perf_hooks: sample delay per event loop iteration

Add a samplePerIteration option to monitorEventLoopDelay that records
event loop delay from libuv event loop iterations instead of the timer
interval sampler. The default remains interval-based; existing uses of
monitorEventLoopDelay() keep behaving the same unless the
samplePerIteration option is passed through.

Signed-off-by: Pablo Erhard <pablo.erhardhernandez@datadoghq.com>
PR-URL: #62935
Reviewed-By: Bryan English <bryan@bryanenglish.com>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Reviewed-By: James M Snell <jasnell@gmail.com>

Author: pabloerhard

doc/api/perf_hooks.md

@@ -1701,23 +1701,32 @@ are not guaranteed to reflect any correct state of the event loop.
 
 <!-- YAML
 added: v11.10.0
+changes:
+  - version: REPLACEME
+    pr-url: https://github.com/nodejs/node/pull/62935
+    description: Added the `samplePerIteration` option.
 -->
 
 * `options` {Object}
-  * `resolution` {number} The sampling rate in milliseconds. Must be greater
-    than zero. **Default:** `10`.
-* Returns: {IntervalHistogram}
+  * `samplePerIteration` {boolean} When `true`, samples are taken once per
+    event loop iteration. **Default:** `false`.
+  * `resolution` {number} The sampling rate in milliseconds for interval-based
+    sampling. Must be greater than zero. This option is ignored when
+    `samplePerIteration` is `true`. **Default:** `10`.
+* Returns: {ELDHistogram}
 
 _This property is an extension by Node.js. It is not available in Web browsers._
 
-Creates an `IntervalHistogram` object that samples and reports the event loop
-delay over time. The delays will be reported in nanoseconds.
+Creates a histogram object that samples and reports the event loop delay over
+time. The delays will be reported in nanoseconds.
 
-Using a timer to detect approximate event loop delay works because the
-execution of timers is tied specifically to the lifecycle of the libuv
-event loop. That is, a delay in the loop will cause a delay in the execution
-of the timer, and those delays are specifically what this API is intended to
-detect.
+By default, the histogram is updated by a timer using the configured
+`resolution`. When `samplePerIteration` is `true`, samples are taken once per
+event loop iteration using `uv_prepare_t` and `uv_check_t` hooks. In that mode,
+the histogram does not keep the loop alive or force additional iterations when
+the application is idle.
+The two sampling modes produce significantly different results and should not
+be compared directly.
 
 ```mjs
 import { monitorEventLoopDelay } from 'node:perf_hooks';
@@ -1992,9 +2001,10 @@ added: v11.10.0
 
 The standard deviation of the recorded event loop delays.
 
-## Class: `IntervalHistogram extends Histogram`
+## Class: `ELDHistogram extends Histogram`
 
-A `Histogram` that is periodically updated on a given interval.
+A `Histogram` that records event loop delay, returned by
+[`perf_hooks.monitorEventLoopDelay()`][].
 
 ### `histogram.disable()`
 
@@ -2004,7 +2014,7 @@ added: v11.10.0
 
 * Returns: {boolean}
 
-Disables the update interval timer. Returns `true` if the timer was
+Disables event loop delay sampling. Returns `true` if sampling was
 stopped, `false` if it was already stopped.
 
 ### `histogram.enable()`
@@ -2015,7 +2025,7 @@ added: v11.10.0
 
 * Returns: {boolean}
 
-Enables the update interval timer. Returns `true` if the timer was
+Enables event loop delay sampling. Returns `true` if sampling was
 started, `false` if it was already started.
 
 ### `histogram[Symbol.dispose]()`
@@ -2024,7 +2034,7 @@ started, `false` if it was already started.
 added: v24.2.0
 -->
 
-Disables the update interval timer when the histogram is disposed.
+Disables event loop delay sampling when the histogram is disposed.
 
