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
// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License in the LICENSE file at the
// root of this repository, or online at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

//! Metrics for materialize systems.
//!
//! The idea here is that each subsystem keeps its metrics in a scoped-to-it struct, which gets
//! registered (once) to the server's (or a test's) prometheus registry.
//!
//! Instead of using prometheus's (very verbose) metrics definitions, we rely on type inference to
//! reduce the verbosity a little bit. A typical subsystem will look like the following:
//!
//! ```rust
//! # use mz_ore::metrics::{MetricsRegistry, IntCounter};
//! # use mz_ore::metric;
//! #[derive(Debug, Clone)] // Note that prometheus metrics can safely be cloned
//! struct Metrics {
//!     pub bytes_sent: IntCounter,
//! }
//!
//! impl Metrics {
//!     pub fn register_into(registry: &MetricsRegistry) -> Metrics {
//!         Metrics {
//!             bytes_sent: registry.register(metric!(
//!                 name: "mz_pg_sent_bytes",
//!                 help: "total number of bytes sent here",
//!             )),
//!         }
//!     }
//! }
//! ```

use std::fmt;
use std::fmt::{Debug, Formatter};
use std::future::Future;
use std::pin::Pin;
use std::sync::{Arc, Mutex};
use std::task::{Context, Poll};
use std::time::{Duration, Instant};

use derivative::Derivative;
use pin_project::pin_project;
use prometheus::core::{
    Atomic, AtomicF64, AtomicI64, AtomicU64, Collector, Desc, GenericCounter, GenericCounterVec,
    GenericGauge, GenericGaugeVec,
};
use prometheus::proto::MetricFamily;
use prometheus::{HistogramOpts, Registry};

mod delete_on_drop;

pub use delete_on_drop::*;
pub use prometheus::Opts as PrometheusOpts;

/// Define a metric for use in materialize.
#[macro_export]
macro_rules! metric {
    (
        name: $name:expr,
        help: $help:expr
        $(, subsystem: $subsystem_name:expr)?
        $(, const_labels: { $($cl_key:expr => $cl_value:expr ),* })?
        $(, var_labels: [ $($vl_name:expr),* ])?
        $(, buckets: $bk_name:expr)?
        $(,)?
    ) => {{
        let const_labels = (&[
            $($(
                ($cl_key.to_string(), $cl_value.to_string()),
            )*)?
        ]).into_iter().cloned().collect();
        let var_labels = vec![
            $(
                $($vl_name.into(),)*
            )?];
        #[allow(unused_mut)]
        let mut mk_opts = $crate::metrics::MakeCollectorOpts {
            opts: $crate::metrics::PrometheusOpts::new($name, $help)
                $(.subsystem( $subsystem_name ))?
                .const_labels(const_labels)
                .variable_labels(var_labels),
            buckets: None,
        };
        // Set buckets if passed
        $(mk_opts.buckets = Some($bk_name);)*
        mk_opts
    }}
}

/// Options for MakeCollector. This struct should be instantiated using the metric macro.
#[derive(Debug, Clone)]
pub struct MakeCollectorOpts {
    /// Common Prometheus options
    pub opts: PrometheusOpts,
    /// Buckets to be used with Histogram and HistogramVec. Must be set to create Histogram types
    /// and must not be set for other types.
    pub buckets: Option<Vec<f64>>,
}

/// The materialize metrics registry.
#[derive(Clone, Derivative)]
#[derivative(Debug)]
pub struct MetricsRegistry {
    inner: Registry,
    #[derivative(Debug = "ignore")]
    postprocessors: Arc<Mutex<Vec<Box<dyn FnMut(&mut Vec<MetricFamily>) + Send + Sync>>>>,
}

