Watchdog now forces all agents' process groups to exit using SIGKILL,…

… to ensure no garbage processes are left behind.

NEWS

@@ -14,6 +14,7 @@ Release 3.9.0
  * [Standalone] Automatic asset pipeline expiry headers support.
  * [Enterprise] Multithreading support for apps
  * Deleting restart.txt will no longer trigger a restart.
+ * The watchdog now forces all agents' process groups to exit using SIGKILL, to ensure no garbage processes are left behind.
 
 
 Release 3.0.17

ext/common/agents/HelperAgent/Main.cpp

@@ -446,6 +446,7 @@ class Server {
 			 * instance directory will be cleaned up, making this helper agent
 			 * inaccessible.
 			 */
+			P_DEBUG("Watchdog seems to be killed; forcing shutdown of all subprocesses");
 			syscalls::killpg(getpgrp(), SIGKILL);
 			_exit(2); // In case killpg() fails.
 		} else {
@@ -486,11 +487,11 @@ main(int argc, char *argv[]) {
 		UPDATE_TRACE_POINT();
 		server.mainLoop();
 	} catch (const tracable_exception &e) {
-		P_ERROR(e.what() << "\n" << e.backtrace());
+		P_ERROR("*** ERROR: " << e.what() << "\n" << e.backtrace());
 		return 1;
 	}
 
 	MultiLibeio::shutdown();
-	P_TRACE(2, "Helper agent exited.");
+	P_TRACE(2, "Helper agent exiting with code 0.");
 	return 0;
 }

ext/common/agents/LoggingAgent/Main.cpp

@@ -283,6 +283,7 @@ main(int argc, char *argv[]) {
 
 		P_DEBUG("Logging agent online, listening at " << socketAddress);
 		ev_run(eventLoop, 0);
+		P_DEBUG("Logging agent exiting with code " << exitCode << ".");
 		return exitCode;
 	} catch (const tracable_exception &e) {
 		P_ERROR("*** ERROR: " << e.what() << "\n" << e.backtrace());

ext/common/agents/Watchdog/Main.cpp

@@ -1,6 +1,6 @@
 /*
  *  Phusion Passenger - http://www.modrails.com/
- *  Copyright (c) 2010 Phusion
+ *  Copyright (c) 2010-2012 Phusion
  *
  *  "Phusion Passenger" is a trademark of Hongli Lai & Ninh Bui.
  *
@@ -982,12 +982,12 @@ cleanupAgentsInBackground(vector<AgentWatcher *> &watchers) {
 			// processes.
 			P_WARN("Some Phusion Passenger agent processes did not exit " <<
 				"in time, forcefully shutting down all.");
-			for (it = watchers.begin(); it != watchers.end(); it++) {
-				(*it)->forceShutdown();
-			}
 		} else {
-			P_DEBUG("All Phusion Passenger agent processes have exited.");
+			P_DEBUG("All Phusion Passenger agent processes have exited. Forcing all subprocesses to shut down.");
 		}
+		for (it = watchers.begin(); it != watchers.end(); it++) {
+				(*it)->forceShutdown();
+			}
 
 		// Now clean up the server instance directory.
 		delete generation.get();
@@ -1126,10 +1126,11 @@ main(int argc, char *argv[]) {
 			 * the background and exit this watchdog process so that we don't block
 			 * the web server.
 			 */
+			P_DEBUG("Web server exited gracefully; gracefully shutting down all agents...");
 			cleanupAgentsInBackground(watchers);
 			return 0;
 		} else {
-			P_DEBUG("Web server did not exit gracefully, forcing shutdown of all service processes...");
+			P_DEBUG("Web server did not exit gracefully, forcing shutdown of all agents...");
 			forceAllAgentsShutdown(watchers);
 			return 1;
 		}

ext/ruby/passenger_native_support.c

@@ -458,6 +458,18 @@ detach_process(VALUE self, VALUE pid) {
 	return Qnil;
 }
 
+/**
+ * Freeze the current process forever. On Ruby 1.9 this never unlocks the GIL.
+ * Useful for testing purposes.
+ */
+static VALUE
+freeze_process(VALUE self) {
+	while (1) {
+		usleep(60 * 1000000);
+	}
+	return Qnil;
+}
+
 #if defined(HAVE_KQUEUE) || defined(IN_DOXYGEN)
 typedef struct {
 	VALUE klass;
@@ -852,6 +864,7 @@ Init_passenger_native_support() {
 	rb_define_singleton_method(mNativeSupport, "writev3", f_writev3, 4);
 	rb_define_singleton_method(mNativeSupport, "process_times", process_times, 0);
 	rb_define_singleton_method(mNativeSupport, "detach_process", detach_process, 1);
+	rb_define_singleton_method(mNativeSupport, "freeze_process", freeze_process, 0);
 
 	#ifdef HAVE_KQUEUE
 		cFileSystemWatcher = rb_define_class_under(mNativeSupport,