second-brain-api 0.13.5

HTTP API server for second-brain: REST endpoints for recall, remember, and ingest
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
//! The drain thread: applies the index outbox every few seconds and, when
//! the outbox is quiet and the interval has passed, repairs the store with
//! the writable indexes it owns. On-demand repairs arrive over a channel so
//! the same thread services them; two writers never exist.

use std::path::PathBuf;
use std::sync::mpsc::{Receiver, RecvTimeoutError, Sender};
use std::sync::{Arc, PoisonError, RwLock};
use std::time::{Duration, Instant};

use anyhow::{Result, anyhow};
use chrono::Utc;
use second_brain_core::config::MaintenanceConfig;
use second_brain_core::embedding::Embedder;
use second_brain_core::index::{IndexClaim, Indexes};
use second_brain_core::maintenance::{MaintenanceSummary, RepairStats};
use second_brain_core::residency::{self, IndexMemory};
use second_brain_core::sealed::{SealKeyId, SealKeys};
use second_brain_core::sqlite_store::{AtRestPurpose, SqliteStore};

use crate::health::Health;
use crate::near_duplicates::{NearDupCount, refresh_count};

const DRAIN_INTERVAL: Duration = Duration::from_secs(5);
/// Peers' deletes ask for a lexical merge, which rewrites every segment,
/// so a burst of them is served by one merge.
const DELETE_PURGE_GAP: Duration = Duration::from_secs(600);

const WRITER_IDLE: Duration = Duration::from_secs(60);
/// How often the drain thread asks the store whether the index files need
/// sealing: after the store's first encryption, or a store key rotation.
const SEAL_CHECK_INTERVAL: Duration = Duration::from_secs(60);

/// Finding chunk work splits every live row, so once a scan comes back
/// empty-handed the next one waits this long unless a drain applies writes.
const CHUNK_IDLE_RESCAN: Duration = Duration::from_secs(600);
/// Per idle tick, so the drain and repair behind it are never held up long.
const CHUNK_TICK_BUDGET: Duration = Duration::from_secs(3);
/// Bounds what one tick holds in memory: at most 16 chunks per memory.
const CHUNK_MEMORIES_PER_TICK: usize = 32;

/// When the drain thread may next look for memories needing chunks.
struct ChunkThrottle {
    next_scan: Instant,
}

impl ChunkThrottle {
    fn new(now: Instant) -> Self {
        Self { next_scan: now }
    }

    fn due(&self, now: Instant) -> bool {
        now >= self.next_scan
    }

    fn after_backfill(&mut self, now: Instant, progressed: bool) {
        self.next_scan = if progressed {
            now
        } else {
            now + CHUNK_IDLE_RESCAN
        };
    }

    /// Applied outbox rows may carry content changes that need re-chunking.
    fn after_drain(&mut self, now: Instant, applied: usize) {
        if applied > 0 {
            self.next_scan = now;
        }
    }
}

pub struct RepairRequest {
    pub dry_run: bool,
    pub reply: Sender<Result<MaintenanceSummary>>,
}

/// Work only the drain thread may do, because it owns the writable indexes.
pub enum DrainRequest {
    Repair(RepairRequest),
    /// Applies the whole outbox, then rewrites the lexical index into fresh
    /// segments (see `Indexes::compact`).
    CompactIndexes(Sender<Result<()>>),
    /// Holds the drain thread, so nothing writes the index files, until the
    /// sender of these commands is dropped.
    Pause(Receiver<PauseCommand>),
}

pub enum PauseCommand {
    /// Applies the whole outbox and lets the lexical writer's merges finish,
    /// so the index files on disk match the store at its watermark.
    Settle(Sender<Result<()>>),
}

/// Runs a repair and re-measures the store afterwards, on whatever index
/// handle the caller owns (the writer on the drain thread, none when the
/// indexes never opened or a released writer could not reopen).
pub fn run_repair(
    store: &SqliteStore,
    mut indexes: Option<&mut Indexes>,
    embedder: Option<&Embedder>,
    dry_run: bool,
) -> Result<MaintenanceSummary> {
    let embed = embedder.map(|e| move |text: &str| e.embed(text));
    let embed_ref = embed
        .as_ref()
        .map(|f| f as &dyn Fn(&str) -> Result<Vec<f32>>);
    let stats = store.repair(indexes.as_deref_mut(), embed_ref, dry_run)?;
    let report_after = store.integrity_report(indexes.as_deref())?;
    Ok(MaintenanceSummary {
        ran_at: Utc::now(),
        stats,
        report_after,
    })
}

