watchdog: Optimize WatchdogImpl::touch in preparation to more frequent petting of the watchdog.

include/envoy/server/watchdog.h

@@ -37,7 +37,6 @@ class WatchDog {
    */
   virtual void touch() PURE;
   virtual Thread::ThreadId threadId() const PURE;
-  virtual MonotonicTime lastTouchTime() const PURE;
 };
 
 using WatchDogSharedPtr = std::shared_ptr<WatchDog>;

source/server/guarddog_impl.cc

@@ -81,11 +81,9 @@ GuardDogImpl::GuardDogImpl(Stats::Scope& stats_scope, const Server::Configuratio
 GuardDogImpl::~GuardDogImpl() { stop(); }
 
 void GuardDogImpl::step() {
-  {
-    Thread::LockGuard guard(mutex_);
-    if (!run_thread_) {
-      return;
-    }
+  Thread::LockGuard guard(mutex_);
+  if (!run_thread_) {
+    return;
   }
 
   const auto now = time_source_.monotonicTime();
@@ -102,7 +100,14 @@ void GuardDogImpl::step() {
                  static_cast<size_t>(ceil(multi_kill_fraction_ * watched_dogs_.size())));
 
     for (auto& watched_dog : watched_dogs_) {
-      const auto ltt = watched_dog->dog_->lastTouchTime();
+      if (watched_dog->dog_->touchCount() > 0) {
+        // Watchdog was touched since guarddog last checked; update time and reset the count.
+        watched_dog->dog_->resetTouchCount();
+        watched_dog->last_touch_time_ = now;
+        continue;
+      }
+
+      const auto ltt = watched_dog->last_touch_time_;
       const auto tid = watched_dog->dog_->threadId();
       const auto delta = now - ltt;
       if (watched_dog->last_alert_time_ && watched_dog->last_alert_time_.value() < ltt) {
@@ -156,12 +161,9 @@ void GuardDogImpl::step() {
     invokeGuardDogActions(WatchDogAction::MISS, miss_threads, now);
   }
 
-  {
-    Thread::LockGuard guard(mutex_);
-    test_interlock_hook_->signalFromImpl(now);
-    if (run_thread_) {
-      loop_timer_->enableTimer(loop_interval_);
-    }
+  test_interlock_hook_->signalFromImpl();
+  if (run_thread_) {
+    loop_timer_->enableTimer(loop_interval_);
   }
 }
 
@@ -172,14 +174,13 @@ WatchDogSharedPtr GuardDogImpl::createWatchDog(Thread::ThreadId thread_id,
   // accessed out of the locked section below is const (time_source_ has no
   // state).
   const auto wd_interval = loop_interval_ / 2;
-  WatchDogSharedPtr new_watchdog =
-      std::make_shared<WatchDogImpl>(std::move(thread_id), time_source_, wd_interval);
+  auto new_watchdog = std::make_shared<WatchDogImpl>(std::move(thread_id), wd_interval);
   WatchedDogPtr watched_dog = std::make_unique<WatchedDog>(stats_scope_, thread_name, new_watchdog);
+  new_watchdog->touch();
   {
     Thread::LockGuard guard(wd_lock_);
     watched_dogs_.push_back(std::move(watched_dog));
   }
-  new_watchdog->touch();
   return new_watchdog;
 }
 
@@ -227,7 +228,7 @@ void GuardDogImpl::invokeGuardDogActions(
 }
 
 GuardDogImpl::WatchedDog::WatchedDog(Stats::Scope& stats_scope, const std::string& thread_name,
-                                     const WatchDogSharedPtr& watch_dog)
+                                     const WatchDogImplSharedPtr& watch_dog)
     : dog_(watch_dog),
       miss_counter_(stats_scope.counterFromStatName(
           Stats::StatNameManagedStorage(fmt::format("server.{}.watchdog_miss", thread_name),

source/server/guarddog_impl.h

@@ -18,6 +18,8 @@
 #include "common/common/thread.h"
 #include "common/event/libevent.h"
 
+#include "server/watchdog_impl.h"
+
 #include "absl/types/optional.h"
 
 namespace Envoy {
@@ -45,16 +47,17 @@ class GuardDogImpl : public GuardDog {
     virtual ~TestInterlockHook() = default;
 
     /**
-     * Called from GuardDogImpl to indicate that it has evaluated all watch-dogs
-     * up to a particular point in time.
+     * Called from GuardDogImpl to indicate that it has evaluated all watch-dogs up to a particular
+     * point in time. Called while the Guarddog mutex is held.
      */
-    virtual void signalFromImpl(MonotonicTime) {}
+    virtual void signalFromImpl() {}
 
