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
//! Synchronized types that allow access to a `Histogram` from multiple threads.

use crate::errors::*;
use crate::{Counter, Histogram};
use std::borrow::Borrow;
use std::borrow::BorrowMut;
use std::marker::PhantomData;
use std::ops::{AddAssign, Deref, DerefMut};
use std::sync::{atomic, Arc, Mutex};
use std::time;

/// A write-only handle to a [`SyncHistogram`].
///
/// This handle allows you to record samples from multiple threads, each with its own `Recorder`,
/// concurrently. Writes to a `Recorder` are wait-free and scalable except for when the
/// [`SyncHistogram`] initiates a _phase shift_. During a phase shift, the next write on each
/// associated `Recorder` merges its results into a shared [`Histogram`] that is then made
/// available to the [`SyncHistogram`] once the phase shift completes. Phase shifts should also be
/// relatively cheap for writers, as they mainly need to perform a channel send on an unbounded,
/// lock-free channel.
///
/// An idle `Recorder` will hold up a phase shift indefinitely, or until it times out (is using
/// [`SyncHistogram::refresh_timeout`]. If a `Recorder` will remain idle for extended periods of
/// time, it should call [`Recorder::idle`], which will tell the reader not to wait for this
/// particular writer.
///
/// When a `Recorder` is dropped, all samples are made visible to the next
/// [`SyncHistogram::refresh`].
#[derive(Debug)]
pub struct Recorder<C: Counter> {
    local: Histogram<C>,
    shared: Arc<Shared<C>>,
    last_phase: usize,
}

// make it more ergonomic to record samples
impl<C: Counter> AddAssign<u64> for Recorder<C> {
    fn add_assign(&mut self, value: u64) {
        self.record(value).unwrap();
    }
}

impl<C: Counter> Clone for Recorder<C> {
    fn clone(&self) -> Self {
        // reader will have to wait for one more recorder
        {
            let mut truth = self.shared.truth.lock().unwrap();
            truth.recorders += 1;
        }

        // new recorder starts at the same phase as we do with an empty histogram
        Recorder {
            local: Histogram::new_from(&self.local),
            shared: self.shared.clone(),
            last_phase: self.last_phase,
        }
    }
}

impl<C: Counter> Drop for Recorder<C> {
    fn drop(&mut self) {
        // we'll need to decrement the # of recorders
        let mut truth = self.shared.truth.lock().unwrap();
        truth.recorders -= 1;

        // we also want to communicate the remainder of our samples to the reader
        // we do this under the lock so that if the reader reads .recorders after we update it
        // above, it is guaranteed to see the samples from this recorder.
        // we also _have_ to do it at some point during drop as the reader _may_ have read
        // .recorders _before_ we decremented it above, in which case it's blocking on us!
        // note that we cannot call self.update() here as it would drop the mutex guard
        let h = Histogram::new_from(&self.local);
        let h = std::mem::replace(&mut self.local, h);
        let _ = self.shared.sender.send(h).is_ok(); // if this is err, the reader went away

        // explicitly drop guard to ensure we don't accidentally drop it above
        drop(truth);
    }
}

#[derive(Debug)]
struct Critical {
    recorders: usize,
}

#[derive(Debug)]
struct Shared<C: Counter> {
    truth: Mutex<Critical>,
    sender: crossbeam_channel::Sender<Histogram<C>>,
    phase: atomic::AtomicUsize,
}

/// See [`IdleRecorder`]. This guard borrows the idle [`Recorder`].
pub type IdleRecorderGuard<'a, C> = IdleRecorder<&'a mut Recorder<C>, C>;

/// This guard denotes that a [`Recorder`] is currently idle, and should not be waited on by a
/// [`SyncHistogram`] phase-shift.
#[derive(Debug)]
pub struct IdleRecorder<T, C: Counter>
where
    T: BorrowMut<Recorder<C>>,
{
    recorder: Option<T>,
    c: PhantomData<C>,
}

impl<T, C: Counter> IdleRecorder<T, C>
where
    T: BorrowMut<Recorder<C>>,
{
    fn reactivate(&mut self) {
        let recorder = if let Some(ref mut r) = self.recorder {
            r
        } else {
            // already reactivated
            return;
        };

        let recorder = recorder.borrow_mut();

        // the Recorder is no longer idle, so the reader has to wait for us again
        // this basically means re-incrementing .recorders
        let mut crit = recorder.shared.truth.lock().unwrap();
        crit.recorders += 1;

