worker: start worker threads from the built-in snapshot

doc/api/cli.md

@@ -2271,6 +2271,17 @@ added: v6.0.0
 
 Silence all process warnings (including deprecations).
 
+### `--no-worker-snapshot`
+
+<!-- YAML
+added: REPLACEME
+-->
+
+> Stability: 1 - Experimental
+
+Start worker threads by running the internal bootstrap from scratch instead of
+deserializing the bootstrapped context from the built-in startup snapshot.
+
 ### `--node-memory-debug`
 
 <!-- YAML
@@ -3937,6 +3948,7 @@ one is included in the list below.
 * `--no-strip-types`
 * `--no-warnings`
 * `--no-webstorage`
+* `--no-worker-snapshot`
 * `--node-memory-debug`
 * `--openssl-config`
 * `--openssl-legacy-provider`

doc/node.1

@@ -1142,6 +1142,10 @@ For more information, see the TypeScript type-stripping documentation.
 .It Fl -no-warnings
 Silence all process warnings (including deprecations).
 .
+.It Fl -no-worker-snapshot
+Start worker threads by running the internal bootstrap from scratch instead of
+deserializing the bootstrapped context from the built-in startup snapshot.
+.
 .It Fl -node-memory-debug
 Enable extra debug checks for memory leaks in Node.js internals. This is
 usually only useful for developers debugging Node.js itself.
@@ -2105,6 +2109,8 @@ one is included in the list below.
 .It
 \fB--no-webstorage\fR
 .It
+\fB--no-worker-snapshot\fR
+.It
 \fB--node-memory-debug\fR
 .It
 \fB--openssl-config\fR

lib/internal/bootstrap/switches/is_not_main_thread.js

@@ -6,6 +6,12 @@ const {
 
 delete process._debugProcess;
 delete process._debugEnd;
+// Also drop the other main-thread-only helpers is_main_thread.js installs, so
+// that this switch can be applied on top of a context bootstrapped for the
+// main thread (as when a worker starts from the built-in snapshot).
+delete process._debugPause;
+delete process._startProfilerIdleNotifier;
+delete process._stopProfilerIdleNotifier;
 
 function defineStream(name, getter) {
   ObjectDefineProperty(process, name, {

src/api/environment.cc

@@ -433,6 +433,11 @@ Environment* CreateEnvironment(
 
   const bool use_snapshot = context.IsEmpty();
   const EnvSerializeInfo* env_snapshot_info = nullptr;
+  // A worker thread (its IsolateData knows its Worker) deserializes the same
+  // bootstrapped principal context the main thread uses and then has the
+  // worker-side bootstrap switches applied on top of it (they are written as
+  // overrides of the main-thread setup).
+  const bool for_worker = isolate_data->worker_context() != nullptr;
   if (use_snapshot) {
     CHECK_NOT_NULL(isolate_data->snapshot_data());
     env_snapshot_info = &isolate_data->snapshot_data()->env_info;
@@ -469,12 +474,31 @@ Environment* CreateEnvironment(
       FreeEnvironment(env);
       return nullptr;
     }
-    SetIsolateErrorHandlers(isolate, {});
+    if (!for_worker) SetIsolateErrorHandlers(isolate, {});
   }
 
   Context::Scope context_scope(context);
   env->InitializeMainContext(context, env_snapshot_info);
 
+  if (use_snapshot && for_worker) {
+    // The deserialized context went through is_main_thread /
+    // does_own_process_state when the snapshot was built; the worker-side
+    // switches redefine exactly those pieces (stdio getters, signal wiring,
+    // process.abort/chdir/umask/..., debug helpers).
+    if (env->principal_realm()
+            ->ExecuteBootstrapper(
+                "internal/bootstrap/switches/is_not_main_thread")
+            .IsEmpty() ||
+        (!env->owns_process_state() &&
+         env->principal_realm()
+             ->ExecuteBootstrapper(
+                 "internal/bootstrap/switches/does_not_own_process_state")
+             .IsEmpty())) {
+      FreeEnvironment(env);
+      return nullptr;
+    }
+  }
+
 #if HAVE_INSPECTOR
   if (env->should_create_inspector()) {
     if (inspector_parent_handle) {

src/node_options.cc

@@ -1275,6 +1275,12 @@ EnvironmentOptionsParser::EnvironmentOptionsParser() {
 
 PerIsolateOptionsParser::PerIsolateOptionsParser(
   const EnvironmentOptionsParser& eop) {
+  AddOption("--worker-snapshot",
+            "start worker threads from the bootstrapped context in the "
+            "built-in startup snapshot",
+            BOOL_FIELD(worker_snapshot),
+            kAllowedInEnvvar,
+            true);
   AddOption("--track-heap-objects",
             "track heap object allocations for heap snapshots",
             BOOL_FIELD(track_heap_objects),

src/node_options.h

@@ -310,6 +310,7 @@ class EnvironmentOptions : public Options {
 
 class PerIsolateOptions : public Options {
  public:
+  bool worker_snapshot = true;  // --[no-]worker-snapshot
   PerIsolateOptions() = default;
   PerIsolateOptions(PerIsolateOptions&&) = default;
 

src/node_worker.cc

@@ -16,6 +16,7 @@
 #include "v8-profiler.h"
 
