vaegram 0.2.0

Telegram MTProto client library for users and bots
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
use std::collections::HashMap;
use std::fmt;
use std::io;
use std::mem;
use std::sync::{Arc, Mutex, MutexGuard, PoisonError};
use std::time::Duration;

use tokio::sync::Notify;
use tokio::task::JoinHandle;
use tokio::time::Instant;

use super::engine::{PeerRecord, Storage, StorageFuture, StorageReporter, StoredState};
use super::{MemoryStorage, Reports, SessionData};

const RETRY_MIN: Duration = Duration::from_secs(1);
const RETRY_MAX: Duration = Duration::from_secs(30);

/// Answers every read from a local engine and copies every write to a remote one in the background, so a slow or
/// unreachable remote never holds the client up.
///
/// The local side is a [`MemoryStorage`] unless [`local`](Self::local) says otherwise. Opening needs the remote. It
/// fills in what the local side lacks: the remote's login and state when the local side has none, and the
/// remote's newest peers unless the local side holds a newer version of them. The local login and state, when
/// there are any, are queued for the remote; local peers aren't sent at open, only when they are next written, so
/// a local peer is never replaced by an older one. After that, writes land locally first and queue for the remote,
/// merged by peer, and are retried from 1 up to every 30 seconds while the remote is down. When more peers queue up
/// than [`queue_size`](Self::queue_size) allows, the oldest peer writes are dropped (they stay local); the login
/// and the state are never dropped. Closing waits up to [`flush_timeout`](Self::flush_timeout) for the queue to
/// drain. Logging out drops every queued write and deletes on both sides; a remote delete that fails stays queued,
/// ahead of anything written since, and is retried like any other write.
pub struct HybridStorage {
    local: Box<dyn Storage>,
    remote: Arc<dyn Storage>,
    warm_peers: usize,
    flush_timeout: Duration,
    shared: Arc<Shared>,
    worker: Mutex<Option<JoinHandle<()>>>,
}

struct Shared {
    pending: Mutex<Pending>,
    queue_size: usize,
    // Wakes the worker for new writes.
    wake: Notify,
    // Cuts the worker's wait between retries short, when closing.
    flush: Notify,
    // Fires when the queue drains.
    idle: Notify,
    // Held for every remote write, so a delete waits for the write in flight.
    remote: tokio::sync::Mutex<()>,
    reports: Reports,
}

#[derive(Default)]
struct Pending {
    queue: Queue,
    // The batch the worker is writing and the epoch it was taken in, put back if the write fails or is cut off.
    in_flight: Option<(Arc<Queue>, u64)>,
    // Bumped by a log-out, so a batch taken before it is never put back.
    epoch: u64,
}

#[derive(Clone, Default)]
struct Queue {
    // Delete everything on the remote before the writes below: a log-out that hasn't reached it yet.
    delete: bool,
    // Forget the remote's peers before writing the ones below.
    clear: bool,
    session: Option<SessionData>,
    state: Option<StoredState>,
    peers: HashMap<i64, PeerRecord>,
}

impl Queue {
    fn is_empty(&self) -> bool {
        !self.delete && !self.clear && self.session.is_none() && self.state.is_none() && self.peers.is_empty()
    }

    /// Puts a batch that didn't reach the remote back under the writes queued since, which are newer.
    fn put_back(&mut self, batch: Queue) {
        if !self.clear {
            for (id, peer) in batch.peers {
                self.peers.entry(id).or_insert(peer);
            }
            self.clear = batch.clear;
        }
        self.delete |= batch.delete;
        self.session = self.session.take().or(batch.session);
        self.state = self.state.take().or(batch.state);
    }

    /// Drops the oldest peer writes over `limit`, and returns how many.
    fn trim(&mut self, limit: usize) -> usize {
        let excess = self.peers.len().saturating_sub(limit);
        if excess == 0 {
            return 0;
        }
        let mut oldest: Vec<(i64, i64)> = self.peers.values().map(|p| (p.updated, p.id)).collect();
        oldest.sort_unstable();
        for (_, id) in &oldest[..excess] {
            self.peers.remove(id);
        }
        excess
    }
}

impl Shared {
    fn pending(&self) -> MutexGuard<'_, Pending> {
        self.pending.lock().unwrap_or_else(PoisonError::into_inner)
    }

