snapdir-stores 1.5.0

snapdir stores: FileStore, S3/B2/GCS native SDK stores + external-store shim.
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
//! Shared concurrent push orchestrator for object-store backends.
//!
//! `S3Store::push` and `GcsStore::push` (and, via S3, `B2Store`) are
//! byte-identical except for the per-object existence check, the per-object
//! upload, and the final manifest write. [`push_objects_concurrent`] factors
//! out that loop so both backends upload objects concurrently (bounded by the
//! store's [`TransferConfig`] concurrency, throttled by a shared
//! [`RateLimiter`]) while each backend injects only those store-specific calls.
//!
//! The orchestrator preserves every Bash/oracle invariant of the original
//! sequential push loops:
//!
//! - **Per-object content-addressed skip:** the injected `key_exists` closure
//!   is consulted for every `File` entry; a present object is skipped with no
//!   read and no upload (content-addressed, so a present object is already the
//!   right bytes).
//! - **Invalid-source guard (verify-before-upload):** for an absent object the
//!   source file at `source.join(rel)` is read and its BLAKE3 is verified
//!   against the manifest checksum *before* the upload; a mismatch is a
//!   [`StoreError::Integrity`] and no manifest is written. The read + verify is
//!   shared here so S3 and GCS never duplicate it.
//! - **Manifest-last / all-or-nothing:** the injected `write_manifest` closure
//!   is called **exactly once and only after every object upload has returned
//!   `Ok`**. If any object upload errors, [`run_concurrent`] returns that error
//!   and the orchestrator returns it *without* writing the manifest — a failed
//!   push leaves no manifest, so a present manifest always implies all of its
//!   objects are present.
//! - **First error wins:** [`run_concurrent`] propagates the first upload error
//!   and cancels the rest.
//!
//! The skip-if-manifest-present early return stays in each store's `push`
//! (a cheap pre-check) *before* this orchestrator is called.

use std::path::Path;
use std::sync::Arc;
use std::time::Instant;

use snapdir_core::manifest::{Manifest, ManifestEntry, PathType};
use snapdir_core::merkle::{Blake3Hasher, Hasher};
use snapdir_core::store::StoreError;
use snapdir_core::Meter;

use crate::adaptive::{
    p95_object_size, AdaptiveGate, AdaptivePolicy as ControllerPolicy, ControllerDriver, OpResult,
    OpSample,
};
use crate::transfer::{
    classify_error, run_adaptive, run_concurrent, AdaptivePolicy, RateLimiter, TransferConfig,
};

/// Strips a leading `./` and a trailing `/` from a manifest path so the
/// remainder can be joined onto a source root (shared with the backends).
fn strip_leading_dot_slash(path: &str) -> &str {
    let trimmed = path.strip_prefix("./").unwrap_or(path);
    trimmed.strip_suffix('/').unwrap_or(trimmed)
}

/// Reads the source file for `entry` under `source`, verifying its BLAKE3
/// against the manifest checksum (the oracle's invalid-source guard) and
/// acquiring rate-limit budget for its byte length before returning the bytes
/// ready to upload.
///
/// Factored out so S3/GCS upload closures inject only `key_exists` +
/// `put_bytes` and never duplicate the read/verify/throttle logic.
///
/// # Errors
///
/// - I/O errors reading `source.join(rel)`.
/// - [`StoreError::Integrity`] when the source bytes no longer hash to the
///   entry's checksum.
async fn read_verified(
    entry: &ManifestEntry,
    source: &Path,
    rate_limiter: &RateLimiter,
    meter: Option<&Meter>,
) -> Result<Vec<u8>, StoreError> {
    let rel = strip_leading_dot_slash(&entry.path);
    let object_source = source.join(rel);
    let bytes = std::fs::read(&object_source)?;
    let actual = Blake3Hasher::new().hash_hex(&bytes);
    if actual != entry.checksum {
        return Err(StoreError::Integrity {
            address: object_source.display().to_string(),
            expected: entry.checksum.clone(),
            actual,
        });
    }
    // Source bytes read (bytes-in). Advisory only.
    if let Some(m) = meter {
        m.add_in(bytes.len() as u64);
    }
    // Throttle by the (verified) object size before the upload.
    rate_limiter.acquire(bytes.len() as u64).await;
    Ok(bytes)
}

