cano 0.14.0

High-performance orchestration engine for building resilient, self-healing systems in Rust. Uses Finite State Machines (FSM) for strict, type-safe transitions.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
//! # Circuit Breaker — Fail-Fast Protection for Flaky Dependencies
//!
//! A [`CircuitBreaker`] short-circuits calls to a failing dependency so the system
//! degrades gracefully instead of hammering a broken backend with retries.
//!
//! State machine:
//!
//! - **`Closed`** — calls flow through. Each consecutive failure increments a counter;
//!   when the counter reaches `failure_threshold` the breaker trips to `Open`.
//!   A success resets the counter.
//! - **`Open { until }`** — every call is rejected with [`CanoError::CircuitOpen`]
//!   without invoking the underlying task. After `reset_timeout` elapses the next
//!   `try_acquire` call lazily transitions the breaker into `HalfOpen`.
//! - **`HalfOpen`** — at most `half_open_max_calls` trial calls are admitted concurrently.
//!   Each trial that returns success bumps a success counter; once that counter
//!   reaches `half_open_max_calls` the breaker closes. Any failure during HalfOpen
//!   immediately reopens the breaker (with a fresh `until = now + reset_timeout`).
//!
//! ## Default integration: wire into `TaskConfig`
//!
//! The recommended path is to attach the breaker to a [`crate::task::TaskConfig`] via
//! [`crate::task::TaskConfig::with_circuit_breaker`]. The retry loop in
//! [`crate::task::run_with_retries`] then:
//!
//! 1. Calls [`CircuitBreaker::try_acquire`] **before** each attempt — an open breaker
//!    short-circuits the entire retry loop with [`CanoError::CircuitOpen`] (no retries
//!    consumed, no task body invocation).
//! 2. Records the attempt's outcome **after** it returns — so the breaker observes the
//!    call's success/failure before the workflow propagates it.
//!
//! ```rust
//! use std::sync::Arc;
//! use std::time::Duration;
//! use cano::prelude::*;
//!
//! let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
//!     failure_threshold: 5,
//!     reset_timeout: Duration::from_secs(30),
//!     half_open_max_calls: 1,
//! }));
//!
//! let config = TaskConfig::new().with_circuit_breaker(Arc::clone(&breaker));
//! # let _ = config;
//! ```
//!
//! Sharing one breaker across multiple tasks lets the trip from any task protect every
//! task that depends on the same flaky resource.
//!
//! ## Advanced: register as a `Resource`
//!
//! [`CircuitBreaker`] also implements [`Resource`] (no-op lifecycle) so it can be
//! registered in [`Resources`] and looked up by key inside a task body. **This bypasses
//! the retry-loop integration**: callers must invoke [`try_acquire`](CircuitBreaker::try_acquire),
//! [`record_success`](CircuitBreaker::record_success), and
//! [`record_failure`](CircuitBreaker::record_failure) themselves, and they lose both
//! (a) the automatic retry-loop short-circuit on `CircuitOpen` and (b) the
//! before-call/after-call ordering guarantee that `with_circuit_breaker` provides. Use
//! this only for advanced cases where one task body needs to share one breaker across
//! several internal sub-calls.
//!
//! [`Resources`]: crate::resource::Resources
//! [`Resource`]: crate::resource::Resource

use crate::error::CanoError;
use crate::resource::Resource;
use cano_macros::resource;
use std::sync::{Arc, Mutex};
use std::time::{Duration, Instant};

#[cfg(feature = "tracing")]
use tracing::info;

/// Policy parameters controlling [`CircuitBreaker`] state transitions.
#[derive(Debug, Clone)]
pub struct CircuitPolicy {
    /// Consecutive failures in `Closed` that trip the breaker into `Open`.
    pub failure_threshold: u32,
    /// Time spent in `Open` before the next `try_acquire` lazily promotes to `HalfOpen`.
    pub reset_timeout: Duration,
    /// Cap on concurrent trial calls admitted while `HalfOpen`, **and** the number of
    /// consecutive trial successes required to close the breaker.
    ///
    /// These two roles are intentionally fused: any failure during `HalfOpen`
    /// immediately reopens the breaker, so to reach the close-threshold every admitted
    /// trial must succeed. Setting this to `N > 1` therefore means "admit up to `N`
    /// concurrent probes; close only after `N` of them have all succeeded". The
    /// default is `1` — one probe, one success closes — which is the right choice for
    /// almost every workload.
    pub half_open_max_calls: u32,
}

impl Default for CircuitPolicy {
    fn default() -> Self {
        Self {
            failure_threshold: 5,
            reset_timeout: Duration::from_secs(30),
            half_open_max_calls: 1,
        }
    }
}