/// A wrapper for metrics to require delete on drop semantics
///
/// The wrapper behaves like regular metrics but only provides functions to create delete-on-drop
/// variants. This way, no metrics of this type can be leaked.
///
/// In situations where the delete-on-drop behavior is not desired or in legacy code, use the raw
/// variants of the metrics, as defined in [self::raw].
#[derive(Clone)]
pub struct DeleteOnDropWrapper<M> {
    inner: M,
}

impl<M: MakeCollector + Debug> Debug for DeleteOnDropWrapper<M> {
    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
        self.inner.fmt(f)
    }
}

impl<M: Collector> Collector for DeleteOnDropWrapper<M> {
    fn desc(&self) -> Vec<&Desc> {
        self.inner.desc()
    }

    fn collect(&self) -> Vec<MetricFamily> {
        self.inner.collect()
    }
}

impl<M: MakeCollector> MakeCollector for DeleteOnDropWrapper<M> {
    fn make_collector(opts: MakeCollectorOpts) -> Self {
        DeleteOnDropWrapper {
            inner: M::make_collector(opts),
        }
    }
}

impl<M: MetricVecExt> DeleteOnDropWrapper<M> {
    /// Returns a metric that deletes its labels from this metrics vector when dropped.
    pub fn get_delete_on_drop_metric<'a, L: PromLabelsExt<'a>>(
        &self,
        labels: L,
    ) -> DeleteOnDropMetric<'a, M, L> {
        self.inner.get_delete_on_drop_metric(labels)
    }
}

/// The unsigned integer version of [`Gauge`]. Provides better performance if
/// metric values are all unsigned integers.
pub type UIntGauge = GenericGauge<AtomicU64>;

/// Delete-on-drop shadow of Prometheus [prometheus::CounterVec].
pub type CounterVec = DeleteOnDropWrapper<prometheus::CounterVec>;
/// Delete-on-drop shadow of Prometheus [prometheus::Gauge].
pub type Gauge = DeleteOnDropWrapper<prometheus::Gauge>;
/// Delete-on-drop shadow of Prometheus [prometheus::CounterVec].
pub type GaugeVec = DeleteOnDropWrapper<prometheus::GaugeVec>;
/// Delete-on-drop shadow of Prometheus [prometheus::HistogramVec].
pub type HistogramVec = DeleteOnDropWrapper<prometheus::HistogramVec>;
/// Delete-on-drop shadow of Prometheus [prometheus::IntCounterVec].
pub type IntCounterVec = DeleteOnDropWrapper<prometheus::IntCounterVec>;
/// Delete-on-drop shadow of Prometheus [prometheus::IntGaugeVec].
pub type IntGaugeVec = DeleteOnDropWrapper<prometheus::IntGaugeVec>;
/// Delete-on-drop shadow of Prometheus [raw::UIntGaugeVec].
pub type UIntGaugeVec = DeleteOnDropWrapper<raw::UIntGaugeVec>;

pub use prometheus::{Counter, Histogram, IntCounter, IntGauge};

use crate::assert_none;

/// Access to non-delete-on-drop vector types
pub mod raw {
    use prometheus::core::{AtomicU64, GenericGaugeVec};

    /// The unsigned integer version of [`GaugeVec`].
    /// Provides better performance if metric values are all unsigned integers.
    pub type UIntGaugeVec = GenericGaugeVec<AtomicU64>;

    pub use prometheus::{CounterVec, Gauge, GaugeVec, HistogramVec, IntCounterVec, IntGaugeVec};
}

impl MetricsRegistry {
    /// Creates a new metrics registry.
    pub fn new() -> Self {
        MetricsRegistry {
            inner: Registry::new(),
            postprocessors: Arc::new(Mutex::new(vec![])),
        }
    }

    /// Register a metric defined with the [`metric`] macro.
    pub fn register<M>(&self, opts: MakeCollectorOpts) -> M
    where
        M: MakeCollector,
    {
        let collector = M::make_collector(opts);
        self.inner.register(Box::new(collector.clone())).unwrap();
        collector
    }