fn log_stats(source: &str, stats: &RepairStats) {
    if stats.is_clean() {
        tracing::info!("{source} store repair: nothing to do");
    } else {
        tracing::info!("{source} store repair: {}", summarize(stats));
    }
}

fn summarize(s: &RepairStats) -> String {
    let mut parts = Vec::new();
    let mut push = |name: &str, n: u64| {
        if n > 0 {
            parts.push(format!("{name}={n}"));
        }
    };
    push("duplicates_superseded", s.duplicates_superseded);
    push("markers_inherited", s.distillation_markers_inherited);
    push("orphaned_markers_dropped", s.orphaned_markers_dropped);
    push("raws_retired", s.raws_retired);
    push("relations_pruned", s.relations_pruned);
    push("vec_keys_deleted", s.vec_keys_deleted);
    push("sync_log_rows_deleted", s.sync_log_rows_deleted);
    push("stale_cursors_dropped", s.stale_cursors_dropped);
    push("embeddings_backfilled", s.embeddings_backfilled);
    push("embedding_failures", s.embedding_failures);
    if s.vec_keys_gc_skipped {
        parts.push("vec_keys_gc=skipped".to_string());
    }
    if s.index_rebuilt {
        parts.push("index=rebuilt".to_string());
    }
    if s.dry_run {
        parts.push("dry_run".to_string());
    }
    parts.join(" ")
}

/// The drain thread's writable indexes. usearch cannot mutate a viewed index,
/// so an open writer holds its own anonymous copy of the vectors beside the
/// reader's mmap; the slot drops it once `WRITER_IDLE` has passed since its
/// last operation finished, and reopens it from the persisted files when work
/// arrives. A writer holding changes its files lack is kept, because a drain
/// commits its store-side deletes before it persists, so only that writer
/// still holds those removals.
///
/// Sealed indexes can't be mapped, so recall reads a reader sharing the
/// writer's vectors; the slot then never drops that writer, and an idle one
/// gives up only its lexical writer's buffers.
pub struct WriterSlot {
    dir: PathBuf,
    open: Option<Indexes>,
    last_used: Instant,
    shared: bool,
    idle_released: bool,
    _claim: IndexClaim,
}

impl WriterSlot {
    pub fn new(writer: Indexes, claim: IndexClaim) -> Self {
        crate::heap::record_index_writer(Some(writer.memory()));
        Self {
            dir: writer.dir().to_path_buf(),
            shared: writer.seal_keys().is_some(),
            open: Some(writer),
            last_used: Instant::now(),
            idle_released: false,
            _claim: claim,
        }
    }

    fn open(&mut self) -> Result<&mut Indexes> {
        if self.open.is_none() {
            if self.shared {
                return Err(anyhow!(
                    "the sealed index writer at {} is gone; the daemon must restart to reopen it",
                    self.dir.display()
                ));
            }
            let writer = Indexes::open(&self.dir)?;
            tracing::info!("index writer reopened at {}", self.dir.display());
            crate::heap::record_index_writer(Some(writer.memory()));
            self.open = Some(writer);
        }
        Ok(self.open.as_mut().expect("writer opened above"))
    }

    fn with<R>(&mut self, op: impl FnOnce(&mut Indexes) -> Result<R>) -> Result<R> {
        let result = op(self.open()?);
        self.last_used = Instant::now();
        self.idle_released = false;
        crate::heap::record_index_writer(self.memory());
        result
    }

    fn release_if_idle(&mut self, now: Instant) -> bool {
        let current = self
            .open
            .as_ref()
            .is_some_and(|w| !w.has_unpersisted_changes());
        if !current || now.saturating_duration_since(self.last_used) < WRITER_IDLE {
            return false;
        }
        if self.shared {
            return self.release_lexical_writer();
        }
        if let Some(writer) = self.open.take()
            && let Err(e) = writer.wait_for_merges()
        {
            tracing::warn!("index writer released without its merges: {e:#}");
        }
        crate::heap::record_index_writer(None);
        residency::release_free_heap();
        tracing::info!(
            "index writer released after {}s idle",
            WRITER_IDLE.as_secs()
        );
        true
    }