 ```js
 const { monitorEventLoopDelay } = require('node:perf_hooks');
@@ -2035,11 +2045,11 @@ const { monitorEventLoopDelay } = require('node:perf_hooks');
 }
 ```
 
-### Cloning an `IntervalHistogram`
+### Cloning an `ELDHistogram`
 
-{IntervalHistogram} instances can be cloned via {MessagePort}. On the receiving
-end, the histogram is cloned as a plain {Histogram} object that does not
-implement the `enable()` and `disable()` methods.
+{ELDHistogram} instances can be cloned via {MessagePort}. On the receiving end,
+the histogram is cloned as a plain {Histogram} object that does not implement
+the `enable()` and `disable()` methods.
 
 ## Class: `RecordableHistogram extends Histogram`
 
@@ -2352,6 +2362,7 @@ dns.promises.resolve('localhost');
 [`'exit'`]: process.md#event-exit
 [`child_process.spawnSync()`]: child_process.md#child_processspawnsynccommand-args-options
 [`perf_hooks.eventLoopUtilization()`]: #perf_hookseventlooputilizationutilization1-utilization2
+[`perf_hooks.monitorEventLoopDelay()`]: #perf_hooksmonitoreventloopdelayoptions
 [`perf_hooks.timerify()`]: #perf_hookstimerifyfn-options
 [`process.hrtime()`]: process.md#processhrtimetime
 [`timeOrigin`]: https://w3c.github.io/hr-time/#dom-performance-timeorigin

lib/internal/perf/event_loop_delay.js

@@ -18,6 +18,7 @@ const {
 } = internalBinding('performance');
 
 const {
+  validateBoolean,
   validateInteger,
   validateObject,
 } = require('internal/validators');
@@ -74,21 +75,23 @@ class ELDHistogram extends Histogram {
 
 /**
  * @param {{
+ *   samplePerIteration : boolean,
  *   resolution : number
  * }} [options]
  * @returns {ELDHistogram}
  */
 function monitorEventLoopDelay(options = kEmptyObject) {
   validateObject(options, 'options');
 
-  const { resolution = 10 } = options;
+  const { samplePerIteration = false, resolution = 10 } = options;
+  validateBoolean(samplePerIteration, 'options.samplePerIteration');
   validateInteger(resolution, 'options.resolution', 1);
 
   return ReflectConstruct(
     function() {
       markTransferMode(this, true, false);
       this[kEnabled] = false;
-      this[kHandle] = createELDHistogram(resolution);
+      this[kHandle] = createELDHistogram(resolution, samplePerIteration);
       this[kMap] = new SafeMap();
     }, [], ELDHistogram);
 }

src/env_properties.h

@@ -418,6 +418,7 @@
   V(http2ping_constructor_template, v8::ObjectTemplate)                        \
   V(i18n_converter_template, v8::ObjectTemplate)                               \
   V(intervalhistogram_constructor_template, v8::FunctionTemplate)              \
+  V(iterationhistogram_constructor_template, v8::FunctionTemplate)             \
   V(iter_template, v8::DictionaryTemplate)                                     \
   V(js_transferable_constructor_template, v8::FunctionTemplate)                \
   V(libuv_stream_wrap_ctor_template, v8::FunctionTemplate)                     \

src/histogram.cc

@@ -25,6 +25,20 @@ using v8::String;
 using v8::Uint32;
 using v8::Value;
 
+template <typename T>
+void StartHandleHistogram(Local<Value> receiver, bool reset) {
+  T* histogram;
+  ASSIGN_OR_RETURN_UNWRAP(&histogram, receiver);
+  histogram->OnStart(reset ? T::StartFlags::RESET : T::StartFlags::NONE);
+}
+
+template <typename T>
+void StopHandleHistogram(Local<Value> receiver) {
+  T* histogram;
+  ASSIGN_OR_RETURN_UNWRAP(&histogram, receiver);
+  histogram->OnStop();
+}
+
 Histogram::Histogram(const Options& options) {
   hdr_histogram* histogram;
   CHECK_EQ(0, hdr_init(options.lowest,
@@ -68,6 +82,10 @@ CFunction IntervalHistogram::fast_start_(
     CFunction::Make(&IntervalHistogram::FastStart));