    /// Reports peer writes `trim` dropped; called with no lock held.
    fn dropped(&self, excess: usize) {
        if excess > 0 {
            let why = format!("the remote storage is behind; {excess} peer writes were dropped (they stay local)");
            self.reports.report(io::Error::other(why));
        }
    }

    fn enqueue(&self, f: impl FnOnce(&mut Queue)) {
        let mut pending = self.pending();
        f(&mut pending.queue);
        let excess = pending.queue.trim(self.queue_size);
        drop(pending);
        self.dropped(excess);
        self.wake.notify_one();
    }

    fn take(&self) -> Option<(Arc<Queue>, u64)> {
        let mut pending = self.pending();
        if pending.queue.is_empty() {
            return None;
        }
        let batch = Arc::new(mem::take(&mut pending.queue));
        pending.in_flight = Some((batch.clone(), pending.epoch));
        Some((batch, pending.epoch))
    }

    fn finish(&self, epoch: u64, failed: bool) {
        let mut pending = self.pending();
        let batch = pending.in_flight.take();
        let mut excess = 0;
        if let Some((batch, _)) = batch.filter(|_| failed && pending.epoch == epoch) {
            pending.queue.put_back(Arc::unwrap_or_clone(batch));
            excess = pending.queue.trim(self.queue_size);
        }
        if pending.queue.is_empty() {
            self.idle.notify_waiters();
        }
        drop(pending);
        self.dropped(excess);
    }

    /// Puts back a batch whose write was cut off, unless a log-out came after it was taken.
    fn recover(&self) {
        let mut pending = self.pending();
        let epoch = pending.epoch;
        let mut excess = 0;
        if let Some((batch, _)) = pending.in_flight.take().filter(|(_, taken)| *taken == epoch) {
            pending.queue.put_back(Arc::unwrap_or_clone(batch));
            excess = pending.queue.trim(self.queue_size);
        }
        drop(pending);
        self.dropped(excess);
    }

    fn drained(&self) -> bool {
        let pending = self.pending();
        pending.queue.is_empty() && pending.in_flight.is_none()
    }
}

async fn write(remote: &dyn Storage, batch: &Queue) -> io::Result<()> {
    if batch.delete {
        remote.delete().await?;
    }
    if batch.clear {
        remote.clear_peers().await?;
    }
    if let Some(session) = &batch.session {
        remote.save_session(session).await?;
    }
    if let Some(state) = &batch.state {
        remote.save_state(state).await?;
    }
    if !batch.peers.is_empty() {
        let peers: Vec<PeerRecord> = batch.peers.values().cloned().collect();
        remote.update_peers(&peers).await?;
    }
    Ok(())
}

async fn run(shared: Arc<Shared>, remote: Arc<dyn Storage>) {
    let mut delay = RETRY_MIN;
    loop {
        let guard = shared.remote.lock().await;
        let Some((batch, epoch)) = shared.take() else {
            drop(guard);
            shared.wake.notified().await;
            continue;
        };
        let written = write(&*remote, &batch).await;
        drop(batch);
        shared.finish(epoch, written.is_err());
        drop(guard);
        match written {
            Ok(()) => delay = RETRY_MIN,
            Err(e) => {
                let why = format!("couldn't write to the remote storage, retrying in {}s: {e}", delay.as_secs());
                shared.reports.report(io::Error::new(e.kind(), why));
                tokio::select! {
                    () = tokio::time::sleep(delay) => {},
                    () = shared.flush.notified() => {},
                }
                delay = (delay * 2).min(RETRY_MAX);
            }
        }
    }
}

