taskvisor 0.6.0

Task supervisor for Tokio: restarts background tasks on failure with exponential backoff and jitter, graceful shutdown, and lifecycle events
Documentation
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
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
use super::*;
use crate::{
    RuntimeError, TaskOutcome, TaskSpec,
    core::actor::ActorExitReason,
    events::{Event, EventKind},
    reasons,
};
use std::{future::Future, pin::Pin, task::Poll};
use tokio::sync::oneshot;

async fn assert_pending_once<F: Future>(mut future: Pin<&mut F>) {
    std::future::poll_fn(|cx| match future.as_mut().poll(cx) {
        Poll::Pending => Poll::Ready(()),
        Poll::Ready(_) => panic!("future completed before the expected ordering point"),
    })
    .await;
}

async fn assert_ready_once<F: Future<Output = ()>>(mut future: Pin<&mut F>) {
    std::future::poll_fn(|cx| match future.as_mut().poll(cx) {
        Poll::Ready(()) => Poll::Ready(()),
        Poll::Pending => panic!("future was not immediately ready"),
    })
    .await;
}

#[tokio::test]
async fn pending_wait_drained_handles_empty_and_resolves_after_last_dec() {
    let p = Arc::new(PendingJoins::default());
    let mut initially_empty = Box::pin(p.wait_drained());
    assert_ready_once(initially_empty.as_mut()).await;
    drop(initially_empty);

    let a = TaskId::next();
    let b = TaskId::next();
    p.inc(a);
    p.inc(b);
    assert!(!p.is_empty());

    let mut drained = Box::pin(p.wait_drained());
    assert_pending_once(drained.as_mut()).await;
    p.dec(a);
    assert_pending_once(drained.as_mut()).await;
    p.dec(b);
    drained.await;
    assert!(p.is_empty(), "no joins should remain after draining");
}

fn registry() -> Arc<Registry> {
    let bus = Bus::new(64);
    let token = CancellationToken::new();
    let (_tx, rx) = mpsc::channel(64);
    Registry::new(
        bus,
        token,
        None,
        Duration::from_secs(5),
        TaskDefaults::default(),
        rx,
    )
}

#[tokio::test(flavor = "current_thread")]
async fn terminal_cleanup_wakes_all_empty_waiters() {
    use tokio::sync::Barrier;

    let registry = registry();
    let id = TaskId::next();
    let label: Arc<str> = Arc::from("empty-waiters");
    let completion = RemovalCompletion::new();
    let mut state = registry.state.write().await;
    state.by_label.insert(Arc::clone(&label), id);
    state.tasks.insert(
        id,
        Entry {
            label: Arc::clone(&label),
            state: EntryState::Removing {
                completion: completion.clone(),
            },
        },
    );
    registry.pending_joins.inc(id);
    registry.pending_joins.label(id, label);

    let ready = Arc::new(Barrier::new(3));
    let first_registry = Arc::clone(&registry);
    let first_ready = Arc::clone(&ready);
    let first = tokio::spawn(async move {
        first_ready.wait().await;
        first_registry.wait_until_empty().await;
    });
    let second_registry = Arc::clone(&registry);
    let second_ready = Arc::clone(&ready);
    let second = tokio::spawn(async move {
        second_ready.wait().await;
        second_registry.wait_until_empty().await;
    });

    ready.wait().await;
    tokio::task::yield_now().await;
    drop(state);

    let state_barrier = registry.state.write().await;
    drop(state_barrier);
    assert!(!first.is_finished());
    assert!(!second.is_finished());

    Registry::finish_removal(
        &registry.state,
        &registry.empty_notify,
        &registry.pending_joins,
        &registry.bus,
        RemovalReport {
            id,
            outcome: None,
            join: JoinCompletion::Joined(Ok(ActorExitReason::Completed)),
            completion,
        },
    )
    .await;

    tokio::time::timeout(Duration::from_secs(1), first)
        .await
        .expect("the first empty waiter must wake")
        .expect("the first empty waiter must not panic");
    tokio::time::timeout(Duration::from_secs(1), second)
        .await
        .expect("the second empty waiter must wake")
        .expect("the second empty waiter must not panic");
    assert!(registry.is_empty().await);
    assert_eq!(registry.id_for_label("empty-waiters").await, None);
    assert!(registry.pending_joins.is_empty());
}

fn started_registry(
    bus_capacity: usize,
    grace: Duration,
) -> (
    Arc<Registry>,
    Bus,
    CancellationToken,
    mpsc::Sender<RegistryCommand>,
) {
    let bus = Bus::new(bus_capacity);
    let token = CancellationToken::new();
    let (tx, rx) = mpsc::channel(64);
    let registry = Registry::new(
        bus.clone(),
        token.clone(),
        None,
        grace,
        TaskDefaults::default(),
        rx,
    );
    registry.clone().spawn_listener();
    (registry, bus, token, tx)
}

struct ControlledCancellationTask {
    task: crate::TaskRef,
    started: Arc<Notify>,
    cancellation_seen: Arc<Notify>,
    release: Arc<Notify>,
}

fn controlled_cancellation_task(label: &'static str) -> ControlledCancellationTask {
    let started = Arc::new(Notify::new());
    let cancellation_seen = Arc::new(Notify::new());
    let release = Arc::new(Notify::new());
    let started_by_task = Arc::clone(&started);
    let seen_by_task = Arc::clone(&cancellation_seen);
    let release_by_task = Arc::clone(&release);
    let task = crate::TaskFn::arc(label, move |ctx: crate::TaskContext| {
        let started = Arc::clone(&started_by_task);
        let cancellation_seen = Arc::clone(&seen_by_task);
        let release = Arc::clone(&release_by_task);
        async move {
            started.notify_one();
            ctx.cancelled().await;
            cancellation_seen.notify_one();
            release.notified().await;
            Err(crate::TaskError::Canceled)
        }
    });