    /// Registers a gauge whose value is computed when observed.
    pub fn register_computed_gauge<F, P>(
        &self,
        opts: MakeCollectorOpts,
        f: F,
    ) -> ComputedGenericGauge<P>
    where
        F: Fn() -> P::T + Send + Sync + 'static,
        P: Atomic + 'static,
    {
        let gauge = ComputedGenericGauge {
            gauge: GenericGauge::make_collector(opts),
            f: Arc::new(f),
        };
        self.inner.register(Box::new(gauge.clone())).unwrap();
        gauge
    }

    /// Register a pre-defined prometheus collector.
    pub fn register_collector<C: 'static + prometheus::core::Collector>(&self, collector: C) {
        self.inner
            .register(Box::new(collector))
            .expect("registering pre-defined metrics collector");
    }

    /// Registers a metric postprocessor.
    ///
    /// Postprocessors are invoked on every call to [`MetricsRegistry::gather`]
    /// in the order that they are registered.
    pub fn register_postprocessor<F>(&self, f: F)
    where
        F: FnMut(&mut Vec<MetricFamily>) + Send + Sync + 'static,
    {
        let mut postprocessors = self.postprocessors.lock().expect("lock poisoned");
        postprocessors.push(Box::new(f));
    }

    /// Gather all the metrics from the metrics registry for reporting.
    ///
    /// This function invokes the postprocessors on all gathered metrics (see
    /// [`MetricsRegistry::register_postprocessor`]) in the order the
    /// postprocessors were registered.
    ///
    /// See also [`prometheus::Registry::gather`].
    pub fn gather(&self) -> Vec<MetricFamily> {
        let mut metrics = self.inner.gather();
        let mut postprocessors = self.postprocessors.lock().expect("lock poisoned");
        for postprocessor in &mut *postprocessors {
            postprocessor(&mut metrics);
        }
        metrics
    }
}

/// A wrapper for creating prometheus metrics more conveniently.
///
/// Together with the [`metric`] macro, this trait is mainly used by [`MetricsRegistry`] and should
/// not normally be used outside the metric registration flow.
pub trait MakeCollector: Collector + Clone + 'static {
    /// Creates a new collector.
    fn make_collector(opts: MakeCollectorOpts) -> Self;
}

impl<T> MakeCollector for GenericCounter<T>
where
    T: Atomic + 'static,
{
    fn make_collector(mk_opts: MakeCollectorOpts) -> Self {
        assert_none!(mk_opts.buckets);
        Self::with_opts(mk_opts.opts).expect("defining a counter")
    }
}

impl<T> MakeCollector for GenericCounterVec<T>
where
    T: Atomic + 'static,
{
    fn make_collector(mk_opts: MakeCollectorOpts) -> Self {
        assert_none!(mk_opts.buckets);
        let labels: Vec<String> = mk_opts.opts.variable_labels.clone();
        let label_refs: Vec<&str> = labels.iter().map(String::as_str).collect();
        Self::new(mk_opts.opts, label_refs.as_slice()).expect("defining a counter vec")
    }
}

impl<T> MakeCollector for GenericGauge<T>
where
    T: Atomic + 'static,
{
    fn make_collector(mk_opts: MakeCollectorOpts) -> Self {
        assert_none!(mk_opts.buckets);
        Self::with_opts(mk_opts.opts).expect("defining a gauge")
    }
}

impl<T> MakeCollector for GenericGaugeVec<T>
where
    T: Atomic + 'static,
{
    fn make_collector(mk_opts: MakeCollectorOpts) -> Self {
        assert_none!(mk_opts.buckets);
        let labels = mk_opts.opts.variable_labels.clone();
        let labels = &labels.iter().map(|x| x.as_str()).collect::<Vec<_>>();
        Self::new(mk_opts.opts, labels).expect("defining a gauge vec")
    }
}

impl MakeCollector for Histogram {
    fn make_collector(mk_opts: MakeCollectorOpts) -> Self {
        assert!(mk_opts.buckets.is_some());
        Self::with_opts(HistogramOpts {
            common_opts: mk_opts.opts,
            buckets: mk_opts.buckets.unwrap(),
        })
        .expect("defining a histogram")
    }
}