/// Uploads every absent object referenced by `manifest` concurrently, then —
/// only if every upload succeeded — writes the manifest via `write_manifest`.
///
/// `upload_one` is the per-object closure each backend injects: given a `File`
/// [`ManifestEntry`], it performs the store-specific existence check + upload.
/// It is handed [`read_verified`] (as the `read`-and-verify step) so each
/// backend only injects `key_exists` + `put_bytes`. `write_manifest` is the
/// backend's manifest-write closure (its own verify + `put_bytes`).
///
/// The orchestrator owns bounded concurrency and the manifest-last /
/// all-or-nothing ordering.
///
/// # Errors
///
/// Propagates the first [`StoreError`] from any object upload (remaining
/// in-flight uploads are cancelled and **no** manifest is written), or the
/// error from `write_manifest`.
pub(crate) async fn push_objects_concurrent<'a, U, UFut, W, WFut>(
    manifest: &'a Manifest,
    config: &TransferConfig,
    rate_limiter: &RateLimiter,
    meter: Option<&Meter>,
    meter_arc: Option<Arc<Meter>>,
    upload_one: U,
    write_manifest: W,
) -> Result<(), StoreError>
where
    U: Fn(&'a ManifestEntry) -> UFut,
    UFut: std::future::Future<Output = Result<(), StoreError>>,
    W: FnOnce() -> WFut,
    WFut: std::future::Future<Output = Result<(), StoreError>>,
{
    // Only `File` entries carry object bytes; directories live only in the
    // manifest text. Order is irrelevant (content-addressed).
    let files: Vec<&ManifestEntry> = manifest
        .entries()
        .iter()
        .filter(|e| e.path_type == PathType::File)
        .collect();

    // Total to push (bytes over the file set), recorded so the bar can track
    // bytes. Advisory: no effect on what is uploaded. No-op without a meter.
    if let Some(m) = meter {
        let total: u64 = files.iter().map(|e| e.size).sum();
        m.set_total(total);
    }

    match config.adaptive {
        AdaptivePolicy::Off => {
            // Concurrent object pass: each task runs the injected per-object work
            // (existence check, then read+verify+upload for absent objects). The
            // first error is propagated and the rest cancelled.
            run_concurrent(files, config.concurrency, upload_one).await?;
        }
        AdaptivePolicy::On { fraction, ceiling } => {
            run_adaptive_objects(
                &files,
                config,
                rate_limiter,
                meter_arc,
                fraction,
                ceiling,
                upload_one,
            )
            .await?;
        }
    }

    // Manifest-last / all-or-nothing: only after every object upload returned
    // Ok do we write the manifest. A failed push (an error above) returns
    // early and leaves no manifest.
    write_manifest().await
}

/// The adaptive sibling of the fixed-concurrency object pass: sizes the in-flight
/// window to the policy ceiling, gates each upload to the controller's live
/// limit, times + classifies + records every op, and runs a background tick
/// driver that resizes the gate and retunes the shared rate limiter. The exact
/// objects uploaded and the first-error-wins semantics are identical to the
/// `Off` path — only scheduling/rate differ.
#[allow(clippy::too_many_arguments)]
async fn run_adaptive_objects<'a, U, UFut>(
    files: &[&'a ManifestEntry],
    config: &TransferConfig,
    rate_limiter: &RateLimiter,
    meter_arc: Option<Arc<Meter>>,
    fraction: f64,
    ceiling: usize,
    upload_one: U,
) -> Result<(), StoreError>
where
    U: Fn(&'a ManifestEntry) -> UFut,
    UFut: std::future::Future<Output = Result<(), StoreError>>,
{
    let gate = AdaptiveGate::new(config.concurrency.get(), ceiling);
    let driver = build_push_driver(
        files,
        config,
        rate_limiter,
        meter_arc,
        fraction,
        ceiling,
        &gate,
    );

