runnerd-core 0.1.5

Config, domain logic and API types shared by runnerd and its clients
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
//! A replay by a `Trace`: each run uses what its samples say, is known when the daemon knew it, and the pooled count
//! reads what the daemon's would at that moment.

use super::*;
use crate::admission::pooled::{self, MOST_CREDIT};
use crate::admission::trace::{Outside, Point, Reading, RunId, RunTrace, STEP, Trace};
use crate::admission::{Brake, GrowthCount, GuardHold};

/// A slice with 1.5G left of two 3.5G reservations and the 512M mirror, under the mark with room for what the credit
/// adds; a host with memory to spare.
const SLICE: u64 = 8 * G + G / 2;

/// The memory of `check` and `site`, 3.5G each once known, as they use it.
const USED: u64 = G / 2;

/// A sample of `memory` every `STEP` for `minutes`.
fn flat(memory: u64, minutes: u64) -> Vec<Point> {
    (0..minutes * MINUTE / STEP)
        .map(|step| Point {
            at: step * STEP,
            memory,
        })
        .collect()
}

fn check(queued: u64, minutes: u64) -> Ran {
    run("halyard/ledger", "check", "jvm", queued, minutes, 3 * G)
}

fn site(queued: u64, minutes: u64) -> Ran {
    run("halyard/storefront", "site", "node", queued, minutes, 3 * G)
}

fn page(queued: u64, minutes: u64) -> Ran {
    run("halyard/storefront", "page", "node", queued, minutes, 2 * G)
}

/// Ten earlier runs of each job, so the replay predicts them from their own.
fn known() -> History {
    let before = NOW - 900 * MINUTE;
    let jobs = (0..10)
        .flat_map(|i| {
            let at = before + i * 40 * MINUTE;
            [check(at, 30), site(at, 30), page(at, 5)]
        })
        .collect();
    History {
        jobs,
        ..History::default()
    }
}

/// `check` and `site` from `NOW` for half an hour, and `page` from `page_at`.
fn jobs(page_at: u64) -> Vec<Ran> {
    let mut jobs = vec![
        in_run(check(NOW, 30), 1, 1),
        in_run(site(NOW, 30), 2, 1),
        in_run(page(page_at, 5), 3, 1),
    ];
    jobs.sort_by_key(|run| run.queued_at);
    jobs
}

/// Every one of `jobs` using `USED` from the start, on a host that keeps well clear of its guards.
fn trace_of(jobs: &[Ran]) -> Trace {
    let runs = jobs
        .iter()
        .map(|run| {
            let trace = RunTrace {
                samples: flat(USED, 40),
                named_after: 0,
            };
            (RunId::of(run), trace)
        })
        .collect();
    Trace {
        runs,
        outside: vec![
            Outside {
                at: NOW - MINUTE,
                memory: 8 * G,
            },
            Outside {
                at: NOW + 60 * MINUTE,
                memory: 8 * G,
            },
        ],
        total: 64 * G,
        overhead: 0,
        guard_overhead: 0,
        host_overhead: 0,
    }
}

fn replay(jobs: &[Ran], trace: &Trace, most_credit: u64) -> Replay {
    replay_held(jobs, trace, most_credit, SLICE, Band::NoWaiterFits)
}

/// A replay under `memory_max`, whose larger margin `band` holds up.
fn replay_held(
    jobs: &[Ran],
    trace: &Trace,
    most_credit: u64,
    memory_max: u64,
    band: Band,
) -> Replay {
    let known = known();
    simulate(
        &config(),
        Some(memory_max),
        None,
        jobs,
        Policy::Measured,
        &BTreeMap::new(),
        Replaying {
            barriers: Barriers::On,
            arrivals: Arrivals::AsQueued,
            known: Some(&known),
            trace: Some(trace),
            most_credit,
            host_reserve: HostReserve::Top2,
            credit_room: CreditRoom::Counted,
            band,
            growth: None,
        },
    )
}

/// How long `page` waited, in seconds.
fn page_waited(replay: &Replay) -> u64 {
    let pool = pool("halyard/storefront", "node");
    *replay.pools[&pool].waits.last().unwrap()
}

