hotpath 0.17.0

One profiler for CPU, time, memory, and async code - quickly find and debug performance bottlenecks.
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
//! RwLock instrumentation module - tracks read/write lock acquisitions and hold durations.

use crossbeam_channel::{bounded, select, unbounded, Receiver as CbReceiver, Sender as CbSender};
use hdrhistogram::Histogram;
use std::collections::HashMap;
use std::sync::atomic::{AtomicU32, Ordering};
use std::sync::{Arc, Mutex, OnceLock, RwLock as StdRwLock};

use crate::batch::{register_thread_batch, BatchRegistry, BatchedMeasurement, MeasurementBatch};
use crate::instant::Instant;
use crate::lib_on::hotpath_guard::{
    WORKER_BATCH_SIZE, WORKER_FLUSH_INTERVAL_MS, WORKER_SHUTDOWN_DRAIN_LIMIT,
};
use crate::lib_on::START_TIME;
use crate::metrics_server::METRICS_SERVER_PORT;

pub(crate) mod wrapper;

// Re-exported to keep the std wrapper reachable at `hotpath::rw_locks::*` for downstream code.
pub use wrapper::std::{RwLock, RwLockReadGuard, RwLockWriteGuard};

static RW_LOCK_ID_COUNTER: AtomicU32 = AtomicU32::new(1);

fn next_rw_lock_id() -> u32 {
    RW_LOCK_ID_COUNTER.fetch_add(1, Ordering::Relaxed)
}

/// Whether an acquisition was a shared (read) or exclusive (write) lock.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub(crate) enum RwLockKind {
    Read,
    Write,
}

/// Events sent to the background lock statistics collection thread.
#[derive(Debug)]
pub(crate) enum RwLockEvent {
    Created {
        id: u32,
        source: &'static str,
        label: Option<String>,
        type_name: &'static str,
    },
    /// Emitted when a guard is dropped. `wait_nanos` is the time blocked
    /// before the lock was granted; `acquire_nanos` is the held duration
    /// (granted -> released).
    Released {
        id: u32,
        kind: RwLockKind,
        wait_nanos: u64,
        acquire_nanos: u64,
        elapsed_ns: u64,
    },
}

/// Statistics for a single instrumented RwLock.
#[derive(Debug, Clone)]
pub(crate) struct RwLockEntry {
    pub(crate) id: u32,
    pub(crate) source: &'static str,
    pub(crate) label: Option<String>,
    pub(crate) type_name: &'static str,
    pub(crate) read_count: u64,
    pub(crate) write_count: u64,
    pub(crate) read_wait_total_nanos: u64,
    pub(crate) write_wait_total_nanos: u64,
    pub(crate) read_acquire_total_nanos: u64,
    pub(crate) write_acquire_total_nanos: u64,
    read_wait_hist: Option<Histogram<u64>>,
    write_wait_hist: Option<Histogram<u64>>,
    read_acquire_hist: Option<Histogram<u64>>,
    write_acquire_hist: Option<Histogram<u64>>,
    pub(crate) iter: u32,
}

impl RwLockEntry {
    const LOW_NS: u64 = 1;
    const HIGH_NS: u64 = 1_000_000_000_000; // 1000s
    const SIGFIGS: u8 = 3;

    fn new_histogram() -> Histogram<u64> {
        Histogram::<u64>::new_with_bounds(Self::LOW_NS, Self::HIGH_NS, Self::SIGFIGS)
            .expect("hdrhistogram init")
    }

    #[inline]
    fn record(hist: &mut Option<Histogram<u64>>, nanos: u64) {
        if let Some(ref mut hist) = hist {
            hist.record(nanos.clamp(Self::LOW_NS, Self::HIGH_NS))
                .unwrap();
        }
    }

    pub(crate) fn count(&self, kind: RwLockKind) -> u64 {
        match kind {
            RwLockKind::Read => self.read_count,
            RwLockKind::Write => self.write_count,
        }
    }

    pub(crate) fn wait_avg_nanos(&self, kind: RwLockKind) -> u64 {
        let (total, count) = match kind {
            RwLockKind::Read => (self.read_wait_total_nanos, self.read_count),
            RwLockKind::Write => (self.write_wait_total_nanos, self.write_count),
        };
        total.checked_div(count).unwrap_or(0)
    }