    // Background tick driver: ~250ms cadence (well below the controller's 15s
    // cooldown/re-probe constants) so the gate/limiter track the controller live.
    let tick_driver = driver.clone();
    let mut ticker = tokio::time::interval(std::time::Duration::from_millis(250));
    ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
    let tick_handle = tokio::spawn(async move {
        loop {
            ticker.tick().await;
            tick_driver.tick();
        }
    });

    let result = run_adaptive(files.iter().copied(), &gate, |entry| {
        let upload_one = &upload_one;
        let driver = &driver;
        async move {
            let started = Instant::now();
            let outcome = upload_one(entry).await;
            let latency = started.elapsed();
            let (bytes, op_result) = match &outcome {
                Ok(()) => (entry.size, OpResult::Ok),
                Err(err) => (0, classify_error(err)),
            };
            driver.record_op(OpSample {
                bytes,
                latency,
                result: op_result,
            });
            outcome
        }
    })
    .await;

    tick_handle.abort();
    result.map(|_| ())
}

/// Builds the [`ControllerDriver`] for the async push path: a controller policy
/// derived from the config (ceiling + fraction, machine RAM for the memory
/// budget, `max_bytes_per_sec` as the rate cap), wired to retune the async
/// [`RateLimiter`] and mirror the limit/rate into the display meter.
fn build_push_driver(
    files: &[&ManifestEntry],
    config: &TransferConfig,
    rate_limiter: &RateLimiter,
    meter_arc: Option<Arc<Meter>>,
    fraction: f64,
    ceiling: usize,
    gate: &AdaptiveGate,
) -> ControllerDriver {
    let sizes: Vec<u64> = files.iter().map(|e| e.size).collect();
    let p95 = p95_object_size(&sizes);
    let total_ram = snapdir_core::resources::total_ram_bytes().unwrap_or(0);
    let policy = ControllerPolicy::new(fraction, ceiling, total_ram, config.max_bytes_per_sec);

    // The async limiter's `set_rate` is async; bridge it to the driver's sync
    // applier by spawning the retune on the current runtime (best-effort,
    // advisory). Cloning shares the same bucket.
    let limiter = rate_limiter.clone();
    let rate_applier: Arc<dyn Fn(Option<u64>) + Send + Sync> = Arc::new(move |rate| {
        let limiter = limiter.clone();
        tokio::spawn(async move {
            limiter.set_rate(rate).await;
        });
    });

    ControllerDriver::new(policy, gate.clone(), p95, Some(rate_applier), meter_arc)
}

/// The shared per-object upload step S3/GCS inject into
/// [`push_objects_concurrent`]: skip if the object key already exists,
/// otherwise read+verify the source (via [`read_verified`]) and upload it.
///
/// Backends supply `key_exists` and `put_bytes` closures over their own
/// `&self`; `object_key` is the backend's content-addressed key for the entry.
///
/// # Errors
///
/// Propagates errors from `key_exists`, the source read/verify, or `put_bytes`.
pub(crate) async fn upload_object<KFut, PFut>(
    entry: &ManifestEntry,
    object_key: String,
    source: &Path,
    rate_limiter: &RateLimiter,
    meter: Option<&Meter>,
    key_exists: impl FnOnce(String) -> KFut,
    put_bytes: impl FnOnce(String, Vec<u8>) -> PFut,
) -> Result<(), StoreError>
where
    KFut: std::future::Future<Output = Result<bool, StoreError>>,
    PFut: std::future::Future<Output = Result<(), StoreError>>,
{
    // Per-object content-addressed skip: a present object is already the right
    // bytes, so no read and no upload.
    if key_exists(object_key.clone()).await? {
        if let Some(m) = meter {
            m.add_skipped(1);
        }
        return Ok(());
    }
    if let Some(m) = meter {
        m.object_started();
    }
    // `read_verified` records bytes-in (the verified source bytes).
    let bytes = read_verified(entry, source, rate_limiter, meter).await?;
    let len = bytes.len() as u64;
    put_bytes(object_key, bytes).await?;
    // Upload succeeded: bytes-out + object done.
    if let Some(m) = meter {
        m.add_out(len);
        m.object_finished();
    }
    Ok(())
}

#[cfg(test)]
mod tests {
    use super::*;
    use snapdir_core::merkle::Hasher;
    use std::collections::HashSet;
    use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
    use std::sync::{Arc, Mutex};

    /// Multi-thread runtime with time enabled so concurrent uploads genuinely
    /// overlap and the in-flight high-water mark is meaningful.
    fn runtime() -> tokio::runtime::Runtime {
        tokio::runtime::Builder::new_multi_thread()
            .worker_threads(4)
            .enable_time()
            .build()
            .expect("build tokio runtime")
    }