impl MakeCollector for raw::HistogramVec {
    fn make_collector(mk_opts: MakeCollectorOpts) -> Self {
        assert!(mk_opts.buckets.is_some());
        let labels = mk_opts.opts.variable_labels.clone();
        let labels = &labels.iter().map(|x| x.as_str()).collect::<Vec<_>>();
        Self::new(
            HistogramOpts {
                common_opts: mk_opts.opts,
                buckets: mk_opts.buckets.unwrap(),
            },
            labels,
        )
        .expect("defining a histogram vec")
    }
}

/// A [`Gauge`] whose value is computed whenever it is observed.
pub struct ComputedGenericGauge<P>
where
    P: Atomic,
{
    gauge: GenericGauge<P>,
    f: Arc<dyn Fn() -> P::T + Send + Sync>,
}

impl<P> fmt::Debug for ComputedGenericGauge<P>
where
    P: Atomic + fmt::Debug,
{
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        f.debug_struct("ComputedGenericGauge")
            .field("gauge", &self.gauge)
            .finish_non_exhaustive()
    }
}

impl<P> Clone for ComputedGenericGauge<P>
where
    P: Atomic,
{
    fn clone(&self) -> ComputedGenericGauge<P> {
        ComputedGenericGauge {
            gauge: self.gauge.clone(),
            f: Arc::clone(&self.f),
        }
    }
}

impl<T> Collector for ComputedGenericGauge<T>
where
    T: Atomic,
{
    fn desc(&self) -> Vec<&prometheus::core::Desc> {
        self.gauge.desc()
    }

    fn collect(&self) -> Vec<MetricFamily> {
        self.gauge.set((self.f)());
        self.gauge.collect()
    }
}

impl<P> ComputedGenericGauge<P>
where
    P: Atomic,
{
    /// Computes the current value of the gauge.
    pub fn get(&self) -> P::T {
        (self.f)()
    }
}

/// A [`ComputedGenericGauge`] for 64-bit floating point numbers.
pub type ComputedGauge = ComputedGenericGauge<AtomicF64>;

/// A [`ComputedGenericGauge`] for 64-bit signed integers.
pub type ComputedIntGauge = ComputedGenericGauge<AtomicI64>;

/// A [`ComputedGenericGauge`] for 64-bit unsigned integers.
pub type ComputedUIntGauge = ComputedGenericGauge<AtomicU64>;

/// Exposes combinators that report metrics related to the execution of a [`Future`] to prometheus.
pub trait MetricsFutureExt<F> {
    /// Records the number of seconds it takes a [`Future`] to complete according to "the clock on
    /// the wall".
    ///
    /// More specifically, it records the instant at which the `Future` was first polled, and the
    /// instant at which the `Future` completes. Then reports the duration between those two
    /// instances to the provided metric.
    ///
    /// # Wall Time vs Execution Time
    ///
    /// There is also [`MetricsFutureExt::exec_time`], which measures how long a [`Future`] spent
    /// executing, instead of how long it took to complete. For example, a network request may have
    /// a wall time of 1 second, meanwhile it's execution time may have only been 50ms. The 950ms
    /// delta would be how long the [`Future`] waited for a response from the network.
    ///
    /// # Uses
    ///
    /// Recording the wall time can be useful for monitoring latency, for example the latency of a
    /// SQL request.
    ///
    /// Note: You must call either [`observe`] to record the execution time to a [`Histogram`] or
    /// [`inc_by`] to record to a [`Counter`]. The following will not compile:
    ///
    /// ```compile_fail
    /// use mz_ore::metrics::MetricsFutureExt;
    ///
    /// # let _ = async {
    /// async { Ok(()) }
    ///     .wall_time()
    ///     .await;
    /// # };
    /// ```
    ///
    /// [`observe`]: WallTimeFuture::observe
    /// [`inc_by`]: WallTimeFuture::inc_by
    fn wall_time(self) -> WallTimeFuture<F, UnspecifiedMetric>;