    ControlledCancellationTask {
        task,
        started,
        cancellation_seen,
        release,
    }
}

fn send_add(
    tx: &mpsc::Sender<RegistryCommand>,
    id: TaskId,
    spec: TaskSpec,
    outcome: Option<OutcomeTx>,
) -> AddReplyRx {
    let (reply, reply_rx) = oneshot::channel();
    tx.try_send(RegistryCommand::Add {
        id,
        spec,
        outcome,
        completion: None,
        reply,
    })
    .expect("registry command channel must stay open");
    reply_rx
}

fn batch_item(id: TaskId, spec: TaskSpec) -> AddBatchItem {
    AddBatchItem {
        id,
        label: Arc::from(spec.task().name()),
        spec,
    }
}

fn send_batch(tx: &mpsc::Sender<RegistryCommand>, items: Vec<AddBatchItem>) -> AddReplyRx {
    let (reply, reply_rx) = oneshot::channel();
    tx.try_send(RegistryCommand::AddBatch { items, reply })
        .expect("registry command channel must stay open");
    reply_rx
}

fn send_remove(tx: &mpsc::Sender<RegistryCommand>, id: TaskId) -> RemoveReplyRx {
    let (reply, reply_rx) = oneshot::channel();
    tx.try_send(RegistryCommand::Remove { id, reply })
        .expect("registry command channel must stay open");
    reply_rx
}

fn send_cancel(tx: &mpsc::Sender<RegistryCommand>, id: TaskId) -> CancelReplyRx {
    let (reply, reply_rx) = oneshot::channel();
    tx.try_send(RegistryCommand::Cancel { id, reply })
        .expect("registry command channel must stay open");
    reply_rx
}

async fn receive_reply<T>(reply: oneshot::Receiver<T>, name: &str) -> T {
    tokio::time::timeout(Duration::from_secs(2), reply)
        .await
        .unwrap_or_else(|_| panic!("{name} timed out"))
        .unwrap_or_else(|_| panic!("{name} sender was dropped"))
}

async fn receive_completion(
    completion_rx: &mut mpsc::UnboundedReceiver<TaskId>,
    name: &str,
) -> TaskId {
    tokio::time::timeout(Duration::from_secs(2), completion_rx.recv())
        .await
        .unwrap_or_else(|_| panic!("{name} timed out"))
        .unwrap_or_else(|| panic!("{name} channel was closed"))
}

async fn stop_registry(registry: &Registry, token: &CancellationToken) {
    token.cancel();
    tokio::time::timeout(Duration::from_secs(2), registry.join_listener())
        .await
        .expect("registry listener must stop");
}

#[tokio::test(flavor = "current_thread")]
async fn add_reply_commits_state_without_event_confirmation() {
    use crate::{TaskContext, TaskFn, TaskRef};
    use tokio::sync::broadcast::error::TryRecvError;

    let (registry, bus, token, tx) = started_registry(1, Duration::from_secs(1));
    let mut stale_events = bus.subscribe();
    let id = TaskId::next();
    let task: TaskRef = TaskFn::arc("reply-add", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Ok(())
    });

    let reply = send_add(&tx, id, TaskSpec::restartable(task), None);
    assert!(
        receive_reply(reply, "add reply").await.is_ok(),
        "registry must accept a unique task"
    );
    assert!(
        registry.contains(id).await,
        "reply requires committed id state"
    );
    assert_eq!(
        registry.id_for_label("reply-add").await,
        Some(id),
        "reply requires committed label state"
    );
    assert_eq!(registry.list().await, vec![(id, Arc::from("reply-add"))]);

    for _ in 0..4 {
        bus.publish(Event::new(EventKind::TaskStarting).with_task("noise"));
    }
    assert!(
        matches!(stale_events.try_recv(), Err(TryRecvError::Lagged(_))),
        "the observer must lag in this regression setup"
    );
    assert!(
        registry.contains(id).await,
        "event lag must not change the authoritative add result"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
async fn single_add_publishes_added_before_starting() {
    use crate::{TaskFn, TaskRef};

    const TASKS: usize = 4_096;

    let bus = Bus::new(TASKS * 8);
    let mut events = bus.subscribe();
    let token = CancellationToken::new();
    let (_tx, rx) = mpsc::channel(1);
    let registry = Registry::new(
        bus,
        token.clone(),
        None,
        Duration::from_secs(1),
        TaskDefaults::default(),
        rx,
    );
    registry.clone().spawn_listener();

    let mut registrations = tokio::task::JoinSet::new();
    for index in 0..TASKS {
        let registry = Arc::clone(&registry);
        registrations.spawn(async move {
            let id = TaskId::next();
            let task: TaskRef =
                TaskFn::arc(format!("ordered-add-{index}"), |_ctx| async { Ok(()) });
            let (reply, reply_rx) = oneshot::channel();
            registry
                .spawn_and_register(id, TaskSpec::once(task), None, None, reply)
                .await;
            assert!(
                matches!(reply_rx.await, Ok(Ok(()))),
                "single registration must succeed"
            );
            id
        });
    }

    while let Some(result) = registrations.join_next().await {
        result.expect("registration worker must not panic");
    }
    tokio::time::timeout(Duration::from_secs(5), registry.wait_until_empty())
        .await
        .expect("all one-shot actors must be reaped");

    let mut order = std::collections::HashMap::<TaskId, (Option<usize>, Option<usize>)>::new();
    for position in 0.. {
        let Ok(event) = events.try_recv() else {
            break;
        };
        let Some(id) = event.id else {
            continue;
        };
        let entry = order.entry(id).or_default();
        match event.kind {
            EventKind::TaskAdded => entry.0 = Some(position),
            EventKind::TaskStarting => {
                entry.1.get_or_insert(position);
            }
            _ => {}
        }
    }

    assert_eq!(order.len(), TASKS, "every registration must be observed");
    for (id, (added, starting)) in order {
        let added = added.unwrap_or_else(|| panic!("{id} is missing TaskAdded"));
        let starting = starting.unwrap_or_else(|| panic!("{id} is missing TaskStarting"));
        assert!(
            added < starting,
            "{id} delivered TaskStarting before TaskAdded: added={added}, starting={starting}"
        );
    }