    /// A temp-dir helper (avoids a dev-dependency) that removes itself on drop.
    struct TempDir {
        path: std::path::PathBuf,
    }

    impl TempDir {
        fn new() -> Self {
            static COUNTER: AtomicUsize = AtomicUsize::new(0);
            let n = COUNTER.fetch_add(1, Ordering::Relaxed);
            let path =
                std::env::temp_dir().join(format!("snapdir-push-test-{}-{n}", std::process::id()));
            std::fs::create_dir_all(&path).expect("create temp dir");
            Self { path }
        }

        fn path(&self) -> &Path {
            &self.path
        }
    }

    impl Drop for TempDir {
        fn drop(&mut self) {
            let _ = std::fs::remove_dir_all(&self.path);
        }
    }

    fn checksum_of(bytes: &[u8]) -> String {
        Blake3Hasher::new().hash_hex(bytes)
    }

    /// Builds a manifest (a root dir entry + one `File` per `(rel, contents)`)
    /// and writes each file's contents under `src` so read+verify is real.
    fn manifest_and_source(files: &[(&str, &[u8])], src: &Path) -> Manifest {
        let mut m = Manifest::new();
        m.push(ManifestEntry::new(
            PathType::Directory,
            "700",
            "0".repeat(64),
            0,
            "./",
        ));
        for (path, contents) in files {
            if let Some(parent) = Path::new(path).parent() {
                std::fs::create_dir_all(src.join(parent)).unwrap();
            }
            std::fs::write(src.join(path), contents).unwrap();
            m.push(ManifestEntry::new(
                PathType::File,
                "600",
                checksum_of(contents),
                contents.len() as u64,
                format!("./{path}"),
            ));
        }
        Manifest::from_entries(m.entries().to_vec())
    }

    /// A fake S3/GCS-like backend: tracks which keys "exist", records uploaded
    /// checksums, the in-flight high-water mark, and whether the manifest was
    /// written (and that it was written after the uploads).
    struct FakeStore {
        /// Object keys that already exist (skip-present source of truth).
        present: HashSet<String>,
        /// Checksums actually uploaded (in completion order).
        uploaded: Mutex<Vec<String>>,
        /// Currently in-flight uploads.
        in_flight: AtomicUsize,
        /// Peak simultaneous in-flight uploads.
        high_water: AtomicUsize,
        /// Set once the manifest write closure runs.
        manifest_written: AtomicBool,
        /// How many uploads had completed when the manifest was written.
        uploads_done_at_manifest: AtomicUsize,
        /// Checksum whose upload must fail (all-or-nothing test); empty = none.
        fail_checksum: String,
    }

    impl FakeStore {
        fn new(present: &[&str], fail_checksum: &str) -> Arc<Self> {
            Arc::new(Self {
                present: present.iter().map(|s| (*s).to_owned()).collect(),
                uploaded: Mutex::new(Vec::new()),
                in_flight: AtomicUsize::new(0),
                high_water: AtomicUsize::new(0),
                manifest_written: AtomicBool::new(false),
                uploads_done_at_manifest: AtomicUsize::new(0),
                fail_checksum: fail_checksum.to_owned(),
            })
        }

        // `async` (despite no await) to model the real backends' async
        // existence check the orchestrator awaits.
        #[allow(clippy::unused_async)]
        async fn key_exists(&self, key: String) -> Result<bool, StoreError> {
            Ok(self.present.contains(&key))
        }

        async fn put_bytes(&self, _key: String, bytes: Vec<u8>) -> Result<(), StoreError> {
            let cur = self.in_flight.fetch_add(1, Ordering::SeqCst) + 1;
            self.high_water.fetch_max(cur, Ordering::SeqCst);
            // Hold the slot so concurrent uploads actually overlap.
            tokio::time::sleep(std::time::Duration::from_millis(20)).await;
            let sum = checksum_of(&bytes);
            self.in_flight.fetch_sub(1, Ordering::SeqCst);
            if !self.fail_checksum.is_empty() && sum == self.fail_checksum {
                return Err(StoreError::Backend {
                    message: "upload blew up".to_owned(),
                    source: None,
                });
            }
            self.uploaded.lock().unwrap().push(sum);
            Ok(())
        }

        #[allow(clippy::unused_async)]
        async fn write_manifest(&self) -> Result<(), StoreError> {
            self.uploads_done_at_manifest
                .store(self.uploaded.lock().unwrap().len(), Ordering::SeqCst);
            self.manifest_written.store(true, Ordering::SeqCst);
            Ok(())
        }
    }