    /// Records the total number of seconds for which a [`Future`] was executing.
    ///
    /// More specifically, every time the `Future` is polled it records how long that individual
    /// call took, and maintains a running sum until the `Future` completes. Then we report that
    /// duration to the provided metric.
    ///
    /// # Wall Time vs Execution Time
    ///
    /// There is also [`MetricsFutureExt::wall_time`], which measures how long a [`Future`] took to
    /// complete, instead of how long it spent executing. For example, a network request may have
    /// a wall time of 1 second, meanwhile it's execution time may have only been 50ms. The 950ms
    /// delta would be how long the [`Future`] waited for a response from the network.
    ///
    /// # Uses
    ///
    /// Recording execution time can be useful if you want to monitor [`Future`]s that could be
    /// sensitive to CPU usage. For example, if you have a single logical control thread you'll
    /// want to make sure that thread never spends too long running a single `Future`. Reporting
    /// the execution time of `Future`s running on this thread can help ensure there is no
    /// unexpected blocking.
    ///
    /// Note: You must call either [`observe`] to record the execution time to a [`Histogram`] or
    /// [`inc_by`] to record to a [`Counter`]. The following will not compile:
    ///
    /// ```compile_fail
    /// use mz_ore::metrics::MetricsFutureExt;
    ///
    /// # let _ = async {
    /// async { Ok(()) }
    ///     .exec_time()
    ///     .await;
    /// # };
    /// ```
    ///
    /// [`observe`]: ExecTimeFuture::observe
    /// [`inc_by`]: ExecTimeFuture::inc_by
    fn exec_time(self) -> ExecTimeFuture<F, UnspecifiedMetric>;
}

impl<F: Future> MetricsFutureExt<F> for F {
    fn wall_time(self) -> WallTimeFuture<F, UnspecifiedMetric> {
        WallTimeFuture {
            fut: self,
            metric: UnspecifiedMetric(()),
            start: None,
            filter: None,
        }
    }

    fn exec_time(self) -> ExecTimeFuture<F, UnspecifiedMetric> {
        ExecTimeFuture {
            fut: self,
            metric: UnspecifiedMetric(()),
            running_duration: Duration::from_millis(0),
            filter: None,
        }
    }
}

/// Future returned by [`MetricsFutureExt::wall_time`].
#[must_use = "futures do nothing unless you `.await` or poll them"]
#[pin_project]
pub struct WallTimeFuture<F, Metric> {
    /// The inner [`Future`] that we're recording the wall time for.
    #[pin]
    fut: F,
    /// Prometheus metric that we'll report to.
    metric: Metric,
    /// [`Instant`] at which the [`Future`] was first polled.
    start: Option<Instant>,
    /// Optional filter that determines if we observe the wall time of this [`Future`].
    filter: Option<Box<dyn FnMut(Duration) -> bool + Send + Sync>>,
}

impl<F: Debug, M: Debug> fmt::Debug for WallTimeFuture<F, M> {
    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
        f.debug_struct("WallTimeFuture")
            .field("fut", &self.fut)
            .field("metric", &self.metric)
            .field("start", &self.start)
            .field("filter", &self.filter.is_some())
            .finish()
    }
}

impl<F> WallTimeFuture<F, UnspecifiedMetric> {
    /// Sets the recored metric to be a [`prometheus::Histogram`].
    ///
    /// ```text
    /// my_future
    ///     .wall_time()
    ///     .observe(metrics.slow_queries_hist.with_label_values(&["select"]))
    /// ```
    pub fn observe(
        self,
        histogram: prometheus::Histogram,
    ) -> WallTimeFuture<F, prometheus::Histogram> {
        WallTimeFuture {
            fut: self.fut,
            metric: histogram,
            start: self.start,
            filter: self.filter,
        }
    }