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn batch_reply_commits_every_task_as_one_registry_decision() {
    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let mut expected = Vec::new();
    let mut items = Vec::new();
    for label in ["batch-a", "batch-b", "batch-c"] {
        let id = TaskId::next();
        let task: TaskRef = TaskFn::arc(label, |ctx: TaskContext| async move {
            ctx.cancelled().await;
            Ok(())
        });
        expected.push((id, Arc::from(label)));
        items.push(batch_item(id, TaskSpec::restartable(task)));
    }
    expected.sort_by_key(|(id, _)| *id);

    let result = receive_reply(send_batch(&tx, items), "batch add reply").await;
    assert!(result.is_ok(), "unique batch must be accepted: {result:?}");
    assert_eq!(registry.list().await, expected);

    let added: Vec<_> = std::iter::from_fn(|| events.try_recv().ok())
        .filter(|event| event.kind == EventKind::TaskAdded)
        .collect();
    assert_eq!(added.len(), 3);

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn dropped_batch_reply_still_starts_after_all_added_events() {
    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let (body_tx, mut body_rx) = mpsc::unbounded_channel();
    let mut items = Vec::new();
    for label in ["dropped-batch-a", "dropped-batch-b"] {
        let id = TaskId::next();
        let body_tx = body_tx.clone();
        let task: TaskRef = TaskFn::arc(label, move |_ctx: TaskContext| {
            let _ = body_tx.send(id);
            async { Ok(()) }
        });
        items.push(batch_item(id, TaskSpec::once(task)));
    }
    drop(body_tx);

    let reply = send_batch(&tx, items);
    drop(reply);
    let first = receive_completion(&mut body_rx, "first batch body").await;
    let second = receive_completion(&mut body_rx, "second batch body").await;
    assert_ne!(first, second);
    tokio::time::timeout(Duration::from_secs(2), registry.wait_until_empty())
        .await
        .expect("both one-shot batch tasks must finish");

    let observed: Vec<_> = std::iter::from_fn(|| events.try_recv().ok()).collect();
    let added: Vec<_> = observed
        .iter()
        .filter(|event| event.kind == EventKind::TaskAdded)
        .collect();
    let starting: Vec<_> = observed
        .iter()
        .filter(|event| event.kind == EventKind::TaskStarting)
        .collect();
    assert_eq!(added.len(), 2);
    assert_eq!(starting.len(), 2);
    let last_added = added.iter().map(|event| event.seq).max().unwrap();
    let first_starting = starting.iter().map(|event| event.seq).min().unwrap();
    assert!(
        last_added < first_starting,
        "the batch start gate must keep bodies behind all TaskAdded events"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn duplicate_inside_batch_rejects_every_item_without_starting_bodies() {
    use std::sync::atomic::{AtomicUsize, Ordering};

    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let runs = Arc::new(AtomicUsize::new(0));
    let mut items = Vec::new();
    let mut ids = Vec::new();
    for label in ["unique", "duplicate", "duplicate"] {
        let runs = Arc::clone(&runs);
        let task: TaskRef = TaskFn::arc(label, move |_ctx: TaskContext| {
            runs.fetch_add(1, Ordering::SeqCst);
            async { Ok(()) }
        });
        let id = TaskId::next();
        ids.push(id);
        items.push(batch_item(id, TaskSpec::once(task)));
    }

    let result = receive_reply(send_batch(&tx, items), "duplicate batch reply").await;
    assert!(
        matches!(
            result,
            Err(RuntimeError::TaskAlreadyExists { ref name }) if name.as_ref() == "duplicate"
        ),
        "the first conflicting input label must reject the batch: {result:?}"
    );
    assert!(registry.list().await.is_empty());
    assert_eq!(runs.load(Ordering::SeqCst), 0);

    let observed: Vec<_> = std::iter::from_fn(|| events.try_recv().ok()).collect();
    assert_eq!(
        observed
            .iter()
            .filter(|event| event.kind == EventKind::TaskAdded)
            .count(),
        0
    );
    let failed: Vec<_> = observed
        .into_iter()
        .filter(|event| event.kind == EventKind::TaskAddFailed)
        .collect();
    assert_eq!(failed.len(), 3);
    assert_eq!(failed[0].id, Some(ids[0]));
    assert_eq!(failed[0].reason.as_deref(), Some(reasons::BATCH_REJECTED));
    assert_eq!(failed[1].id, Some(ids[1]));
    assert_eq!(failed[1].reason.as_deref(), Some(reasons::BATCH_REJECTED));
    assert_eq!(failed[2].id, Some(ids[2]));
    assert_eq!(failed[2].reason.as_deref(), Some(reasons::ALREADY_EXISTS));

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn batch_conflict_with_registered_or_removing_label_starts_no_new_body() {
    use std::sync::atomic::{AtomicUsize, Ordering};