    fn release_lexical_writer(&mut self) -> bool {
        if self.idle_released {
            return false;
        }
        if let Some(writer) = self.open.as_mut()
            && let Err(e) = writer.close_lexical_writer()
        {
            tracing::warn!("lexical writer released without its merges: {e:#}");
        }
        self.idle_released = true;
        residency::release_free_heap();
        tracing::info!(
            "lexical index writer released after {}s idle",
            WRITER_IDLE.as_secs()
        );
        true
    }

    fn memory(&self) -> Option<IndexMemory> {
        self.open.as_ref().map(Indexes::memory)
    }

    fn sealed_under(&self) -> Option<SealKeyId> {
        self.open
            .as_ref()
            .and_then(Indexes::seal_keys)
            .map(|keys| keys.current_id())
    }

    /// Seals every index file under `keys`' current key. A plaintext writer
    /// is persisted and reopened sealed, and recall moves to a reader that
    /// shares its vectors.
    fn seal_under(&mut self, keys: SealKeys, reader: &RwLock<Indexes>) -> Result<()> {
        if self.shared {
            let writer = self
                .open
                .as_mut()
                .ok_or_else(|| anyhow!("the sealed index writer is gone"))?;
            writer.set_seal_keys(keys)?;
            return writer.reseal();
        }
        if let Some(writer) = self.open.as_mut()
            && writer.has_unpersisted_changes()
        {
            writer.persist()?;
        }
        if let Some(writer) = self.open.take() {
            crate::heap::record_index_writer(None);
            writer.wait_for_merges()?;
        }
        let mut writer = Indexes::open_keyed(&self.dir, &keys)?;
        writer.reseal()?;
        let shared = writer.shared_reader()?;
        *reader.write().unwrap_or_else(PoisonError::into_inner) = shared;
        crate::heap::record_index_writer(Some(writer.memory()));
        self.open = Some(writer);
        self.shared = true;
        self.idle_released = false;
        self.last_used = Instant::now();
        Ok(())
    }
}

pub struct DrainLoop {
    pub store: Arc<SqliteStore>,
    pub reader: Arc<RwLock<Indexes>>,
    pub writer: WriterSlot,
    pub embedder: Option<Arc<Embedder>>,
    pub summary: Arc<RwLock<Option<MaintenanceSummary>>>,
    pub near_duplicates: Arc<RwLock<Option<NearDupCount>>>,
    pub near_duplicate_threshold: f32,
    pub config: MaintenanceConfig,
    pub requests: Receiver<DrainRequest>,
    pub health: Arc<Health>,
}

impl DrainLoop {
    pub fn spawn(self) {
        std::thread::Builder::new()
            .name("index-drain".to_string())
            .spawn(move || self.run())
            .expect("spawning the index drain thread");
    }

    fn run(mut self) {
        let mut last_repair: Option<Instant> = None;
        let mut chunks = ChunkThrottle::new(Instant::now());
        let mut last_purge: Option<Instant> = None;
        let mut last_seal_check: Option<Instant> = None;
        loop {
            match self.requests.recv_timeout(DRAIN_INTERVAL) {
                Ok(DrainRequest::Repair(request)) => {
                    let result = self.repair("requested", request.dry_run);
                    if result.is_ok() && !request.dry_run {
                        last_repair = Some(Instant::now());
                    }
                    // A caller that gave up waiting is not an error here.
                    let _ = request.reply.send(result);
                }
                Ok(DrainRequest::CompactIndexes(reply)) => {
                    let _ = reply.send(self.compact_indexes());
                }
                Ok(DrainRequest::Pause(commands)) => {
                    for PauseCommand::Settle(reply) in commands {
                        let _ = reply.send(self.settle());
                    }
                }
                Err(RecvTimeoutError::Timeout) | Err(RecvTimeoutError::Disconnected) => {
                    let applied = self.drain();
                    chunks.after_drain(Instant::now(), applied.unwrap_or(0));
                    let due = last_repair.is_none_or(|t| {
                        t.elapsed() >= Duration::from_secs(self.config.interval_secs)
                    });
                    if applied == Some(0) && self.config.enabled && due {
                        if self.repair("scheduled", false).is_ok() {
                            last_repair = Some(Instant::now());
                        } else {
                            // Retry sooner than the full interval, but not
                            // every tick against a store that keeps failing.
                            last_repair = Some(
                                Instant::now()
                                    - Duration::from_secs(
                                        self.config.interval_secs.saturating_sub(600),
                                    ),
                            );
                        }
                    }
                    if applied == Some(0)
                        && last_purge.is_none_or(|t| t.elapsed() >= DELETE_PURGE_GAP)
                        && self.purge_applied_deletes()
                    {
                        last_purge = Some(Instant::now());
                    }
                    if applied == Some(0) && chunks.due(Instant::now()) {
                        let progressed = self.backfill_chunks();
                        chunks.after_backfill(Instant::now(), progressed);
                    }
                }
            }
            if last_seal_check.is_none_or(|t| t.elapsed() >= SEAL_CHECK_INTERVAL) {
                self.seal_if_due();
                last_seal_check = Some(Instant::now());
            }
            self.writer.release_if_idle(Instant::now());
        }
    }