    /// Sets the recored metric to be a [`prometheus::Counter`].
    ///
    /// ```text
    /// my_future
    ///     .wall_time()
    ///     .inc_by(metrics.slow_queries.with_label_values(&["select"]))
    /// ```
    pub fn inc_by(self, counter: prometheus::Counter) -> WallTimeFuture<F, prometheus::Counter> {
        WallTimeFuture {
            fut: self.fut,
            metric: counter,
            start: self.start,
            filter: self.filter,
        }
    }

    /// Sets the recorded duration in a specific f64.
    pub fn set_at(self, place: &mut f64) -> WallTimeFuture<F, &mut f64> {
        WallTimeFuture {
            fut: self.fut,
            metric: place,
            start: self.start,
            filter: self.filter,
        }
    }
}

impl<F, M> WallTimeFuture<F, M> {
    /// Specifies a filter which much return `true` for the wall time to be recorded.
    ///
    /// This can be particularly useful if you have a high volume `Future` and you only want to
    /// record ones that take a long time to complete.
    pub fn with_filter(
        mut self,
        filter: impl FnMut(Duration) -> bool + Send + Sync + 'static,
    ) -> Self {
        self.filter = Some(Box::new(filter));
        self
    }
}

impl<F: Future, M: DurationMetric> Future for WallTimeFuture<F, M> {
    type Output = F::Output;

    fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
        let this = self.project();

        if this.start.is_none() {
            *this.start = Some(Instant::now());
        }

        let result = match this.fut.poll(cx) {
            Poll::Ready(r) => r,
            Poll::Pending => return Poll::Pending,
        };
        let duration = Instant::now().duration_since(this.start.expect("timer to be started"));

        let pass = this
            .filter
            .as_mut()
            .map(|filter| filter(duration))
            .unwrap_or(true);
        if pass {
            this.metric.record(duration.as_secs_f64())
        }

        Poll::Ready(result)
    }
}

/// Future returned by [`MetricsFutureExt::exec_time`].
#[must_use = "futures do nothing unless you `.await` or poll them"]
#[pin_project]
pub struct ExecTimeFuture<F, Metric> {
    /// The inner [`Future`] that we're recording the wall time for.
    #[pin]
    fut: F,
    /// Prometheus metric that we'll report to.
    metric: Metric,
    /// Total [`Duration`] for which this [`Future`] has been executing.
    running_duration: Duration,
    /// Optional filter that determines if we observe the execution time of this [`Future`].
    filter: Option<Box<dyn FnMut(Duration) -> bool + Send + Sync>>,
}

impl<F: Debug, M: Debug> fmt::Debug for ExecTimeFuture<F, M> {
    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
        f.debug_struct("ExecTimeFuture")
            .field("fut", &self.fut)
            .field("metric", &self.metric)
            .field("running_duration", &self.running_duration)
            .field("filter", &self.filter.is_some())
            .finish()
    }
}

impl<F> ExecTimeFuture<F, UnspecifiedMetric> {
    /// Sets the recored metric to be a [`prometheus::Histogram`].
    ///
    /// ```text
    /// my_future
    ///     .exec_time()
    ///     .observe(metrics.slow_queries_hist.with_label_values(&["select"]))
    /// ```
    pub fn observe(
        self,
        histogram: prometheus::Histogram,
    ) -> ExecTimeFuture<F, prometheus::Histogram> {
        ExecTimeFuture {
            fut: self.fut,
            metric: histogram,
            running_duration: self.running_duration,
            filter: self.filter,
        }
    }