    /// Runs the orchestrator over `manifest`/`src` with `fake` injected.
    fn run_push(
        fake: &Arc<FakeStore>,
        manifest: &Manifest,
        src: &Path,
        concurrency: usize,
    ) -> Result<(), StoreError> {
        let cfg = TransferConfig::new(concurrency, None);
        let limiter = RateLimiter::new(None);
        let rt = runtime();
        rt.block_on(async {
            push_objects_concurrent(
                manifest,
                &cfg,
                &limiter,
                None,
                None,
                |entry| {
                    let fake = Arc::clone(fake);
                    let limiter = &limiter;
                    async move {
                        // Use the entry's checksum as the object key so the
                        // fake's `present` set keys on checksums directly.
                        upload_object(
                            entry,
                            entry.checksum.clone(),
                            src,
                            limiter,
                            None,
                            |key| {
                                let fake = Arc::clone(&fake);
                                async move { fake.key_exists(key).await }
                            },
                            |key, bytes| {
                                let fake = Arc::clone(&fake);
                                async move { fake.put_bytes(key, bytes).await }
                            },
                        )
                        .await
                    }
                },
                || {
                    let fake = Arc::clone(fake);
                    async move { fake.write_manifest().await }
                },
            )
            .await
        })
    }

    #[test]
    fn concurrent_upload_all_objects_then_manifest() {
        let files: &[(&str, &[u8])] = &[
            ("a.txt", b"alpha" as &[u8]),
            ("nested/b.txt", b"bravo"),
            ("nested/deep/c.txt", b"charlie"),
            ("d.txt", b"delta"),
        ];

        for concurrency in [1usize, 4] {
            let src = TempDir::new();
            let manifest = manifest_and_source(files, src.path());
            let fake = FakeStore::new(&[], "");

            run_push(&fake, &manifest, src.path(), concurrency).expect("push must succeed");

            // Every absent object was uploaded exactly once.
            let mut uploaded = fake.uploaded.lock().unwrap().clone();
            uploaded.sort();
            let mut expected: Vec<String> = files.iter().map(|(_, c)| checksum_of(c)).collect();
            expected.sort();
            assert_eq!(uploaded, expected, "all absent objects must be uploaded");

            // Concurrency bound: peak in-flight == min(concurrency, n), == 1 at
            // concurrency 1 (strictly sequential).
            let hw = fake.high_water.load(Ordering::SeqCst);
            let want = concurrency.min(files.len());
            assert_eq!(
                hw, want,
                "concurrency={concurrency}: peak in-flight {hw} != expected {want}"
            );

            // Manifest written EXACTLY once, and only AFTER all uploads.
            assert!(
                fake.manifest_written.load(Ordering::SeqCst),
                "manifest must be written"
            );
            assert_eq!(
                fake.uploads_done_at_manifest.load(Ordering::SeqCst),
                files.len(),
                "manifest must be written only after every object upload completed"
            );
        }
    }

    #[test]
    fn adaptive_upload_all_objects_then_manifest_within_ceiling() {
        // An adaptive push (policy On, ceiling 2) uploads every absent object,
        // writes the manifest last, and caps effective concurrency at the ceiling.
        use crate::transfer::AdaptivePolicy;

        let files: &[(&str, &[u8])] = &[
            ("a.txt", b"alpha" as &[u8]),
            ("nested/b.txt", b"bravo"),
            ("nested/deep/c.txt", b"charlie"),
            ("d.txt", b"delta"),
            ("e.txt", b"echo"),
        ];
        let src = TempDir::new();
        let manifest = manifest_and_source(files, src.path());
        let src_path = src.path().to_path_buf();
        let fake = FakeStore::new(&[], "");

        let cfg = TransferConfig::new(4, None).with_adaptive(AdaptivePolicy::On {
            fraction: 0.8,
            ceiling: 2,
        });
        let limiter = RateLimiter::new(None);
        let rt = runtime();
        let fake_ref = Arc::clone(&fake);
        rt.block_on(async {
            push_objects_concurrent(
                &manifest,
                &cfg,
                &limiter,
                None,
                None,
                |entry| {
                    let fake = Arc::clone(&fake_ref);
                    let limiter = &limiter;
                    let src_path = &src_path;
                    async move {
                        upload_object(
                            entry,
                            entry.checksum.clone(),
                            src_path,
                            limiter,
                            None,
                            |key| {
                                let fake = Arc::clone(&fake);
                                async move { fake.key_exists(key).await }
                            },
                            |key, bytes| {
                                let fake = Arc::clone(&fake);
                                async move { fake.put_bytes(key, bytes).await }
                            },
                        )
                        .await
                    }
                },
                || {
                    let fake = Arc::clone(&fake_ref);
                    async move { fake.write_manifest().await }
                },
            )
            .await
        })
        .expect("adaptive push must succeed");