    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, _bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let controlled = controlled_cancellation_task("reserved-batch-name");
    let existing_id = TaskId::next();
    assert!(
        receive_reply(
            send_add(
                &tx,
                existing_id,
                TaskSpec::restartable(controlled.task),
                None,
            ),
            "existing add reply",
        )
        .await
        .is_ok()
    );
    tokio::time::timeout(Duration::from_secs(2), controlled.started.notified())
        .await
        .expect("the existing task body must start before removal");

    let candidate_runs = Arc::new(AtomicUsize::new(0));
    let make_candidate = |label: &'static str| {
        let runs = Arc::clone(&candidate_runs);
        let task: TaskRef = TaskFn::arc(label, move |_ctx: TaskContext| {
            runs.fetch_add(1, Ordering::SeqCst);
            async { Ok(()) }
        });
        batch_item(TaskId::next(), TaskSpec::once(task))
    };

    let registered_result = receive_reply(
        send_batch(
            &tx,
            vec![
                make_candidate("registered-peer"),
                make_candidate("reserved-batch-name"),
            ],
        ),
        "registered conflict batch",
    )
    .await;
    assert!(matches!(
        registered_result,
        Err(RuntimeError::TaskAlreadyExists { name })
            if name.as_ref() == "reserved-batch-name"
    ));
    assert_eq!(candidate_runs.load(Ordering::SeqCst), 0);
    assert_eq!(
        registry.list().await,
        vec![(existing_id, Arc::from("reserved-batch-name"))]
    );

    assert!(matches!(
        receive_reply(send_remove(&tx, existing_id), "existing remove reply").await,
        Ok(true)
    ));
    tokio::time::timeout(
        Duration::from_secs(2),
        controlled.cancellation_seen.notified(),
    )
    .await
    .expect("the existing task must enter Removing");

    let removing_result = receive_reply(
        send_batch(
            &tx,
            vec![
                make_candidate("removing-peer"),
                make_candidate("reserved-batch-name"),
            ],
        ),
        "removing conflict batch",
    )
    .await;
    assert!(matches!(
        removing_result,
        Err(RuntimeError::TaskAlreadyExists { name })
            if name.as_ref() == "reserved-batch-name"
    ));
    assert_eq!(candidate_runs.load(Ordering::SeqCst), 0);
    assert_eq!(
        registry.list().await,
        vec![(existing_id, Arc::from("reserved-batch-name"))]
    );

    controlled.release.notify_one();
    tokio::time::timeout(Duration::from_secs(2), registry.wait_until_empty())
        .await
        .expect("the existing removing task must finish");
    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn duplicate_add_reply_rejects_without_starting_body() {
    use std::sync::atomic::{AtomicUsize, Ordering};

    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, _bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let first_id = TaskId::next();
    let first: TaskRef = TaskFn::arc("duplicate", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Ok(())
    });
    assert!(
        receive_reply(
            send_add(&tx, first_id, TaskSpec::restartable(first), None),
            "first add reply",
        )
        .await
        .is_ok()
    );

    let runs = Arc::new(AtomicUsize::new(0));
    let duplicate_runs = Arc::clone(&runs);
    let duplicate: TaskRef = TaskFn::arc("duplicate", move |_ctx: TaskContext| {
        duplicate_runs.fetch_add(1, Ordering::SeqCst);
        async { Ok(()) }
    });
    let second_id = TaskId::next();
    let (outcome, outcome_rx) = oneshot::channel();
    let duplicate_reply = receive_reply(
        send_add(&tx, second_id, TaskSpec::once(duplicate), Some(outcome)),
        "duplicate add reply",
    )
    .await;

    assert!(
        matches!(
            duplicate_reply,
            Err(RuntimeError::TaskAlreadyExists { name }) if name.as_ref() == "duplicate"
        ),
        "duplicate add must return its authoritative rejection"
    );
    assert!(!registry.contains(second_id).await);
    assert_eq!(registry.id_for_label("duplicate").await, Some(first_id));
    assert_eq!(runs.load(Ordering::SeqCst), 0, "rejected body must not run");
    assert!(matches!(
        receive_reply(outcome_rx, "duplicate outcome").await,
        TaskOutcome::Rejected { reason } if reason.as_ref() == reasons::ALREADY_EXISTS
    ));

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn remove_reply_claims_once_before_terminal_completion() {
    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let controlled = controlled_cancellation_task("remove-once");
    let id = TaskId::next();
    assert!(
        receive_reply(
            send_add(&tx, id, TaskSpec::restartable(controlled.task), None),
            "setup add reply",
        )
        .await
        .is_ok()
    );
    while events.try_recv().is_ok() {}

    assert!(
        matches!(
            receive_reply(send_remove(&tx, id), "first remove reply").await,
            Ok(true)
        ),
        "the first remove must claim the task"
    );
    tokio::time::timeout(
        Duration::from_secs(2),
        controlled.cancellation_seen.notified(),
    )
    .await
    .expect("the task must observe cancellation");
    assert!(registry.pending_joins.contains(id));
    assert!(
        registry.contains(id).await,
        "a removing task must keep its registry identity"
    );
    assert_eq!(
        registry.list().await,
        vec![(id, Arc::from("remove-once"))],
        "a removing task must stay visible in registry listings"
    );
    assert_eq!(registry.id_for_label("remove-once").await, Some(id));
    let mut empty = Box::pin(registry.wait_until_empty());
    assert_pending_once(empty.as_mut()).await;
    drop(empty);
    while let Ok(event) = events.try_recv() {
        assert_ne!(
            event.kind,
            EventKind::TaskRemoved,
            "remove reply must not wait for or invent terminal completion"
        );
    }

    assert!(
        matches!(
            receive_reply(send_remove(&tx, id), "second remove reply").await,
            Ok(false)
        ),
        "a second remove cannot claim the same task"
    );