#[test]
fn a_pooled_replay_starts_what_only_the_pooled_count_fits() {
    let jobs = jobs(NOW + 5 * MINUTE);
    let trace = trace_of(&jobs);
    let plain = replay(&jobs, &trace, 0);
    assert_eq!(page_waited(&plain), 25 * 60, "until one of the two ends");
    assert_eq!(plain.hazard.most_credit, 0);
    let pooled = replay(&jobs, &trace, MOST_CREDIT);
    assert!(page_waited(&pooled) <= 15, "at the next plan");
    let (check, site, page) = (Key::of(&jobs[0]), Key::of(&jobs[1]), Key::of(&jobs[2]));
    let runner = |key, takes| pooled::Runner {
        key: Some(key),
        takes,
        used: Reading::Used(USED),
        ran: Some(pooled::POOL_AFTER),
        basis: Some(Basis::Job {
            runs: 10,
            peak: JobPeak::P95,
        }),
    };
    let direct = pooled::count(&pooled::Snapshot {
        runners: &[
            runner(&check, 3584 * M),
            runner(&site, 3584 * M),
            runner(&page, 2560 * M),
        ],
        committed: 0,
        slice: Some(pooled::Slice {
            memory: 3 * USED,
            memory_max: Some(SLICE),
        }),
        host: Some(pooled::Host {
            available: 50 * G,
            keeps: 20 * G,
            stalled: false,
        }),
        most: MOST_CREDIT,
        host_reserve: true,
        growth: None,
    });
    assert_eq!(
        pooled.hazard.most_credit, direct.credit,
        "the replay's count is the daemon's"
    );
}

#[test]
fn a_replayed_run_counts_in_full_for_its_first_minute() {
    let jobs = jobs(NOW + 20_000);
    let trace = trace_of(&jobs);
    let waited = page_waited(&replay(&jobs, &trace, MOST_CREDIT));
    assert!(
        (40..=60).contains(&waited),
        "from 20 s to a plan past the minute, waited {waited}"
    );
}

#[test]
fn a_replayed_run_is_not_pooled_before_the_daemon_could_name_its_job() {
    let jobs = jobs(NOW + 5 * MINUTE);
    let mut trace = trace_of(&jobs);
    let check = RunId::of(&jobs[0]);
    trace.runs.get_mut(&check).unwrap().named_after = 10 * MINUTE;
    let waited = page_waited(&replay(&jobs, &trace, MOST_CREDIT));
    assert!(
        (300..=320).contains(&waited),
        "until check is known at minute 10, waited {waited}"
    );
}

#[test]
fn a_sampler_gap_takes_the_credit_back_and_the_replay_says_so() {
    let jobs = jobs(NOW + 5 * MINUTE);
    let mut trace = trace_of(&jobs);
    let check = RunId::of(&jobs[0]);
    let run = trace.runs.get_mut(&check).unwrap();
    run.samples
        .retain(|point| !(8 * MINUTE..9 * MINUTE).contains(&point.at));
    let hazard = replay(&jobs, &trace, MOST_CREDIT).hazard;
    assert!(hazard.credit_falls >= 1, "the gap took it back: {hazard:?}");
}

#[test]
fn a_credit_the_host_takes_back_leaves_the_count_past_the_budget_and_the_replay_says_so() {
    let jobs = jobs(NOW + 5 * MINUTE);
    let mut trace = trace_of(&jobs);
    // At minute 8, with page running in what the credit left, the desktop takes what the host kept.
    trace.outside.insert(
        1,
        Outside {
            at: NOW + 8 * MINUTE,
            memory: 43 * G,
        },
    );
    let hazard = replay(&jobs, &trace, MOST_CREDIT).hazard;
    assert_eq!(hazard.overdrawn, 1, "{hazard:?}");
}

#[test]
fn a_host_short_of_memory_keeps_a_replay_from_pooling() {
    let jobs = jobs(NOW + 5 * MINUTE);
    let mut trace = trace_of(&jobs);
    for reading in &mut trace.outside {
        // 21G available, under the margin and a quarter of the host with the two runners' 1G.
        reading.memory = 43 * G;
    }
    let replay = replay(&jobs, &trace, MOST_CREDIT);
    assert!(replay.hazard.most_credit < G, "{:?}", replay.hazard);
    assert_eq!(page_waited(&replay), 25 * 60);
}

/// `jobs` on a 64G host whose other programs leave the runners `limit` by its plain margin of 4G, 2G less by the larger.
fn host_limited(jobs: &[Ran], limit: u64) -> Trace {
    let mut trace = trace_of(jobs);
    for reading in &mut trace.outside {
        reading.memory = 60 * G - limit;
    }
    trace
}

/// A job of a repository with no runs yet: it reserves what its profile's jobs do, 3.5G, more than these hosts leave.
fn report(queued: u64) -> Ran {
    run("halyard/other", "report", "jvm", queued, 30, 3 * G)
}

