Server: Add deadlock checker

Crashes the server when the main thread is idle for too long

include/CommonLib/HealthCheckerAppComponent.hpp

@@ -0,0 +1,41 @@
+// Copyright (C) 2024 Jérôme "SirLynix" Leclercq ([email protected])
+// This file is part of the "This Space Of Mine" project
+// For conditions of distribution and use, see copyright notice in LICENSE
+
+#pragma once
+
+#ifndef TSOM_COMMONLIB_HEALTHCHECKERAPPCOMPONENT_HPP
+#define TSOM_COMMONLIB_HEALTHCHECKERAPPCOMPONENT_HPP
+
+#include <CommonLib/Export.hpp>
+#include <Nazara/Core/ApplicationComponent.hpp>
+#include <atomic>
+#include <semaphore>
+#include <thread>
+
+namespace tsom
+{
+	class TSOM_COMMONLIB_API HealthCheckerAppComponent : public Nz::ApplicationComponent
+	{
+		public:
+			HealthCheckerAppComponent(Nz::ApplicationBase& app, unsigned int maxHangSeconds);
+			HealthCheckerAppComponent(const HealthCheckerAppComponent&) = delete;
+			HealthCheckerAppComponent(HealthCheckerAppComponent&&) = delete;
+			~HealthCheckerAppComponent();
+
+			HealthCheckerAppComponent& operator=(const HealthCheckerAppComponent&) = delete;
+			HealthCheckerAppComponent& operator=(HealthCheckerAppComponent&&) = delete;
+
+		private:
+			void Update(Nz::Time elapsedTime) override;
+
+			std::atomic_uint m_updateCounter;
+			std::binary_semaphore m_exitChecker;
+			std::thread m_healthCheckThread;
+			unsigned int m_maxHangSeconds;
+	};
+}
+
+#include <CommonLib/HealthCheckerAppComponent.inl>
+
+#endif // TSOM_COMMONLIB_HEALTHCHECKERAPPCOMPONENT_HPP

include/CommonLib/HealthCheckerAppComponent.inl

@@ -0,0 +1,7 @@
+// Copyright (C) 2024 Jérôme "SirLynix" Leclercq ([email protected])
+// This file is part of the "This Space Of Mine" project
+// For conditions of distribution and use, see copyright notice in LICENSE
+
+namespace tsom
+{
+}

src/CommonLib/HealthCheckerAppComponent.cpp

@@ -0,0 +1,56 @@
+// Copyright (C) 2024 Jérôme "SirLynix" Leclercq ([email protected])
+// This file is part of the "This Space Of Mine" project
+// For conditions of distribution and use, see copyright notice in LICENSE
+
+#include <CommonLib/HealthCheckerAppComponent.hpp>
+#include <fmt/color.h>
+#include <chrono>
+#include <csignal>
+
+namespace tsom
+{
+	HealthCheckerAppComponent::HealthCheckerAppComponent(Nz::ApplicationBase& app, unsigned int maxHangSeconds) :
+	ApplicationComponent(app),
+	m_updateCounter(0),
+	m_exitChecker(0),
+	m_maxHangSeconds(maxHangSeconds)
+	{
+		m_healthCheckThread = std::thread([this]
+		{
+			unsigned int lastUpdateCounter = 0;
+			unsigned int hangCounter = 0;
+
+			while (!m_exitChecker.try_acquire_for(std::chrono::seconds(1)))
+			{
+				unsigned int updateCount = m_updateCounter.load(std::memory_order_relaxed);
+				if NAZARA_LIKELY(updateCount != lastUpdateCounter)
+				{
+					hangCounter = 0;
+					lastUpdateCounter = updateCount;
+				}
+				else
+				{
+					if (++hangCounter >= m_maxHangSeconds)
+					{
+						fmt::print(fg(fmt::color::red), "main loop has been unresponsive for {} seconds, exiting...", hangCounter);
+						std::abort();
+						break; //< just in case
+					}
+				}
+			}
+
+			// Normal exit
+		});
+	}
+
+	HealthCheckerAppComponent::~HealthCheckerAppComponent()
+	{
+		m_exitChecker.release();
+		m_healthCheckThread.join();
+	}
+
+	void HealthCheckerAppComponent::Update(Nz::Time elapsedTime)
+	{
+		m_updateCounter.fetch_add(1, std::memory_order_relaxed);
+	}
+}

src/Server/ServerConfigAppComponent.cpp

@@ -15,6 +15,7 @@ namespace tsom
 		RegisterStringOption("Api.Url");
 		RegisterStringOption("ConnectionToken.EncryptionKey", "");
 		RegisterIntegerOption("Server.Port", 1, 0xFFFF, 29536);
+		RegisterIntegerOption("Server.MaxStuckSeconds", 0, 60, 10);
 		RegisterBoolOption("Server.SleepWhenEmpty", true);
 		RegisterStringOption("Save.Directory", "saves/chunks");
 		RegisterIntegerOption("Save.Interval", 0, 60 * 60, 30);

src/Server/main.cpp

@@ -2,6 +2,7 @@
 // This file is part of the "This Space Of Mine" project
 // For conditions of distribution and use, see copyright notice in LICENSE
 
+#include <CommonLib/HealthCheckerAppComponent.hpp>
 #include <CommonLib/InternalConstants.hpp>
 #include <Server/ServerConfigAppComponent.hpp>
 #include <ServerLib/PlayerTokenAppComponent.hpp>
@@ -43,6 +44,9 @@ int ServerMain(int argc, char* argv[])
 
 	auto& config = configAppComponent.GetConfig();
 
+	if (Nz::UInt32 maxStuckTime = config.GetIntegerValue<Nz::UInt32>("Server.MaxStuckSeconds"))
+		app.AddComponent<tsom::HealthCheckerAppComponent>(maxStuckTime);
+
 	Nz::UInt16 serverPort = config.GetIntegerValue<Nz::UInt16>("Server.Port");
 	std::filesystem::path saveDirectory = Nz::Utf8Path(config.GetStringValue("Save.Directory"));