    let joined_cancel = receive_reply(send_cancel(&tx, id), "joined cancel reply")
        .await
        .expect("the cancel command must succeed")
        .expect("the removing task must expose its completion");
    assert!(
        !joined_cancel.claimed,
        "cancel must join an existing Remove instead of claiming again"
    );
    assert!(
        !joined_cancel.is_complete(),
        "joining cancellation cannot complete before the actor join"
    );

    controlled.release.notify_one();
    tokio::time::timeout(Duration::from_secs(2), joined_cancel.wait())
        .await
        .expect("joined cancellation must finish with the Remove owner");
    tokio::time::timeout(
        Duration::from_secs(2),
        registry.pending_joins.wait_drained(),
    )
    .await
    .expect("the released task must finish its join");
    assert!(!registry.contains(id).await);
    assert_eq!(registry.id_for_label("remove-once").await, None);
    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn concurrent_cancel_commands_share_one_terminal_completion() {
    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(5));
    let mut events = bus.subscribe();
    let controlled = controlled_cancellation_task("shared-cancel");
    let id = TaskId::next();
    assert!(
        receive_reply(
            send_add(&tx, id, TaskSpec::restartable(controlled.task), None),
            "shared cancel add reply",
        )
        .await
        .is_ok()
    );
    while events.try_recv().is_ok() {}

    const CALLERS: usize = 8;
    let replies: Vec<_> = (0..CALLERS).map(|_| send_cancel(&tx, id)).collect();
    let mut decisions = Vec::with_capacity(CALLERS);
    for reply in replies {
        decisions.push(
            receive_reply(reply, "concurrent cancel reply")
                .await
                .expect("cancel command must succeed")
                .expect("the task must still be removing"),
        );
    }
    tokio::time::timeout(
        Duration::from_secs(2),
        controlled.cancellation_seen.notified(),
    )
    .await
    .expect("the task must observe one cancellation");

    assert_eq!(
        decisions.iter().filter(|decision| decision.claimed).count(),
        1,
        "exactly one cancellation command may claim the task"
    );
    assert!(decisions.iter().all(|decision| !decision.is_complete()));
    assert!(registry.contains(id).await);
    assert!(
        std::iter::from_fn(|| events.try_recv().ok())
            .all(|event| event.id != Some(id) || event.kind != EventKind::TaskRemoved),
        "terminal cleanup cannot happen before the task is released"
    );

    controlled.release.notify_one();
    for decision in &decisions {
        tokio::time::timeout(Duration::from_secs(2), decision.wait())
            .await
            .expect("all cancel callers must share terminal completion");
    }
    tokio::time::timeout(Duration::from_secs(2), registry.wait_until_empty())
        .await
        .expect("terminal cleanup must remove the task");
    let removed = std::iter::from_fn(|| events.try_recv().ok())
        .filter(|event| event.id == Some(id) && event.kind == EventKind::TaskRemoved)
        .count();
    assert_eq!(
        removed, 1,
        "shared cancellation must publish one terminal event"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn removing_task_keeps_label_reserved_until_terminal_join() {
    use std::sync::atomic::{AtomicUsize, Ordering};

    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, _bus, token, tx) = started_registry(64, Duration::from_secs(5));
    let controlled = controlled_cancellation_task("reserved-name");
    let first_id = TaskId::next();
    assert!(
        receive_reply(
            send_add(&tx, first_id, TaskSpec::restartable(controlled.task), None,),
            "reserved-name add reply",
        )
        .await
        .is_ok()
    );
    assert!(matches!(
        receive_reply(send_remove(&tx, first_id), "reserved-name remove reply").await,
        Ok(true)
    ));
    tokio::time::timeout(
        Duration::from_secs(2),
        controlled.cancellation_seen.notified(),
    )
    .await
    .expect("the old task must observe cancellation");

    let duplicate_runs = Arc::new(AtomicUsize::new(0));
    let runs_by_task = Arc::clone(&duplicate_runs);
    let duplicate: TaskRef = TaskFn::arc("reserved-name", move |_ctx: TaskContext| {
        runs_by_task.fetch_add(1, Ordering::SeqCst);
        async { Ok(()) }
    });
    let duplicate_id = TaskId::next();
    let duplicate_reply = receive_reply(
        send_add(&tx, duplicate_id, TaskSpec::once(duplicate), None),
        "removing duplicate add reply",
    )
    .await;
    assert!(
        matches!(
            duplicate_reply,
            Err(RuntimeError::TaskAlreadyExists { name })
                if name.as_ref() == "reserved-name"
        ),
        "a removing task must keep its label reserved"
    );
    assert_eq!(
        duplicate_runs.load(Ordering::SeqCst),
        0,
        "a rejected replacement body must not run"
    );
    assert_eq!(registry.id_for_label("reserved-name").await, Some(first_id));
    assert_eq!(
        registry.list().await,
        vec![(first_id, Arc::from("reserved-name"))]
    );
    let mut empty = Box::pin(registry.wait_until_empty());
    assert_pending_once(empty.as_mut()).await;
    drop(empty);

    controlled.release.notify_one();
    tokio::time::timeout(Duration::from_secs(2), registry.wait_until_empty())
        .await
        .expect("terminal join must release the old task identity");
    assert_eq!(registry.id_for_label("reserved-name").await, None);
    assert!(!registry.pending_joins.contains(first_id));