    /// Sets the recored metric to be a [`prometheus::Counter`].
    ///
    /// ```text
    /// my_future
    ///     .exec_time()
    ///     .inc_by(metrics.slow_queries.with_label_values(&["select"]))
    /// ```
    pub fn inc_by(self, counter: prometheus::Counter) -> ExecTimeFuture<F, prometheus::Counter> {
        ExecTimeFuture {
            fut: self.fut,
            metric: counter,
            running_duration: self.running_duration,
            filter: self.filter,
        }
    }
}

impl<F, M> ExecTimeFuture<F, M> {
    /// Specifies a filter which much return `true` for the execution time to be recorded.
    pub fn with_filter(
        mut self,
        filter: impl FnMut(Duration) -> bool + Send + Sync + 'static,
    ) -> Self {
        self.filter = Some(Box::new(filter));
        self
    }
}

impl<F: Future, M: DurationMetric> Future for ExecTimeFuture<F, M> {
    type Output = F::Output;

    fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
        let this = self.project();

        let start = Instant::now();
        let result = this.fut.poll(cx);
        let duration = Instant::now().duration_since(start);

        *this.running_duration = this.running_duration.saturating_add(duration);

        let result = match result {
            Poll::Ready(result) => result,
            Poll::Pending => return Poll::Pending,
        };

        let duration = *this.running_duration;
        let pass = this
            .filter
            .as_mut()
            .map(|filter| filter(duration))
            .unwrap_or(true);
        if pass {
            this.metric.record(duration.as_secs_f64());
        }

        Poll::Ready(result)
    }
}

/// A type level flag used to ensure callers specify the kind of metric to record for
/// [`MetricsFutureExt`].
///
/// For example, `WallTimeFuture<F, M>` only implements [`Future`] for `M` that implements
/// `DurationMetric` which [`UnspecifiedMetric`] does not. This forces users at build time to
/// call [`WallTimeFuture::observe`] or [`WallTimeFuture::inc_by`].
#[derive(Debug)]
pub struct UnspecifiedMetric(());

/// A trait makes recording a duration generic over different prometheus metrics. This allows us to
/// de-dupe the implemenation of [`Future`] for our wrapper Futures like [`WallTimeFuture`] and
/// [`ExecTimeFuture`] over different kinds of prometheus metrics.
trait DurationMetric {
    fn record(&mut self, seconds: f64);
}

impl DurationMetric for prometheus::Histogram {
    fn record(&mut self, seconds: f64) {
        self.observe(seconds)
    }
}

impl DurationMetric for prometheus::Counter {
    fn record(&mut self, seconds: f64) {
        self.inc_by(seconds)
    }
}

// An implementation of `DurationMetric` that lets the user take the recorded
// value and use it elsewhere.
impl DurationMetric for &'_ mut f64 {
    fn record(&mut self, seconds: f64) {
        **self = seconds;
    }
}

#[cfg(test)]
mod tests {
    use std::time::Duration;

    use prometheus::{CounterVec, HistogramVec};

    use crate::stats::histogram_seconds_buckets;

    use super::{MetricsFutureExt, MetricsRegistry};

    struct Metrics {
        pub wall_time_hist: HistogramVec,
        pub wall_time_cnt: CounterVec,
        pub exec_time_hist: HistogramVec,
        pub exec_time_cnt: CounterVec,
    }

    impl Metrics {
        pub fn register_into(registry: &MetricsRegistry) -> Self {
            Self {
                wall_time_hist: registry.register(metric!(
                    name: "wall_time_hist",
                    help: "help",
                    var_labels: ["action"],
                    buckets: histogram_seconds_buckets(0.000_128, 8.0),
                )),
                wall_time_cnt: registry.register(metric!(
                    name: "wall_time_cnt",
                    help: "help",
                    var_labels: ["action"],
                )),
                exec_time_hist: registry.register(metric!(
                    name: "exec_time_hist",
                    help: "help",
                    var_labels: ["action"],
                    buckets: histogram_seconds_buckets(0.000_128, 8.0),
                )),
                exec_time_cnt: registry.register(metric!(
                    name: "exec_time_cnt",
                    help: "help",
                    var_labels: ["action"],
                )),
            }
        }
    }