/// The current state of a [`CircuitBreaker`].
///
/// Returned from [`CircuitBreaker::state`] for inspection / observability.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum CircuitState {
    /// Calls flow through. Failures are counted toward the trip threshold.
    Closed,
    /// All calls are rejected until `until`.
    Open {
        /// The earliest instant at which the breaker may transition to `HalfOpen`.
        until: Instant,
    },
    /// A bounded number of trial calls are admitted to probe the dependency.
    HalfOpen,
}

#[derive(Debug)]
struct Inner {
    state: CircuitState,
    consecutive_failures: u32,
    half_open_in_flight: u32,
    half_open_successes: u32,
    /// Monotonic epoch bumped on every state transition. Permits capture the
    /// epoch at issuance; outcomes from a stale epoch are ignored at
    /// consumption time. This prevents a slow caller — whose call straddles
    /// one or more `Open ↔ HalfOpen ↔ Closed` transitions — from corrupting
    /// the *next* state-machine session by bumping its counters.
    epoch: u64,
}

impl Inner {
    fn new() -> Self {
        Self {
            state: CircuitState::Closed,
            consecutive_failures: 0,
            half_open_in_flight: 0,
            half_open_successes: 0,
            epoch: 0,
        }
    }

    fn reset_half_open(&mut self) {
        self.half_open_in_flight = 0;
        self.half_open_successes = 0;
    }

    fn bump_epoch(&mut self) {
        self.epoch = self.epoch.wrapping_add(1);
    }

    /// Apply a state transition: install the new state, reset the half-open
    /// counters, and bump the epoch so any in-flight permits from the prior
    /// session are filtered as stale.
    fn transition(&mut self, new_state: CircuitState) {
        self.state = new_state;
        self.reset_half_open();
        self.bump_epoch();
    }
}

/// A reusable circuit breaker.
///
/// Cheap to clone (`Arc` internally). Share one breaker across every task that
/// depends on the same external resource so a trip from any caller protects every
/// caller. See the [module documentation](self) for the state-machine semantics.
#[derive(Debug, Clone)]
pub struct CircuitBreaker {
    inner: Arc<Mutex<Inner>>,
    policy: CircuitPolicy,
}

/// Upper bound on `CircuitPolicy::half_open_max_calls`.
///
/// Pathological values (very close to [`u32::MAX`]) would either silently never close
/// the breaker — once `half_open_successes` reaches its `saturating_add` ceiling no
/// further increment can land — or just take effectively forever to do so. Either way
/// it's a configuration mistake, so we reject it at construction.
const HALF_OPEN_MAX_CALLS_LIMIT: u32 = u32::MAX / 2;

impl CircuitBreaker {
    /// Construct a breaker in the `Closed` state.
    ///
    /// # Panics
    ///
    /// Panics if `policy.failure_threshold == 0`,
    /// `policy.half_open_max_calls == 0`, or
    /// `policy.half_open_max_calls > u32::MAX / 2`:
    ///
    /// - `failure_threshold == 0` is nonsensical: every failed call would meet the
    ///   trip threshold before any positive failure count exists.
    /// - `0` would deadlock the breaker permanently in `HalfOpen` (no probe could ever
    ///   be admitted, so the success counter could never advance).
    /// - Values approaching `u32::MAX` interact badly with the `saturating_add` on
    ///   `half_open_successes` — either the counter saturates before the threshold is
    ///   reached and the breaker never closes, or recovery takes effectively forever.
    ///
    /// Misconfigured policies are programmer errors caught at construction, before any
    /// task runs — consistent with the panic in `Resources::insert` for duplicate keys.
    pub fn new(policy: CircuitPolicy) -> Self {
        assert!(
            policy.failure_threshold >= 1,
            "CircuitPolicy::failure_threshold must be >= 1; 0 would make the breaker trip semantics nonsensical"
        );
        assert!(
            policy.half_open_max_calls >= 1,
            "CircuitPolicy::half_open_max_calls must be >= 1; 0 deadlocks the breaker in HalfOpen (no trial would ever be admitted)"
        );
        assert!(
            policy.half_open_max_calls <= HALF_OPEN_MAX_CALLS_LIMIT,
            "CircuitPolicy::half_open_max_calls must be <= {HALF_OPEN_MAX_CALLS_LIMIT}; got {} (values near u32::MAX may prevent the breaker from ever closing)",
            policy.half_open_max_calls
        );
        Self {
            inner: Arc::new(Mutex::new(Inner::new())),
            policy,
        }
    }