     /**
-     * Called from GuardDog tests to block until the implementation has reached
-     * the desired point in time.
+     * Called from GuardDog tests to block until the implementation has reached the desired
+     * condition. Called while the Guarddog mutex is held.
+     * @param mutex The Guarddog's mutex for use by Thread::CondVar::wait.
      */
-    virtual void waitFromTest(Thread::MutexBasicLockable&, MonotonicTime) {}
+    virtual void waitFromTest(Thread::MutexBasicLockable& /*mutex*/) {}
   };
 
   /**
@@ -84,9 +87,8 @@ class GuardDogImpl : public GuardDog {
    */
   void forceCheckForTest() {
     Thread::LockGuard guard(mutex_);
-    MonotonicTime now = time_source_.monotonicTime();
     loop_timer_->enableTimer(std::chrono::milliseconds(0));
-    test_interlock_hook_->waitFromTest(mutex_, now);
+    test_interlock_hook_->waitFromTest(mutex_);
   }
 
   // Server::GuardDog
@@ -110,11 +112,13 @@ class GuardDogImpl : public GuardDog {
                         std::vector<std::pair<Thread::ThreadId, MonotonicTime>> thread_ltt_pairs,
                         MonotonicTime now);
 
+  using WatchDogImplSharedPtr = std::shared_ptr<WatchDogImpl>;
   struct WatchedDog {
     WatchedDog(Stats::Scope& stats_scope, const std::string& thread_name,
-               const WatchDogSharedPtr& watch_dog);
+               const WatchDogImplSharedPtr& watch_dog);
 
-    const WatchDogSharedPtr dog_;
+    const WatchDogImplSharedPtr dog_;
+    MonotonicTime last_touch_time_;
     absl::optional<MonotonicTime> last_alert_time_;
     bool miss_alerted_{};
     bool megamiss_alerted_{};

source/server/watchdog_impl.h

@@ -19,26 +19,24 @@ class WatchDogImpl : public WatchDog {
   /**
    * @param interval WatchDog timer interval (used after startWatchdog())
    */
-  WatchDogImpl(Thread::ThreadId thread_id, TimeSource& tsource, std::chrono::milliseconds interval)
-      : thread_id_(thread_id), time_source_(tsource),
-        latest_touch_time_since_epoch_(tsource.monotonicTime().time_since_epoch()),
-        timer_interval_(interval) {}
+  WatchDogImpl(Thread::ThreadId thread_id, std::chrono::milliseconds interval)
+      : thread_id_(thread_id), timer_interval_(interval) {}
 
   Thread::ThreadId threadId() const override { return thread_id_; }
-  MonotonicTime lastTouchTime() const override {
-    return MonotonicTime(latest_touch_time_since_epoch_.load());
-  }
+  // Accessors for use by GuardDogImpl to tell if the watchdog was touched recently and reset the
+  // count back to 0 when it has.
+  uint64_t touchCount() const { return touch_count_.load(); }
+  void resetTouchCount() { touch_count_.store(0); }
 
   // Server::WatchDog
   void startWatchdog(Event::Dispatcher& dispatcher) override;
-  void touch() override {
-    latest_touch_time_since_epoch_.store(time_source_.monotonicTime().time_since_epoch());
-  }
+  void touch() override { touch_count_.fetch_add(1); }
 
 private:
   const Thread::ThreadId thread_id_;
-  TimeSource& time_source_;
-  std::atomic<std::chrono::steady_clock::duration> latest_touch_time_since_epoch_;
+  // Use a 32-bit atomic counter since 64-bit atomics are not generally available in 32-bit ARM and
+  // 32-bit is large enough for our purposes.
+  std::atomic<uint32_t> touch_count_ = 0;
   Event::TimerPtr timer_;
   const std::chrono::milliseconds timer_interval_;
 };

test/mocks/server/watch_dog.h

@@ -15,7 +15,6 @@ class MockWatchDog : public WatchDog {
   MOCK_METHOD(void, startWatchdog, (Event::Dispatcher & dispatcher));
   MOCK_METHOD(void, touch, ());
   MOCK_METHOD(Thread::ThreadId, threadId, (), (const));
-  MOCK_METHOD(MonotonicTime, lastTouchTime, (), (const));
 };
 } // namespace Server
 } // namespace Envoy

test/server/guarddog_impl_test.cc

@@ -48,21 +48,23 @@ const int DISABLE_MEGAMISS = 1000000;
 class DebugTestInterlock : public GuardDogImpl::TestInterlockHook {
 public:
   // GuardDogImpl::TestInterlockHook
-  void signalFromImpl(MonotonicTime time) override {
-    impl_reached_ = time;
+  void signalFromImpl() override {
+    waiting_for_signal_ = false;
     impl_.notifyAll();
   }
 