/// `check` from `NOW`, beside which a 7G limit leaves 3G; `report` waiting from minute 5; `page`, 2.5G, from minute 10,
/// which fits the plain margin and not the larger.
fn one_job_never_fits() -> (Vec<Ran>, Trace) {
    let jobs = vec![
        in_run(check(NOW, 30), 1, 1),
        in_run(report(NOW + 5 * MINUTE), 2, 1),
        in_run(page(NOW + 10 * MINUTE, 5), 3, 1),
    ];
    let trace = host_limited(&jobs, 7 * G);
    (jobs, trace)
}

#[test]
fn a_job_no_margin_fits_holds_up_the_larger_margin_until_a_waiter_fits_the_plain_one() {
    let (jobs, trace) = one_job_never_fits();
    let replay = replay_held(&jobs, &trace, 0, 40 * G, Band::NoWaiterFits);
    let waited = page_waited(&replay);
    assert!(
        (1..=30).contains(&waited),
        "refused by the larger margin, then it fits at the next plan, waited {waited}"
    );
    assert!(
        replay.hazard.in_band >= 20 * MINUTE,
        "the margin stood from minute 5 on: {:?}",
        replay.hazard
    );
}

#[test]
fn any_job_waiting_for_the_host_holds_up_the_larger_margin_when_a_replay_is_asked_to() {
    let (jobs, trace) = one_job_never_fits();
    let replay = replay_held(&jobs, &trace, 0, 40 * G, Band::AnyWaiter);
    assert_eq!(
        page_waited(&replay),
        50 * 60,
        "report starts into the empty slice at minute 30, and while anything waits the margin stands, so page waits it out"
    );
    assert!(
        replay.hazard.in_band >= 25 * MINUTE,
        "the margin stood from minute 5 on: {:?}",
        replay.hazard
    );
}

#[test]
fn a_replay_starts_at_the_plain_margin() {
    let jobs = vec![in_run(check(NOW, 30), 1, 1), in_run(page(NOW, 5), 2, 1)];
    let trace = host_limited(&jobs, 7 * G);
    let replay = replay_held(&jobs, &trace, 0, 40 * G, Band::NoWaiterFits);
    assert_eq!(
        page_waited(&replay),
        0,
        "page fits beside check by the plain margin at the first plan"
    );
}

#[test]
fn the_first_runner_into_an_empty_slice_needs_only_the_plain_margin() {
    let jobs = vec![
        in_run(check(NOW, 30), 1, 1),
        in_run(site(NOW + 5 * MINUTE, 30), 2, 1),
    ];
    // Beside the mirror a 5G limit leaves 4.5G by the plain margin, past site's 3.5G, and 2.5G by the larger.
    let trace = host_limited(&jobs, 5 * G);
    let replay = replay_held(&jobs, &trace, 0, 40 * G, Band::NoWaiterFits);
    assert_eq!(
        page_waited(&replay),
        25 * 60,
        "the plan that finds the slice empty starts it, though the larger margin stood"
    );
}

#[test]
fn a_host_credits_fall_is_judged_by_the_margin_that_stands() {
    let jobs = vec![
        in_run(check(NOW, 30), 1, 1),
        in_run(site(NOW, 30), 2, 1),
        in_run(page(NOW, 30), 3, 1),
        in_run(report(NOW + 3 * MINUTE), 4, 1),
    ];
    // Page comes to use all it takes at minute 8, which ends the host's 2G credit: 10G counted, up from 8G, under the
    // plain limit of 11G and past the 9G that stands while report waits.
    let mut trace = host_limited(&jobs, 11 * G);
    let page_run = trace.runs.get_mut(&RunId::of(&jobs[2])).unwrap();
    for point in &mut page_run.samples {
        if point.at >= 8 * MINUTE {
            point.memory = 2560 * M;
        }
    }
    let hazard = replay_held(&jobs, &trace, 0, 40 * G, Band::NoWaiterFits).hazard;
    assert_eq!(hazard.overdrawn, 1, "{hazard:?}");
}

#[test]
fn a_replays_credit_keeps_clear_of_the_larger_margin_while_it_stands() {
    let jobs = vec![
        in_run(check(NOW, 30), 1, 1),
        in_run(site(NOW, 30), 2, 1),
        in_run(report(NOW), 3, 1),
    ];
    // A 24G host with 10G available beside the two runners, and report waiting from the first plan: the credit is
    // what is left past the larger margin of 3G and a quarter of the host, 1G of the 1.75G their use would allow.
    let mut trace = trace_of(&jobs);
    trace.total = 24 * G;
    for reading in &mut trace.outside {
        reading.memory = 13 * G;
    }
    let hazard = replay_held(&jobs, &trace, MOST_CREDIT, 40 * G, Band::NoWaiterFits).hazard;
    assert_eq!(hazard.most_credit, G, "{hazard:?}");
}

