async_hooks: use typed array stack as fast path

- Communicate the current async stack length through a
  typed array field rather than a native binding method
- Add a new fixed-size `async_ids_fast_stack` typed array
  that contains the async ID stack up to a fixed limit.
  This increases performance noticeably, since most of the time
  the async ID stack will not be more than a handful of
  levels deep.
- Make the JS `pushAsyncIds()` and `popAsyncIds()` functions
  do the same thing as the native ones if the fast path
  is applicable.

Benchmarks:

    $ ./node benchmark/compare.js --new ./node --old ./node-master --runs 10 --filter next-tick process | Rscript benchmark/compare.R
    [00:03:25|% 100| 6/6 files | 20/20 runs | 1/1 configs]: Done
                                                   improvement confidence      p.value
     process/next-tick-breadth-args.js millions=4     19.72 %        *** 3.013913e-06
     process/next-tick-breadth.js millions=4          27.33 %        *** 5.847983e-11
     process/next-tick-depth-args.js millions=12      40.08 %        *** 1.237127e-13
     process/next-tick-depth.js millions=12           77.27 %        *** 1.413290e-11
     process/next-tick-exec-args.js millions=5        13.58 %        *** 1.245180e-07
     process/next-tick-exec.js millions=5             16.80 %        *** 2.961386e-07

lib/internal/async_hooks.js

@@ -19,14 +19,20 @@ const async_wrap = process.binding('async_wrap');
  *    retrieving the triggerAsyncId value is passing directly to the
  *    constructor -> value set in kDefaultTriggerAsyncId -> executionAsyncId of
  *    the current resource.
+ *
+ * async_ids_fast_stack is a Float64Array that contains part of the async ID
+ * stack. Each pushAsyncIds() call adds two doubles to it, and each
+ * popAsyncIds() call removes two doubles from it.
+ * It has a fixed size, so if that is exceeded, calls to the native
+ * side are used instead in pushAsyncIds() and popAsyncIds().
  */
-const { async_hook_fields, async_id_fields } = async_wrap;
+const { async_hook_fields, async_id_fields, async_ids_fast_stack } = async_wrap;
 // Store the pair executionAsyncId and triggerAsyncId in a std::stack on
 // Environment::AsyncHooks::ids_stack_ tracks the resource responsible for the
 // current execution stack. This is unwound as each resource exits. In the case
 // of a fatal exception this stack is emptied after calling each hook's after()
 // callback.
-const { pushAsyncIds, popAsyncIds } = async_wrap;
+const { pushAsyncIds: pushAsyncIds_, popAsyncIds: popAsyncIds_ } = async_wrap;
 // For performance reasons, only track Proimses when a hook is enabled.
 const { enablePromiseHook, disablePromiseHook } = async_wrap;
 // Properties in active_hooks are used to keep track of the set of hooks being
@@ -60,8 +66,9 @@ const active_hooks = {
 // async execution. These are tracked so if the user didn't include callbacks
 // for a given step, that step can bail out early.
 const { kInit, kBefore, kAfter, kDestroy, kPromiseResolve,
-        kCheck, kExecutionAsyncId, kAsyncIdCounter,
-        kDefaultTriggerAsyncId } = async_wrap.constants;
+        kCheck, kExecutionAsyncId, kAsyncIdCounter, kTriggerAsyncId,
+        kDefaultTriggerAsyncId, kStackLength,
+        kFastStackCapacity } = async_wrap.constants;
 
 // Used in AsyncHook and AsyncResource.
 const init_symbol = Symbol('init');
@@ -332,6 +339,39 @@ function emitDestroyScript(asyncId) {
 }
 
