src: always compile and store code cache for native modules

This patch changes the NativeModuleLoader to always try to find
code cache for native modules when it compiles them, and always
produce and store the code cache after compilation. The cache
map is protected by a mutex and can be accessed by different
threads - including the worker threads and the main thread. Hence any
thread can reuse the code cache if the native module has already
been compiled by another thread - in particular the cache of the
bootstrappers and per_context.js will always be hit when a new thread
is spun.

This results in a ~6% startup overhead in the worst case
(when only the main thread is launched without requiring any additional
native module - it now needs to do the extra work of finding and
storing caches), which balances out the recent improvements by moving
the compilation to C++, but it also leads to a ~60% improvement in
the best case (when a worker thread is spun and requires a lot of native
modules thus hitting the cache compiled by the main thread).

PR-URL: #24950
Reviewed-By: Anna Henningsen <anna@addaleax.net>

lib/internal/bootstrap/cache.js

@@ -6,7 +6,9 @@
 // cannot be tampered with even with --expose-internals.
 
 const { NativeModule } = require('internal/bootstrap/loaders');
-const { source, compileCodeCache } = internalBinding('native_module');
+const {
+  source, getCodeCache, compileFunction
+} = internalBinding('native_module');
 const { hasTracing } = process.binding('config');
 
 const depsModule = Object.keys(source).filter(
@@ -69,6 +71,7 @@ module.exports = {
     (key) => !cannotUseCache.includes(key)
   ),
   getSource(id) { return source[id]; },
-  getCodeCache: compileCodeCache,
+  getCodeCache,
+  compileFunction,
   cannotUseCache
 };

src/node.cc

@@ -110,7 +110,6 @@ typedef int mode_t;
 
 namespace node {
 
-using native_module::NativeModuleLoader;
 using options_parser::kAllowedInEnvironment;
 using options_parser::kDisallowedInEnvironment;
 using v8::Array;
@@ -162,7 +161,6 @@ double prog_start_time;
 Mutex per_process_opts_mutex;
 std::shared_ptr<PerProcessOptions> per_process_opts {
     new PerProcessOptions() };
-NativeModuleLoader per_process_loader;
 static Mutex node_isolate_mutex;
 static Isolate* node_isolate;
 
@@ -1187,7 +1185,7 @@ static MaybeLocal<Value> ExecuteBootstrapper(
     const char* id,
     std::vector<Local<String>>* parameters,
     std::vector<Local<Value>>* arguments) {
-  MaybeLocal<Value> ret = per_process_loader.CompileAndCall(
+  MaybeLocal<Value> ret = per_process::native_module_loader.CompileAndCall(
       env->context(), id, parameters, arguments, env);
 
   // If there was an error during bootstrap then it was either handled by the
@@ -1906,7 +1904,7 @@ Local<Context> NewContext(Isolate* isolate,
     std::vector<Local<String>> parameters = {
         FIXED_ONE_BYTE_STRING(isolate, "global")};
     std::vector<Local<Value>> arguments = {context->Global()};
-    MaybeLocal<Value> result = per_process_loader.CompileAndCall(
+    MaybeLocal<Value> result = per_process::native_module_loader.CompileAndCall(
         context, "internal/per_context", &parameters, &arguments, nullptr);
     if (result.IsEmpty()) {
       // Execution failed during context creation.

src/node_binding.cc

@@ -411,13 +411,14 @@ void GetInternalBinding(const FunctionCallbackInfo<Value>& args) {
         exports->SetPrototype(env->context(), Null(env->isolate())).FromJust());
     DefineConstants(env->isolate(), exports);
   } else if (!strcmp(*module_v, "natives")) {
-    exports = per_process_loader.GetSourceObject(env->context());
+    exports = per_process::native_module_loader.GetSourceObject(env->context());
     // Legacy feature: process.binding('natives').config contains stringified
     // config.gypi
     CHECK(exports
               ->Set(env->context(),
                     env->config_string(),
-                    per_process_loader.GetConfigString(env->isolate()))
+                    per_process::native_module_loader.GetConfigString(
+                        env->isolate()))
               .FromJust());
   } else {
     return ThrowIfNoSuchModule(env, *module_v);

src/node_internals.h

@@ -91,7 +91,6 @@ extern bool v8_initialized;
 
 extern Mutex per_process_opts_mutex;
 extern std::shared_ptr<PerProcessOptions> per_process_opts;
-extern native_module::NativeModuleLoader per_process_loader;
 