    #[crate::test]
    #[cfg_attr(miri, ignore)] // unsupported operation: integer-to-pointer casts and `ptr::from_exposed_addr` are not supported with `-Zmiri-strict-provenance`
    fn smoke_test_metrics_future_ext() {
        let runtime = tokio::runtime::Builder::new_current_thread()
            .enable_time()
            .build()
            .expect("failed to start runtime");
        let registry = MetricsRegistry::new();
        let metrics = Metrics::register_into(&registry);

        // Record the walltime and execution time of an async sleep.
        let async_sleep_future = async {
            tokio::time::sleep(tokio::time::Duration::from_secs(1)).await;
        };
        runtime.block_on(
            async_sleep_future
                .wall_time()
                .observe(metrics.wall_time_hist.with_label_values(&["async_sleep_w"]))
                .exec_time()
                .observe(metrics.exec_time_hist.with_label_values(&["async_sleep_e"])),
        );

        let reports = registry.gather();

        let exec_family = reports
            .iter()
            .find(|m| m.get_name() == "exec_time_hist")
            .expect("metric not found");
        let exec_metric = exec_family.get_metric();
        assert_eq!(exec_metric.len(), 1);
        assert_eq!(exec_metric[0].get_label()[0].get_value(), "async_sleep_e");

        let exec_histogram = exec_metric[0].get_histogram();
        assert_eq!(exec_histogram.get_sample_count(), 1);
        // The 4th bucket is 1ms, which we should complete faster than, but is still much quicker
        // than the 1 second we slept for.
        assert_eq!(exec_histogram.get_bucket()[3].get_cumulative_count(), 1);

        let wall_family = reports
            .iter()
            .find(|m| m.get_name() == "wall_time_hist")
            .expect("metric not found");
        let wall_metric = wall_family.get_metric();
        assert_eq!(wall_metric.len(), 1);
        assert_eq!(wall_metric[0].get_label()[0].get_value(), "async_sleep_w");

        let wall_histogram = wall_metric[0].get_histogram();
        assert_eq!(wall_histogram.get_sample_count(), 1);
        // The 13th bucket is 512ms, which the wall time should be longer than, but is also much
        // faster than the actual execution time of the async sleep.
        assert_eq!(wall_histogram.get_bucket()[12].get_cumulative_count(), 0);

        // Reset the registery to make collecting metrics easier.
        let registry = MetricsRegistry::new();
        let metrics = Metrics::register_into(&registry);

        // Record the walltime and execution time of a thread sleep.
        let thread_sleep_future = async {
            std::thread::sleep(std::time::Duration::from_secs(1));
        };
        runtime.block_on(
            thread_sleep_future
                .wall_time()
                .with_filter(|duration| duration < Duration::from_millis(10))
                .inc_by(metrics.wall_time_cnt.with_label_values(&["thread_sleep_w"]))
                .exec_time()
                .inc_by(metrics.exec_time_cnt.with_label_values(&["thread_sleep_e"])),
        );

        let reports = registry.gather();

        let exec_family = reports
            .iter()
            .find(|m| m.get_name() == "exec_time_cnt")
            .expect("metric not found");
        let exec_metric = exec_family.get_metric();
        assert_eq!(exec_metric.len(), 1);
        assert_eq!(exec_metric[0].get_label()[0].get_value(), "thread_sleep_e");

        let exec_counter = exec_metric[0].get_counter();
        // Since we're synchronously sleeping the execution time will be long.
        assert!(exec_counter.get_value() >= 1.0);

        let wall_family = reports
            .iter()
            .find(|m| m.get_name() == "wall_time_cnt")
            .expect("metric not found");
        let wall_metric = wall_family.get_metric();
        assert_eq!(wall_metric.len(), 1);

        let wall_counter = wall_metric[0].get_counter();
        // We filtered wall time to < 10ms, so our wall time metric should be filtered out.
        assert_eq!(wall_counter.get_value(), 0.0);
    }
}