#[test]
fn a_replay_finds_the_slice_past_its_lines_and_how_long() {
    let jobs = jobs(NOW + 40 * MINUTE);
    let mut trace = trace_of(&jobs);
    let check = RunId::of(&jobs[0]);
    let run = trace.runs.get_mut(&check).unwrap();
    for point in &mut run.samples {
        if (10 * MINUTE..11 * MINUTE).contains(&point.at) {
            point.memory = 9 * G;
        }
    }
    let hazard = replay(&jobs, &trace, 0).hazard;
    assert_eq!(hazard.over_mark.count, 1, "{hazard:?}");
    assert_eq!(hazard.over_max.count, 1, "{hazard:?}");
    assert_eq!(hazard.over_max.time, MINUTE);
    assert_eq!(hazard.over_max.worst, 9 * G + USED - SLICE);
    assert_eq!(hazard.slice_peak, 9 * G + USED);
}

#[test]
fn a_cold_window_gives_every_run_its_jobs_heaviest_samples() {
    let first = in_run(check(NOW, 30), 1, 1);
    let second = in_run(check(NOW + MINUTE, 30), 2, 1);
    let other = in_run(site(NOW, 30), 3, 1);
    let jobs = [first, second, other];
    let mut trace = trace_of(&jobs);
    trace.runs.get_mut(&RunId::of(&jobs[1])).unwrap().samples = flat(4 * G, 1);
    let cold = trace.cold(&jobs);
    let peak = |at: usize| cold.runs[&RunId::of(&jobs[at])].at(0);
    assert_eq!(peak(0), Reading::Used(4 * G), "the job's heaviest run");
    assert_eq!(peak(1), Reading::Used(4 * G));
    assert_eq!(peak(2), Reading::Used(USED), "another job keeps its own");
}

#[test]
fn a_cold_run_shorter_than_the_heaviest_reaches_its_peak_in_the_time_it_has() {
    // A job's short runs are most of them: a peak past their end would leave the stress with nothing to stress.
    let long = in_run(check(NOW, 30), 1, 1);
    let short = in_run(check(NOW + MINUTE, 5), 2, 1);
    let jobs = [long, short];
    let mut trace = trace_of(&jobs);
    let heavy = &mut trace.runs.get_mut(&RunId::of(&jobs[0])).unwrap().samples;
    heavy
        .iter_mut()
        .find(|point| point.at == 20 * MINUTE)
        .unwrap()
        .memory = 4 * G;
    let cold = trace.cold(&jobs);
    let lasted = jobs[1].finished_at.unwrap() - jobs[1].started_at.unwrap();
    let run = &cold.runs[&RunId::of(&jobs[1])];
    let at = run.samples.iter().find(|point| point.memory == 4 * G);
    assert!(
        at.is_some_and(|point| point.at < lasted),
        "the peak falls inside its {lasted} ms: {at:?}"
    );
    assert_eq!(run.at(0), Reading::Used(USED), "it reads from its start");
}

#[test]
fn a_shifted_host_keeps_every_reading_and_comes_round_the_window() {
    let trace = Trace {
        outside: [(0, 1), (10, 2), (20, 3), (30, 4)]
            .map(|(at, memory)| Outside { at, memory })
            .to_vec(),
        ..Trace::default()
    };
    let shifted = trace.shifted(10);
    assert_eq!(shifted.outside_at(0), Some(4), "what ended the window");
    assert_eq!(shifted.outside_at(10), Some(1));
    assert_eq!(shifted.outside_at(30), Some(3));
    assert_eq!(shifted.outside.len(), 4);
}

#[test]
fn a_slice_that_reads_low_at_a_decision_goes_past_the_mark_it_did_not_see() {
    // The hazard is read at what the slice holds, the guards at what a sample showed them.
    let jobs = jobs(NOW + 40 * MINUTE);
    let mut trace = trace_of(&jobs);
    trace.overhead = 8 * G;
    trace.guard_overhead = 0;
    let replay = replay(&jobs, &trace, 0);
    assert!(replay.hazard.over_mark.count > 0, "{:?}", replay.hazard);
    assert_eq!(replay.held_back, 0, "the guards saw room");
}