    /// Seals the index files when the store's keys for them changed: on the
    /// store's first encryption, after a store key rotation, or when a
    /// reseal was cut short. Retired keys are forgotten only once every file
    /// is under the current one.
    fn seal_if_due(&mut self) {
        let keys = match self.store.at_rest_keys(AtRestPurpose::Indexes) {
            Ok(Some(keys)) => keys,
            Ok(None) => return,
            Err(e) => {
                tracing::warn!("reading the search indexes' keys failed: {e:#}");
                return;
            }
        };
        let current = keys.keys.current_id();
        if keys.settled && self.writer.sealed_under() == Some(current) {
            return;
        }
        let started = Instant::now();
        if let Err(e) = self.writer.seal_under(keys.keys, &self.reader) {
            tracing::warn!("sealing the search indexes failed and will retry: {e:#}");
            return;
        }
        match self
            .store
            .settle_at_rest_keys(AtRestPurpose::Indexes, current)
        {
            Ok(()) => tracing::info!(
                "search indexes sealed under key {current} in {:.1}s",
                started.elapsed().as_secs_f64()
            ),
            Err(e) => tracing::warn!("recording the sealed search indexes failed: {e:#}"),
        }
    }

    /// Embeds pending chunk sets within one tick's budget; true when any set
    /// was stored, so the next idle tick keeps going.
    fn backfill_chunks(&self) -> bool {
        let Some(embedder) = self.embedder.as_deref() else {
            return false;
        };
        let embed = |texts: &[&str]| embedder.embed_batch(texts);
        match self.store.backfill_chunk_embeddings(
            &embed,
            CHUNK_MEMORIES_PER_TICK,
            CHUNK_TICK_BUDGET,
        ) {
            Ok(r) => {
                if r.memories > 0 || r.failed > 0 {
                    tracing::info!(
                        "chunk backfill: memories={} chunks={} failed={} pending={}",
                        r.memories,
                        r.chunks,
                        r.failed,
                        r.pending
                    );
                }
                r.memories > 0
            }
            Err(e) => {
                tracing::warn!("chunk backfill failed: {e:#}");
                false
            }
        }
    }

    /// Applies pending outbox rows; None when the drain itself failed. An
    /// empty outbox never touches the writer, so a released one stays closed.
    fn drain(&mut self) -> Option<usize> {
        let store = &self.store;
        let writer = &mut self.writer;
        let result = store.outbox_pending().and_then(|pending| {
            if pending == 0 {
                Ok(0)
            } else {
                writer.with(|w| w.drain(store))
            }
        });
        match result {
            Ok(0) => {
                self.health.drained(0, Utc::now());
                Some(0)
            }
            Ok(n) => {
                tracing::info!("index drain applied {n} outbox rows");
                self.reload_reader();
                match self.store.outbox_pending() {
                    Ok(pending) => self.health.drained(pending, Utc::now()),
                    Err(e) => tracing::warn!("counting the index outbox failed: {e:#}"),
                }
                Some(n)
            }
            Err(e) => {
                tracing::warn!("index drain failed: {e:#}");
                self.health.drain_failed(&format!("{e:#}"), Utc::now());
                None
            }
        }
    }