    pub(crate) fn acquire_avg_nanos(&self, kind: RwLockKind) -> u64 {
        let (total, count) = match kind {
            RwLockKind::Read => (self.read_acquire_total_nanos, self.read_count),
            RwLockKind::Write => (self.write_acquire_total_nanos, self.write_count),
        };
        total.checked_div(count).unwrap_or(0)
    }

    fn percentile(hist: &Option<Histogram<u64>>, count: u64, p: f64) -> u64 {
        match hist {
            Some(hist) if count > 0 => hist.value_at_percentile(p.clamp(0.0, 100.0)),
            _ => 0,
        }
    }

    pub(crate) fn wait_percentile_nanos(&self, kind: RwLockKind, p: f64) -> u64 {
        let (hist, count) = match kind {
            RwLockKind::Read => (&self.read_wait_hist, self.read_count),
            RwLockKind::Write => (&self.write_wait_hist, self.write_count),
        };
        Self::percentile(hist, count, p)
    }

    pub(crate) fn acquire_percentile_nanos(&self, kind: RwLockKind, p: f64) -> u64 {
        let (hist, count) = match kind {
            RwLockKind::Read => (&self.read_acquire_hist, self.read_count),
            RwLockKind::Write => (&self.write_acquire_hist, self.write_count),
        };
        Self::percentile(hist, count, p)
    }
}

pub(crate) struct RwLocksInternalState {
    pub(crate) stats: HashMap<u32, RwLockEntry>,
}

pub(crate) struct RwLocksState {
    pub(crate) event_tx: CbSender<Vec<RwLockEvent>>,
    pub(crate) inner: Arc<StdRwLock<RwLocksInternalState>>,
    pub(crate) shutdown_tx: Mutex<Option<CbSender<()>>>,
    pub(crate) completion_rx: Mutex<Option<CbReceiver<()>>>,
}

pub(crate) static RW_LOCKS_STATE: OnceLock<RwLocksState> = OnceLock::new();

pub(crate) fn get_sorted_rw_lock_entries() -> Vec<RwLockEntry> {
    let Some(state) = RW_LOCKS_STATE.get() else {
        return Vec::new();
    };
    let guard = state.inner.read().unwrap();
    let mut stats: Vec<RwLockEntry> = guard.stats.values().cloned().collect();
    stats.sort_by(compare_rw_lock_entries);
    stats
}

pub(crate) fn get_rw_locks_json() -> crate::json::JsonRwLocksList {
    let entries = get_sorted_rw_lock_entries();
    let elapsed = std::time::Duration::from_nanos(crate::lib_on::current_elapsed_ns());
    crate::lib_on::report::collect_rw_locks_json(
        &entries,
        elapsed,
        &crate::lib_on::hotpath_guard::configured_percentiles(),
    )
}

#[inline]
pub(crate) fn elapsed_nanos(start: Instant) -> u64 {
    start.elapsed().as_nanos() as u64
}

static EVENT_REGISTRY: BatchRegistry<RwLockEvent> = BatchRegistry::new();

thread_local! {
    static EVENT_BATCH: std::sync::Arc<std::sync::Mutex<MeasurementBatch<RwLockEvent>>> =
        register_thread_batch(&EVENT_REGISTRY);
}

#[inline]
pub(crate) fn send_rw_lock_event(event: RwLockEvent) {
    let _suspend = crate::lib_on::SuspendAllocTracking::new();
    EVENT_BATCH.with(|b| {
        if let Ok(mut b) = b.lock() {
            b.add(event);
        }
    });
}