-  void waitFromTest(Thread::MutexBasicLockable& mutex, MonotonicTime time) override
+  void waitFromTest(Thread::MutexBasicLockable& mutex) override
       ABSL_EXCLUSIVE_LOCKS_REQUIRED(mutex) {
-    while (impl_reached_ < time) {
+    ASSERT(!waiting_for_signal_);
+    waiting_for_signal_ = true;
+    while (waiting_for_signal_) {
       impl_.wait(mutex);
     }
   }
 
 private:
   Thread::CondVar impl_;
-  MonotonicTime impl_reached_;
+  bool waiting_for_signal_ = false;
 };
 
 // We want to make sure guard-dog is tested with both simulated time and real
@@ -307,6 +309,7 @@ TEST_P(GuardDogMissTest, MissTest) {
   initGuardDog(stats_store_, config_miss_);
   auto unpet_dog =
       guard_dog_->createWatchDog(api_->threadFactory().currentThreadId(), "test_thread");
+  guard_dog_->forceCheckForTest();
   // We'd better start at 0:
   checkMiss(0, "MissTest check 1");
   // At 300ms we shouldn't have hit the timeout yet:
@@ -332,6 +335,7 @@ TEST_P(GuardDogMissTest, MegaMissTest) {
   initGuardDog(stats_store_, config_mega_);
   auto unpet_dog =
       guard_dog_->createWatchDog(api_->threadFactory().currentThreadId(), "test_thread");
+  guard_dog_->forceCheckForTest();
   // We'd better start at 0:
   checkMegaMiss(0, "MegaMissTest check 1");
   // This shouldn't be enough to increment the stat:
@@ -358,6 +362,7 @@ TEST_P(GuardDogMissTest, MissCountTest) {
   initGuardDog(stats_store_, config_miss_);
   auto sometimes_pet_dog =
       guard_dog_->createWatchDog(api_->threadFactory().currentThreadId(), "test_thread");
+  guard_dog_->forceCheckForTest();
   // These steps are executed once without ever touching the watchdog.
   // Then the last step is to touch the watchdog and repeat the steps.
   // This verifies that the behavior is reset back to baseline after a touch.
@@ -380,9 +385,11 @@ TEST_P(GuardDogMissTest, MissCountTest) {
     // When we finally touch the dog we should get one more increment once the
     // timeout value expires:
     sometimes_pet_dog->touch();
+    guard_dog_->forceCheckForTest();
   }
   time_system_->advanceTimeWait(std::chrono::milliseconds(1000));
   sometimes_pet_dog->touch();
+  guard_dog_->forceCheckForTest();
   // Make sure megamiss still works:
   checkMegaMiss(0UL, "MissCountTest check 5");
   time_system_->advanceTimeWait(std::chrono::milliseconds(1500));
@@ -655,6 +662,7 @@ TEST_P(GuardDogActionsTest, MissShouldSaturateOnMissEvent) {
 
   // Touch the watchdog, which should allow the event to trigger again.
   first_dog_->touch();
+  guard_dog_->forceCheckForTest();
 
   time_system_->advanceTimeWait(std::chrono::milliseconds(101));
   guard_dog_->forceCheckForTest();
@@ -717,6 +725,7 @@ TEST_P(GuardDogActionsTest, MegaMissShouldSaturateOnMegaMissEvent) {
 
   // Touch the watchdog, which should allow the event to trigger again.
   first_dog_->touch();
+  guard_dog_->forceCheckForTest();
 
   time_system_->advanceTimeWait(std::chrono::milliseconds(101));
   guard_dog_->forceCheckForTest();
@@ -732,6 +741,7 @@ TEST_P(GuardDogActionsTest, ShouldRespectEventPriority) {
     initGuardDog(fake_stats_, config);
     auto first_dog = guard_dog_->createWatchDog(Thread::ThreadId(10), "test_thread");
     auto second_dog = guard_dog_->createWatchDog(Thread::ThreadId(11), "test_thread");
+    guard_dog_->forceCheckForTest();
     time_system_->advanceTimeWait(std::chrono::milliseconds(101));
     guard_dog_->forceCheckForTest();
   };
@@ -746,6 +756,7 @@ TEST_P(GuardDogActionsTest, ShouldRespectEventPriority) {
     initGuardDog(fake_stats_, config);
     auto first_dog = guard_dog_->createWatchDog(Thread::ThreadId(10), "test_thread");
     auto second_dog = guard_dog_->createWatchDog(Thread::ThreadId(11), "test_thread");
+    guard_dog_->forceCheckForTest();
     time_system_->advanceTimeWait(std::chrono::milliseconds(101));
     guard_dog_->forceCheckForTest();
   };