 // Forward declaration
 class Environment;

src/node_native_module.cc

@@ -3,6 +3,11 @@
 #include "node_internals.h"
 
 namespace node {
+
+namespace per_process {
+native_module::NativeModuleLoader native_module_loader;
+}  // namespace per_process
+
 namespace native_module {
 
 using v8::Array;
@@ -78,13 +83,14 @@ void NativeModuleLoader::GetCacheUsage(
 void NativeModuleLoader::SourceObjectGetter(
     Local<Name> property, const PropertyCallbackInfo<Value>& info) {
   Local<Context> context = info.GetIsolate()->GetCurrentContext();
-  info.GetReturnValue().Set(per_process_loader.GetSourceObject(context));
+  info.GetReturnValue().Set(
+      per_process::native_module_loader.GetSourceObject(context));
 }
 
 void NativeModuleLoader::ConfigStringGetter(
     Local<Name> property, const PropertyCallbackInfo<Value>& info) {
   info.GetReturnValue().Set(
-      per_process_loader.GetConfigString(info.GetIsolate()));
+      per_process::native_module_loader.GetConfigString(info.GetIsolate()));
 }
 
 Local<Object> NativeModuleLoader::GetSourceObject(
@@ -96,41 +102,62 @@ Local<String> NativeModuleLoader::GetConfigString(Isolate* isolate) const {
   return config_.ToStringChecked(isolate);
 }
 
-Local<String> NativeModuleLoader::GetSource(Isolate* isolate,
-                                            const char* id) const {
-  const auto it = source_.find(id);
-  CHECK_NE(it, source_.end());
-  return it->second.ToStringChecked(isolate);
-}
-
 NativeModuleLoader::NativeModuleLoader() : config_(GetConfig()) {
   LoadJavaScriptSource();
   LoadCodeCache();
 }
 
-void NativeModuleLoader::CompileCodeCache(
-    const FunctionCallbackInfo<Value>& args) {
+// This is supposed to be run only by the main thread in
+// tools/generate_code_cache.js
+void NativeModuleLoader::GetCodeCache(const FunctionCallbackInfo<Value>& args) {
   Environment* env = Environment::GetCurrent(args);
+  Isolate* isolate = env->isolate();
+  CHECK(env->is_main_thread());
+
   CHECK(args[0]->IsString());
-  node::Utf8Value id(env->isolate(), args[0].As<String>());
+  node::Utf8Value id_v(isolate, args[0].As<String>());
+  const char* id = *id_v;
 
-  // TODO(joyeecheung): allow compiling cache for bootstrapper by
-  // switching on id
-  MaybeLocal<Value> result =
-      CompileAsModule(env, *id, CompilationResultType::kCodeCache);
-  if (!result.IsEmpty()) {
-    args.GetReturnValue().Set(result.ToLocalChecked());
+  const NativeModuleLoader& loader = per_process::native_module_loader;
+  MaybeLocal<Uint8Array> ret = loader.GetCodeCache(isolate, id);
+  if (!ret.IsEmpty()) {
+    args.GetReturnValue().Set(ret.ToLocalChecked());
   }
 }
 
+// This is supposed to be run only by the main thread in
+// tools/generate_code_cache.js
+MaybeLocal<Uint8Array> NativeModuleLoader::GetCodeCache(Isolate* isolate,
+                                                        const char* id) const {
+  EscapableHandleScope scope(isolate);
+  Mutex::ScopedLock lock(code_cache_mutex_);
+
+  ScriptCompiler::CachedData* cached_data = nullptr;
+  const auto it = code_cache_.find(id);
+  if (it == code_cache_.end()) {
+    // The module has not been compiled before.
+    return MaybeLocal<Uint8Array>();
+  }
+
+  cached_data = it->second.get();
+
+  MallocedBuffer<uint8_t> copied(cached_data->length);
+  memcpy(copied.data, cached_data->data, cached_data->length);
+  Local<ArrayBuffer> buf =
+      ArrayBuffer::New(isolate,
+                       copied.release(),
+                       cached_data->length,
+                       ArrayBufferCreationMode::kInternalized);
+  return scope.Escape(Uint8Array::New(buf, 0, cached_data->length));
+}
+
 void NativeModuleLoader::CompileFunction(
     const FunctionCallbackInfo<Value>& args) {
   Environment* env = Environment::GetCurrent(args);
   CHECK(args[0]->IsString());
   node::Utf8Value id(env->isolate(), args[0].As<String>());
 