#[test]
fn a_run_between_samples_reads_the_last_and_a_gap_reads_nothing() {
    let run = RunTrace {
        samples: vec![
            Point {
                at: 1000,
                memory: G,
            },
            Point {
                at: 3000,
                memory: 2 * G,
            },
            Point {
                at: 30_000,
                memory: 3 * G,
            },
        ],
        named_after: 0,
    };
    assert_eq!(run.at(0), Reading::NotYet);
    assert_eq!(run.at(2000), Reading::Used(G));
    assert_eq!(run.at(9000), Reading::Used(2 * G), "at the edge of fresh");
    assert_eq!(run.at(9001), Reading::Stale);
    assert_eq!(run.at(90_000), Reading::Used(3 * G), "its end is not a gap");
}

/// A replay of `jobs` as `trace` says they ran, with six slots a profile, under `memory_max`, counted by `growth` or,
/// without it, plainly.
fn replay_counted(
    jobs: &[Ran],
    trace: &Trace,
    memory_max: u64,
    growth: Option<GrowthCount>,
) -> Replay {
    let known = known();
    let slots = BTreeMap::from([("jvm".to_owned(), 6), ("node".to_owned(), 6)]);
    simulate(
        &config(),
        Some(memory_max),
        None,
        jobs,
        Policy::Measured,
        &slots,
        Replaying {
            barriers: Barriers::On,
            arrivals: Arrivals::AsQueued,
            known: Some(&known),
            trace: Some(trace),
            most_credit: 0,
            host_reserve: HostReserve::Plain,
            credit_room: CreditRoom::Counted,
            band: Band::NoWaiterFits,
            growth,
        },
    )
}

const TOP3: GrowthCount = GrowthCount {
    growth: pooled::Growth {
        jobs: 3,
        bound: pooled::Bound::Open,
        young: pooled::Young::Ranked,
        below: pooled::Below::Plain,
        hold: pooled::Hold::Free,
        ceiling: None,
    },
    brake: Brake::Off,
    guard: GuardHold::Off,
};

const TOP3_BRAKED: GrowthCount = GrowthCount {
    brake: Brake::On,
    ..TOP3
};

/// Four `check` runners and four `site` runners from `NOW` for `minutes`, 3.5G reserved each, and `page`, 2.5G, from
/// minute 12: the runner sets the growth count's replays read.
fn eight_and_a_page(minutes: u64) -> Vec<Ran> {
    let mut jobs: Vec<Ran> = (0..4)
        .flat_map(|i| {
            [
                in_run(check(NOW, minutes), 10 + i, 1),
                in_run(site(NOW, minutes), 20 + i, 1),
            ]
        })
        .collect();
    jobs.push(in_run(page(NOW + 12 * MINUTE, 5), 30, 1));
    jobs
}

/// `jobs` each using `USED` until minute 10 and `grown` after.
fn growing(jobs: &[Ran], grown: u64) -> Trace {
    let mut trace = trace_of(jobs);
    for run in trace.runs.values_mut() {
        run.samples = (0..70 * MINUTE / STEP)
            .map(|step| Point {
                at: step * STEP,
                memory: if step * STEP < 10 * MINUTE {
                    USED
                } else {
                    grown
                },
            })
            .collect();
    }
    trace
}

/// How long the last job of `pool` waited, in seconds.
fn last_waited(replay: &Replay, repository: &str, profile: &str) -> u64 {
    *replay.pools[&pool(repository, profile)]
        .waits
        .last()
        .unwrap()
}

#[test]
fn the_growth_count_starts_what_only_it_fits() {
    // Eight runners using 0.5G each: the plain count fits four of the eight's 3.5G reservations in 15G, the rest waiting
    // for the first to end. Under the growth count a start is all growth: kept among the three largest, the fourth and
    // later add nothing, so the eight start at once. Once they use 0.5G each, the count is 0.5G + 4G + the three largest
    // growths, 13.5G: page's 2.5G does not fit the 1.5G left before it starts, and it waits for the first to end. As in
    // the daemon's plan, the room before a start must fit its whole take, and the count is read again after it.
    let jobs = eight_and_a_page(30);
    let trace = trace_of(&jobs);
    let plain = replay_counted(&jobs, &trace, 15 * G, None);
    let counted = replay_counted(&jobs, &trace, 15 * G, Some(TOP3));
    let at_once = |replay: &Replay| {
        replay
            .pools
            .values()
            .flat_map(|pool| &pool.waits)
            .filter(|waited| **waited <= 15)
            .count()
    };
    assert_eq!(at_once(&plain), 4, "{:?}", plain.pools);
    assert_eq!(at_once(&counted), 8, "{:?}", counted.pools);
    assert_eq!(last_waited(&counted, "halyard/storefront", "node"), 18 * 60);
    assert_eq!(counted.hazard.braked, 0, "no brake to engage");
}