impl HybridStorage {
    /// Copies every write to `remote` and answers reads from memory until [`local`](Self::local) says otherwise.
    pub fn new(remote: impl Storage) -> HybridStorage {
        HybridStorage {
            local: Box::new(MemoryStorage::new()),
            remote: Arc::new(remote),
            warm_peers: 4096,
            flush_timeout: Duration::from_secs(10),
            shared: Arc::new(Shared {
                pending: Mutex::new(Pending::default()),
                queue_size: 1024,
                wake: Notify::new(),
                flush: Notify::new(),
                idle: Notify::new(),
                remote: tokio::sync::Mutex::new(()),
                reports: Reports::default(),
            }),
            worker: Mutex::new(None),
        }
    }

    /// Answers reads from `engine` instead of from memory.
    #[must_use]
    pub fn local(mut self, engine: impl Storage) -> HybridStorage {
        self.local = Box::new(engine);
        self
    }

    /// How many peer writes may wait for the remote before the oldest are dropped; 1024 unless set. Set it before
    /// opening: once the storage has been opened it has no effect.
    #[must_use]
    pub fn queue_size(mut self, peers: usize) -> HybridStorage {
        if let Some(shared) = Arc::get_mut(&mut self.shared) {
            shared.queue_size = peers;
        }
        self
    }

    /// How many of the remote's newest peers opening copies to the local side; 4096 unless set. Set it before
    /// opening.
    #[must_use]
    pub fn warm_peers(mut self, peers: usize) -> HybridStorage {
        self.warm_peers = peers;
        self
    }

    /// How long closing waits for the queue to reach the remote; 10 seconds unless set. `Duration::MAX` waits
    /// for as long as it takes.
    #[must_use]
    pub fn flush_timeout(mut self, timeout: Duration) -> HybridStorage {
        self.flush_timeout = timeout;
        self
    }

    fn queued(&self) -> usize {
        let pending = self.shared.pending();
        let q = &pending.queue;
        q.peers.len() + usize::from(q.session.is_some()) + usize::from(q.state.is_some()) + usize::from(q.clear)
    }

    /// Fills in what the local side lacks from the remote, and queues what the local side has for it.
    async fn warm(&self) -> io::Result<()> {
        match self.local.load_session().await? {
            Some(session) => self.shared.enqueue(|q| q.session = Some(session)),
            None => {
                if let Some(session) = self.remote.load_session().await? {
                    self.local.save_session(&session).await?;
                }
            }
        }
        match self.local.load_state().await? {
            Some(state) => self.shared.enqueue(|q| q.state = Some(state)),
            None => {
                if let Some(state) = self.remote.load_state().await? {
                    self.local.save_state(&state).await?;
                }
            }
        }
        let mut newer = Vec::new();
        for peer in self.remote.export_peers(Some(self.warm_peers)).await? {
            match self.local.get_peer_by_id(peer.id).await? {
                Some(known) if known.updated >= peer.updated => {}
                _ => newer.push(peer),
            }
        }
        if !newer.is_empty() {
            self.local.update_peers(&newer).await?;
        }
        Ok(())
    }

    async fn open(&self) -> io::Result<()> {
        self.local.open().await?;
        self.remote.open().await?;
        self.warm().await?;
        let mut worker = self.worker.lock().unwrap_or_else(PoisonError::into_inner);
        if worker.is_none() {
            *worker = Some(tokio::spawn(run(self.shared.clone(), self.remote.clone())));
        }
        Ok(())
    }

    async fn close(&self) -> io::Result<()> {
        // A timeout too long to add to the clock means no deadline.
        let deadline = Instant::now().checked_add(self.flush_timeout);
        self.shared.flush.notify_one();
        let mut drained = true;
        loop {
            let idle = self.shared.idle.notified();
            tokio::pin!(idle);
            idle.as_mut().enable();
            if self.shared.drained() {
                break;
            }
            let woke = match deadline {
                Some(deadline) => tokio::time::timeout_at(deadline, idle).await.is_ok(),
                None => {
                    idle.await;
                    true
                }
            };
            if !woke {
                drained = self.shared.drained();
                break;
            }
        }
        let worker = self.worker.lock().unwrap_or_else(PoisonError::into_inner).take();
        if let Some(worker) = worker {
            worker.abort();
            let _ = worker.await;
        }
        self.shared.recover();
        let local = self.local.close().await;
        let remote = self.remote.close().await;
        if !drained {
            let left = self.queued();
            return Err(io::Error::new(
                io::ErrorKind::TimedOut,
                format!("{left} writes didn't reach the remote storage; they stay queued until it is opened again"),
            ));
        }
        local.and(remote)
    }