    /// Return the policy this breaker was constructed with.
    pub fn policy(&self) -> &CircuitPolicy {
        &self.policy
    }

    /// Inspect the current state. Useful for observability and tests.
    pub fn state(&self) -> CircuitState {
        let inner = self.inner.lock().expect("circuit breaker mutex poisoned");
        inner.state
    }

    /// Attempt to acquire permission to make a call.
    ///
    /// Returns a [`Permit`] that the caller must consume by calling
    /// [`record_success`](Self::record_success) or [`record_failure`](Self::record_failure)
    /// once the call completes. If neither is called the permit's `Drop` impl treats
    /// the call as a failure — this keeps the in-flight counter accurate even on panic
    /// or early return.
    ///
    /// # Errors
    ///
    /// Returns [`CanoError::CircuitOpen`] when the breaker is `Open` (and the cool-down
    /// has not elapsed) or when `HalfOpen` already has `half_open_max_calls` trials in
    /// flight.
    pub fn try_acquire(self: &Arc<Self>) -> Result<Permit, CanoError> {
        let mut inner = self.inner.lock().expect("circuit breaker mutex poisoned");

        // Lazy Open -> HalfOpen transition. We only flip when a caller actually
        // tries to acquire; no background task is required. The epoch bump
        // here invalidates any still-in-flight permits issued during the prior
        // Open / Closed session, so their outcomes won't corrupt this probe.
        if let CircuitState::Open { until } = inner.state
            && Instant::now() >= until
        {
            inner.transition(CircuitState::HalfOpen);
            #[cfg(feature = "tracing")]
            info!(
                half_open_max_calls = self.policy.half_open_max_calls,
                "circuit breaker transition: Open -> HalfOpen"
            );
            #[cfg(feature = "metrics")]
            crate::metrics::circuit_transition("open_to_halfopen");
        }

        let epoch = inner.epoch;
        let result = match inner.state {
            CircuitState::Closed => Ok(Permit::new(Arc::clone(self), false, epoch)),
            CircuitState::Open { .. } => Err(CanoError::circuit_open(
                "circuit breaker open: rejecting call",
            )),
            CircuitState::HalfOpen => {
                if inner.half_open_in_flight >= self.policy.half_open_max_calls {
                    Err(CanoError::circuit_open(
                        "circuit breaker half-open: trial slot exhausted",
                    ))
                } else {
                    inner.half_open_in_flight += 1;
                    Ok(Permit::new(Arc::clone(self), true, epoch))
                }
            }
        };
        #[cfg(feature = "metrics")]
        {
            drop(inner);
            crate::metrics::circuit_acquire(if result.is_ok() {
                "acquired"
            } else {
                "rejected"
            });
        }
        result
    }

    /// Record a successful call, consuming the permit.
    ///
    /// In `Closed` this resets the consecutive-failure counter. In `HalfOpen` it bumps
    /// the success counter and closes the breaker once `half_open_max_calls` successes
    /// have been observed.
    pub fn record_success(&self, mut permit: Permit) {
        permit.consumed = true;
        #[cfg(feature = "metrics")]
        crate::metrics::circuit_outcome("success");
        let mut inner = self.inner.lock().expect("circuit breaker mutex poisoned");

        // Stale outcome from a prior epoch: the in-flight slot, if any, was
        // already wiped by `reset_half_open()` at the transition. Skip both
        // counter accounting and the in-flight decrement.
        if permit.epoch != inner.epoch {
            return;
        }

        if permit.was_half_open && inner.half_open_in_flight > 0 {
            inner.half_open_in_flight -= 1;
        }

        match inner.state {
            CircuitState::Closed => {
                inner.consecutive_failures = 0;
            }
            CircuitState::HalfOpen => {
                inner.half_open_successes = inner.half_open_successes.saturating_add(1);
                if inner.half_open_successes >= self.policy.half_open_max_calls {
                    inner.consecutive_failures = 0;
                    inner.transition(CircuitState::Closed);
                    #[cfg(feature = "tracing")]
                    info!(
                        successes = self.policy.half_open_max_calls,
                        "circuit breaker transition: HalfOpen -> Closed"
                    );
                    #[cfg(feature = "metrics")]
                    crate::metrics::circuit_transition("halfopen_to_closed");
                }
            }
            CircuitState::Open { .. } => {
                // Unreachable when epoch is current — any transition out of
                // Closed/HalfOpen would have bumped the epoch.
                debug_assert!(
                    false,
                    "record_success on Open with current epoch is unreachable; epoch tracking should have filtered the stale outcome"
                );
            }
        }
    }