    fn repair(&mut self, source: &str, dry_run: bool) -> Result<MaintenanceSummary> {
        let started = Instant::now();
        let store = &self.store;
        let embedder = self.embedder.as_deref();
        let repair = |w: &mut Indexes| run_repair(store, Some(w), embedder, dry_run);
        let result = match self.writer.open().map(|_| ()) {
            Ok(()) => self.writer.with(repair),
            Err(e) => {
                tracing::warn!(
                    "index writer unavailable ({e:#}); repairing the store without the indexes"
                );
                match run_repair(store, None, embedder, dry_run) {
                    Ok(summary) => log_stats(source, &summary.stats),
                    Err(store_error) => {
                        tracing::warn!("{source} store repair failed: {store_error:#}")
                    }
                }
                Err(e
                    .context("the index writer could not reopen, so the indexes were not repaired"))
            }
        };
        match &result {
            Ok(summary) => {
                self.health.repaired(Utc::now());
                log_stats(source, &summary.stats);
                tracing::info!(
                    "{source} store repair finished in {:.1}s; drift remaining: {:?}",
                    started.elapsed().as_secs_f64(),
                    summary.report_after.drift()
                );
                if !dry_run {
                    self.reload_reader();
                    let mut guard = match self.summary.write() {
                        Ok(g) => g,
                        Err(poisoned) => poisoned.into_inner(),
                    };
                    *guard = Some(summary.clone());
                    drop(guard);
                    let threshold = self.near_duplicate_threshold;
                    let near_duplicates = &self.near_duplicates;
                    let counted = self.writer.with(|w| {
                        refresh_count(store, w, threshold, near_duplicates);
                        Ok(())
                    });
                    if let Err(e) = counted {
                        tracing::warn!("near-duplicate count failed: {e:#}");
                    }
                    if summary.stats.index_rebuilt {
                        residency::release_free_heap();
                    }
                }
            }
            Err(e) => {
                tracing::warn!("{source} store repair failed: {e:#}");
                self.health.repair_failed(&format!("{e:#}"), Utc::now());
            }
        }
        result
    }

    /// Runs the lexical merge and cold purge a peer's applied delete asked
    /// for, so its text leaves this disk as it left the deleting one's;
    /// true when it ran.
    fn purge_applied_deletes(&mut self) -> bool {
        let token = match self.store.delete_purge_requested() {
            Ok(Some(token)) => token,
            Ok(None) => return false,
            Err(e) => {
                tracing::warn!("reading the delete purge request failed: {e:#}");
                return false;
            }
        };
        let result = self.compact_indexes().and_then(|()| {
            let cold = self.store.purge_cold_segments()?;
            if cold != Default::default() {
                tracing::info!(
                    "deleted records purged from the cold tier: {} retired, {} segments rewritten, {} files removed",
                    cold.retired,
                    cold.rewritten,
                    cold.removed_files
                );
            }
            self.store.delete_purge_done(&token)
        });
        if let Err(e) = result {
            tracing::warn!("purging a peer's deletes failed and will retry: {e:#}");
        }
        true
    }

    fn compact_indexes(&mut self) -> Result<()> {
        let store = &self.store;
        let took = self.writer.with(|w| {
            while w.drain(store)? > 0 {}
            let started = Instant::now();
            w.compact()?;
            Ok(started.elapsed())
        })?;
        self.reload_reader();
        residency::release_free_heap();
        tracing::info!("lexical index compacted in {:.1}s", took.as_secs_f64());
        Ok(())
    }

    fn settle(&mut self) -> Result<()> {
        let store = &self.store;
        self.writer.with(|w| {
            while w.drain(store)? > 0 {}
            if w.has_unpersisted_changes() {
                w.persist()?;
            }
            w.close_lexical_writer()
        })?;
        self.reload_reader();
        Ok(())
    }

    fn reload_reader(&self) {
        let mut guard = match self.reader.write() {
            Ok(g) => g,
            Err(poisoned) => poisoned.into_inner(),
        };
        if let Err(e) = guard.reload_view() {
            tracing::warn!("index reader reload failed: {e:#}");
        }
    }
}

#[cfg(test)]
mod tests {
    use std::fs::Permissions;
    use std::os::unix::fs::PermissionsExt;

    use super::*;
    use second_brain_core::schema::{Memory, MemoryType};
    use second_brain_core::store::Store;

    #[test]
    fn a_fresh_throttle_scans_on_the_first_idle_tick() {
        let now = Instant::now();
        assert!(ChunkThrottle::new(now).due(now));
    }

    #[test]
    fn a_scan_that_finds_nothing_waits_out_the_idle_interval() {
        let now = Instant::now();
        let mut t = ChunkThrottle::new(now);
        t.after_backfill(now, false);
        assert!(!t.due(now + CHUNK_IDLE_RESCAN - Duration::from_secs(1)));
        assert!(t.due(now + CHUNK_IDLE_RESCAN));
    }