    async fn delete(&self) -> io::Result<()> {
        let epoch = {
            let mut pending = self.shared.pending();
            pending.queue = Queue { delete: true, ..Queue::default() };
            pending.epoch += 1;
            pending.epoch
        };
        let writing = self.shared.remote.lock().await;
        self.local.delete().await?;
        match self.remote.delete().await {
            Ok(()) => {
                let mut pending = self.shared.pending();
                if pending.epoch == epoch {
                    pending.queue.delete = false;
                }
                Ok(())
            }
            Err(e) => {
                drop(writing);
                self.shared.wake.notify_one();
                let why = format!("couldn't delete on the remote storage, it stays queued and is retried: {e}");
                Err(io::Error::new(e.kind(), why))
            }
        }
    }
}

impl Drop for HybridStorage {
    fn drop(&mut self) {
        if let Some(worker) = self.worker.get_mut().unwrap_or_else(PoisonError::into_inner).take() {
            worker.abort();
        }
    }
}

impl fmt::Debug for HybridStorage {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_struct("HybridStorage").field("queued", &self.queued()).finish_non_exhaustive()
    }
}

impl Storage for HybridStorage {
    // The wrapped engines see it too: the remote's own warnings (a Redis that evicts keys) come out of `open`.
    fn set_reporter(&self, reporter: StorageReporter) {
        self.shared.reports.set(reporter.clone());
        self.local.set_reporter(reporter.clone());
        self.remote.set_reporter(reporter);
    }

    fn open(&self) -> StorageFuture<'_, ()> {
        Box::pin(self.open())
    }

    fn close(&self) -> StorageFuture<'_, ()> {
        Box::pin(self.close())
    }

    fn delete(&self) -> StorageFuture<'_, ()> {
        Box::pin(self.delete())
    }

    fn load_session(&self) -> StorageFuture<'_, Option<SessionData>> {
        self.local.load_session()
    }

    fn save_session<'a>(&'a self, session: &'a SessionData) -> StorageFuture<'a, ()> {
        Box::pin(async move {
            self.local.save_session(session).await?;
            self.shared.enqueue(|q| q.session = Some(session.clone()));
            Ok(())
        })
    }

    fn load_state(&self) -> StorageFuture<'_, Option<StoredState>> {
        self.local.load_state()
    }

    fn save_state<'a>(&'a self, state: &'a StoredState) -> StorageFuture<'a, ()> {
        Box::pin(async move {
            self.local.save_state(state).await?;
            self.shared.enqueue(|q| q.state = Some(state.clone()));
            Ok(())
        })
    }

    fn update_peers<'a>(&'a self, peers: &'a [PeerRecord]) -> StorageFuture<'a, ()> {
        Box::pin(async move {
            if peers.is_empty() {
                return Ok(());
            }
            self.local.update_peers(peers).await?;
            self.shared.enqueue(|q| q.peers.extend(peers.iter().map(|p| (p.id, p.clone()))));
            Ok(())
        })
    }

    fn get_peer_by_id(&self, id: i64) -> StorageFuture<'_, Option<PeerRecord>> {
        self.local.get_peer_by_id(id)
    }

    fn get_peer_by_username<'a>(&'a self, username: &'a str) -> StorageFuture<'a, Option<PeerRecord>> {
        self.local.get_peer_by_username(username)
    }

    fn get_peer_by_phone_number<'a>(&'a self, phone_number: &'a str) -> StorageFuture<'a, Option<PeerRecord>> {
        self.local.get_peer_by_phone_number(phone_number)
    }

    fn export_peers(&self, limit: Option<usize>) -> StorageFuture<'_, Vec<PeerRecord>> {
        self.local.export_peers(limit)
    }

    fn clear_peers(&self) -> StorageFuture<'_, ()> {
        Box::pin(async move {
            self.local.clear_peers().await?;
            self.shared.enqueue(|q| {
                q.peers.clear();
                q.clear = true;
            });
            Ok(())
        })
    }
}