    /// Record a failed call, consuming the permit.
    ///
    /// In `Closed` this increments the consecutive-failure counter and trips the breaker
    /// when the threshold is reached. In `HalfOpen` it immediately reopens the breaker
    /// with a fresh cool-down deadline.
    pub fn record_failure(&self, mut permit: Permit) {
        permit.consumed = true;
        self.do_record_failure(permit.was_half_open, permit.epoch);
    }

    fn do_record_failure(&self, was_half_open: bool, permit_epoch: u64) {
        #[cfg(feature = "metrics")]
        crate::metrics::circuit_outcome("failure");
        let mut inner = self.inner.lock().expect("circuit breaker mutex poisoned");

        // Stale outcome from a prior epoch — see `record_success` for rationale.
        if permit_epoch != inner.epoch {
            return;
        }

        if was_half_open && inner.half_open_in_flight > 0 {
            inner.half_open_in_flight -= 1;
        }

        match inner.state {
            CircuitState::Closed => {
                inner.consecutive_failures = inner.consecutive_failures.saturating_add(1);
                if inner.consecutive_failures >= self.policy.failure_threshold {
                    inner.transition(CircuitState::Open {
                        until: Instant::now() + self.policy.reset_timeout,
                    });
                    #[cfg(feature = "tracing")]
                    info!(
                        failure_threshold = self.policy.failure_threshold,
                        reset_timeout_ms = self.policy.reset_timeout.as_millis() as u64,
                        "circuit breaker transition: Closed -> Open"
                    );
                    #[cfg(feature = "metrics")]
                    crate::metrics::circuit_transition("closed_to_open");
                }
            }
            CircuitState::HalfOpen => {
                // `consecutive_failures` is only read in the Closed arm; the path back
                // to Closed (via HalfOpen -> Closed) resets it. No write needed here.
                inner.transition(CircuitState::Open {
                    until: Instant::now() + self.policy.reset_timeout,
                });
                #[cfg(feature = "tracing")]
                info!(
                    reset_timeout_ms = self.policy.reset_timeout.as_millis() as u64,
                    "circuit breaker transition: HalfOpen -> Open"
                );
                #[cfg(feature = "metrics")]
                crate::metrics::circuit_transition("halfopen_to_open");
            }
            CircuitState::Open { .. } => {
                // Unreachable when epoch is current; the epoch check above filters
                // stale outcomes from prior state-machine sessions.
                debug_assert!(
                    false,
                    "record_failure on Open with current epoch is unreachable; epoch tracking should have filtered the stale outcome"
                );
            }
        }
    }
}

#[resource]
impl Resource for CircuitBreaker {}

/// A token that authorises one call against a [`CircuitBreaker`].
///
/// Consume it by passing it to [`CircuitBreaker::record_success`] or
/// [`CircuitBreaker::record_failure`]. If a `Permit` is dropped without being
/// consumed it is treated as a failure — this keeps the in-flight counter accurate
/// when the caller panics or returns early.
#[must_use = "drop a Permit only as a deliberate failure signal; pass it to record_success or record_failure to indicate the call outcome"]
pub struct Permit {
    breaker: Arc<CircuitBreaker>,
    was_half_open: bool,
    consumed: bool,
    /// Epoch the permit was issued in. Used at consumption time to ignore
    /// outcomes whose state-machine session is no longer current.
    epoch: u64,
}

impl Permit {
    fn new(breaker: Arc<CircuitBreaker>, was_half_open: bool, epoch: u64) -> Self {
        Self {
            breaker,
            was_half_open,
            consumed: false,
            epoch,
        }
    }
}

impl std::fmt::Debug for Permit {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("Permit")
            .field("was_half_open", &self.was_half_open)
            .field("consumed", &self.consumed)
            .field("epoch", &self.epoch)
            .finish()
    }
}

impl Drop for Permit {
    fn drop(&mut self) {
        if !self.consumed {
            self.breaker
                .do_record_failure(self.was_half_open, self.epoch);
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn fast_policy() -> CircuitPolicy {
        CircuitPolicy {
            failure_threshold: 3,
            reset_timeout: Duration::from_millis(20),
            half_open_max_calls: 2,
        }
    }