 
+// This is the equivalent of the native push_async_ids() call.
+function pushAsyncIds(asyncId, triggerAsyncId) {
+  const stackLength = async_hook_fields[kStackLength];
+  if (stackLength >= kFastStackCapacity)
+    return pushAsyncIds_(asyncId, triggerAsyncId);
+  const offset = stackLength;
+  async_ids_fast_stack[offset * 2] = async_id_fields[kExecutionAsyncId];
+  async_ids_fast_stack[offset * 2 + 1] = async_id_fields[kTriggerAsyncId];
+  async_hook_fields[kStackLength]++;
+  async_id_fields[kExecutionAsyncId] = asyncId;
+  async_id_fields[kTriggerAsyncId] = triggerAsyncId;
+}
+
+
+// This is the equivalent of the native pop_async_ids() call.
+function popAsyncIds(asyncId) {
+  if (async_hook_fields[kStackLength] === 0) return false;
+  const stackLength = async_hook_fields[kStackLength];
+
+  if (stackLength > kFastStackCapacity ||
+        (async_hook_fields[kCheck] > 0 &&
+         async_id_fields[kExecutionAsyncId] !== asyncId)) {
+    return popAsyncIds_(asyncId);
+  }
+
+  const offset = stackLength - 1;
+  async_id_fields[kExecutionAsyncId] = async_ids_fast_stack[2 * offset];
+  async_id_fields[kTriggerAsyncId] = async_ids_fast_stack[2 * offset + 1];
+  async_hook_fields[kStackLength] = offset;
+  return offset > 0;
+}
+
+
 module.exports = {
   // Private API
   getHookArrays,

lib/internal/bootstrap_node.js

@@ -367,9 +367,9 @@
     // Arrays containing hook flags and ids for async_hook calls.
     const { async_hook_fields, async_id_fields } = async_wrap;
     // Internal functions needed to manipulate the stack.
-    const { clearAsyncIdStack, asyncIdStackSize } = async_wrap;
+    const { clearAsyncIdStack } = async_wrap;
     const { kAfter, kExecutionAsyncId,
-            kDefaultTriggerAsyncId } = async_wrap.constants;
+            kDefaultTriggerAsyncId, kStackLength } = async_wrap.constants;
 
     process._fatalException = function(er) {
       var caught;
@@ -407,7 +407,7 @@
           do {
             NativeModule.require('internal/async_hooks').emitAfter(
               async_id_fields[kExecutionAsyncId]);
-          } while (asyncIdStackSize() > 0);
+          } while (async_hook_fields[kStackLength] > 0);
         // Or completely empty the id stack.
         } else {
           clearAsyncIdStack();

src/aliased_buffer.h

@@ -111,11 +111,17 @@ class AliasedBuffer {
           index_(that.index_) {
     }
 
-    inline Reference& operator=(const NativeT &val) {
+    template <typename T>
+    inline Reference& operator=(const T& val) {
       aliased_buffer_->SetValue(index_, val);
       return *this;
     }
 
+    // This is not caught by the template operator= above.
+    inline Reference& operator=(const Reference& val) {
+      return *this = static_cast<NativeT>(val);
+    }
+
     operator NativeT() const {
       return aliased_buffer_->GetValue(index_);
     }

src/async_wrap.cc

@@ -468,13 +468,6 @@ void AsyncWrap::PopAsyncIds(const FunctionCallbackInfo<Value>& args) {
 }
 
 
-void AsyncWrap::AsyncIdStackSize(const FunctionCallbackInfo<Value>& args) {
-  Environment* env = Environment::GetCurrent(args);
-  args.GetReturnValue().Set(
-      static_cast<double>(env->async_hooks()->stack_size()));
-}
-
-
 void AsyncWrap::ClearAsyncIdStack(const FunctionCallbackInfo<Value>& args) {
   Environment* env = Environment::GetCurrent(args);
   env->async_hooks()->clear_async_id_stack();
@@ -513,7 +506,6 @@ void AsyncWrap::Initialize(Local<Object> target,
   env->SetMethod(target, "setupHooks", SetupHooks);
   env->SetMethod(target, "pushAsyncIds", PushAsyncIds);
   env->SetMethod(target, "popAsyncIds", PopAsyncIds);
-  env->SetMethod(target, "asyncIdStackSize", AsyncIdStackSize);
   env->SetMethod(target, "clearAsyncIdStack", ClearAsyncIdStack);
   env->SetMethod(target, "queueDestroyAsyncId", QueueDestroyAsyncId);
   env->SetMethod(target, "enablePromiseHook", EnablePromiseHook);
@@ -551,6 +543,11 @@ void AsyncWrap::Initialize(Local<Object> target,
                          "async_id_fields",
                          env->async_hooks()->async_id_fields().GetJSArray());
 