#[cfg(test)]
mod tests {
    use std::io;
    use std::sync::Arc;
    use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
    use std::time::Duration;

    use tokio::time::{Instant, sleep};

    use super::{HybridStorage, Queue};
    use crate::session::Salt;
    use crate::storage::{MemoryStorage, PeerRecord, SessionData, Storage, StorageFuture, StoredState, conformance};

    /// A remote that can go down and counts the calls it gets.
    #[derive(Default)]
    struct Remote {
        inner: MemoryStorage,
        down: AtomicBool,
        calls: AtomicUsize,
    }

    impl Remote {
        fn call<'a, T: Send + 'a>(&'a self, f: StorageFuture<'a, T>) -> StorageFuture<'a, T> {
            self.calls.fetch_add(1, Ordering::SeqCst);
            if self.down.load(Ordering::SeqCst) {
                return Box::pin(async { Err(io::Error::new(io::ErrorKind::ConnectionRefused, "down")) });
            }
            f
        }
    }

    impl Storage for Remote {
        fn open(&self) -> StorageFuture<'_, ()> {
            self.call(self.inner.open())
        }

        fn delete(&self) -> StorageFuture<'_, ()> {
            self.call(self.inner.delete())
        }

        fn load_session(&self) -> StorageFuture<'_, Option<SessionData>> {
            self.call(self.inner.load_session())
        }

        fn save_session<'a>(&'a self, session: &'a SessionData) -> StorageFuture<'a, ()> {
            self.call(self.inner.save_session(session))
        }

        fn load_state(&self) -> StorageFuture<'_, Option<StoredState>> {
            self.call(self.inner.load_state())
        }

        fn save_state<'a>(&'a self, state: &'a StoredState) -> StorageFuture<'a, ()> {
            self.call(self.inner.save_state(state))
        }

        fn update_peers<'a>(&'a self, peers: &'a [PeerRecord]) -> StorageFuture<'a, ()> {
            self.call(self.inner.update_peers(peers))
        }

        fn get_peer_by_id(&self, id: i64) -> StorageFuture<'_, Option<PeerRecord>> {
            self.call(self.inner.get_peer_by_id(id))
        }

        fn get_peer_by_username<'a>(&'a self, username: &'a str) -> StorageFuture<'a, Option<PeerRecord>> {
            self.call(self.inner.get_peer_by_username(username))
        }

        fn get_peer_by_phone_number<'a>(&'a self, phone: &'a str) -> StorageFuture<'a, Option<PeerRecord>> {
            self.call(self.inner.get_peer_by_phone_number(phone))
        }

        fn export_peers(&self, limit: Option<usize>) -> StorageFuture<'_, Vec<PeerRecord>> {
            self.call(self.inner.export_peers(limit))
        }

        fn clear_peers(&self) -> StorageFuture<'_, ()> {
            self.call(self.inner.clear_peers())
        }
    }

    fn session(user_id: i64) -> SessionData {
        let mut s = SessionData::new(4, [9; 256]);
        s.user_id = user_id;
        s
    }

    fn peer(id: i64, updated: i64) -> PeerRecord {
        let mut p = PeerRecord::new(id, id * 3);
        (p.usernames, p.updated) = (vec![format!("user{id}")], updated);
        p
    }

    fn state(owner: i64) -> StoredState {
        StoredState::new(owner, Salt { value: 1, time_offset: 0 }, None)
    }

