fix release and add bench

Author: k-morozov

bench/bench_pool.cpp

@@ -46,19 +46,19 @@ void bench_logic(NExecutors::IExecutor& pool) {
     });
 }
 
-static void IntrusiveThreadPool(benchmark::State& state) {
-    for (auto _ : state) {
-        NExecutors::IntrusiveThreadPool pool{CountThreads};
-        pool.Start();
-        bench_logic(pool);
-        pool.WaitIdle();
-    }
-}
-BENCHMARK(IntrusiveThreadPool)
-    ->Name(std::format("IntrusiveThreadPool_task_{}", kTasks))
-    ->Repetitions(CountRepetitions)
-    ->Iterations(CountIteration)
-    ->Unit(benchmark::kMillisecond);
+//static void IntrusiveThreadPool(benchmark::State& state) {
+//    for (auto _ : state) {
+//        NExecutors::IntrusiveThreadPool pool{CountThreads};
+//        pool.Start();
+//        bench_logic(pool);
+//        pool.WaitIdle();
+//    }
+//}
+//BENCHMARK(IntrusiveThreadPool)
+//    ->Name(std::format("IntrusiveThreadPool_task_{}", kTasks))
+//    ->Repetitions(CountRepetitions)
+//    ->Iterations(CountIteration)
+//    ->Unit(benchmark::kMillisecond);
 
 static void DistributedPool(benchmark::State& state) {
     for (auto _ : state) {

bench/readme.md

@@ -28,6 +28,20 @@
 | DistributedPool_task_100000/iterations:10/repeats:5_stddev    | 8.96 ms| 0.477 ms| 5          |
 | DistributedPool_task_100000/iterations:10/repeats:5_cv        | 7.05 % | 0.80 %  | 5          |
 
+### DistributedPool Task 100,000 (queue spinlock)
+
+| Benchmark                                                     | Time   | CPU    | Iterations |
+|---------------------------------------------------------------|--------|--------|------------|
+| DistributedPool_task_100000/iterations:10/repeats:5           | 304 ms | 126 ms | 10         |
+| DistributedPool_task_100000/iterations:10/repeats:5           | 263 ms | 107 ms | 10         |
+| DistributedPool_task_100000/iterations:10/repeats:5           | 205 ms | 79.7 ms| 10         |
+| DistributedPool_task_100000/iterations:10/repeats:5           | 190 ms | 78.7 ms| 10         |
+| DistributedPool_task_100000/iterations:10/repeats:5           | 460 ms | 189 ms | 10         |
+| DistributedPool_task_100000/iterations:10/repeats:5_mean      | 285 ms | 116 ms | 5          |
+| DistributedPool_task_100000/iterations:10/repeats:5_median    | 263 ms | 107 ms | 5          |
+| DistributedPool_task_100000/iterations:10/repeats:5_stddev    | 108 ms | 45.5 ms| 5          |
+| DistributedPool_task_100000/iterations:10/repeats:5_cv        | 37.98 %| 39.18 %| 5          |
+
 ### DistributedPool Task 100,000 (std::mutex)
 
 | Benchmark                                                       | Time    | CPU    | Iterations |

src/components/sync/queue_spinlock.h

@@ -12,42 +12,32 @@ class QueueSpinLock final {
 public:
     class Guard final {
     public:
-        explicit Guard(QueueSpinLock& host) : host(host) {
-            host.Acquire(this);
-        }
+        explicit Guard(QueueSpinLock& host) : host(host) { host.Acquire(this); }
         ~Guard() {
-            Release();
+            if (is_owner) Release();
         }
 
         void Release() {
             host.Release(this);
+            is_owner.store(false);
         }
 
-        void SetOwner() {
-            is_owner.store(true);
-        }
+        void SetOwner() { is_owner.store(true); }
 
-        void SetNext(Guard* guard) {
-            next.store(guard);
-        }
+        void SetNext(Guard* guard) { next.store(guard); }
 
-        bool IsOwner() const {
-            return is_owner.load();
-        }
+        bool IsOwner() const { return is_owner.load(); }
 
-        bool HasNext() const {
-            return next.load() != nullptr;
-        }
+        bool HasNext() const { return next.load() != nullptr; }
 
-        void SetNextOwner() {
-            next.load()->SetOwner();
-        }
+        void SetNextOwner() { next.load()->SetOwner(); }
 
     private:
         QueueSpinLock& host;
         std::atomic<Guard*> next{};
         std::atomic<bool> is_owner{};
     };
+
 private:
     void Acquire(Guard* guard) {
         auto ancestor = tail_.exchange(guard);
@@ -57,7 +47,8 @@ class QueueSpinLock final {
         }
 
         ancestor->SetNext(guard);
-        while(!guard->IsOwner()) {}
+        while (!guard->IsOwner()) {
+        }
     }
 
     void Release(Guard* guard) {
@@ -67,7 +58,7 @@ class QueueSpinLock final {
         }
 
         Guard* old_guard = guard;
-        while(!tail_.compare_exchange_weak(old_guard, nullptr)) {
+        while (!tail_.compare_exchange_weak(old_guard, nullptr)) {
             if (guard->HasNext()) {
                 guard->SetNextOwner();
                 return;

src/fiber/awaiter/mutex_awaiter.h

@@ -28,7 +28,6 @@ class AsyncMutexWaiter : public IAwaiter,
         stopped_handle = handle;
         async_mutex->Park(this);
 
-//        guard.release()->unlock();
         guard.Release();
     }