dispatcher: faster runOnAllThreads

source/common/thread_local/thread_local_impl.cc

@@ -98,15 +98,15 @@ void InstanceImpl::runOnAllThreads(Event::PostCb cb, Event::PostCb all_threads_c
   // all_threads_complete_cb method. Parallelism of main thread execution is being traded off
   // for programming simplicity here.
   cb();
-  std::shared_ptr<std::atomic<uint64_t>> worker_count =
-      std::make_shared<std::atomic<uint64_t>>(registered_threads_.size());
-  for (Event::Dispatcher& dispatcher : registered_threads_) {
-    dispatcher.post([this, worker_count, cb, all_threads_complete_cb]() -> void {
-      cb();
-      if (--*worker_count == 0) {
+
+  std::shared_ptr<Event::PostCb> shared_function(
+      new Event::PostCb(cb), [this, all_threads_complete_cb](Event::PostCb* cb) {
         main_thread_dispatcher_->post(all_threads_complete_cb);
-      }
-    });
+        delete cb;
+      });
+
+  for (Event::Dispatcher& dispatcher : registered_threads_) {
+    dispatcher.post([cb_guard = shared_function]() -> void { cb_guard->operator()(); });
   }
 }