    #[tokio::test]
    async fn conformance_on_memory_sides() {
        let remote = Arc::new(MemoryStorage::new());
        conformance::check(&HybridStorage::new(remote.clone())).await;
        assert!(remote.load_session().await.unwrap().is_none(), "delete reached the remote");
    }

    #[tokio::test]
    async fn writes_reach_the_remote_and_reads_never_touch_it() {
        let remote = Arc::new(Remote::default());
        let hybrid = HybridStorage::new(remote.clone());
        hybrid.open().await.unwrap();
        hybrid.save_session(&session(7)).await.unwrap();
        hybrid.update_peers(&[peer(1, 1)]).await.unwrap();
        let before = remote.calls.load(Ordering::SeqCst);
        remote.down.store(true, Ordering::SeqCst);
        assert_eq!(hybrid.load_session().await.unwrap().map(|s| s.user_id), Some(7));
        assert_eq!(hybrid.get_peer_by_username("user1").await.unwrap(), Some(peer(1, 1)));
        assert_eq!(hybrid.export_peers(None).await.unwrap(), [peer(1, 1)]);
        assert_eq!(remote.calls.load(Ordering::SeqCst), before, "reads stayed local");

        remote.down.store(false, Ordering::SeqCst);
        hybrid.close().await.unwrap();
        assert_eq!(remote.inner.load_session().await.unwrap().map(|s| s.user_id), Some(7));
        assert_eq!(remote.inner.export_peers(None).await.unwrap(), [peer(1, 1)]);
    }

    #[tokio::test(start_paused = true)]
    async fn remote_outage_keeps_reads_and_retries() {
        let remote = Arc::new(Remote::default());
        let hybrid = HybridStorage::new(remote.clone());
        hybrid.open().await.unwrap();
        remote.down.store(true, Ordering::SeqCst);
        let calls = remote.calls.load(Ordering::SeqCst);
        hybrid.save_state(&state(7)).await.unwrap();
        hybrid.update_peers(&[peer(1, 1)]).await.unwrap();
        assert_eq!(hybrid.load_state().await.unwrap(), Some(state(7)));

        // Tries at 0, 1, 3, 7 and 15 seconds: the waits double.
        sleep(Duration::from_millis(15_500)).await;
        assert_eq!(remote.calls.load(Ordering::SeqCst) - calls, 5);
        remote.down.store(false, Ordering::SeqCst);
        sleep(Duration::from_secs(16)).await;
        assert_eq!(remote.inner.load_state().await.unwrap(), Some(state(7)));
        assert_eq!(remote.inner.export_peers(None).await.unwrap(), [peer(1, 1)]);
        hybrid.close().await.unwrap();
    }

    #[tokio::test(start_paused = true)]
    async fn full_queue_drops_peers_first() {
        let remote = Arc::new(Remote::default());
        let hybrid = HybridStorage::new(remote.clone()).queue_size(2);
        hybrid.open().await.unwrap();
        remote.down.store(true, Ordering::SeqCst);
        hybrid.save_session(&session(7)).await.unwrap();
        hybrid.save_state(&state(7)).await.unwrap();
        hybrid.update_peers(&[peer(1, 1), peer(2, 2)]).await.unwrap();
        sleep(Duration::from_secs(2)).await;
        hybrid.update_peers(&[peer(3, 3)]).await.unwrap();
        assert_eq!(hybrid.export_peers(None).await.unwrap().len(), 3, "every peer stays local");

        remote.down.store(false, Ordering::SeqCst);
        hybrid.close().await.unwrap();
        assert_eq!(remote.inner.load_session().await.unwrap().map(|s| s.user_id), Some(7));
        assert_eq!(remote.inner.load_state().await.unwrap(), Some(state(7)));
        assert_eq!(remote.inner.export_peers(None).await.unwrap(), [peer(3, 3), peer(2, 2)]);
    }