    #[test]
    fn progress_keeps_the_next_idle_tick_scanning() {
        let now = Instant::now();
        let mut t = ChunkThrottle::new(now);
        t.after_backfill(now, true);
        assert!(t.due(now));
    }

    #[test]
    fn an_applied_drain_resets_the_idle_wait() {
        let now = Instant::now();
        let mut t = ChunkThrottle::new(now);
        t.after_backfill(now, false);
        t.after_drain(now, 0);
        assert!(!t.due(now));
        t.after_drain(now, 3);
        assert!(t.due(now));
    }

    fn remember(store: &SqliteStore, content: &str) -> uuid::Uuid {
        let m = Memory::new(
            content.to_string(),
            MemoryType::Semantic,
            "test".to_string(),
            String::new(),
        );
        store.store_memory(&m).unwrap();
        m.id
    }

    fn drain_loop(
        store: Arc<SqliteStore>,
        reader: Arc<RwLock<Indexes>>,
        writer: Indexes,
    ) -> DrainLoop {
        let (_tx, requests) = std::sync::mpsc::channel();
        let claim = IndexClaim::acquire(writer.dir()).unwrap();
        DrainLoop {
            store,
            reader,
            writer: WriterSlot::new(writer, claim),
            embedder: None,
            summary: Arc::new(RwLock::new(None)),
            near_duplicates: Arc::new(RwLock::new(None)),
            near_duplicate_threshold: 0.97,
            config: MaintenanceConfig {
                enabled: false,
                interval_secs: 3600,
            },
            requests,
            health: Arc::new(Health::new(Utc::now(), None)),
        }
    }

    fn idle_loop(dir: &std::path::Path) -> DrainLoop {
        let store = Arc::new(SqliteStore::in_memory("m".to_string()).unwrap());
        let mut writer = Indexes::open(dir).unwrap();
        writer.persist().unwrap();
        let reader = Arc::new(RwLock::new(Indexes::open_readonly(dir).unwrap()));
        drain_loop(store, reader, writer)
    }

    #[test]
    fn idleness_counts_from_when_an_operation_finishes() {
        let dir = tempfile::tempdir().unwrap();
        let mut lp = idle_loop(dir.path());
        let mut finishing = None;
        lp.writer
            .with(|_| {
                finishing = Some(Instant::now());
                Ok(())
            })
            .unwrap();
        let finishing = finishing.unwrap();
        assert!(lp.writer.last_used >= finishing);
        assert!(
            !lp.writer
                .release_if_idle(finishing + WRITER_IDLE - Duration::from_secs(1))
        );
    }

    #[test]
    fn a_writer_holding_changes_its_files_lack_is_never_released() {
        let dir = tempfile::tempdir().unwrap();
        let mut lp = idle_loop(dir.path());
        remember(&lp.store, "the zanzibar ferry leaves at noon");
        let tantivy = dir.path().join("tantivy");
        let set_mode =
            |mode| std::fs::set_permissions(&tantivy, Permissions::from_mode(mode)).unwrap();
        set_mode(0o500);
        let drained = lp.drain();
        set_mode(0o700);
        assert_eq!(
            drained, None,
            "the drain should fail to write the lexical index"
        );
        assert!(
            !lp.writer
                .release_if_idle(lp.writer.last_used + WRITER_IDLE * 10)
        );
        lp.repair("test", true).unwrap();
        assert!(!lp.writer.release_if_idle(lp.writer.last_used + WRITER_IDLE));

        assert_eq!(lp.drain(), Some(1));
        assert!(lp.writer.release_if_idle(lp.writer.last_used + WRITER_IDLE));
    }

    #[test]
    fn a_failure_that_left_the_files_current_still_releases() {
        let dir = tempfile::tempdir().unwrap();
        let mut lp = idle_loop(dir.path());
        let failed: Result<()> = lp.writer.with(|_| anyhow::bail!("a store step failed"));
        assert!(failed.is_err());
        assert!(lp.writer.release_if_idle(lp.writer.last_used + WRITER_IDLE));
    }

    #[test]
    fn a_writer_that_cannot_reopen_fails_the_repair() {
        let dir = tempfile::tempdir().unwrap();
        let mut lp = idle_loop(dir.path());
        assert!(lp.writer.release_if_idle(lp.writer.last_used + WRITER_IDLE));
        let blocker = dir.path().join("not-a-dir");
        std::fs::write(&blocker, b"x").unwrap();
        lp.writer.dir = blocker.join("indexes");

        let err = lp.repair("test", false).unwrap_err();
        assert!(
            format!("{err:#}").contains("indexes were not repaired"),
            "{err:#}"
        );
        assert!(lp.writer.memory().is_none());
    }