    #[test]
    fn closed_initial_state() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        assert_eq!(breaker.state(), CircuitState::Closed);
    }

    #[test]
    fn opens_after_threshold_failures() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..3 {
            let permit = breaker.try_acquire().unwrap();
            breaker.record_failure(permit);
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));

        let err = breaker.try_acquire().unwrap_err();
        assert_eq!(err.category(), "circuit_open");
    }

    #[test]
    fn success_resets_consecutive_failures() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..2 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        // One success resets the counter, so two more failures should not trip yet.
        let p = breaker.try_acquire().unwrap();
        breaker.record_success(p);
        for _ in 0..2 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert_eq!(breaker.state(), CircuitState::Closed);
    }

    #[tokio::test]
    async fn open_to_half_open_after_reset_timeout() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..3 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));

        tokio::time::sleep(Duration::from_millis(40)).await;

        // Lazy transition: try_acquire flips Open -> HalfOpen.
        let permit = breaker.try_acquire().unwrap();
        assert_eq!(breaker.state(), CircuitState::HalfOpen);
        breaker.record_success(permit);
    }

    #[tokio::test]
    async fn half_open_full_success_closes() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..3 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        tokio::time::sleep(Duration::from_millis(40)).await;

        // half_open_max_calls = 2, so two consecutive successes close the breaker.
        let p1 = breaker.try_acquire().unwrap();
        breaker.record_success(p1);
        let p2 = breaker.try_acquire().unwrap();
        breaker.record_success(p2);
        assert_eq!(breaker.state(), CircuitState::Closed);
    }

    #[tokio::test]
    async fn half_open_failure_reopens() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..3 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        tokio::time::sleep(Duration::from_millis(40)).await;

        let p = breaker.try_acquire().unwrap();
        assert_eq!(breaker.state(), CircuitState::HalfOpen);
        breaker.record_failure(p);
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
    }

    #[tokio::test]
    async fn half_open_caps_concurrent_trials() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..3 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        tokio::time::sleep(Duration::from_millis(40)).await;

        // half_open_max_calls = 2: two trial permits in flight, third must be rejected.
        let p1 = breaker.try_acquire().unwrap();
        let p2 = breaker.try_acquire().unwrap();
        let err = breaker.try_acquire().unwrap_err();
        assert_eq!(err.category(), "circuit_open");
        breaker.record_success(p1);
        breaker.record_success(p2);
    }

    #[test]
    fn dropped_permit_counts_as_failure() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..3 {
            let _p = breaker.try_acquire().unwrap();
            // _p drops here without consume → treated as failure.
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
    }

    #[test]
    fn shared_breaker_trips_for_all_callers() {
        // Two callers share one breaker; failures from caller A trip it for caller B.
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        let breaker_a = Arc::clone(&breaker);
        let breaker_b = Arc::clone(&breaker);

        for _ in 0..3 {
            let p = breaker_a.try_acquire().unwrap();
            breaker_a.record_failure(p);
        }

        let err = breaker_b.try_acquire().unwrap_err();
        assert_eq!(err.category(), "circuit_open");
    }

    #[test]
    #[should_panic(expected = "half_open_max_calls must be >= 1")]
    fn rejects_zero_half_open_max_calls() {
        // Misconfigured policy: zero trial slots would deadlock the breaker
        // permanently in HalfOpen. Construction must fail loudly.
        let _ = CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 1,
            reset_timeout: Duration::from_millis(1),
            half_open_max_calls: 0,
        });
    }

    #[test]
    #[should_panic(expected = "failure_threshold must be >= 1")]
    fn rejects_zero_failure_threshold() {
        let _ = CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 0,
            reset_timeout: Duration::from_millis(1),
            half_open_max_calls: 1,
        });
    }

    #[tokio::test]
    async fn stale_closed_permit_does_not_close_a_later_half_open() {
        // Reproduces the epoch-tracking concern: a Closed permit issued before
        // a trip is consumed during a later HalfOpen probe. Without epoch
        // tracking, `record_success` would see `state == HalfOpen` and bump
        // half_open_successes — closing the breaker on a stale outcome that
        // was never a probe call. With epoch tracking the stale outcome must
        // be ignored and the HalfOpen probe must still need its own success
        // to close the breaker.
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 2,
            reset_timeout: Duration::from_millis(20),
            half_open_max_calls: 1,
        }));

