Enables clean VM shutdown.

runtime/vm/thread_pool.cc

Index: runtime/vm/thread_pool.cc
diff --git a/runtime/vm/thread_pool.cc b/runtime/vm/thread_pool.cc
index 5b3a713d2d09b2c32be7605f622e7671669e36c6..2541eacd32a95bafcc4438fd7dc6cbf66f9ebeb2 100644
--- a/runtime/vm/thread_pool.cc
+++ b/runtime/vm/thread_pool.cc
@@ -11,24 +11,27 @@ namespace dart {
 
 DEFINE_FLAG(int, worker_timeout_millis, 5000,
             "Free workers when they have been idle for this amount of time.");
+#if defined(USING_SIMULATOR)
+// The simulator can take much longer to finish running Dart code.
+DEFINE_FLAG(int, shutdown_timeout_millis, 15000,
+            "Amount of time to wait for a worker to stop during shutdown.");
+#else
+DEFINE_FLAG(int, shutdown_timeout_millis, 1000,
+            "Amount of time to wait for a worker to stop during shutdown.");
+#endif
 
-Monitor* ThreadPool::exit_monitor_ = NULL;
-int* ThreadPool::exit_count_ = NULL;
-
-ThreadPool::ThreadPool()
-  : shutting_down_(false),
-    all_workers_(NULL),
-    idle_workers_(NULL),
-    count_started_(0),
-    count_stopped_(0),
-    count_running_(0),
-    count_idle_(0) {
-}
-
+Mutex ThreadPool::mutex_;
+bool ThreadPool::shutting_down_ = false;
+ThreadPool::Worker* ThreadPool::all_workers_ = NULL;
+ThreadPool::Worker* ThreadPool::idle_workers_ = NULL;
+uint64_t ThreadPool::count_started_ = 0;
+uint64_t ThreadPool::count_stopped_ = 0;
+uint64_t ThreadPool::count_running_ = 0;
+uint64_t ThreadPool::count_idle_ = 0;
+bool ThreadPool::shutdown_timeout_fatal_ = true;
 
-ThreadPool::~ThreadPool() {
-  Shutdown();
-}
+Monitor ThreadPool::exit_monitor_;
+ThreadPool::Worker* ThreadPool::shutting_down_workers_ = NULL;
 
 
 void ThreadPool::Run(Task* task) {
@@ -42,7 +45,7 @@ void ThreadPool::Run(Task* task) {
       return;
     }
     if (idle_workers_ == NULL) {
-      worker = new Worker(this);
+      worker = new Worker();
       ASSERT(worker != NULL);
       new_worker = true;
       count_started_++;
@@ -70,7 +73,7 @@ void ThreadPool::Run(Task* task) {
 }
 
 
-void ThreadPool::Shutdown() {
+bool ThreadPool::Shutdown() {
   Worker* saved = NULL;
   {
     MutexLocker ml(&mutex_);
@@ -94,14 +97,48 @@ void ThreadPool::Shutdown() {
   }
   // Release ThreadPool::mutex_ before calling Worker functions.
 
-  Worker* current = saved;
-  while (current != NULL) {
-    // We may access all_next_ without holding ThreadPool::mutex_ here
-    // because the worker is no longer owned by the ThreadPool.
-    Worker* next = current->all_next_;
-    current->all_next_ = NULL;
-    current->Shutdown();
-    current = next;
+  {
+    MonitorLocker eml(&ThreadPool::exit_monitor_);
+
+    // First tell all the workers to shut down.
+    Worker* current = saved;
+    while (current != NULL) {
+      Worker* next = current->all_next_;
+      if (current->id_ != OSThread::GetCurrentThreadId()) {
+        AddWorkerToShutdownList(current);
+      }
+      current->Shutdown();
+      current = next;
+    }
+    saved = NULL;
+
+    // Give workers a chance to exit gracefully.
+    const int64_t start_wait = OS::GetCurrentTimeMillis();
+    int timeout = FLAG_shutdown_timeout_millis;
+    while (shutting_down_workers_ != NULL) {
+      if (timeout > 0) {
+        // Here, we are waiting for workers to exit. When a worker exits we will
+        // be notified.
+        eml.Wait(timeout);
+
+        // We decrement the timeout for the next wait by the amount of time
+        // we've already waited. If the new timeout drops below zero, we break
+        // out of this loop, which triggers the termination code below.
+        const int64_t after_wait = OS::GetCurrentTimeMillis();
+        timeout = FLAG_shutdown_timeout_millis - (after_wait - start_wait);
+      } else {
+        break;
+      }
+    }
+
+    // It is an error if all workers have not exited within the timeout. We
+    // assume that they have run off into the weeds, and it is a bug.
+    bool timely_shutdown = shutting_down_workers_ == NULL;
+    if (!timely_shutdown && shutdown_timeout_fatal_) {
+      FATAL("Thread pool worker threads failed to exit.");
+    }
+
+    return timely_shutdown;
   }
 }
 
@@ -156,7 +193,7 @@ bool ThreadPool::RemoveWorkerFromAllList(Worker* worker) {
     all_workers_ = worker->all_next_;
     worker->all_next_ = NULL;
     worker->owned_ = false;
-    worker->pool_ = NULL;
+    worker->done_ = true;
     return true;
   }
 