#[test]
fn the_brake_holds_starts_to_the_plain_count_once_what_runs_outgrows_its_room_and_then_releases() {
    // At minute 10 each runner grows 1.3G, 10.4G together, past the 9G the count kept: page, at minute 12, fits the
    // growth count (0.5G + 14.4G + 6G kept) but not the plain one. Every look for the next 10 minutes still has a look
    // from before the growth in its window, so the brake stands until 30 minutes after the last of them, minute 50.
    let jobs = eight_and_a_page(60);
    let trace = growing(&jobs, 1800 * M);
    let free = replay_counted(&jobs, &trace, 24 * G, Some(TOP3));
    assert!(
        last_waited(&free, "halyard/storefront", "node") <= 15,
        "without the brake it starts at once"
    );
    let braked = replay_counted(&jobs, &trace, 24 * G, Some(TOP3_BRAKED));
    let waited = last_waited(&braked, "halyard/storefront", "node");
    assert!(
        (38 * 60 - 15..=38 * 60 + 15).contains(&waited),
        "from minute 12 to the brake's end at minute 50, waited {waited}"
    );
    assert!(
        (40 * MINUTE - 30_000..=40 * MINUTE + 30_000).contains(&braked.hazard.braked),
        "it stood from minute 10 to 50, then released: {:?}",
        braked.hazard
    );
}

#[test]
fn a_host_under_its_line_keeps_the_growth_count_plain() {
    let jobs = eight_and_a_page(30);
    let mut trace = trace_of(&jobs);
    for reading in &mut trace.outside {
        // 64G less 44G outside leaves just under 20G, the 4G margin and the 16G quarter, with no runner yet.
        reading.memory = 44 * G + 1;
    }
    let counted = replay_counted(&jobs, &trace, 24 * G, Some(TOP3));
    let plain = replay_counted(&jobs, &trace, 24 * G, None);
    assert_eq!(
        counted.pools, plain.pools,
        "the same starts as the plain count"
    );
    assert!(counted.hazard.guarded > 0, "{:?}", counted.hazard);
}

#[test]
fn a_cold_window_engages_the_brake_where_its_warm_one_does_not() {
    // One check run grows to 3.4G at minute 10: within the room the count keeps. Cold, every check run does, and
    // four together outgrow it.
    let jobs = eight_and_a_page(30);
    let mut trace = trace_of(&jobs);
    let heavy = growing(&jobs[..1], 3400 * M);
    trace.runs.extend(heavy.runs);
    let warm = replay_counted(&jobs, &trace, 24 * G, Some(TOP3_BRAKED));
    assert_eq!(warm.hazard.braked, 0, "{:?}", warm.hazard);
    let cold = replay_counted(&jobs, &trace.cold(&jobs), 24 * G, Some(TOP3_BRAKED));
    assert!(cold.hazard.braked > 0, "{:?}", cold.hazard);
}

#[test]
fn a_replayed_runner_counts_its_pools_largest_cpu_until_its_job_is_known() {
    // Site's runner starts while build, 6.5 CPUs, waits in its pool, so it reserves 6.5 CPUs until site is named at
    // minute 10, as the daemon's does; then 1. Report, 2 CPUs, fits beside it only from then.
    let cpu = |run: Ran, average| Ran {
        cpu_average: Some(average),
        ..run
    };
    let build = |queued| {
        cpu(
            run("halyard/storefront", "build", "node", queued, 30, G),
            6500,
        )
    };
    let report = |queued| cpu(run("halyard/ledger", "report", "jvm", queued, 30, G), 2000);
    let mut known = known();
    let before = NOW - 900 * MINUTE;
    known
        .jobs
        .extend((0..10).flat_map(|i| [build(before + i * MINUTE), report(before + i * MINUTE)]));
    known.jobs.sort_by_key(|run| run.queued_at);
    let jobs = vec![
        in_run(site(NOW, 30), 1, 1),
        in_run(build(NOW), 2, 1),
        in_run(report(NOW + MINUTE), 3, 1),
    ];
    let mut trace = trace_of(&jobs);
    trace
        .runs
        .get_mut(&RunId::of(&jobs[0]))
        .unwrap()
        .named_after = 10 * MINUTE;
    let replay = simulate(
        &config(),
        Some(40 * G),
        Some(7000),
        &jobs,
        Policy::Measured,
        &BTreeMap::new(),
        Replaying {
            barriers: Barriers::Off,
            arrivals: Arrivals::AsQueued,
            known: Some(&known),
            trace: Some(&trace),
            most_credit: 0,
            host_reserve: HostReserve::Plain,
            credit_room: CreditRoom::Counted,
            band: Band::NoWaiterFits,
            growth: None,
        },
    );
    let waited = last_waited(&replay, "halyard/ledger", "jvm");
    assert!(
        (9 * 60..=9 * 60 + 15).contains(&waited),
        "from minute 1 to site's name at minute 10, waited {waited}"
    );
}

