(hint: &CheckHint, detail: String) -> Check {
    Check {
        name: hint.name.clone(),
        label: hint.label.clone(),
        status: CheckStatus::Unknown,
        detail: Some(detail),
        troubleshoot: hint.troubleshoot.clone(),
    }
}

/// Render a detail field value as a string: pass non-empty strings
/// through, stringify scalars, and compact-JSON-encode arrays /
/// objects (so an operator who puts structured data in `detail` sees
/// *something* on the row instead of a silently-blank column). Only
/// `null` / empty-string yield `None`.
fn json_to_detail(v: &serde_json::Value) -> Option<String> {
    match v {
        serde_json::Value::Null => None,
        serde_json::Value::String(s) if s.is_empty() => None,
        serde_json::Value::String(s) => Some(s.clone()),
        serde_json::Value::Number(n) => Some(n.to_string()),
        serde_json::Value::Bool(b) => Some(b.to_string()),
        // Array / Object → compact JSON, e.g. `["C:","D:"]`.
        other => Some(other.to_string()),
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn hint(name: &str) -> CheckHint {
        CheckHint {
            name: name.into(),
            label: None,
            status_field: "status".into(),
            detail_field: "detail".into(),
            troubleshoot: None,
            fleet: true,
            alert: None,
        }
    }

    #[test]
    fn build_check_reads_status_and_detail() {
        let c = build_check(
            &hint("bitlocker"),
            r#"{"status":"warn","detail":"D: unprotected"}"#,
        );
        assert_eq!(c.name, "bitlocker");
        assert_eq!(c.status, CheckStatus::Warn);
        assert_eq!(c.detail.as_deref(), Some("D: unprotected"));
    }

    #[test]
    fn build_check_honours_custom_fields_and_troubleshoot() {
        let h = CheckHint {
            name: "patch".into(),
            label: Some("Windows 更新プログラム".into()),
            status_field: "compliance".into(),
            detail_field: "summary".into(),
            troubleshoot: Some("fix-patch".into()),
            fleet: true,
            alert: None,
        };
        let c = build_check(
            &h,
            r#"{"compliance":"fail","summary":"12 missing","extra":[1,2]}"#,
        );
        assert_eq!(c.status, CheckStatus::Fail);
        assert_eq!(c.label.as_deref(), Some("Windows 更新プログラム"));
        assert_eq!(c.detail.as_deref(), Some("12 missing"));
        assert_eq!(c.troubleshoot.as_deref(), Some("fix-patch"));
    }

    #[test]
    fn build_check_detail_optional() {
        let c = build_check(&hint("x"), r#"{"status":"ok"}"#);
        assert_eq!(c.status, CheckStatus::Ok);
        assert!(c.detail.is_none());
    }

    #[test]
    fn build_check_degrades_on_bad_input() {
        // Not JSON.
        assert_eq!(
            build_check(&hint("x"), "not json").status,
            CheckStatus::Unknown
        );
        // Not an object.
        assert_eq!(
            build_check(&hint("x"), "[1,2,3]").status,
            CheckStatus::Unknown
        );
        // Missing status field.
        assert_eq!(
            build_check(&hint("x"), r#"{"detail":"hi"}"#).status,
            CheckStatus::Unknown
        );
        // Unrecognised status value — and the reason is surfaced.
        let bad = build_check(&hint("x"), r#"{"status":"green"}"#);
        assert_eq!(bad.status, CheckStatus::Unknown);
        assert!(bad.detail.unwrap().contains("green"));
    }

    #[test]
    fn merge_checks_overrides_builtin_by_name() {
        let base = vec![Check {
            name: "agent_self_update".into(),
            label: None,
            status: CheckStatus::Ok,
            detail: None,
            troubleshoot: None,
        }];
        let extra = vec![
            // operator's own check named like the built-in replaces it
            Check {
                name: "agent_self_update".into(),
                label: None,
                status: CheckStatus::Warn,
                detail: Some("override".into()),
                troubleshoot: None,
            },
            Check {
                name: "disk_space".into(),
                label: None,
                status: CheckStatus::Warn,
                detail: Some("8% free".into()),
                troubleshoot: None,
            },
            Check {
                name: "bitlocker".into(),
                label: None,
                status: CheckStatus::Ok,
                detail: None,
                troubleshoot: None,
            },
        ];
        let merged = merge_checks(base, &extra);
        // agent_self_update overridden (Warn, not duplicated); disk_space
        // + bitlocker appended.
        assert_eq!(merged.len(), 3);
        let asu = merged
            .iter()
            .find(|c| c.name == "agent_self_update")
            .unwrap();
        assert_eq!(asu.status, CheckStatus::Warn);
        assert!(merged.iter().any(|c| c.name == "bitlocker"));
        assert_eq!(
            merged
                .iter()
                .filter(|c| c.name == "agent_self_update")
                .count(),
            1
        );
    }

    #[test]
    fn build_check_failed_surfaces_exit_and_stderr() {
        let c = build_check_failed(&hint("av"), 1, "  Get-MpComputerStatus: access denied  ");
        assert_eq!(c.status, CheckStatus::Unknown);
        let d = c.detail.unwrap();
        assert!(d.contains("exited 1"), "detail: {d}");
        assert!(d.contains("access denied"), "detail: {d}");
    }

    #[test]
    fn json_to_detail_renders_arrays_as_compact_json() {
        let v: serde_json::Value = serde_json::json!(["C:", "D:"]);
        assert_eq!(json_to_detail(&v).as_deref(), Some(r#"["C:","D:"]"#));
        assert!(json_to_detail(&serde_json::Value::Null).is_none());
    }