    #[tokio::test(start_paused = true)]
    async fn dropped_peers_and_failed_writes_are_reported() {
        let remote = Arc::new(Remote::default());
        let hybrid = HybridStorage::new(remote.clone()).queue_size(2);
        let reports = conformance::collect_reports(&hybrid);
        hybrid.open().await.unwrap();
        remote.down.store(true, Ordering::SeqCst);
        hybrid.update_peers(&[peer(1, 1), peer(2, 2), peer(3, 3)]).await.unwrap();
        // The first retry waits a second, so only the first failed try has happened.
        sleep(Duration::from_millis(500)).await;
        let reports = reports.lock().unwrap().clone();
        remote.down.store(false, Ordering::SeqCst);
        hybrid.close().await.unwrap();
        let count = |what: &str| reports.iter().filter(|r| r.contains(what)).count();
        assert_eq!((count("peer writes were dropped"), count("couldn't write to the remote storage")), (1, 1));
        assert_eq!(reports.len(), 2, "{reports:?}");
    }

    #[tokio::test(start_paused = true)]
    async fn close_waits_at_most_flush_timeout() {
        let remote = Arc::new(Remote::default());
        let hybrid = HybridStorage::new(remote.clone()).flush_timeout(Duration::from_secs(10));
        hybrid.open().await.unwrap();
        remote.down.store(true, Ordering::SeqCst);
        hybrid.save_state(&state(7)).await.unwrap();
        let start = Instant::now();
        let closed = hybrid.close().await;
        let waited = start.elapsed();
        let error = closed.unwrap_err();
        assert_eq!(error.kind(), io::ErrorKind::TimedOut);
        assert!(waited >= Duration::from_secs(10) && waited < Duration::from_secs(11), "{waited:?}");

        remote.down.store(false, Ordering::SeqCst);
        hybrid.open().await.unwrap();
        hybrid.close().await.unwrap();
        assert_eq!(remote.inner.load_state().await.unwrap(), Some(state(7)), "the queue outlived the close");
    }

    #[tokio::test(start_paused = true)]
    async fn a_flush_timeout_of_duration_max_waits_for_the_remote() {
        let remote = Arc::new(Remote::default());
        let hybrid = Arc::new(HybridStorage::new(remote.clone()).flush_timeout(Duration::MAX));
        hybrid.open().await.unwrap();
        remote.down.store(true, Ordering::SeqCst);
        hybrid.save_state(&state(7)).await.unwrap();
        let closing = tokio::spawn({
            let hybrid = hybrid.clone();
            async move { hybrid.close().await }
        });
        tokio::time::sleep(Duration::from_secs(3600)).await;
        assert!(!closing.is_finished(), "close gave up on the queue");
        remote.down.store(false, Ordering::SeqCst);
        closing.await.unwrap().unwrap();
        assert_eq!(remote.inner.load_state().await.unwrap(), Some(state(7)));
    }

    #[tokio::test]
    async fn warms_from_the_remote() {
        let remote = Arc::new(MemoryStorage::new());
        remote.save_session(&session(7)).await.unwrap();
        remote.save_state(&state(7)).await.unwrap();
        remote.update_peers(&[peer(1, 1), peer(2, 2)]).await.unwrap();
        let local = Arc::new(MemoryStorage::new());
        local.update_peers(&[peer(9, 9)]).await.unwrap();

        let hybrid = HybridStorage::new(remote.clone()).local(local.clone()).warm_peers(1);
        hybrid.open().await.unwrap();
        assert_eq!(local.load_session().await.unwrap().map(|s| s.user_id), Some(7));
        assert_eq!(local.load_state().await.unwrap(), Some(state(7)));
        assert_eq!(local.export_peers(None).await.unwrap(), [peer(9, 9), peer(2, 2)], "only the newest one");
        hybrid.close().await.unwrap();
    }