#[test]
fn a_runner_still_booting_at_a_look_grows_from_nothing_and_engages_the_brake() {
    // Eight runners boot for five minutes before their jobs start, unsampled and so using nothing at the looks then;
    // each job uses 1.8G from its start, 14.4G together, past the 10.5G the count kept for three 3.5G starts. Taking a
    // booting runner's first sample as where it began would hide all of it.
    let booting = |run: Ran| Ran {
        started_at: Some(run.queued_at + 5 * MINUTE),
        ..run
    };
    let jobs: Vec<Ran> = eight_and_a_page(30)
        .into_iter()
        .take(8)
        .map(booting)
        .collect();
    let mut trace = trace_of(&jobs);
    for run in trace.runs.values_mut() {
        run.samples = flat(1800 * M, 40);
    }
    let free = replay_counted(&jobs, &trace, 40 * G, Some(TOP3_BRAKED));
    assert!(free.hazard.braked > 0, "{:?}", free.hazard);
    let steady = replay_counted(&jobs, &trace_of(&jobs), 40 * G, Some(TOP3_BRAKED));
    assert_eq!(
        steady.hazard.braked, 0,
        "0.5G each is within the room: {:?}",
        steady.hazard
    );
}

/// The daemon knows a runner's job, and so when it usually ends, only once it can name it: until then the runner may
/// end at any moment as far as a backfill knows, so nothing passes by its run time (ADR 0097), and a job the barrier
/// would let pass waits behind it rather than for the host (ADR 0151). A replay that knew more would start jobs, and
/// drop the host's larger margin for them, where the daemon does not. Two node runners end at minute 20; the 9G owner
/// holds from minute 11; a 3-minute docs job, too long to pass as short, queues at minute 12.
#[test]
fn a_replay_reads_no_usual_end_of_a_runner_whose_job_has_no_name_yet() {
    let (mut jobs, mut known) = a_barrier_with_room_behind_it();
    jobs.truncate(3);
    let docs = |queued, minutes| run("halyard/lint", "docs", "light", queued, minutes, G / 2);
    known
        .jobs
        .extend((0..3).map(|i| docs(NOW - 600 * MINUTE + i * 40 * MINUTE, 3)));
    jobs.push(docs(NOW + 12 * MINUTE, 3));
    let docs_waited = |named_after| {
        let runs = jobs
            .iter()
            .filter(|run| run.name == "site")
            .map(|run| {
                let trace = RunTrace {
                    samples: flat(G / 2, 20),
                    named_after,
                };
                (RunId::of(run), trace)
            })
            .collect();
        let trace = Trace {
            runs,
            outside: Vec::new(),
            total: 64 * G,
            overhead: 0,
            guard_overhead: 0,
            host_overhead: 0,
        };
        let replay = simulate(
            &with_light(),
            Some(17 * G),
            None,
            &jobs,
            Policy::Caps,
            &BTreeMap::new(),
            Replaying {
                barriers: Barriers::On,
                arrivals: Arrivals::AsQueued,
                known: Some(&known),
                trace: Some(&trace),
                most_credit: 0,
                host_reserve: HostReserve::Top2,
                credit_room: CreditRoom::Counted,
                band: Band::NoWaiterFits,
                growth: None,
            },
        );
        replay.pools[&pool("halyard/lint", "light")].waits.clone()
    };
    assert_eq!(
        docs_waited(0),
        [0],
        "named at once, the runners end at minute 20"
    );
    let unnamed = docs_waited(15 * MINUTE);
    assert!(
        unnamed[0] > 0,
        "unnamed until minute 15, nothing is known of their end: {unnamed:?}"
    );
}

/// The replay and the daemon's plan start the same jobs under an active growth credit: both fit a runner by the room
/// the count leaves before it starts, count again after it, and count a runner that starts as having no job nor sample
/// yet. Eight 3.5G jobs queue beside the 512M mirror, on a host with room to spare; each start is all growth. In a 13G
/// slice the count after three starts is 11G and leaves 2G, so both start three: fitting a start by the count with it
/// already in, as the replay once did, would let all eight in, as a fourth adds nothing past the three largest. In
/// 15G it leaves 4G after any number, so both start all eight: reading each start's first 0.5G sample at once, as the
/// replay once did, would leave 3.5G after the fifth and start five.
#[test]
fn the_replay_starts_what_the_daemons_plan_starts_under_the_growth_count() {
    for (memory_max, started) in [(13 * G, 3), (15 * G, 8)] {
        let planned = planned_starts(memory_max);
        assert_eq!(
            replayed_starts(memory_max),
            planned,
            "in {}G",
            memory_max / G
        );
        assert_eq!(planned.iter().sum::<usize>(), started);
    }
}