 #include <memory>
+#include <optional>
 #include <string>
 #include <vector>
 
@@ -293,6 +294,18 @@ size_t Worker::NearHeapLimit(void* data, size_t current_heap_limit,
   return new_limit;
 }
 
+// Can this worker start by deserializing the bootstrapped principal context
+// from the embedded snapshot (plus the worker-side switches) instead of
+// bootstrapping from scratch? Only that snapshot qualifies: an embedder's own
+// or a --snapshot-blob one has run application code in its main context.
+// --no-worker-snapshot opts out; kNoBrowserGlobals changes the bootstrap.
+bool Worker::UseWorkerContextSnapshot() const {
+  return snapshot_data_ != nullptr &&
+         snapshot_data_ == SnapshotBuilder::GetEmbeddedSnapshotData() &&
+         !(environment_flags_ & EnvironmentFlags::kNoBrowserGlobals) &&
+         per_process::cli_options->per_isolate->worker_snapshot;
+}
+
 void Worker::Run() {
   std::string trace_name = "[worker " + std::to_string(thread_id_.id) + "]" +
                            (name_ == "" ? "" : " " + name_);
@@ -341,7 +354,15 @@ void Worker::Run() {
         // resource constraints, we need something in place to handle it,
         // though.
         TryCatch try_catch(isolate_);
-        if (snapshot_data_ != nullptr) {
+        if (UseWorkerContextSnapshot()) {
+          // Leave `context` empty: CreateEnvironment() deserializes the
+          // bootstrapped principal context (kNodeMainContextIndex) and the
+          // Environment state that goes with it, applies the worker-side
+          // switches, and skips RunBootstrapping().
+          Debug(this,
+                "Worker %llu deserializes the bootstrapped context\n",
+                thread_id_.id);
+        } else if (snapshot_data_ != nullptr) {
           Debug(this,
                 "Worker %llu uses context from snapshot %d\n",
                 thread_id_.id,
@@ -358,7 +379,7 @@ void Worker::Run() {
               this, "Worker %llu builds context from scratch\n", thread_id_.id);
           context = NewContext(isolate_);
         }
-        if (context.IsEmpty()) {
+        if (context.IsEmpty() && !UseWorkerContextSnapshot()) {
           // TODO(joyeecheung): maybe this should be kBootstrapFailure instead?
           Exit(ExitCode::kGenericUserError,
                "ERR_WORKER_INIT_FAILED",
@@ -368,8 +389,8 @@ void Worker::Run() {
       }
 
       if (is_stopped()) return;
-      CHECK(!context.IsEmpty());
-      Context::Scope context_scope(context);
+      std::optional<Context::Scope> context_scope;
+      if (!context.IsEmpty()) context_scope.emplace(context);
       {
 #if HAVE_INSPECTOR
         environment_flags_ |= EnvironmentFlags::kNoWaitForInspectorFrontend;
@@ -385,6 +406,7 @@ void Worker::Run() {
             name_));
         if (is_stopped()) return;
         CHECK_NOT_NULL(env_);
+        if (!context_scope) context_scope.emplace(env_->context());
         env_->set_env_vars(std::move(env_vars_));
         SetProcessExitHandler(env_.get(), [this](Environment*, int exit_code) {
           Exit(static_cast<ExitCode>(exit_code));
@@ -1414,8 +1436,70 @@ void GetEnvMessagePort(const FunctionCallbackInfo<Value>& args) {
   }
 }
 