    #[test]
    fn an_idle_writer_is_released_and_reopened_by_the_next_write() {
        let dir = tempfile::tempdir().unwrap();
        let store = Arc::new(SqliteStore::in_memory("m".to_string()).unwrap());
        let mut writer = Indexes::open(dir.path()).unwrap();
        writer.persist().unwrap();
        let reader = Arc::new(RwLock::new(Indexes::open_readonly(dir.path()).unwrap()));
        let mut lp = drain_loop(store.clone(), reader.clone(), writer);

        let first = remember(&store, "the zanzibar ferry leaves at noon");
        assert_eq!(lp.drain(), Some(1));
        assert!(lp.writer.memory().is_some());

        let used = lp.writer.last_used;
        assert!(
            !lp.writer
                .release_if_idle(used + WRITER_IDLE - Duration::from_secs(1))
        );
        assert!(lp.writer.memory().is_some());
        assert!(lp.writer.release_if_idle(used + WRITER_IDLE));
        assert!(lp.writer.memory().is_none());

        assert_eq!(lp.drain(), Some(0));
        assert!(
            lp.writer.memory().is_none(),
            "an empty tick reopened the writer"
        );

        let second = remember(&store, "the kilimanjaro trail opens in june");
        assert_eq!(lp.drain(), Some(1));
        assert!(lp.writer.memory().is_some());
        assert_eq!(lp.writer.open().unwrap().lexical_len(), 2);

        let guard = reader.read().unwrap();
        let hits = |q: &str| -> Vec<uuid::Uuid> {
            guard
                .bm25_search(q, 5)
                .unwrap()
                .into_iter()
                .map(|(id, _)| id)
                .collect()
        };
        assert_eq!(hits("kilimanjaro"), vec![second]);
        assert_eq!(hits("zanzibar"), vec![first]);
    }

    fn keyed_store(dir: &std::path::Path) -> Arc<SqliteStore> {
        let key = second_brain_core::keys::StoreKey::generate().unwrap();
        Arc::new(SqliteStore::open_with_key(&dir.join("brain.db"), "m".to_string(), &key).unwrap())
    }

    fn index_files(dir: &std::path::Path) -> Vec<std::path::PathBuf> {
        let mut out = Vec::new();
        let mut stack = vec![dir.to_path_buf()];
        while let Some(d) = stack.pop() {
            for entry in std::fs::read_dir(d).unwrap() {
                let path = entry.unwrap().path();
                let name = path.file_name().unwrap().to_string_lossy().to_string();
                if path.is_dir() {
                    stack.push(path);
                } else if !name.ends_with(".lock") && name != second_brain_core::index::CLAIM_FILE {
                    out.push(path);
                }
            }
        }
        out
    }

    fn assert_all_sealed_under(dir: &std::path::Path, key: SealKeyId) {
        let files = index_files(dir);
        assert!(!files.is_empty());
        for path in files {
            assert_eq!(
                second_brain_core::sealed::sealed_key_id(&path).unwrap(),
                Some(key),
                "{}",
                path.display()
            );
        }
    }

    #[test]
    fn indexes_opened_before_the_store_was_encrypted_are_sealed_and_recall_moves_to_a_shared_reader()
     {
        let data = tempfile::tempdir().unwrap();
        let store = keyed_store(data.path());
        let dir = data.path().join("brain.indexes");
        let mut writer = Indexes::open(&dir).unwrap();
        writer.persist().unwrap();
        let reader = Arc::new(RwLock::new(Indexes::open_readonly(&dir).unwrap()));
        let mut lp = drain_loop(store.clone(), reader.clone(), writer);
        let first = remember(&store, "the zanzibar ferry leaves at noon");
        assert_eq!(lp.drain(), Some(1));

        lp.seal_if_due();
        let keys = store.at_rest_keys(AtRestPurpose::Indexes).unwrap().unwrap();
        assert!(keys.settled);
        assert_all_sealed_under(&dir, keys.keys.current_id());
        assert!(Indexes::open_readonly(&dir).is_err());

        let second = remember(&store, "the kilimanjaro trail opens in june");
        assert_eq!(lp.drain(), Some(1));
        let guard = reader.read().unwrap();
        assert_eq!(guard.bm25_search("zanzibar", 5).unwrap()[0].0, first);
        assert_eq!(guard.bm25_search("kilimanjaro", 5).unwrap()[0].0, second);
        assert_eq!(guard.vector_len(), lp.writer.open().unwrap().vector_len());
    }