        // Acquire a permit while still Closed — this is the "slow caller".
        let stale_permit = breaker.try_acquire().unwrap();
        assert_eq!(breaker.state(), CircuitState::Closed);

        // Trip the breaker via two other failures (independent of the stale permit).
        for _ in 0..2 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));

        // Wait out the cool-down so the next try_acquire promotes to HalfOpen.
        tokio::time::sleep(Duration::from_millis(40)).await;
        // Provoke the lazy promotion without consuming the trial slot.
        let probe_permit = breaker.try_acquire().unwrap();
        assert_eq!(breaker.state(), CircuitState::HalfOpen);

        // Now consume the *stale* permit with a success. With epoch tracking
        // this is a no-op — it must NOT close the breaker.
        breaker.record_success(stale_permit);
        assert_eq!(
            breaker.state(),
            CircuitState::HalfOpen,
            "stale Closed-epoch success must not close a later HalfOpen"
        );

        // The actual probe still needs its own success to close.
        breaker.record_success(probe_permit);
        assert_eq!(breaker.state(), CircuitState::Closed);
    }

    #[test]
    fn open_state_is_debug_loggable() {
        // Operators typically log breaker state via Debug; pin the format so
        // accidental refactors that drop the `Open { until }` shape are caught.
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        for _ in 0..3 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        let dbg = format!("{:?}", breaker.state());
        assert!(
            dbg.starts_with("Open { until:"),
            "expected Debug to start with `Open {{ until:`, got: {dbg}"
        );
    }

    #[test]
    #[should_panic(expected = "must be <=")]
    fn rejects_pathological_half_open_max_calls() {
        // Values near u32::MAX would either saturate half_open_successes before
        // crossing the threshold or take effectively forever — both classed as
        // misconfiguration and rejected at construction.
        let _ = CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 1,
            reset_timeout: Duration::from_millis(1),
            half_open_max_calls: u32::MAX,
        });
    }

    #[tokio::test]
    async fn open_half_open_open_round_trip() {
        // Full lifecycle: trip → cool-down → HalfOpen → probe fails → re-Open
        // → cool-down → HalfOpen → probe succeeds → Closed.
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 2,
            reset_timeout: Duration::from_millis(20),
            half_open_max_calls: 1,
        }));

        // 1. Trip from Closed.
        for _ in 0..2 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));

        // 2. Cool-down → HalfOpen → failed probe → re-Open with fresh deadline.
        tokio::time::sleep(Duration::from_millis(40)).await;
        let p = breaker.try_acquire().unwrap();
        assert_eq!(breaker.state(), CircuitState::HalfOpen);
        breaker.record_failure(p);
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));

        // 3. Second cool-down → HalfOpen → successful probe → Closed.
        tokio::time::sleep(Duration::from_millis(40)).await;
        let p = breaker.try_acquire().unwrap();
        assert_eq!(breaker.state(), CircuitState::HalfOpen);
        breaker.record_success(p);
        assert_eq!(breaker.state(), CircuitState::Closed);

        // 4. Sanity: the closed breaker once again admits calls and fresh
        //    failures count toward a new threshold (counters were reset).
        for _ in 0..2 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
    }

    // ----- edge cases -----

    #[test]
    fn policy_accessor_returns_configuration() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy()));
        assert_eq!(breaker.policy().failure_threshold, 3);
        assert_eq!(breaker.policy().half_open_max_calls, 2);
    }

    #[test]
    fn default_policy_trips_after_five_failures() {
        // Default: failure_threshold 5, reset_timeout 30s, half_open_max_calls 1.
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy::default()));
        assert_eq!(breaker.state(), CircuitState::Closed);
        for _ in 0..5 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
    }

    #[test]
    fn closed_failures_below_threshold_do_not_trip() {
        let breaker = Arc::new(CircuitBreaker::new(fast_policy())); // threshold 3
        for _ in 0..2 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert_eq!(breaker.state(), CircuitState::Closed); // 2 < 3
    }