    #[test]
    fn sink_record_and_read_round_trip() {
        let sink = CheckSink::new();
        assert!(sink.checks().is_empty());
        sink.record(build_check(&hint("bitlocker"), r#"{"status":"ok"}"#));
        let checks = sink.checks();
        assert_eq!(checks.len(), 1);
        assert_eq!(checks[0].name, "bitlocker");
        assert_eq!(checks[0].status, CheckStatus::Ok);
    }

    // ---- active-slug filtering (deleted-job prune) ----

    #[test]
    fn checks_unfiltered_until_active_set_is_known() {
        // Before the scheduler's first resync the active set is `None`,
        // so every cached check shows — a broker-down boot must not
        // blank the Health tab.
        let sink = CheckSink::new();
        sink.record(build_check(&hint("bitlocker"), r#"{"status":"ok"}"#));
        sink.record(build_check(&hint("av_signature"), r#"{"status":"ok"}"#));
        assert_eq!(sink.checks().len(), 2);
    }

    #[test]
    fn active_slugs_prunes_checks_whose_job_was_deleted() {
        let sink = CheckSink::new();
        sink.record(build_check(&hint("bitlocker"), r#"{"status":"ok"}"#));
        sink.record(build_check(&hint("av_signature"), r#"{"status":"warn"}"#));

        // Job behind `av_signature` deleted → only `bitlocker` is still
        // a live job, so the tab shows just that one.
        sink.set_active_slugs(HashSet::from(["bitlocker".to_string()]));
        let names: Vec<String> = sink.checks().into_iter().map(|c| c.name).collect();
        assert_eq!(names, vec!["bitlocker".to_string()]);
    }

    #[test]
    fn active_slugs_empty_hides_all_but_keeps_data() {
        // Every check job deleted → nothing shows, but the cache still
        // holds the results (read-time filter, not a delete): restoring
        // the active set brings the rows back, no re-run needed.
        let sink = CheckSink::new();
        sink.record(build_check(&hint("bitlocker"), r#"{"status":"ok"}"#));
        sink.set_active_slugs(HashSet::new());
        assert!(sink.checks().is_empty());

        sink.set_active_slugs(HashSet::from(["bitlocker".to_string()]));
        assert_eq!(sink.checks().len(), 1);
    }

    #[tokio::test]
    async fn set_active_slugs_wakes_the_evaluator() {
        // A deletion must prune promptly (next push), not wait for the
        // 30 s tick — set_active_slugs stores a notify permit just like
        // record().
        let sink = CheckSink::new();
        sink.set_active_slugs(HashSet::new());
        tokio::time::timeout(std::time::Duration::from_secs(1), sink.wait())
            .await
            .expect("set_active_slugs must wake the evaluator");
    }

    #[tokio::test]
    async fn wait_resolves_when_a_result_is_recorded() {
        let sink = CheckSink::new();
        // notify_one stores a permit even though we record BEFORE
        // wait() is polled — the next wait() must still complete.
        sink.record(build_check(&hint("x"), r#"{"status":"ok"}"#));
        tokio::time::timeout(std::time::Duration::from_secs(1), sink.wait())
            .await
            .expect("wait() must observe the stored notify permit");
    }

    // ---- persistence (#290: survive restart / offline) ----

    #[test]
    fn load_missing_file_is_empty() {
        let dir = tempfile::tempdir().unwrap();
        let sink = CheckSink::load(dir.path().join("check_results.json"));
        assert!(sink.checks().is_empty());
    }

    #[test]
    fn record_persists_and_reloads_across_restart() {
        let dir = tempfile::tempdir().unwrap();
        let path = dir.path().join("check_results.json");

        // First "boot": record two checks.
        let sink = CheckSink::load(path.clone());
        sink.record(build_check(
            &hint("bitlocker"),
            r#"{"status":"warn","detail":"D: off"}"#,
        ));
        sink.record(build_check(&hint("av_signature"), r#"{"status":"ok"}"#));
        drop(sink);
        assert!(path.exists(), "record must persist the cache file");

        // Second "boot": a fresh sink loads the persisted results, so
        // the Health tab is populated before any check re-runs.
        let reloaded = CheckSink::load(path);
        let mut checks = reloaded.checks();
        checks.sort_by(|a, b| a.name.cmp(&b.name));
        assert_eq!(checks.len(), 2);
        assert_eq!(checks[0].name, "av_signature");
        assert_eq!(checks[1].name, "bitlocker");
        assert_eq!(checks[1].status, CheckStatus::Warn);
        assert_eq!(checks[1].detail.as_deref(), Some("D: off"));
    }

    #[test]
    fn load_corrupt_file_is_empty_not_fatal() {
        let dir = tempfile::tempdir().unwrap();
        let path = dir.path().join("check_results.json");
        std::fs::write(&path, b"{ this is not valid json").unwrap();
        // Must boot with an empty cache rather than panicking.
        let sink = CheckSink::load(path);
        assert!(sink.checks().is_empty());
    }

    #[test]
    fn new_does_not_persist() {
        // The in-memory constructor writes nothing to disk.
        let sink = CheckSink::new();
        sink.record(build_check(&hint("x"), r#"{"status":"ok"}"#));
        assert_eq!(sink.checks().len(), 1);
        // (no path → persist() is a no-op; nothing to assert beyond
        // not panicking and the value still being in memory)
    }
}