    let replacement: TaskRef = TaskFn::arc("reserved-name", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Ok(())
    });
    let replacement_id = TaskId::next();
    assert!(
        receive_reply(
            send_add(
                &tx,
                replacement_id,
                TaskSpec::restartable(replacement),
                None,
            ),
            "replacement add reply",
        )
        .await
        .is_ok(),
        "the label must be reusable after terminal cleanup"
    );
    assert_eq!(
        registry.id_for_label("reserved-name").await,
        Some(replacement_id)
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn unknown_remove_replies_false_without_pending_join() {
    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let unknown = TaskId::next();

    assert!(
        matches!(
            receive_reply(send_remove(&tx, unknown), "unknown remove reply").await,
            Ok(false)
        ),
        "unknown remove must return false"
    );
    assert!(matches!(
        receive_reply(
            send_remove(&tx, TaskId::next()),
            "unknown remove barrier reply",
        )
        .await,
        Ok(false)
    ));
    assert!(
        registry.pending_joins.is_empty(),
        "unknown removal must not leak pending join state"
    );
    assert!(
        std::iter::from_fn(|| events.try_recv().ok())
            .all(|event| event.id != Some(unknown) || event.kind != EventKind::TaskRemoved),
        "unknown removal must not invent a terminal event"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn dropped_add_reply_does_not_stop_command_processing() {
    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let first_id = TaskId::next();
    let first: TaskRef = TaskFn::arc("dropped-add-a", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Ok(())
    });
    drop(send_add(&tx, first_id, TaskSpec::restartable(first), None));

    let second_id = TaskId::next();
    let second: TaskRef = TaskFn::arc("dropped-add-b", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Ok(())
    });
    assert!(
        receive_reply(
            send_add(&tx, second_id, TaskSpec::restartable(second), None),
            "second add reply",
        )
        .await
        .is_ok()
    );

    assert!(registry.contains(first_id).await);
    assert!(registry.contains(second_id).await);
    let mut added = 0;
    while let Ok(event) = events.try_recv() {
        if event.kind == EventKind::TaskAdded {
            added += 1;
        }
    }
    assert_eq!(added, 2, "a dropped reply must not suppress TaskAdded");

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn dropped_remove_reply_does_not_skip_join_cleanup() {
    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let controlled = controlled_cancellation_task("dropped-remove");
    let id = TaskId::next();
    assert!(
        receive_reply(
            send_add(&tx, id, TaskSpec::restartable(controlled.task), None),
            "setup add reply",
        )
        .await
        .is_ok()
    );
    while events.try_recv().is_ok() {}

    drop(send_remove(&tx, id));
    assert!(
        matches!(
            receive_reply(
                send_remove(&tx, TaskId::next()),
                "synchronizing remove reply",
            )
            .await,
            Ok(false)
        ),
        "the listener must process commands after a dropped reply"
    );
    tokio::time::timeout(
        Duration::from_secs(2),
        controlled.cancellation_seen.notified(),
    )
    .await
    .expect("dropped receiver must not suppress cancellation");

    controlled.release.notify_one();
    tokio::time::timeout(
        Duration::from_secs(2),
        registry.pending_joins.wait_drained(),
    )
    .await
    .expect("dropped receiver must not suppress join cleanup");
    let mut saw_removed = false;
    while let Ok(event) = events.try_recv() {
        if event.kind == EventKind::TaskRemoved && event.id == Some(id) {
            saw_removed = true;
        }
    }
    assert!(saw_removed, "join cleanup must still publish TaskRemoved");

    stop_registry(&registry, &token).await;
}

#[tokio::test(start_paused = true)]
async fn wait_joins_within_reports_stuck_labels_then_drains() {
    let reg = registry();

    assert!(
        reg.wait_joins_within(Duration::from_millis(50))
            .await
            .is_empty(),
        "an empty join set must drain immediately"
    );

    let id = TaskId::next();
    reg.pending_joins.inc(id);
    reg.pending_joins.label(id, Arc::from("stuck-task"));
    let stuck = reg.wait_joins_within(Duration::from_millis(30)).await;
    assert_eq!(
        stuck,
        vec![Arc::<str>::from("stuck-task")],
        "an in-flight join must be reported with its label on timeout"
    );

    let mut draining = Box::pin(reg.wait_joins_within(Duration::from_secs(1)));
    assert_pending_once(draining.as_mut()).await;
    reg.pending_joins.dec(id);
    assert!(
        draining.await.is_empty(),
        "must drain once the in-flight join is decremented"
    );
}

#[tokio::test(flavor = "current_thread")]
async fn completion_guard_signals_on_panic_and_abort_before_first_poll() {
    use std::sync::atomic::{AtomicBool, Ordering};

    let (completion_tx, mut completion_rx) = mpsc::unbounded_channel();

    let panic_id = TaskId::next();
    let panic_handle = Registry::spawn_tracked_actor(panic_id, completion_tx.clone(), async move {
        panic!("outer actor panic")
    });
    let panic_result = panic_handle.await;
    assert!(
        panic_result.is_err_and(|error| error.is_panic()),
        "outer actor panic must stay visible through JoinError"
    );
    assert_eq!(
        receive_completion(&mut completion_rx, "panic completion").await,
        panic_id
    );

    let polled = Arc::new(AtomicBool::new(false));
    let polled_by_task = Arc::clone(&polled);
    let abort_id = TaskId::next();
    let abort_handle = Registry::spawn_tracked_actor(abort_id, completion_tx, async move {
        polled_by_task.store(true, Ordering::SeqCst);
        std::future::pending::<()>().await;
        ActorExitReason::Completed
    });
    abort_handle.abort();
    let abort_result = abort_handle.await;
    assert!(
        abort_result.is_err_and(|error| error.is_cancelled()),
        "aborted actor must return a cancelled JoinError"
    );
    assert!(
        !polled.load(Ordering::SeqCst),
        "the abort regression requires abort-before-first-poll"
    );
    assert_eq!(
        receive_completion(&mut completion_rx, "abort completion").await,
        abort_id
    );
    assert!(
        completion_rx.try_recv().is_err(),
        "each actor exit must send one completion identity"
    );
}