+  FORCE_SET_TARGET_FIELD(target,
+                         "async_ids_fast_stack",
+                         env->async_hooks()->async_ids_fast_stack()
+                            .GetJSArray());
+
   Local<Object> constants = Object::New(isolate);
 #define SET_HOOKS_CONSTANT(name)                                              \
   FORCE_SET_TARGET_FIELD(                                                     \
@@ -567,6 +564,8 @@ void AsyncWrap::Initialize(Local<Object> target,
   SET_HOOKS_CONSTANT(kTriggerAsyncId);
   SET_HOOKS_CONSTANT(kAsyncIdCounter);
   SET_HOOKS_CONSTANT(kDefaultTriggerAsyncId);
+  SET_HOOKS_CONSTANT(kStackLength);
+  SET_HOOKS_CONSTANT(kFastStackCapacity);
 #undef SET_HOOKS_CONSTANT
   FORCE_SET_TARGET_FIELD(target, "constants", constants);
 

src/async_wrap.h

@@ -122,7 +122,6 @@ class AsyncWrap : public BaseObject {
   static void GetAsyncId(const v8::FunctionCallbackInfo<v8::Value>& args);
   static void PushAsyncIds(const v8::FunctionCallbackInfo<v8::Value>& args);
   static void PopAsyncIds(const v8::FunctionCallbackInfo<v8::Value>& args);
-  static void AsyncIdStackSize(const v8::FunctionCallbackInfo<v8::Value>& args);
   static void ClearAsyncIdStack(
     const v8::FunctionCallbackInfo<v8::Value>& args);
   static void AsyncReset(const v8::FunctionCallbackInfo<v8::Value>& args);

src/env-inl.h

@@ -55,6 +55,7 @@ inline MultiIsolatePlatform* IsolateData::platform() const {
 
 inline Environment::AsyncHooks::AsyncHooks(v8::Isolate* isolate)
     : isolate_(isolate),
+      async_ids_fast_stack_(isolate, kFastStackCapacity * 2),
       fields_(isolate, kFieldsCount),
       async_id_fields_(isolate, kUidFieldsCount) {
   v8::HandleScope handle_scope(isolate_);
@@ -101,6 +102,11 @@ Environment::AsyncHooks::async_id_fields() {
   return async_id_fields_;
 }
 
+inline AliasedBuffer<double, v8::Float64Array>&
+Environment::AsyncHooks::async_ids_fast_stack() {
+  return async_ids_fast_stack_;
+}
+
 inline v8::Local<v8::String> Environment::AsyncHooks::provider_string(int idx) {
   return providers_[idx].Get(isolate_);
 }
@@ -110,6 +116,7 @@ inline void Environment::AsyncHooks::no_force_checks() {
   fields_[kCheck] = fields_[kCheck] - 1;
 }
 
+// Remember to keep this code aligned with pushAsyncIds() in JS.
 inline void Environment::AsyncHooks::push_async_ids(double async_id,
                                               double trigger_async_id) {
   // Since async_hooks is experimental, do only perform the check
@@ -119,16 +126,26 @@ inline void Environment::AsyncHooks::push_async_ids(double async_id,
     CHECK_GE(trigger_async_id, -1);
   }
 
-  async_ids_stack_.push({ async_id_fields_[kExecutionAsyncId],
-                    async_id_fields_[kTriggerAsyncId] });
+  uint32_t offset = fields_[kStackLength];
+  if (offset < kFastStackCapacity) {
+    async_ids_fast_stack_[2 * offset] = async_id_fields_[kExecutionAsyncId];
+    async_ids_fast_stack_[2 * offset + 1] = async_id_fields_[kTriggerAsyncId];
+  } else {
+    async_ids_stack_.push({
+      async_id_fields_[kExecutionAsyncId],
+      async_id_fields_[kTriggerAsyncId]
+    });
+  }
+  fields_[kStackLength] = fields_[kStackLength] + 1;
   async_id_fields_[kExecutionAsyncId] = async_id;
   async_id_fields_[kTriggerAsyncId] = trigger_async_id;
 }
 
+// Remember to keep this code aligned with popAsyncIds() in JS.
 inline bool Environment::AsyncHooks::pop_async_id(double async_id) {
   // In case of an exception then this may have already been reset, if the
   // stack was multiple MakeCallback()'s deep.
-  if (async_ids_stack_.empty()) return false;
+  if (fields_[kStackLength] == 0) return false;
 