    #[test]
    fn a_store_key_rotation_reseals_the_indexes_under_a_new_key() {
        let data = tempfile::tempdir().unwrap();
        let store = keyed_store(data.path());
        let dir = data.path().join("brain.indexes");
        let keys = store
            .at_rest_keys(AtRestPurpose::Indexes)
            .unwrap()
            .unwrap()
            .keys;
        let handles = crate::models::open_index_handles(&dir, &store, Some(&keys), |_| {}).unwrap();
        let crate::models::IndexHandles {
            reader,
            writer,
            claim,
        } = handles;
        drop(claim);
        let mut lp = drain_loop(store.clone(), reader.clone(), writer);
        let id = remember(&store, "the zanzibar ferry leaves at noon");
        assert_eq!(lp.drain(), Some(1));
        lp.seal_if_due();
        let before = store.at_rest_keys(AtRestPurpose::Indexes).unwrap().unwrap();
        assert!(before.settled);

        let rotated = second_brain_core::keys::StoreKey::generate().unwrap();
        store
            .rewrite_under_key(
                second_brain_core::sqlite_store::encrypt_migration::RewriteTo::Key(&rotated),
            )
            .unwrap();
        lp.seal_if_due();
        let after = store.at_rest_keys(AtRestPurpose::Indexes).unwrap().unwrap();
        assert_ne!(after.keys.current_id(), before.keys.current_id());
        assert!(after.settled);
        assert!(after.keys.retired_ids().is_empty());
        assert_all_sealed_under(&dir, after.keys.current_id());
        assert_eq!(
            reader.read().unwrap().bm25_search("zanzibar", 5).unwrap()[0].0,
            id
        );
    }

    #[test]
    fn an_idle_sealed_writer_keeps_its_vectors_for_recall() {
        let data = tempfile::tempdir().unwrap();
        let store = keyed_store(data.path());
        let dir = data.path().join("brain.indexes");
        let keys = store
            .at_rest_keys(AtRestPurpose::Indexes)
            .unwrap()
            .unwrap()
            .keys;
        let handles = crate::models::open_index_handles(&dir, &store, Some(&keys), |_| {}).unwrap();
        let crate::models::IndexHandles {
            reader,
            writer,
            claim,
        } = handles;
        drop(claim);
        let mut lp = drain_loop(store.clone(), reader.clone(), writer);
        let first = remember(&store, "the zanzibar ferry leaves at noon");
        assert_eq!(lp.drain(), Some(1));
        let used = lp.writer.last_used;
        assert!(lp.writer.release_if_idle(used + WRITER_IDLE));
        assert!(!lp.writer.release_if_idle(used + WRITER_IDLE * 2));
        assert!(lp.writer.memory().is_some());

        let second = remember(&store, "the kilimanjaro trail opens in june");
        assert_eq!(lp.drain(), Some(1));
        let guard = reader.read().unwrap();
        assert_eq!(guard.bm25_search("zanzibar", 5).unwrap()[0].0, first);
        assert_eq!(guard.bm25_search("kilimanjaro", 5).unwrap()[0].0, second);
    }

    #[test]
    fn indexes_that_open_under_no_held_key_are_rebuilt_from_the_store() {
        let data = tempfile::tempdir().unwrap();
        let store = keyed_store(data.path());
        let dir = data.path().join("brain.indexes");
        let id = remember(&store, "the zanzibar ferry leaves at noon");
        let foreign = SealKeys::single(second_brain_core::sealed::SealKey::generate().unwrap());
        let handles =
            crate::models::open_index_handles(&dir, &store, Some(&foreign), |_| {}).unwrap();
        drop(handles);

        let keys = store
            .at_rest_keys(AtRestPurpose::Indexes)
            .unwrap()
            .unwrap()
            .keys;
        let handles = crate::models::open_index_handles(&dir, &store, Some(&keys), |_| {}).unwrap();
        assert_eq!(
            handles
                .reader
                .read()
                .unwrap()
                .bm25_search("zanzibar", 5)
                .unwrap()[0]
                .0,
            id
        );
    }
}