#[tokio::test(flavor = "current_thread")]
async fn natural_completion_cleans_registry_when_event_observer_lags() {
    use crate::{TaskContext, TaskFn, TaskRef};
    use tokio::sync::broadcast::error::TryRecvError;

    let (registry, bus, token, tx) = started_registry(1, Duration::from_secs(1));
    let mut stale_events = bus.subscribe();
    let task: TaskRef = TaskFn::arc("completion-no-bus", |_ctx: TaskContext| async { Ok(()) });
    let id = TaskId::next();
    let (outcome, outcome_rx) = oneshot::channel();

    assert!(
        receive_reply(
            send_add(&tx, id, TaskSpec::once(task), Some(outcome)),
            "fast add reply",
        )
        .await
        .is_ok()
    );
    tokio::time::timeout(Duration::from_secs(2), registry.wait_until_empty())
        .await
        .expect("completion channel must remove the finished task");
    assert!(
        registry
            .wait_joins_within(Duration::from_secs(2))
            .await
            .is_empty()
    );
    assert!(matches!(
        receive_reply(outcome_rx, "fast task outcome").await,
        TaskOutcome::Completed
    ));
    assert_eq!(registry.id_for_label("completion-no-bus").await, None);
    assert!(
        matches!(stale_events.try_recv(), Err(TryRecvError::Lagged(_))),
        "the observer must lose terminal events in this regression setup"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn forged_terminal_event_does_not_remove_running_actor() {
    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let _observer = bus.subscribe();
    assert_eq!(
        bus.receiver_count(),
        1,
        "the registry listener must not subscribe to the event bus"
    );
    let id = TaskId::next();
    let task: TaskRef = TaskFn::arc("ignore-terminal-event", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Ok(())
    });
    assert!(
        receive_reply(
            send_add(&tx, id, TaskSpec::restartable(task), None),
            "running add reply",
        )
        .await
        .is_ok()
    );

    bus.publish(
        Event::new(EventKind::ActorExhausted)
            .with_task("ignore-terminal-event")
            .with_id(id),
    );

    let barrier_id = TaskId::next();
    let barrier: TaskRef = TaskFn::arc("event-barrier", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Ok(())
    });
    assert!(
        receive_reply(
            send_add(&tx, barrier_id, TaskSpec::restartable(barrier), None,),
            "barrier add reply",
        )
        .await
        .is_ok()
    );
    assert!(
        registry.contains(id).await,
        "terminal events are observability and cannot trigger cleanup"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn outer_actor_panic_is_reaped_by_completion_channel() {
    let (registry, bus, token, _tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let id = TaskId::next();
    let label: Arc<str> = Arc::from("outer-panic");
    let (done, done_rx) = oneshot::channel();

    let mut state = registry.state.write().await;
    let join =
        Registry::spawn_tracked_actor(id, registry.listener.completion_tx.clone(), async move {
            panic!("outer actor panic")
        });
    state.by_label.insert(Arc::clone(&label), id);
    state.tasks.insert(
        id,
        Entry {
            label: Arc::clone(&label),
            state: EntryState::Registered(Handle {
                join,
                cancel: CancellationToken::new(),
                done: Some(done),
                completion: RemovalCompletion::new(),
            }),
        },
    );
    drop(state);

    assert!(matches!(
        receive_reply(done_rx, "panic outcome").await,
        TaskOutcome::Panicked
    ));
    tokio::time::timeout(Duration::from_secs(2), registry.wait_until_empty())
        .await
        .expect("panicked actor must leave the registry");
    assert!(
        registry
            .wait_joins_within(Duration::from_secs(2))
            .await
            .is_empty()
    );

    let mut actor_dead = 0;
    let mut task_removed = 0;
    while let Ok(event) = events.try_recv() {
        if event.id == Some(id) && event.kind == EventKind::ActorDead {
            actor_dead += 1;
        }
        if event.id == Some(id) && event.kind == EventKind::TaskRemoved {
            task_removed += 1;
        }
    }
    assert_eq!(actor_dead, 1);
    assert_eq!(task_removed, 1);

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn remove_path_owns_cleanup_when_completion_signal_arrives() {
    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(1));
    let mut events = bus.subscribe();
    let controlled = controlled_cancellation_task("remove-completion-race");
    let id = TaskId::next();
    let (done, done_rx) = oneshot::channel();
    assert!(
        receive_reply(
            send_add(&tx, id, TaskSpec::restartable(controlled.task), Some(done),),
            "race add reply",
        )
        .await
        .is_ok()
    );
    while events.try_recv().is_ok() {}

    assert!(matches!(
        receive_reply(send_remove(&tx, id), "race remove reply").await,
        Ok(true)
    ));
    tokio::time::timeout(
        Duration::from_secs(2),
        controlled.cancellation_seen.notified(),
    )
    .await
    .expect("removed task must observe cancellation");
    controlled.release.notify_one();
    assert!(matches!(
        receive_reply(done_rx, "race outcome").await,
        TaskOutcome::Canceled
    ));
    tokio::time::timeout(
        Duration::from_secs(2),
        registry.pending_joins.wait_drained(),
    )
    .await
    .expect("remove-owned join must drain");