        // All objects uploaded exactly once.
        let mut uploaded = fake.uploaded.lock().unwrap().clone();
        uploaded.sort();
        let mut expected: Vec<String> = files.iter().map(|(_, c)| checksum_of(c)).collect();
        expected.sort();
        assert_eq!(uploaded, expected, "all objects uploaded under adaptive");

        // Effective concurrency capped at the ceiling (2).
        let hw = fake.high_water.load(Ordering::SeqCst);
        assert!(
            hw <= 2,
            "effective concurrency must not exceed ceiling 2, got {hw}"
        );

        // Manifest written exactly once, after every upload.
        assert!(fake.manifest_written.load(Ordering::SeqCst));
        assert_eq!(
            fake.uploads_done_at_manifest.load(Ordering::SeqCst),
            files.len(),
            "manifest written only after every object upload"
        );
    }

    #[test]
    fn concurrent_upload_skips_present_objects() {
        let files: &[(&str, &[u8])] = &[
            ("present.txt", b"already-here" as &[u8]),
            ("missing.txt", b"needs-upload"),
        ];
        let src = TempDir::new();
        let manifest = manifest_and_source(files, src.path());

        // Mark `present.txt`'s object key (its checksum) as already present.
        let present_sum = checksum_of(b"already-here");
        let fake = FakeStore::new(&[present_sum.as_str()], "");

        run_push(&fake, &manifest, src.path(), 4).expect("push must succeed");

        let uploaded = fake.uploaded.lock().unwrap().clone();
        let missing_sum = checksum_of(b"needs-upload");
        assert!(
            !uploaded.contains(&present_sum),
            "present object must never be uploaded"
        );
        assert_eq!(
            uploaded,
            vec![missing_sum],
            "only the absent object should be uploaded"
        );
        assert!(fake.manifest_written.load(Ordering::SeqCst));
    }

    #[test]
    fn concurrent_upload_all_or_nothing_on_failure() {
        let files: &[(&str, &[u8])] = &[
            ("ok1.txt", b"one" as &[u8]),
            ("boom.txt", b"two"),
            ("ok2.txt", b"three"),
        ];
        let src = TempDir::new();
        let manifest = manifest_and_source(files, src.path());

        // The upload of `boom.txt`'s object fails.
        let boom_sum = checksum_of(b"two");
        let fake = FakeStore::new(&[], boom_sum.as_str());

        let result = run_push(&fake, &manifest, src.path(), 4);

        let err = result.expect_err("a failing object upload must surface");
        assert!(
            matches!(err, StoreError::Backend { ref message, .. } if message == "upload blew up"),
            "unexpected error: {err:?}"
        );
        // THE invariant: a failed push writes NO manifest.
        assert!(
            !fake.manifest_written.load(Ordering::SeqCst),
            "write_manifest must NEVER be called when an object upload fails"
        );
    }

    #[test]
    fn concurrent_upload_rejects_corrupt_source() {
        let files: &[(&str, &[u8])] = &[("good.txt", b"good" as &[u8]), ("bad.txt", b"bad")];
        let src = TempDir::new();
        let manifest = manifest_and_source(files, src.path());

        // Corrupt `bad.txt` on disk so its bytes no longer match the manifest
        // checksum: the verify-before-upload guard must fire.
        std::fs::write(src.path().join("bad.txt"), b"tampered").unwrap();
        let fake = FakeStore::new(&[], "");

        let result = run_push(&fake, &manifest, src.path(), 4);

        let err = result.expect_err("a corrupt source must surface an Integrity error");
        assert!(
            matches!(err, StoreError::Integrity { .. }),
            "unexpected error: {err:?}"
        );
        assert!(
            !fake.manifest_written.load(Ordering::SeqCst),
            "a corrupt source must leave no manifest"
        );
    }
}