/// Flushes every thread's buffered rw_lock events into the worker channel.
/// Called at shutdown before the worker is signalled to stop.
pub(crate) fn flush_rw_lock_batch() {
    EVENT_REGISTRY.flush_all();
}

impl BatchedMeasurement for RwLockEvent {
    fn elapsed_since_start_ns(&self) -> u64 {
        match self {
            RwLockEvent::Released { elapsed_ns, .. } => *elapsed_ns,
            _ => 0,
        }
    }

    fn fetch_sender() -> Option<CbSender<Vec<Self>>> {
        Some(RW_LOCKS_STATE.get()?.event_tx.clone())
    }

    fn is_flush_boundary(&self) -> bool {
        matches!(self, RwLockEvent::Created { .. })
    }
}

fn process_rw_lock_event(state: &mut RwLocksInternalState, event: RwLockEvent) {
    match event {
        RwLockEvent::Created {
            id,
            source,
            label,
            type_name,
        } => {
            let iter = state.stats.values().filter(|s| s.source == source).count() as u32;
            state.stats.insert(
                id,
                RwLockEntry {
                    id,
                    source,
                    label,
                    type_name,
                    read_count: 0,
                    write_count: 0,
                    read_wait_total_nanos: 0,
                    write_wait_total_nanos: 0,
                    read_acquire_total_nanos: 0,
                    write_acquire_total_nanos: 0,
                    read_wait_hist: Some(RwLockEntry::new_histogram()),
                    write_wait_hist: Some(RwLockEntry::new_histogram()),
                    read_acquire_hist: Some(RwLockEntry::new_histogram()),
                    write_acquire_hist: Some(RwLockEntry::new_histogram()),
                    iter,
                },
            );
        }
        RwLockEvent::Released {
            id,
            kind,
            wait_nanos,
            acquire_nanos,
            elapsed_ns: _,
        } => {
            if let Some(entry) = state.stats.get_mut(&id) {
                match kind {
                    RwLockKind::Read => {
                        entry.read_count += 1;
                        entry.read_wait_total_nanos += wait_nanos;
                        entry.read_acquire_total_nanos += acquire_nanos;
                        RwLockEntry::record(&mut entry.read_wait_hist, wait_nanos);
                        RwLockEntry::record(&mut entry.read_acquire_hist, acquire_nanos);
                    }
                    RwLockKind::Write => {
                        entry.write_count += 1;
                        entry.write_wait_total_nanos += wait_nanos;
                        entry.write_acquire_total_nanos += acquire_nanos;
                        RwLockEntry::record(&mut entry.write_wait_hist, wait_nanos);
                        RwLockEntry::record(&mut entry.write_acquire_hist, acquire_nanos);
                    }
                }
            }
        }
    }
}

/// Registers a new RwLock with the profiling subsystem.
pub(crate) fn register_rw_lock<T>(source: &'static str, label: Option<String>) -> u32 {
    let type_name = std::any::type_name::<T>();
    init_rw_locks_state();
    let id = next_rw_lock_id();

    send_rw_lock_event(RwLockEvent::Created {
        id,
        source,
        label,
        type_name,
    });

    id
}

/// Initialize the lock statistics collection system (called on first instrumented lock).
pub(crate) fn init_rw_locks_state() -> &'static RwLocksState {
    RW_LOCKS_STATE.get_or_init(|| {
        START_TIME.get_or_init(Instant::now);

        let (event_tx, event_rx) = unbounded::<Vec<RwLockEvent>>();
        let (shutdown_tx, shutdown_rx) = bounded::<()>(1);
        let (completion_tx, completion_rx) = bounded::<()>(1);

        let inner = Arc::new(StdRwLock::new(RwLocksInternalState {
            stats: HashMap::new(),
        }));
        let inner_clone = Arc::clone(&inner);

        std::thread::Builder::new()
            .name("hp-rw-locks".into())
            .spawn(move || {
                let mut local_buffer: Vec<RwLockEvent> = Vec::with_capacity(WORKER_BATCH_SIZE);
                let flush_interval = std::time::Duration::from_millis(WORKER_FLUSH_INTERVAL_MS);

                loop {
                    select! {
                        recv(event_rx) -> result => {
                            match result {
                                Ok(events) => {
                                    local_buffer.extend(events);
                                    if local_buffer.len() >= WORKER_BATCH_SIZE {
                                        if let Ok(mut shared) = inner_clone.write() {
                                            for e in local_buffer.drain(..) {
                                                process_rw_lock_event(&mut shared, e);
                                            }
                                        }
                                    }
                                }
                                Err(_) => {
                                    if !local_buffer.is_empty() {
                                        if let Ok(mut shared) = inner_clone.write() {
                                            for e in local_buffer.drain(..) {
                                                process_rw_lock_event(&mut shared, e);
                                            }
                                        }
                                    }
                                    break;
                                }
                            }
                        }
                        recv(shutdown_rx) -> _ => {
                            let mut drained_events = Vec::with_capacity(WORKER_BATCH_SIZE);
                            for _ in 0..WORKER_SHUTDOWN_DRAIN_LIMIT {
                                match event_rx.try_recv() {
                                    Ok(events) => drained_events.extend(events),
                                    Err(_) => break,
                                }
                            }