    #[tokio::test]
    async fn a_newer_local_side_catches_the_remote_up() {
        let remote = Arc::new(MemoryStorage::new());
        remote.save_session(&session(1)).await.unwrap();
        remote.update_peers(&[peer(5, 9)]).await.unwrap();
        let local = Arc::new(MemoryStorage::new());
        local.save_session(&session(2)).await.unwrap();
        local.update_peers(&[peer(5, 10)]).await.unwrap();

        let hybrid = HybridStorage::new(remote.clone()).local(local.clone());
        hybrid.open().await.unwrap();
        hybrid.close().await.unwrap();
        assert_eq!(local.get_peer_by_id(5).await.unwrap(), Some(peer(5, 10)), "an older remote peer is skipped");
        assert_eq!(remote.load_session().await.unwrap().map(|s| s.user_id), Some(2));
    }

    #[tokio::test(start_paused = true)]
    async fn a_failed_remote_delete_is_retried_before_later_writes() {
        let remote = Arc::new(Remote::default());
        remote.inner.save_session(&session(1)).await.unwrap();
        remote.inner.update_peers(&[peer(1, 1)]).await.unwrap();
        let hybrid = HybridStorage::new(remote.clone());
        hybrid.open().await.unwrap();

        remote.down.store(true, Ordering::SeqCst);
        let error = hybrid.delete().await.unwrap_err();
        assert!(error.to_string().contains("stays queued"), "{error}");
        assert!(hybrid.load_session().await.unwrap().is_none(), "the local side is gone at once");
        hybrid.save_session(&session(9)).await.unwrap();

        remote.down.store(false, Ordering::SeqCst);
        sleep(Duration::from_secs(2)).await;
        assert_eq!(remote.inner.load_session().await.unwrap().map(|s| s.user_id), Some(9), "the new login landed");
        assert!(remote.inner.export_peers(None).await.unwrap().is_empty(), "after the delete");
        hybrid.close().await.unwrap();
    }

    #[test]
    fn recover_never_restores_a_batch_from_before_a_log_out() {
        let hybrid = HybridStorage::new(MemoryStorage::new());
        let shared = &hybrid.shared;
        shared.enqueue(|q| q.session = Some(session(1)));
        let _cut_off = shared.take().unwrap();
        shared.recover();
        assert!(shared.pending().queue.session.is_some(), "a batch from this epoch comes back");

        let _cut_off = shared.take().unwrap();
        {
            let mut pending = shared.pending();
            pending.queue = Queue { delete: true, ..Queue::default() };
            pending.epoch += 1;
        }
        shared.recover();
        let pending = shared.pending();
        assert!(pending.queue.session.is_none() && pending.queue.delete && pending.in_flight.is_none());
    }

    #[tokio::test(start_paused = true)]
    async fn log_out_stops_queued_writes() {
        let remote = Arc::new(Remote::default());
        let hybrid = HybridStorage::new(remote.clone());
        hybrid.open().await.unwrap();
        remote.down.store(true, Ordering::SeqCst);
        hybrid.save_session(&session(7)).await.unwrap();
        hybrid.update_peers(&[peer(1, 1)]).await.unwrap();
        sleep(Duration::from_millis(100)).await;

        remote.down.store(false, Ordering::SeqCst);
        hybrid.delete().await.unwrap();
        sleep(Duration::from_secs(60)).await;
        assert!(remote.inner.load_session().await.unwrap().is_none(), "the queued login never landed");
        assert!(remote.inner.export_peers(None).await.unwrap().is_empty());
        assert!(hybrid.load_session().await.unwrap().is_none());
        hybrid.close().await.unwrap();
    }

    #[cfg(feature = "redis")]
    #[tokio::test]
    #[ignore = "needs VAEGRAM_REDIS_URL, a Redis server it may write to"]
    async fn conformance_over_a_live_redis() {
        let Ok(url) = std::env::var("VAEGRAM_REDIS_URL") else {
            assert!(std::env::var_os("CI").is_none(), "set VAEGRAM_REDIS_URL");
            return;
        };
        let prefix = format!("vaegram:live-hybrid:{}", std::process::id());
        let remote = crate::storage::RedisStorage::new(&url, "bot").unwrap().prefix(prefix);
        conformance::check(&HybridStorage::new(remote)).await;
    }
}