 CFunction IntervalHistogram::fast_stop_(
     CFunction::Make(&IntervalHistogram::FastStop));
+CFunction IterationHistogram::fast_start_(
+    CFunction::Make(&IterationHistogram::FastStart));
+CFunction IterationHistogram::fast_stop_(
+    CFunction::Make(&IterationHistogram::FastStop));
 
 void HistogramImpl::AddMethods(Isolate* isolate, Local<FunctionTemplate> tmpl) {
   // TODO(@jasnell): The bigint API variations do not yet support fast
@@ -416,29 +434,157 @@ void IntervalHistogram::OnStop() {
 }
 
 void IntervalHistogram::Start(const FunctionCallbackInfo<Value>& args) {
-  IntervalHistogram* histogram;
-  ASSIGN_OR_RETURN_UNWRAP(&histogram, args.This());
-  histogram->OnStart(args[0]->IsTrue() ? StartFlags::RESET : StartFlags::NONE);
+  StartHandleHistogram<IntervalHistogram>(args.This(), args[0]->IsTrue());
 }
 
 void IntervalHistogram::FastStart(Local<Value> receiver, bool reset) {
   TRACK_V8_FAST_API_CALL("histogram.start");
-  IntervalHistogram* histogram;
-  ASSIGN_OR_RETURN_UNWRAP(&histogram, receiver);
-  histogram->OnStart(reset ? StartFlags::RESET : StartFlags::NONE);
+  StartHandleHistogram<IntervalHistogram>(receiver, reset);
 }
 
 void IntervalHistogram::Stop(const FunctionCallbackInfo<Value>& args) {
-  IntervalHistogram* histogram;
-  ASSIGN_OR_RETURN_UNWRAP(&histogram, args.This());
-  histogram->OnStop();
+  StopHandleHistogram<IntervalHistogram>(args.This());
 }
 
 void IntervalHistogram::FastStop(Local<Value> receiver) {
   TRACK_V8_FAST_API_CALL("histogram.stop");
-  IntervalHistogram* histogram;
-  ASSIGN_OR_RETURN_UNWRAP(&histogram, receiver);
-  histogram->OnStop();
+  StopHandleHistogram<IntervalHistogram>(receiver);
+}
+
+Local<FunctionTemplate> IterationHistogram::GetConstructorTemplate(
+    Environment* env) {
+  Local<FunctionTemplate> tmpl = env->iterationhistogram_constructor_template();
+  if (tmpl.IsEmpty()) {
+    Isolate* isolate = env->isolate();
+    tmpl = NewFunctionTemplate(isolate, nullptr);
+    tmpl->Inherit(HandleWrap::GetConstructorTemplate(env));
+    tmpl->SetClassName(FIXED_ONE_BYTE_STRING(isolate, "Histogram"));
+    auto instance = tmpl->InstanceTemplate();
+    instance->SetInternalFieldCount(IterationHistogram::kInternalFieldCount);
+    HistogramImpl::AddMethods(isolate, tmpl);
+    SetFastMethod(isolate, instance, "start", Start, &fast_start_);
+    SetFastMethod(isolate, instance, "stop", Stop, &fast_stop_);
+    env->set_iterationhistogram_constructor_template(tmpl);
+  }
+  return tmpl;
+}
+
+void IterationHistogram::RegisterExternalReferences(
+    ExternalReferenceRegistry* registry) {
+  registry->Register(Start);
+  registry->Register(Stop);
+  registry->Register(fast_start_);
+  registry->Register(fast_stop_);
+  HistogramImpl::RegisterExternalReferences(registry);
+}
+
+IterationHistogram::IterationHistogram(Environment* env,
+                                       Local<Object> wrap,
+                                       AsyncWrap::ProviderType type,
+                                       const Histogram::Options& options)
+    : HandleWrap(
+          env, wrap, reinterpret_cast<uv_handle_t*>(&check_handle_), type),
+      HistogramImpl(options) {
+  MakeWeak();
+  wrap->SetAlignedPointerInInternalField(
+      HistogramImpl::InternalFields::kImplField,
+      static_cast<HistogramImpl*>(this),
+      EmbedderDataTag::kDefault);
+  uv_check_init(env->event_loop(), &check_handle_);
+  uv_prepare_init(env->event_loop(), &prepare_handle_);
+  uv_unref(reinterpret_cast<uv_handle_t*>(&check_handle_));
+  uv_unref(reinterpret_cast<uv_handle_t*>(&prepare_handle_));
