Changed how sponsor processes are managed

We can now monitor more than one process and if a new process is started on the same port we will hand over the owner pid to the process which is already executing before exiting. Note that this is only done if there is actually already an owner process in the instance list.

1 files changed, 66 insertions, 21 deletions

diff --git a/zenserver/zenserver.cpp b/zenserver/zenserver.cpp
index 3b56d8683..aa4a42fd7 100644
--- a/zenserver/zenserver.cpp
+++ b/zenserver/zenserver.cpp
@@ -84,9 +84,12 @@
 
 class ZenServer
 {
+	ZenServerState::ZenServerEntry* m_ServerEntry = nullptr;
+
 public:
-	void Initialize(ZenServiceConfig& ServiceConfig, int BasePort, int ParentPid)
+	void Initialize(ZenServiceConfig& ServiceConfig, int BasePort, int ParentPid, ZenServerState::ZenServerEntry* ServerEntry)
 	{
+		m_ServerEntry = ServerEntry;
 		using namespace fmt::literals;
 		ZEN_INFO(ZEN_APP_NAME " initializing");
 
@@ -94,23 +97,29 @@ public:
 
 		if (ParentPid)
 		{
-			m_Process.Initialize(ParentPid);
+			zen::ProcessHandle OwnerProcess;
+			OwnerProcess.Initialize(ParentPid);
 
-			if (!m_Process.IsValid())
+			if (!OwnerProcess.IsValid())
 			{
 				ZEN_WARN("Unable to initialize process handle for specified parent pid #{}", ParentPid);
+
+				// If the pid is not reachable should we just shut down immediately? the intended owner process
+				// could have been killed or somehow crashed already
 			}
 			else
 			{
 				ZEN_INFO("Using parent pid #{} to control process lifetime", ParentPid);
 			}
+
+			m_ProcessMonitor.AddPid(ParentPid);
 		}
 
 		// Initialize/check mutex based on base port
 
 		std::string MutexName = "zen_{}"_format(BasePort);
 
-		if (zen::NamedMutex::Exists(MutexName) || (m_ServerMutex.Create(MutexName) == false))
+		if (zen::NamedMutex::Exists(MutexName) || ((m_ServerMutex.Create(MutexName) == false)))
 		{
 			throw std::runtime_error("Failed to create mutex '{}' - is another instance already running?"_format(MutexName).c_str());
 		}
@@ -267,7 +276,7 @@ public:
 
 	void Run()
 	{
-		if (m_Process.IsValid())
+		if (m_ProcessMonitor.IsActive())
 		{
 			EnqueueTimer();
 		}
@@ -282,7 +291,7 @@ public:
 			ZEN_INFO("        \\/   \\/     \\/          \\/                         \\/ ");
 		}
 
-		ZEN_INFO(ZEN_APP_NAME " now running");
+		ZEN_INFO(ZEN_APP_NAME " now running (pid: {})", zen::GetCurrentProcessId());
 
 #if USE_SENTRY
 		sentry_clear_modulecache();
@@ -332,13 +341,33 @@ public:
 
 	void CheckOwnerPid()
 	{
-		if (m_Process.IsRunning())
+		// Pick up any new "owner" processes
+
+		std::set<uint32_t> AddedPids;
+
+		for (auto& PidEntry : m_ServerEntry->SponsorPids)
+		{
+			if (uint32_t ThisPid = PidEntry.load(std::memory_order::memory_order_relaxed))
+			{
+				if (PidEntry.compare_exchange_strong(ThisPid, 0))
+				{
+					if (AddedPids.insert(ThisPid).second)
+					{
+						m_ProcessMonitor.AddPid(ThisPid);
+
+						ZEN_INFO("added process with pid #{} as a sponsor process", ThisPid);
+					}
+				}
+			}
+		}
+
+		if (m_ProcessMonitor.IsRunning())
 		{
 			EnqueueTimer();
 		}
 		else
 		{
-			ZEN_INFO(ZEN_APP_NAME " exiting since parent process id {} is gone", m_Process.Pid());
+			ZEN_INFO(ZEN_APP_NAME " exiting since sponsor processes are all gone");
 
 			RequestExit(0);
 		}
@@ -366,7 +395,7 @@ private:
 	std::jthread		  m_IoRunner;
 	asio::io_context	  m_IoContext;
 	asio::steady_timer	  m_PidCheckTimer{m_IoContext};
-	zen::ProcessHandle	  m_Process;
+	zen::ProcessMonitor	  m_ProcessMonitor;
 	zen::NamedMutex		  m_ServerMutex;
 
 	zen::Ref<zen::HttpServer>						 m_Http;
@@ -411,27 +440,41 @@ main(int argc, char* argv[])
 	auto _ = zen::MakeGuard([&] { sentry_close(); });
 #endif
 
-	// Prototype config system, let's see how this pans out
+	try
+	{
+		// Prototype config system, we'll see how this pans out
+		//
+		// TODO: we need to report any parse errors here
 
-	ParseServiceConfig(GlobalOptions.DataDir, /* out */ ServiceConfig);
+		ParseServiceConfig(GlobalOptions.DataDir, /* out */ ServiceConfig);
 
-	ZEN_INFO("zen cache server starting on port {}", GlobalOptions.BasePort);
+		ZEN_INFO("zen cache server starting on port {}", GlobalOptions.BasePort);
 
-	try
-	{
 		ZenServerState ServerState;
 		ServerState.Initialize();
 		ServerState.Sweep();
 
-		if (ZenServerState::ZenServerEntry* Entry = ServerState.Lookup(GlobalOptions.BasePort))
+		ZenServerState::ZenServerEntry* Entry = ServerState.Lookup(GlobalOptions.BasePort);
+
+		if (Entry)
 		{
 			// Instance already running for this port? Should double check pid
 
 			ZEN_WARN("Looks like there is already a process listening to this port (pid: {})", Entry->Pid);
+
+			if (GlobalOptions.OwnerPid)
+			{
+				Entry->AddSponsorProcess(GlobalOptions.OwnerPid);
+
+				std::exit(0);
+			}
 		}
-		else
+
+		Entry = ServerState.Register(GlobalOptions.BasePort);
+
+		if (GlobalOptions.OwnerPid)
 		{
-			ServerState.Register(GlobalOptions.BasePort);
+			Entry->AddSponsorProcess(GlobalOptions.OwnerPid);
 		}
 
 		std::unique_ptr<std::thread>	 ShutdownThread;
@@ -445,15 +488,17 @@ main(int argc, char* argv[])
 		Server.SetDataRoot(GlobalOptions.DataDir);
 		Server.SetTestMode(GlobalOptions.IsTest);
 		Server.SetDedicatedMode(GlobalOptions.IsDedicated);
-		Server.Initialize(ServiceConfig, GlobalOptions.BasePort, GlobalOptions.OwnerPid);
+		Server.Initialize(ServiceConfig, GlobalOptions.BasePort, GlobalOptions.OwnerPid, Entry);
 
 		// Monitor shutdown signals
 
 		ShutdownThread.reset(new std::thread{[&] {
 			ZEN_INFO("shutdown monitor thread waiting for shutdown signal '{}'", ShutdownEventName);
-			ShutdownEvent->Wait();
-			ZEN_INFO("shutdown signal received");
-			Server.RequestExit(0);
+			if (ShutdownEvent->Wait())
+			{
+				ZEN_INFO("shutdown signal received");
+				Server.RequestExit(0);
+			}
 		}});
 
 		// If we have a parent process, establish the mechanisms we need