    #[tokio::test]
    async fn dropped_half_open_trial_permit_reopens() {
        // An unconsumed trial permit dropped during HalfOpen is treated as a failure -> reopen
        // (the RAII failure path, exercised in HalfOpen rather than Closed).
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 1,
            reset_timeout: Duration::from_millis(20),
            half_open_max_calls: 1,
        }));
        let p = breaker.try_acquire().unwrap();
        breaker.record_failure(p);
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
        tokio::time::sleep(Duration::from_millis(40)).await;
        {
            let _trial = breaker.try_acquire().unwrap();
            assert_eq!(breaker.state(), CircuitState::HalfOpen);
            // `_trial` drops here unconsumed -> recorded as a failure -> reopen.
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
    }

    #[tokio::test]
    async fn consuming_a_half_open_trial_frees_a_slot() {
        // max_calls = 3: fill all three trial slots, the fourth is rejected; consuming one
        // (success) decrements the in-flight count so a fresh trial is admitted into the slot.
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 1,
            reset_timeout: Duration::from_millis(20),
            half_open_max_calls: 3,
        }));
        let p = breaker.try_acquire().unwrap();
        breaker.record_failure(p);
        tokio::time::sleep(Duration::from_millis(40)).await;

        let p1 = breaker.try_acquire().unwrap(); // promotes to HalfOpen, slot 1
        let p2 = breaker.try_acquire().unwrap(); // slot 2
        let p3 = breaker.try_acquire().unwrap(); // slot 3 (full)
        assert_eq!(breaker.state(), CircuitState::HalfOpen);
        assert_eq!(
            breaker.try_acquire().unwrap_err().category(),
            "circuit_open" // fourth rejected — slots exhausted
        );

        breaker.record_success(p1); // frees a slot; successes 1 of 3 -> still HalfOpen
        assert_eq!(breaker.state(), CircuitState::HalfOpen);
        let p4 = breaker
            .try_acquire()
            .expect("a slot freed by the consumed trial admits a fresh probe");

        // Close it out: the third success crosses the threshold.
        breaker.record_success(p2);
        breaker.record_success(p3);
        assert_eq!(breaker.state(), CircuitState::Closed);
        // p4 is now stale (the close bumped the epoch); consuming it is a no-op.
        breaker.record_success(p4);
        assert_eq!(breaker.state(), CircuitState::Closed);
    }

    #[tokio::test]
    async fn half_open_partial_success_does_not_carry_across_a_reopen() {
        // max_calls = 2: one success then a failure reopens. The prior success must NOT persist,
        // so after the next promotion two FRESH successes are required to close.
        let breaker = Arc::new(CircuitBreaker::new(fast_policy())); // threshold 3, max_calls 2
        for _ in 0..3 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        tokio::time::sleep(Duration::from_millis(40)).await;

        let p1 = breaker.try_acquire().unwrap(); // HalfOpen, slot 1
        let p2 = breaker.try_acquire().unwrap(); // slot 2
        breaker.record_success(p1); // successes 1 of 2
        breaker.record_failure(p2); // any failure reopens; the success counter is discarded
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));

        tokio::time::sleep(Duration::from_millis(40)).await;
        let q1 = breaker.try_acquire().unwrap(); // fresh HalfOpen, successes start at 0
        breaker.record_success(q1);
        assert_eq!(
            breaker.state(),
            CircuitState::HalfOpen,
            "one fresh success must not close — the pre-reopen success did not carry over"
        );
        let q2 = breaker.try_acquire().unwrap();
        breaker.record_success(q2);
        assert_eq!(breaker.state(), CircuitState::Closed);
    }

    #[tokio::test]
    async fn stale_half_open_permit_ignored_after_reopen() {
        // Two trials in flight; one fails and reopens (epoch bumps). The other, consumed late, is
        // stale and must be ignored — it cannot close or otherwise mutate the reopened breaker.
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 1,
            reset_timeout: Duration::from_millis(20),
            half_open_max_calls: 2,
        }));
        let p = breaker.try_acquire().unwrap();
        breaker.record_failure(p);
        tokio::time::sleep(Duration::from_millis(40)).await;

        let p1 = breaker.try_acquire().unwrap(); // HalfOpen
        let p2 = breaker.try_acquire().unwrap();
        breaker.record_failure(p1); // reopens; epoch bumps -> p2 is now stale
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
        breaker.record_success(p2); // stale -> no-op
        assert!(
            matches!(breaker.state(), CircuitState::Open { .. }),
            "a stale HalfOpen success must not mutate the reopened breaker"
        );
    }