    assert!(matches!(
        receive_reply(send_remove(&tx, TaskId::next()), "completion barrier reply",).await,
        Ok(false)
    ));
    let removed_count = std::iter::from_fn(|| events.try_recv().ok())
        .filter(|event| event.id == Some(id) && event.kind == EventKind::TaskRemoved)
        .count();
    assert_eq!(
        removed_count, 1,
        "stale completion signal must not duplicate terminal cleanup"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test(flavor = "current_thread")]
async fn completion_claim_before_remove_emits_one_terminal_event() {
    use crate::{TaskContext, TaskFn, TaskRef};

    let (registry, bus, token, tx) = started_registry(64, Duration::from_secs(5));
    let mut events = bus.subscribe();
    let release = Arc::new(Notify::new());
    let task_release = Arc::clone(&release);
    let task: TaskRef = TaskFn::arc("completion-first", move |_ctx: TaskContext| {
        let release = Arc::clone(&task_release);
        async move {
            release.notified().await;
            Ok(())
        }
    });
    let id = TaskId::next();
    let (done, mut done_rx) = oneshot::channel();
    assert!(
        receive_reply(
            send_add(&tx, id, TaskSpec::once(task), Some(done)),
            "completion-first add reply",
        )
        .await
        .is_ok()
    );
    while events.try_recv().is_ok() {}

    registry
        .listener
        .completion_tx
        .send(id)
        .expect("completion receiver must be open");
    assert!(matches!(
        receive_reply(
            send_remove(&tx, TaskId::next()),
            "completion claim barrier reply",
        )
        .await,
        Ok(false)
    ));
    assert!(registry.pending_joins.contains(id));
    assert!(registry.contains(id).await);
    assert_eq!(registry.id_for_label("completion-first").await, Some(id));
    let mut empty = Box::pin(registry.wait_until_empty());
    assert_pending_once(empty.as_mut()).await;
    drop(empty);
    assert!(matches!(
        receive_reply(send_remove(&tx, id), "completion-first remove reply").await,
        Ok(false)
    ));
    let joined_cancel = receive_reply(send_cancel(&tx, id), "completion-first cancel reply")
        .await
        .expect("the cancel command must succeed")
        .expect("the completion-owned removal must still exist");
    assert!(
        !joined_cancel.claimed,
        "cancel must join the completion-plane owner"
    );
    assert!(!joined_cancel.is_complete());
    assert!(
        std::iter::from_fn(|| events.try_recv().ok())
            .all(|event| event.id != Some(id) || event.kind != EventKind::TaskRemoved),
        "remove must not report termination while cleanup is still joining"
    );

    release.notify_one();
    tokio::time::timeout(Duration::from_secs(2), joined_cancel.wait())
        .await
        .expect("cancel must finish with the completion-plane owner");
    assert!(
        matches!(done_rx.try_recv(), Ok(TaskOutcome::Completed)),
        "watched outcome must be ready before terminal completion is signalled"
    );
    tokio::time::timeout(Duration::from_secs(2), registry.wait_until_empty())
        .await
        .expect("completion-owned join must finish registry cleanup");
    assert!(!registry.pending_joins.contains(id));

    assert!(matches!(
        receive_reply(
            send_remove(&tx, TaskId::next()),
            "duplicate completion barrier reply",
        )
        .await,
        Ok(false)
    ));
    let removed_count = std::iter::from_fn(|| events.try_recv().ok())
        .filter(|event| event.id == Some(id) && event.kind == EventKind::TaskRemoved)
        .count();
    assert_eq!(
        removed_count, 1,
        "completion-first race must publish one terminal event"
    );

    stop_registry(&registry, &token).await;
}

#[tokio::test]
async fn shutdown_drains_buffered_command_and_never_silently_drops() {
    use crate::{TaskContext, TaskError, TaskFn, TaskRef};

    let bus = Bus::new(64);
    let token = CancellationToken::new();
    let (tx, rx) = mpsc::channel(1);
    let reg = Registry::new(
        bus,
        token.clone(),
        None,
        Duration::from_millis(50),
        TaskDefaults::default(),
        rx,
    );

    let task: TaskRef = TaskFn::arc("buffered", |ctx: TaskContext| async move {
        ctx.cancelled().await;
        Err(TaskError::Canceled)
    });
    let (done_tx, done_rx) = oneshot::channel();
    let (reply_tx, reply_rx) = oneshot::channel();
    let id = TaskId::next();
    tx.try_send(RegistryCommand::Add {
        id,
        spec: TaskSpec::restartable(task),
        outcome: Some(done_tx),
        completion: None,
        reply: reply_tx,
    })
    .expect("channel is open before shutdown");

    token.cancel();
    reg.clone().spawn_listener();
    tokio::time::timeout(Duration::from_secs(2), reg.join_listener())
        .await
        .expect("join_listener must not hang");

    let reply = tokio::time::timeout(Duration::from_secs(1), reply_rx)
        .await
        .expect("buffered Add reply must resolve")
        .expect("buffered Add reply sender must not be dropped");
    assert!(
        reply.is_ok(),
        "buffered Add must be registered before drain"
    );

    let outcome = tokio::time::timeout(Duration::from_secs(1), done_rx)
        .await
        .expect("watcher must resolve")
        .expect("watcher sender must not be dropped — the buffered Add must be acted on");
    assert!(
        matches!(outcome, TaskOutcome::Canceled | TaskOutcome::ForceAborted),
        "a buffered task drained at shutdown must terminate, got {outcome:?}"
    );

    assert!(
        reg.pending_joins.is_empty(),
        "wait_drained must leave no in-flight joins after shutdown"
    );

    let (reply, _reply_rx) = oneshot::channel();
    assert!(
        tx.try_send(RegistryCommand::Remove {
            id: TaskId::next(),
            reply,
        })
        .is_err(),
        "after shutdown the command channel is closed; sends must return Err"
    );
}