+// Per-thread values of the `worker` binding are lazy properties of the
+// per-isolate template, so that a bootstrapped context carries none of them
+// and can be deserialized by any thread.
+void ThreadIdGetter(Local<v8::Name>,
+                    const v8::PropertyCallbackInfo<Value>& info) {
+  Environment* env = Environment::GetCurrent(info);
+  info.GetReturnValue().Set(static_cast<double>(env->thread_id()));
+}
+
+void ThreadNameGetter(Local<v8::Name>,
+                      const v8::PropertyCallbackInfo<Value>& info) {
+  Environment* env = Environment::GetCurrent(info);
+  Local<String> name;
+  if (String::NewFromUtf8(info.GetIsolate(),
+                          env->thread_name().data(),
+                          NewStringType::kNormal,
+                          env->thread_name().size())
+          .ToLocal(&name)) {
+    info.GetReturnValue().Set(name);
+  }
+}
+
+void IsMainThreadGetter(Local<v8::Name>,
+                        const v8::PropertyCallbackInfo<Value>& info) {
+  info.GetReturnValue().Set(Environment::GetCurrent(info)->is_main_thread());
+}
+
+void IsInternalThreadGetter(Local<v8::Name>,
+                            const v8::PropertyCallbackInfo<Value>& info) {
+  Worker* worker =
+      Environment::GetCurrent(info)->isolate_data()->worker_context();
+  info.GetReturnValue().Set(worker != nullptr && worker->is_internal());
+}
+
+void OwnsProcessStateGetter(Local<v8::Name>,
+                            const v8::PropertyCallbackInfo<Value>& info) {
+  info.GetReturnValue().Set(
+      Environment::GetCurrent(info)->owns_process_state());
+}
+
+void ResourceLimitsGetter(Local<v8::Name>,
+                          const v8::PropertyCallbackInfo<Value>& info) {
+  Environment* env = Environment::GetCurrent(info);
+  if (env->worker_context() != nullptr) {
+    info.GetReturnValue().Set(
+        env->worker_context()->GetResourceLimits(info.GetIsolate()));
+  }
+}
+
 void CreateWorkerPerIsolateProperties(IsolateData* isolate_data,
                                       Local<ObjectTemplate> target) {
+  {
+    Isolate* isolate = isolate_data->isolate();
+    auto lazy = [&](const char* name, v8::AccessorNameGetterCallback getter) {
+      target->SetLazyDataProperty(OneByteString(isolate, name), getter);
+    };
+    lazy("threadId", ThreadIdGetter);
+    lazy("threadName", ThreadNameGetter);
+    lazy("isMainThread", IsMainThreadGetter);
+    lazy("isInternalThread", IsInternalThreadGetter);
+    lazy("ownsProcessState", OwnsProcessStateGetter);
+    lazy("resourceLimits", ResourceLimitsGetter);
+  }
+
   Isolate* isolate = isolate_data->isolate();
 
   {
@@ -1520,55 +1604,9 @@ void CreateWorkerPerContextProperties(Local<Object> target,
                                       Local<Value> unused,
                                       Local<Context> context,
                                       void* priv) {
-  Environment* env = Environment::GetCurrent(context);
-  Isolate* isolate = env->isolate();
-
-  target
-      ->Set(env->context(),
-            env->thread_id_string(),
-            Number::New(isolate, static_cast<double>(env->thread_id())))
-      .Check();
-
-  target
-      ->Set(env->context(),
-            env->thread_name_string(),
-            String::NewFromUtf8(isolate,
-                                env->thread_name().data(),
-                                NewStringType::kNormal,
-                                env->thread_name().size())
-                .ToLocalChecked())
-      .Check();
-
-  target
-      ->Set(env->context(),
-            FIXED_ONE_BYTE_STRING(isolate, "isMainThread"),
-            Boolean::New(isolate, env->is_main_thread()))
-      .Check();
-
-  Worker* worker = env->isolate_data()->worker_context();
-  bool is_internal = worker != nullptr && worker->is_internal();
-
-  // Set the is_internal property
-  target
-      ->Set(env->context(),
-            FIXED_ONE_BYTE_STRING(isolate, "isInternalThread"),
-            Boolean::New(isolate, is_internal))
-      .Check();
-
-  target
-      ->Set(env->context(),
-            FIXED_ONE_BYTE_STRING(isolate, "ownsProcessState"),
-            Boolean::New(isolate, env->owns_process_state()))
-      .Check();
-
-  if (!env->is_main_thread()) {
-    target
-        ->Set(env->context(),
-              FIXED_ONE_BYTE_STRING(isolate, "resourceLimits"),
-              env->worker_context()->GetResourceLimits(isolate))
-        .Check();
-  }
-
+  // threadId, threadName, isMainThread, isInternalThread, ownsProcessState
+  // and resourceLimits are lazy properties of the per-isolate template (see
+  // CreateWorkerPerIsolateProperties).
   NODE_DEFINE_CONSTANT(target, kMaxYoungGenerationSizeMb);
   NODE_DEFINE_CONSTANT(target, kMaxOldGenerationSizeMb);
   NODE_DEFINE_CONSTANT(target, kCodeRangeSizeMb);
@@ -1578,6 +1616,12 @@ void CreateWorkerPerContextProperties(Local<Object> target,
 
 void RegisterExternalReferences(ExternalReferenceRegistry* registry) {
   registry->Register(GetEnvMessagePort);
+  registry->Register(ThreadIdGetter);
+  registry->Register(ThreadNameGetter);
+  registry->Register(IsMainThreadGetter);
+  registry->Register(IsInternalThreadGetter);
+  registry->Register(OwnsProcessStateGetter);
+  registry->Register(ResourceLimitsGetter);
   registry->Register(Worker::New);
   registry->Register(Worker::StartThread);
   registry->Register(Worker::StopThread);

src/node_worker.h

@@ -42,6 +42,7 @@ class Worker : public AsyncWrap {
 
   // Run the worker. This is only called from the worker thread.
   void Run();
+  bool UseWorkerContextSnapshot() const;
 
   // Forcibly exit the thread with a specified exit code. This may be called
   // from any thread. `error_code` and `error_message` can be used to create