    #[test]
    fn stale_closed_failure_ignored_after_trip() {
        // A Closed permit consumed *as a failure* after the breaker has already tripped (via other
        // failures) is stale and must be ignored — this also guards the `unreachable` Open-arm
        // debug_assert in do_record_failure from firing on a stale outcome.
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 2,
            reset_timeout: Duration::from_secs(60),
            half_open_max_calls: 1,
        }));
        let stale = breaker.try_acquire().unwrap(); // Closed, epoch 0
        for _ in 0..2 {
            let p = breaker.try_acquire().unwrap();
            breaker.record_failure(p);
        }
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
        breaker.record_failure(stale); // epoch mismatch -> no-op, no panic
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
    }

    #[test]
    fn concurrent_failures_trip_without_corruption() {
        // Many threads hammer try_acquire + record_failure on one shared breaker. The state
        // machine (a std Mutex) must stay consistent: the breaker ends Open and never panics or
        // deadlocks. Plain threads (the breaker is synchronous) keep the load light.
        let breaker = Arc::new(CircuitBreaker::new(CircuitPolicy {
            failure_threshold: 5,
            reset_timeout: Duration::from_secs(60),
            half_open_max_calls: 1,
        }));
        std::thread::scope(|s| {
            for _ in 0..6 {
                let b = Arc::clone(&breaker);
                s.spawn(move || {
                    for _ in 0..30 {
                        if let Ok(permit) = b.try_acquire() {
                            b.record_failure(permit);
                        }
                    }
                });
            }
        });
        // 6 × 30 attempts ≫ threshold 5 -> Open; state() reads cleanly (no poison / deadlock).
        assert!(matches!(breaker.state(), CircuitState::Open { .. }));
    }
}

#[cfg(all(test, feature = "metrics"))]
mod metrics_tests {
    use super::*;
    use crate::metrics::test_support::*;
    use std::sync::Arc;
    use std::time::Duration;

    #[test]
    fn closed_to_open_and_outcomes_are_recorded() {
        let ((), rows) = run_with_recorder(|| async {
            let b = Arc::new(CircuitBreaker::new(CircuitPolicy {
                failure_threshold: 2,
                reset_timeout: Duration::from_secs(60),
                half_open_max_calls: 1,
            }));
            let p = b.try_acquire().unwrap();
            b.record_success(p);
            let p = b.try_acquire().unwrap();
            b.record_failure(p);
            let p = b.try_acquire().unwrap();
            b.record_failure(p);
            assert!(b.try_acquire().is_err());
        });
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_acquires_total",
                &[("result", "acquired")]
            ),
            3
        );
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_acquires_total",
                &[("result", "rejected")]
            ),
            1
        );
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_outcomes_total",
                &[("outcome", "success")]
            ),
            1
        );
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_outcomes_total",
                &[("outcome", "failure")]
            ),
            2
        );
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_transitions_total",
                &[("transition", "closed_to_open")]
            ),
            1
        );
    }

    #[test]
    fn dropped_unconsumed_permit_counts_as_failure() {
        let ((), rows) = run_with_recorder(|| async {
            let b = Arc::new(CircuitBreaker::new(CircuitPolicy::default()));
            let _p = b.try_acquire().unwrap();
        });
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_outcomes_total",
                &[("outcome", "failure")]
            ),
            1
        );
    }

    #[test]
    fn half_open_transitions_are_recorded() {
        // Exercise the recovery-side transitions the closed_to_open test doesn't reach:
        // open_to_halfopen (lazy promotion) and halfopen_to_closed (successful probe).
        let ((), rows) = run_with_recorder(|| async {
            let b = Arc::new(CircuitBreaker::new(CircuitPolicy {
                failure_threshold: 1,
                reset_timeout: Duration::from_millis(20),
                half_open_max_calls: 1,
            }));
            let p = b.try_acquire().unwrap();
            b.record_failure(p); // closed -> open
            tokio::time::sleep(Duration::from_millis(40)).await;
            let p = b.try_acquire().unwrap(); // open -> halfopen (lazy)
            b.record_success(p); // halfopen -> closed
        });
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_transitions_total",
                &[("transition", "closed_to_open")]
            ),
            1
        );
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_transitions_total",
                &[("transition", "open_to_halfopen")]
            ),
            1
        );
        assert_eq!(
            counter(
                &rows,
                "cano_circuit_transitions_total",
                &[("transition", "halfopen_to_closed")]
            ),
            1
        );
    }
}