        // we need to figure out what phase we're joining
        // the easiest way to do that is to adopt the current phase
        //
        // note that we have to load the phase while holding the lock.
        // if we did not, the reader could come along, read our ++'d .recorders (and so wait for us
        // to send), and bump the phase, all before we read it, which would lead us to believe that
        // we were already synchronized when in reality we were not, which would stall the reader
        // even if we issued more writes.
        recorder.last_phase = recorder.shared.phase.load(atomic::Ordering::Acquire);

        // explicitly drop guard to ensure we don't accidentally drop it above
        drop(crit);
    }
}

impl<C: Counter> IdleRecorder<Recorder<C>, C> {
    /// Mark the wrapped [`Recorder`] as active again and return it.
    pub fn activate(mut self) -> Recorder<C> {
        self.reactivate();
        self.recorder.take().unwrap()
    }

    /// Clone the wrapped [`Recorder`].
    pub fn recorder(&self) -> Recorder<C> {
        self.recorder.as_ref().unwrap().clone()
    }
}

impl<T, C: Counter> Drop for IdleRecorder<T, C>
where
    T: BorrowMut<Recorder<C>>,
{
    fn drop(&mut self) {
        self.reactivate()
    }
}

impl<C: Counter> Recorder<C> {
    fn with_hist<F, R>(&mut self, f: F) -> R
    where
        F: FnOnce(&mut Histogram<C>) -> R,
    {
        let r = f(&mut self.local);
        let phase = self.shared.phase.load(atomic::Ordering::Acquire);
        if phase != self.last_phase {
            self.update();
            self.last_phase = phase;
        }
        r
    }

    // return our current histogram and leave a cleared one in its place
    fn shed(&mut self) -> Histogram<C> {
        let h = Histogram::new_from(&self.local);
        std::mem::replace(&mut self.local, h)
    }

    fn update(&mut self) {
        let h = self.shed();
        let _ = self.shared.sender.send(h).is_ok(); // if this is err, the reader went away
    }

    fn deactivate(&mut self) {
        let phase;
        {
            // we're leaving rotation, so we need to decrement .recorders
            let mut crit = self.shared.truth.lock().unwrap();
            crit.recorders -= 1;

            // make sure we don't hold up the current phase shift (if any)
            phase = self.shared.phase.load(atomic::Ordering::Acquire);
            if phase != self.last_phase {
                // can't call self.update() due to borrow of self.shared above
                let h = Histogram::new_from(&self.local);
                let h = std::mem::replace(&mut self.local, h);
                let _ = self.shared.sender.send(h).is_ok(); // if this is err, the reader went away
            }
        }
        self.last_phase = phase;
    }

    /// Call this method if the Recorder will be idle for a while.
    ///
    /// Until the returned guard is dropped, the associated [`SyncHistogram`] will not wait for
    /// this recorder on a phase shift.
    pub fn idle(&mut self) -> IdleRecorderGuard<C> {
        self.deactivate();
        IdleRecorder {
            recorder: Some(self),
            c: PhantomData,
        }
    }

    /// Mark this `Recorder` as inactive.
    ///
    /// Until the returned guard is consumed, either by calling [`IdleRecorder::activate`] or by
    /// dropping it, the associated [`SyncHistogram`] will not wait for this recorder on a phase
    /// shift.
    pub fn into_idle(mut self) -> IdleRecorder<Self, C> {
        self.deactivate();
        IdleRecorder {
            recorder: Some(self),
            c: PhantomData,
        }
    }

    /// See [`Histogram::add`].
    pub fn add<B: Borrow<Histogram<C>>>(&mut self, source: B) -> Result<(), AdditionError> {
        self.with_hist(move |h| h.add(source))
    }

    /// See [`Histogram::add_correct`].
    pub fn add_correct<B: Borrow<Histogram<C>>>(
        &mut self,
        source: B,
        interval: u64,
    ) -> Result<(), RecordError> {
        self.with_hist(move |h| h.add_correct(source, interval))
    }

    /// See [`Histogram::subtract`].
    pub fn subtract<B: Borrow<Histogram<C>>>(
        &mut self,
        subtrahend: B,
    ) -> Result<(), SubtractionError> {
        self.with_hist(move |h| h.subtract(subtrahend))
    }

    /// See [`Histogram::record`].
    pub fn record(&mut self, value: u64) -> Result<(), RecordError> {
        self.with_hist(move |h| h.record(value))
    }

    /// See [`Histogram::saturating_record`].
    pub fn saturating_record(&mut self, value: u64) {
        self.with_hist(move |h| h.saturating_record(value))
    }

    /// See [`Histogram::record_n`].
    pub fn record_n(&mut self, value: u64, count: C) -> Result<(), RecordError> {
        self.with_hist(move |h| h.record_n(value, count))
    }

    /// See [`Histogram::saturating_record_n`].
    pub fn saturating_record_n(&mut self, value: u64, count: C) {
        self.with_hist(move |h| h.saturating_record_n(value, count))
    }