+  prepare_handle_.data = this;
+}
+
+BaseObjectPtr<IterationHistogram> IterationHistogram::Create(
+    Environment* env, const Histogram::Options& options) {
+  Local<Object> obj;
+  if (!GetConstructorTemplate(env)
+           ->InstanceTemplate()
+           ->NewInstance(env->context())
+           .ToLocal(&obj)) {
+    return nullptr;
+  }
+
+  return MakeBaseObject<IterationHistogram>(
+      env, obj, AsyncWrap::PROVIDER_ELDHISTOGRAM, options);
+}
+
+void IterationHistogram::PrepareCB(uv_prepare_t* handle) {
+  IterationHistogram* self = static_cast<IterationHistogram*>(handle->data);
+  if (!self->enabled_) return;
+  self->prepare_time_ = uv_hrtime();
+  self->timeout_ = uv_backend_timeout(handle->loop);
+}
+
+void IterationHistogram::CheckCB(uv_check_t* handle) {
+  IterationHistogram* self =
+      ContainerOf(&IterationHistogram::check_handle_, handle);
+  if (!self->enabled_) return;
+
+  uint64_t check_time = uv_hrtime();
+  uint64_t poll_time = check_time - self->prepare_time_;
+  uint64_t latency = self->prepare_time_ - self->check_time_;
+
+  if (self->timeout_ >= 0) {
+    uint64_t timeout_ns = static_cast<uint64_t>(self->timeout_) * 1000 * 1000;
+    if (poll_time > timeout_ns) {
+      latency += poll_time - timeout_ns;
+    }
+  }
+
+  self->histogram()->Record(latency == 0 ? 1 : latency);
+  self->check_time_ = check_time;
+}
+
+void IterationHistogram::MemoryInfo(MemoryTracker* tracker) const {
+  tracker->TrackField("histogram", histogram());
+}
+
+void IterationHistogram::OnStart(StartFlags flags) {
+  if (enabled_ || IsHandleClosing()) return;
+  enabled_ = true;
+  if (flags == StartFlags::RESET) histogram()->Reset();
+  check_time_ = uv_hrtime();
+  prepare_time_ = check_time_;
+  timeout_ = 0;
+  uv_check_start(&check_handle_, CheckCB);
+  uv_prepare_start(&prepare_handle_, PrepareCB);
+  uv_unref(reinterpret_cast<uv_handle_t*>(&check_handle_));
+  uv_unref(reinterpret_cast<uv_handle_t*>(&prepare_handle_));
+}
+
+void IterationHistogram::OnStop() {
+  if (!enabled_ || IsHandleClosing()) return;
+  enabled_ = false;
+  uv_check_stop(&check_handle_);
+  uv_prepare_stop(&prepare_handle_);
+}
+
+void IterationHistogram::Close(Local<Value> close_callback) {
+  if (IsHandleClosing()) return;
+  OnStop();
+  HandleWrap::Close(close_callback);
+  uv_close(reinterpret_cast<uv_handle_t*>(&prepare_handle_), nullptr);
+}
+
+void IterationHistogram::Start(const FunctionCallbackInfo<Value>& args) {
+  StartHandleHistogram<IterationHistogram>(args.This(), args[0]->IsTrue());
+}
+
+void IterationHistogram::FastStart(Local<Value> receiver, bool reset) {
+  TRACK_V8_FAST_API_CALL("histogram.eventLoopDelay.start");
+  StartHandleHistogram<IterationHistogram>(receiver, reset);
+}
+
+void IterationHistogram::Stop(const FunctionCallbackInfo<Value>& args) {
+  StopHandleHistogram<IterationHistogram>(args.This());
+}
+
+void IterationHistogram::FastStop(Local<Value> receiver) {
+  TRACK_V8_FAST_API_CALL("histogram.eventLoopDelay.stop");
+  StopHandleHistogram<IterationHistogram>(receiver);
 }
 
 void HistogramImpl::GetCount(const FunctionCallbackInfo<Value>& args) {
@@ -604,6 +750,11 @@ HistogramImpl* HistogramImpl::FromJSObject(Local<Value> value) {
       obj->GetAlignedPointerFromInternalField(HistogramImpl::kImplField));
 }
 
+std::unique_ptr<worker::TransferData> IterationHistogram::CloneForMessaging()
+    const {
+  return std::make_unique<HistogramBase::HistogramTransferData>(histogram());
+}
+
 std::unique_ptr<worker::TransferData>
 IntervalHistogram::CloneForMessaging() const {
   return std::make_unique<HistogramBase::HistogramTransferData>(histogram());