/// How many jobs of `jvm` and of `node` the replay of `eight_and_a_page` starts at `NOW` in a slice of `memory_max`.
fn replayed_starts(memory_max: u64) -> Vec<usize> {
    let jobs = eight_and_a_page(30);
    let trace = trace_of(&jobs);
    let replay = replay_counted(&jobs, &trace, memory_max, Some(TOP3));
    ["jvm", "node"]
        .iter()
        .map(|profile| {
            replay
                .pools
                .iter()
                .filter(|(pool, _)| pool.profile == *profile)
                .flat_map(|(_, pool)| &pool.waits)
                .filter(|waited| **waited == 0)
                .count()
        })
        .collect()
}

/// How many jobs of `jvm` and of `node` the daemon's plan starts at once in a slice of `memory_max`, reading what the
/// replay reads at `NOW`.
fn planned_starts(memory_max: u64) -> Vec<usize> {
    use crate::admission::host;
    use crate::api;
    use crate::scheduler::{self, Budget, Profile, Recount, Slot, SlotRef};

    // The replay's reading at `NOW`: nothing runs, the host's other programs hold 8G of 64G.
    let memory = api::HostMemory {
        at: NOW / 1000,
        total: 64 * G,
        available: 56 * G,
        slice: 0,
        some: None,
        full: None,
        holders: Vec::new(),
        ceiling: None,
        swap: None,
    };
    let remembered = host::Remembered::new(host::HostBand::Plain, host::Occupancy::Occupied);
    let snapshot = pooled::Snapshot {
        runners: &[],
        committed: G / 2,
        slice: Some(pooled::Slice {
            memory: 0,
            memory_max: Some(memory_max),
        }),
        host: Some(host::pooling(&memory, remembered)),
        most: 0,
        host_reserve: false,
        growth: Some(TOP3.growth),
    };
    let tally = pooled::Tally::<SlotRef>::new(&snapshot, []);
    let count = tally.count();
    let limit = host::limit(&memory, false);
    let budget = Budget {
        memory: Some(i64::try_from(memory_max - count.pooled()).unwrap()),
        host: Some(host::left(limit, count.hosted(), false)),
        recount: Some(Recount {
            tally,
            memory_max: Some(memory_max),
            host: Some(scheduler::HostLimit {
                limit,
                stalled: false,
            }),
        }),
        ..Budget::default()
    };
    let config = config();
    let profiles: BTreeMap<String, Profile> = ["jvm", "node"]
        .iter()
        .map(|name| {
            let caps = &config.profiles[*name];
            let profile = Profile {
                memory: caps.memory.0,
                cpus: 0,
                preparing: None,
                upkeep_due: false,
                count: 6,
                ran: Vec::new(),
                slots: vec![Slot::Free; 6],
            };
            ((*name).to_owned(), profile)
        })
        .collect();
    let repositories = [
        parity_pool("halyard/ledger", "jvm", "jvm", 10),
        parity_pool("halyard/storefront", "node", "node", 11),
    ];
    let plan = scheduler::plan(
        &repositories,
        &profiles,
        budget,
        std::time::UNIX_EPOCH + Duration::from_millis(NOW),
    );
    ["jvm", "node"]
        .iter()
        .map(|profile| {
            plan.decisions
                .iter()
                .filter(|decision| {
                    matches!(decision, Decision::Start { slot, .. } if slot.profile == *profile)
                })
                .count()
        })
        .collect()
}

/// The pool `name` of `profile` as the plan reads it: four 3.5G jobs of `label` queued at `NOW`, from `first` by twos.
fn parity_pool(name: &str, profile: &str, label: &str, first: u64) -> Repository {
    Repository {
        name: name.to_owned(),
        profile: profile.to_owned(),
        labels: vec![label.to_owned()],
        paused: false,
        held: None,
        // Queued at once, check and site by turns as `eight_and_a_page` lists them.
        queued: (0..4)
            .map(|turn| first + 2 * turn)
            .map(|id| QueuedJob {
                id,
                run_id: id,
                labels: vec![label.to_owned()],
                created_at: NOW,
                ..QueuedJob::default()
            })
            .collect(),
        runner_memory: Some(3 * G + G / 2),
        runner_cpu: Some(0),
        waits: Waits::default(),
        usual: None,
        warm: 0,
    }
}