    /// See [`Histogram::record_correct`].
    pub fn record_correct(&mut self, value: u64, interval: u64) -> Result<(), RecordError> {
        self.with_hist(move |h| h.record_correct(value, interval))
    }

    /// See [`Histogram::record_n_correct`].
    pub fn record_n_correct(
        &mut self,
        value: u64,
        count: C,
        interval: u64,
    ) -> Result<(), RecordError> {
        self.with_hist(move |h| h.record_n_correct(value, count, interval))
    }
}

/// A `Histogram` that can be written to by multiple threads concurrently.
///
/// Each writer thread should have a [`Recorder`], which allows it to record new samples without
/// synchronization. New recorded samples are made available through this histogram by calling
/// [`SyncHistogram::refresh`], which blocks until it has synchronized with every recorder.
#[derive(Debug)]
pub struct SyncHistogram<C: Counter> {
    merged: Histogram<C>,
    shared: Arc<Shared<C>>,
    receiver: crossbeam_channel::Receiver<Histogram<C>>,
}

impl<C: Counter> SyncHistogram<C> {
    fn refresh_inner(&mut self, timeout: Option<time::Duration>) {
        let end = timeout.map(|dur| time::Instant::now() + dur);

        // time to start a phase change
        // we first want to drain any histograms left over by dropped recorders
        // note that we do this _before_ incrementing the phase, so we know they're "old"
        while let Ok(h) = self.receiver.try_recv() {
            self.merged
                .add(&h)
                .expect("TODO: failed to merge histogram");
        }

        // make sure no recorders can join or leave in the middle of this
        let recorders = self.shared.truth.lock().unwrap().recorders;

        // then, we tell writers to phase
        let _ = self.shared.phase.fetch_add(1, atomic::Ordering::AcqRel);

        // we want to wait for writers to all have phased
        let mut phased = 0;

        // at this point, we expect to get at least truth.recorders histograms
        while phased < recorders {
            let h = if let Some(end) = end {
                let now = time::Instant::now();
                if now > end {
                    break;
                }

                match self.receiver.recv_timeout(end - now) {
                    Ok(h) => h,
                    Err(crossbeam_channel::RecvTimeoutError::Timeout) => break,
                    Err(crossbeam_channel::RecvTimeoutError::Disconnected) => unreachable!(),
                }
            } else {
                self.receiver
                    .recv()
                    .expect("SyncHistogram has an Arc<Shared> with a Receiver")
            };

            self.merged
                .add(&h)
                .expect("TODO: failed to merge histogram");
            phased += 1;
        }

        // we also gobble up extra histograms we may have been sent from more dropped writers
        while let Ok(h) = self.receiver.try_recv() {
            self.merged
                .add(&h)
                .expect("TODO: failed to merge histogram");
        }
    }

    /// Block until writes from all [`Recorder`] instances for this histogram have been
    /// incorporated.
    pub fn refresh(&mut self) {
        self.refresh_inner(None)
    }

    /// Block until writes from all [`Recorder`] instances for this histogram have been
    /// incorporated, or until the given amount of time has passed.
    pub fn refresh_timeout(&mut self, timeout: time::Duration) {
        self.refresh_inner(Some(timeout))
    }

    /// Obtain another multi-threaded writer for this histogram.
    ///
    /// Note that writes made to the `Recorder` will not be visible until the next call to
    /// [`SyncHistogram::refresh`].
    pub fn recorder(&self) -> Recorder<C> {
        // we will have to wait for one more recorder
        {
            let mut truth = self.shared.truth.lock().unwrap();
            truth.recorders += 1;
        }

        // new recorder starts at the current phase with an empty histogram
        Recorder {
            local: Histogram::new_from(&self.merged),
            shared: self.shared.clone(),
            last_phase: self.shared.phase.load(atomic::Ordering::Acquire),
        }
    }
}

impl<C: Counter> From<Histogram<C>> for SyncHistogram<C> {
    fn from(h: Histogram<C>) -> Self {
        let (tx, rx) = crossbeam_channel::unbounded();
        SyncHistogram {
            merged: h,
            receiver: rx,
            shared: Arc::new(Shared {
                truth: Mutex::new(Critical { recorders: 0 }),
                sender: tx,
                phase: atomic::AtomicUsize::new(0),
            }),
        }
    }
}

impl<C: Counter> Deref for SyncHistogram<C> {
    type Target = Histogram<C>;
    fn deref(&self) -> &Self::Target {
        &self.merged
    }
}

impl<C: Counter> DerefMut for SyncHistogram<C> {
    fn deref_mut(&mut self) -> &mut Self::Target {
        &mut self.merged
    }
}