@@ -206,6 +243,40 @@ bool ThreadPool::ReleaseIdleWorker(Worker* worker) {
 }
 
 
+// Only call while holding the exit_monitor_
+void ThreadPool::AddWorkerToShutdownList(Worker* worker) {
+  worker->shutdown_next_ = shutting_down_workers_;
+  shutting_down_workers_ = worker;
+}
+
+
+// Only call while holding the exit_monitor_
+bool ThreadPool::RemoveWorkerFromShutdownList(Worker* worker) {
+  ASSERT(worker != NULL);
+  if (shutting_down_workers_ == NULL) {
+    return false;
+  }
+
+  // Special case head of list.
+  if (shutting_down_workers_ == worker) {
+    shutting_down_workers_ = worker->shutdown_next_;
+    worker->shutdown_next_ = NULL;
+    return true;
+  }
+
+  for (Worker* current = shutting_down_workers_;
+       current->shutdown_next_ != NULL;
+       current = current->shutdown_next_) {
+    if (current->shutdown_next_ == worker) {
+      current->shutdown_next_ = worker->shutdown_next_;
+      worker->shutdown_next_ = NULL;
+      return true;
+    }
+  }
+  return false;
+}
+
+
 ThreadPool::Task::Task() {
 }
 
@@ -214,12 +285,15 @@ ThreadPool::Task::~Task() {
 }
 
 
-ThreadPool::Worker::Worker(ThreadPool* pool)
-  : pool_(pool),
+ThreadPool::Worker::Worker()
+  : done_(false),
     task_(NULL),
+    id_(OSThread::kInvalidThreadId),
+    started_(false),
     owned_(false),
     all_next_(NULL),
-    idle_next_(NULL) {
+    idle_next_(NULL),
+    shutdown_next_(NULL) {
 }
 
 
@@ -264,7 +338,7 @@ static int64_t ComputeTimeout(int64_t idle_start) {
 }
 
 
-void ThreadPool::Worker::Loop() {
+bool ThreadPool::Worker::Loop() {
   MonitorLocker ml(&monitor_);
   int64_t idle_start;
   while (true) {
@@ -281,10 +355,10 @@ void ThreadPool::Worker::Loop() {
 
     ASSERT(task_ == NULL);
     if (IsDone()) {
-      return;
+      return false;
     }
-    ASSERT(pool_ != NULL);
-    pool_->SetIdle(this);
+    ASSERT(!done_);
+    ThreadPool::SetIdle(this);
     idle_start = OS::GetCurrentTimeMillis();
     while (true) {
       Monitor::WaitResult result = ml.Wait(ComputeTimeout(idle_start));
@@ -294,21 +368,22 @@ void ThreadPool::Worker::Loop() {
         break;
       }
       if (IsDone()) {
-        return;
+        return false;
       }
-      if (result == Monitor::kTimedOut &&
-          pool_->ReleaseIdleWorker(this)) {
-        return;
+      if ((result == Monitor::kTimedOut) &&
+           ThreadPool::ReleaseIdleWorker(this)) {
+        return true;
       }
     }
   }
   UNREACHABLE();
+  return false;
 }
 
 
 void ThreadPool::Worker::Shutdown() {
   MonitorLocker ml(&monitor_);
-  pool_ = NULL;  // Fail fast if someone tries to access pool_.
+  done_ = true;
   ml.Notify();
 }
 
@@ -317,20 +392,55 @@ void ThreadPool::Worker::Shutdown() {
 void ThreadPool::Worker::Main(uword args) {
   Thread::EnsureInit();
   Worker* worker = reinterpret_cast<Worker*>(args);
-  worker->Loop();
+  bool delete_self = false;
+
+  {
+    MonitorLocker ml(&(worker->monitor_));
+    if (worker->IsDone()) {
+      // id_ hasn't been set, yet, but the ThreadPool is being shutdown.
+      // Delete the task, and return.
+      ASSERT(worker->task_);
+      delete worker->task_;
+      worker->task_ = NULL;
+      delete_self = true;
+    } else {
+      worker->id_ = OSThread::GetCurrentThreadId();
+      worker->started_ = true;
+    }
+  }
+
+  // We aren't able to delete the worker while holding the worker's monitor.
+  // Now that we have released it, and we know that ThreadPool::Shutdown
+  // won't touch it again, we can delete it and return.
+  if (delete_self) {
+    MonitorLocker eml(&ThreadPool::exit_monitor_);
+    RemoveWorkerFromShutdownList(worker);
+    delete worker;
+    eml.Notify();
+    return;
+  }
+
+  bool released = worker->Loop();
 
   // It should be okay to access these unlocked here in this assert.
-  ASSERT(!worker->owned_ &&
-         worker->all_next_ == NULL &&
-         worker->idle_next_ == NULL);
-
-  // The exit monitor is only used during testing.
-  if (ThreadPool::exit_monitor_) {
-    MonitorLocker ml(ThreadPool::exit_monitor_);
-    (*ThreadPool::exit_count_)++;
-    ml.Notify();
+  // worker->all_next_ is retained by the pool for shutdown monitoring.
+  ASSERT(!worker->owned_ && (worker->idle_next_ == NULL));
+
+  if (!released) {
+    // This worker is exiting because the thread pool is being shut down.
+    // Inform the thread pool that we are exiting. We remove this worker from
+    // shutting_down_workers_ list because there will be no need for the
+    // ThreadPool to take action for this worker.
+    MonitorLocker eml(&ThreadPool::exit_monitor_);
+    worker->id_ = OSThread::kInvalidThreadId;
+    RemoveWorkerFromShutdownList(worker);
+    delete worker;
+    eml.Notify();
+  } else {
+    // This worker is going down because it was idle for too long. This case
+    // is not due to a ThreadPool Shutdown. Thus, we simply delete the worker.
+    delete worker;
   }
-  delete worker;
 #if defined(TARGET_OS_WINDOWS)
   Thread::CleanUp();
 #endif