-  MaybeLocal<Value> result =
-      CompileAsModule(env, *id, CompilationResultType::kFunction);
+  MaybeLocal<Function> result = CompileAsModule(env, *id);
   if (!result.IsEmpty()) {
     args.GetReturnValue().Set(result.ToLocalChecked());
   }
@@ -145,57 +172,43 @@ MaybeLocal<Value> NativeModuleLoader::CompileAndCall(
     std::vector<Local<Value>>* arguments,
     Environment* optional_env) {
   Isolate* isolate = context->GetIsolate();
-  MaybeLocal<Value> compiled = per_process_loader.LookupAndCompile(
-      context, id, parameters, CompilationResultType::kFunction, nullptr);
+  MaybeLocal<Function> compiled =
+      per_process::native_module_loader.LookupAndCompile(
+          context, id, parameters, nullptr);
   if (compiled.IsEmpty()) {
-    return compiled;
+    return MaybeLocal<Value>();
   }
   Local<Function> fn = compiled.ToLocalChecked().As<Function>();
   return fn->Call(
       context, v8::Null(isolate), arguments->size(), arguments->data());
 }
 
-MaybeLocal<Value> NativeModuleLoader::CompileAsModule(
-    Environment* env, const char* id, CompilationResultType result) {
+MaybeLocal<Function> NativeModuleLoader::CompileAsModule(Environment* env,
+                                                         const char* id) {
   std::vector<Local<String>> parameters = {env->exports_string(),
                                            env->require_string(),
                                            env->module_string(),
                                            env->process_string(),
                                            env->internal_binding_string()};
-  return per_process_loader.LookupAndCompile(
-      env->context(), id, &parameters, result, env);
-}
-
-// Returns nullptr if there is no code cache corresponding to the id
-ScriptCompiler::CachedData* NativeModuleLoader::GetCachedData(
-    const char* id) const {
-  const auto it = per_process_loader.code_cache_.find(id);
-  // This could be false if the module cannot be cached somehow.
-  // See lib/internal/bootstrap/cache.js on the modules that cannot be cached
-  if (it == per_process_loader.code_cache_.end()) {
-    return nullptr;
-  }
-
-  const uint8_t* code_cache_value = it->second.one_bytes_data();
-  size_t code_cache_length = it->second.length();
-
-  return new ScriptCompiler::CachedData(code_cache_value, code_cache_length);
+  return per_process::native_module_loader.LookupAndCompile(
+      env->context(), id, &parameters, env);
 }
 
 // Returns Local<Function> of the compiled module if return_code_cache
 // is false (we are only compiling the function).
 // Otherwise return a Local<Object> containing the cache.
-MaybeLocal<Value> NativeModuleLoader::LookupAndCompile(
+MaybeLocal<Function> NativeModuleLoader::LookupAndCompile(
     Local<Context> context,
     const char* id,
     std::vector<Local<String>>* parameters,
-    CompilationResultType result_type,
     Environment* optional_env) {
   Isolate* isolate = context->GetIsolate();
   EscapableHandleScope scope(isolate);
   Local<Value> ret;  // Used to convert to MaybeLocal before return
 
-  Local<String> source = GetSource(isolate, id);
+  const auto source_it = source_.find(id);
+  CHECK_NE(source_it, source_.end());
+  Local<String> source = source_it->second.ToStringChecked(isolate);
 
   std::string filename_s = id + std::string(".js");
   Local<String> filename =
@@ -204,31 +217,24 @@ MaybeLocal<Value> NativeModuleLoader::LookupAndCompile(
   Local<Integer> column_offset = Integer::New(isolate, 0);
   ScriptOrigin origin(filename, line_offset, column_offset);
 
-  bool use_cache = false;
-  ScriptCompiler::CachedData* cached_data = nullptr;
+  Mutex::ScopedLock lock(code_cache_mutex_);
 