                            if let Ok(mut shared) = inner_clone.write() {
                                for e in local_buffer.drain(..) {
                                    process_rw_lock_event(&mut shared, e);
                                }
                                for event in drained_events {
                                    process_rw_lock_event(&mut shared, event);
                                }
                            }
                            break;
                        }
                        default(flush_interval) => {
                            if !local_buffer.is_empty() {
                                if let Ok(mut shared) = inner_clone.write() {
                                    for e in local_buffer.drain(..) {
                                        process_rw_lock_event(&mut shared, e);
                                    }
                                }
                            }
                        }
                    }
                }

                let _ = completion_tx.send(());
            })
            .expect("Failed to spawn rw_lock-stats-collector thread");

        crate::metrics_server::start_metrics_server_once(*METRICS_SERVER_PORT);

        RwLocksState {
            event_tx,
            inner,
            shutdown_tx: Mutex::new(Some(shutdown_tx)),
            completion_rx: Mutex::new(Some(completion_rx)),
        }
    })
}

/// Compare two lock stats for sorting. Custom labels first, then by source and iter.
pub(crate) fn compare_rw_lock_entries(a: &RwLockEntry, b: &RwLockEntry) -> std::cmp::Ordering {
    match (a.label.is_some(), b.label.is_some()) {
        (true, false) => std::cmp::Ordering::Less,
        (false, true) => std::cmp::Ordering::Greater,
        (true, true) => a
            .label
            .as_ref()
            .unwrap()
            .cmp(b.label.as_ref().unwrap())
            .then_with(|| a.iter.cmp(&b.iter)),
        (false, false) => a.source.cmp(b.source).then_with(|| a.iter.cmp(&b.iter)),
    }
}

/// Trait for instrumenting RwLocks. Dispatches on the type of the wrapped lock
/// (e.g. [`std::sync::RwLock`] or [`parking_lot::RwLock`]).
///
/// This trait is not intended for direct use. Use the `rw_lock!` macro instead.
#[doc(hidden)]
pub trait InstrumentRwLock {
    type Output;
    fn instrument(self, source: &'static str, label: Option<String>) -> Self::Output;
}

/// Instrument an [`std::sync::RwLock`], [`parking_lot::RwLock`], `async_lock::RwLock`, or
/// `tokio::sync::RwLock` for read/write profiling.
///
/// Returns an instrumented drop-in replacement that proxies to the wrapped lock and records
/// how long read and write locks are held. The wrapper type matches the API of the underlying
/// lock (`std::sync::RwLock` returns `LockResult`s; `parking_lot::RwLock` returns guards directly;
/// `async_lock::RwLock` and `tokio::sync::RwLock` expose async `read`/`write` returning guards).
///
/// `parking_lot::RwLock` support requires the `parking_lot` feature; `async_lock::RwLock`
/// support requires the `async-lock` feature; `tokio::sync::RwLock` support requires the
/// `tokio` feature.
///
/// # Examples
///
/// ```rust,no_run
/// let lock = hotpath::rw_lock!(std::sync::RwLock::new(0u32));
/// *lock.write().unwrap() += 1;
/// let _ = *lock.read().unwrap();
/// ```
#[macro_export]
macro_rules! rw_lock {
    ($expr:expr) => {{
        const RW_LOCK_ID: &'static str = concat!(file!(), ":", line!());
        $crate::InstrumentRwLock::instrument($expr, RW_LOCK_ID, None)
    }};

    ($expr:expr, label = $label:expr) => {{
        const RW_LOCK_ID: &'static str = concat!(file!(), ":", line!());
        $crate::InstrumentRwLock::instrument($expr, RW_LOCK_ID, Some($label.to_string()))
    }};
}