   // Ask for the async_id to be restored as a check that the stack
   // hasn't been corrupted.
@@ -150,22 +167,26 @@ inline bool Environment::AsyncHooks::pop_async_id(double async_id) {
     ABORT_NO_BACKTRACE();
   }
 
-  auto async_ids = async_ids_stack_.top();
-  async_ids_stack_.pop();
-  async_id_fields_[kExecutionAsyncId] = async_ids.async_id;
-  async_id_fields_[kTriggerAsyncId] = async_ids.trigger_async_id;
-  return !async_ids_stack_.empty();
-}
-
-inline size_t Environment::AsyncHooks::stack_size() {
-  return async_ids_stack_.size();
+  uint32_t offset = fields_[kStackLength] - 1;
+  if (offset >= kFastStackCapacity) {
+    auto async_ids = async_ids_stack_.top();
+    async_ids_stack_.pop();
+    async_id_fields_[kExecutionAsyncId] = async_ids.async_id;
+    async_id_fields_[kTriggerAsyncId] = async_ids.trigger_async_id;
+  } else {
+    async_id_fields_[kExecutionAsyncId] = async_ids_fast_stack_[2 * offset];
+    async_id_fields_[kTriggerAsyncId] = async_ids_fast_stack_[2 * offset + 1];
+  }
+  fields_[kStackLength] = offset;
+  return fields_[kStackLength] > 0;
 }
 
 inline void Environment::AsyncHooks::clear_async_id_stack() {
   while (!async_ids_stack_.empty())
     async_ids_stack_.pop();
   async_id_fields_[kExecutionAsyncId] = 0;
   async_id_fields_[kTriggerAsyncId] = 0;
+  fields_[kStackLength] = 0;
 }
 
 inline Environment::AsyncHooks::DefaultTriggerAsyncIdScope

src/env.h

@@ -369,6 +369,7 @@ class Environment {
       kPromiseResolve,
       kTotals,
       kCheck,
+      kStackLength,
       kFieldsCount,
     };
 
@@ -380,18 +381,22 @@ class Environment {
       kUidFieldsCount,
     };
 
+    enum FastStackFields {
+      kFastStackCapacity = 64
+    };
+
     AsyncHooks() = delete;
 
     inline AliasedBuffer<uint32_t, v8::Uint32Array>& fields();
     inline AliasedBuffer<double, v8::Float64Array>& async_id_fields();
+    inline AliasedBuffer<double, v8::Float64Array>& async_ids_fast_stack();
 
     inline v8::Local<v8::String> provider_string(int idx);
 
     inline void no_force_checks();
 
     inline void push_async_ids(double async_id, double trigger_async_id);
     inline bool pop_async_id(double async_id);
-    inline size_t stack_size();
     inline void clear_async_id_stack();  // Used in fatal exceptions.
 
     // Used to set the kDefaultTriggerAsyncId in a scope. This is instead of
@@ -420,6 +425,9 @@ class Environment {
     v8::Isolate* isolate_;
     // Stores the ids of the current execution context stack.
     std::stack<async_context> async_ids_stack_;
+    // Stores the ids of the current execution context stack with limited
+    // capacity, but with the benefit of much faster access from JS.
+    AliasedBuffer<double, v8::Float64Array> async_ids_fast_stack_;
     // Attached to a Uint32Array that tracks the number of active hooks for
     // each type.
     AliasedBuffer<uint32_t, v8::Uint32Array> fields_;

test/parallel/test-async-hooks-recursive-stack.js

@@ -0,0 +1,20 @@
+'use strict';
+require('../common');
+const assert = require('assert');
+const async_hooks = require('async_hooks');
+
+// This test verifies that the async ID stack can grow indefinitely.
+
+function recurse(n) {
+  const a = new async_hooks.AsyncResource('foobar');
+  a.emitBefore();
+  assert.strictEqual(a.asyncId(), async_hooks.executionAsyncId());
+  assert.strictEqual(a.triggerAsyncId(), async_hooks.triggerAsyncId());
+  if (n >= 0)
+    recurse(n - 1);
+  assert.strictEqual(a.asyncId(), async_hooks.executionAsyncId());
+  assert.strictEqual(a.triggerAsyncId(), async_hooks.triggerAsyncId());
+  a.emitAfter();
+}
+
+recurse(1000);