-  // 1. We won't even check the existence of the cache if the binary is not
-  //    built with them.
-  // 2. If we are generating code cache for tools/general_code_cache.js, we
-  //    are not going to use any cache ourselves.
-  if (has_code_cache_ && result_type == CompilationResultType::kFunction) {
-    cached_data = GetCachedData(id);
-    if (cached_data != nullptr) {
-      use_cache = true;
+  ScriptCompiler::CachedData* cached_data = nullptr;
+  {
+    auto cache_it = code_cache_.find(id);
+    if (cache_it != code_cache_.end()) {
+      // Transfer ownership to ScriptCompiler::Source later.
+      cached_data = cache_it->second.release();
+      code_cache_.erase(cache_it);
     }
   }
 
+  const bool use_cache = cached_data != nullptr;
+  ScriptCompiler::CompileOptions options =
+      use_cache ? ScriptCompiler::kConsumeCodeCache
+                : ScriptCompiler::kEagerCompile;
   ScriptCompiler::Source script_source(source, origin, cached_data);
 
-  ScriptCompiler::CompileOptions options;
-  if (result_type == CompilationResultType::kCodeCache) {
-    options = ScriptCompiler::kEagerCompile;
-  } else if (use_cache) {
-    options = ScriptCompiler::kConsumeCodeCache;
-  } else {
-    options = ScriptCompiler::kNoCompileOptions;
-  }
-
   MaybeLocal<Function> maybe_fun =
       ScriptCompiler::CompileFunctionInContext(context,
                                                &script_source,
@@ -244,10 +250,14 @@ MaybeLocal<Value> NativeModuleLoader::LookupAndCompile(
     // In the case of early errors, v8 is already capable of
     // decorating the stack for us - note that we use CompileFunctionInContext
     // so there is no need to worry about wrappers.
-    return MaybeLocal<Value>();
+    return MaybeLocal<Function>();
   }
 
   Local<Function> fun = maybe_fun.ToLocalChecked();
+  // XXX(joyeecheung): this bookkeeping is not exactly accurate because
+  // it only starts after the Environment is created, so the per_context.js
+  // will never be in any of these two sets, but the two sets are only for
+  // testing anyway.
   if (use_cache) {
     if (optional_env != nullptr) {
       // This could happen when Node is run with any v8 flag, but
@@ -264,29 +274,15 @@ MaybeLocal<Value> NativeModuleLoader::LookupAndCompile(
     }
   }
 
-  if (result_type == CompilationResultType::kCodeCache) {
-    std::unique_ptr<ScriptCompiler::CachedData> cached_data(
-        ScriptCompiler::CreateCodeCacheForFunction(fun));
-    CHECK_NE(cached_data, nullptr);
-    size_t cached_data_length = cached_data->length;
-    // Since we have no special allocator to create an ArrayBuffer
-    // from a new'ed pointer, we will need to copy it - but this
-    // code path is only run by the tooling that generates the code
-    // cache to be bundled in the binary
-    // so it should be fine.
-    MallocedBuffer<uint8_t> copied(cached_data->length);
-    memcpy(copied.data, cached_data->data, cached_data_length);
-    Local<ArrayBuffer> buf =
-        ArrayBuffer::New(isolate,
-                         copied.release(),
-                         cached_data_length,
-                         ArrayBufferCreationMode::kInternalized);
-    ret = Uint8Array::New(buf, 0, cached_data_length);
-  } else {
-    ret = fun;
-  }
+  // Generate new cache for next compilation
+  std::unique_ptr<ScriptCompiler::CachedData> new_cached_data(
+      ScriptCompiler::CreateCodeCacheForFunction(fun));
+  CHECK_NE(new_cached_data, nullptr);
 
-  return scope.Escape(ret);
+  // The old entry should've been erased by now so we can just emplace
+  code_cache_.emplace(id, std::move(new_cached_data));
+
+  return scope.Escape(fun);
 }
 
 void NativeModuleLoader::Initialize(Local<Object> target,
@@ -320,8 +316,7 @@ void NativeModuleLoader::Initialize(Local<Object> target,
       target, "getCacheUsage", NativeModuleLoader::GetCacheUsage);
   env->SetMethod(
       target, "compileFunction", NativeModuleLoader::CompileFunction);
-  env->SetMethod(
-      target, "compileCodeCache", NativeModuleLoader::CompileCodeCache);
+  env->SetMethod(target, "getCodeCache", NativeModuleLoader::GetCodeCache);
   // internalBinding('native_module') should be frozen
   target->SetIntegrityLevel(context, IntegrityLevel::kFrozen).FromJust();
 }