launch command worker earlier

features/steps/naemon.py

@@ -145,7 +145,7 @@ def naemon_started_and_ready(context, timeout_s):
             # When we see this line in the log, we'll wait 1 more second and
             # then Naemon should be ready, with signal handlers setup so that a
             # test can SIGTERM it.
-            if 'Successfully launched command file worker with pid' in log:
+            if 'Naemon successfully initialized' in log:
                 ready = True
                 time.sleep(1)
                 break

src/naemon/commands.c

@@ -19,6 +19,7 @@
 #include "globals.h"
 #include "logging.h"
 #include "nm_alloc.h"
+#include "query-handler.h"
 #include "lib/libnaemon.h"
 #include <string.h>
 #include <sys/types.h>
@@ -388,6 +389,13 @@ int launch_command_file_worker(void)
 	/* make our own process-group so we can be traced into and stuff */
 	setpgid(0, 0);
 
+
+	// close inherited file handles
+	close_log_file();
+	close_standard_fds();
+	qh_close_socket();
+	close_lockfile_fd();
+
 	str = nm_strdup(command_file);
 	free_memory(get_global_macros());
 	command_file = str;

src/naemon/naemon.c

@@ -556,6 +556,33 @@ int main(int argc, char **argv)
 		nerd_init();
 		timing_point("Initialized NERD\n");
 
+		/*
+		 * the queue has to be initialized before loading the neb modules
+		 * to give them the chance to register user events.
+		 * (initializing event queue requires number of objects, so do
+		 * this after parsing the objects)
+		 */
+		timing_point("Initializing Event queue\n");
+		init_event_queue();
+		timing_point("Initialized Event queue\n");
+
+		registered_commands_init(200);
+		register_core_commands();
+		/* fire up command file worker */
+		timing_point("Launching command file worker\n");
+		launch_command_file_worker();
+		timing_point("Launched command file worker\n");
+
+		/* read in all object config data after launching the command worker
+		 * to keep the memory footprint of the command worker smaller
+		 */
+		if (result == OK) {
+			nm_log(NSLOG_INFO_MESSAGE, "Reading all config object data\n");
+			timing_point("Reading all object data\n");
+			result = read_all_object_data(config_file);
+			timing_point("Read all object data\n");
+		}
+
 		/* initialize check workers */
 		timing_point("Spawning %u workers\n", wproc_num_workers_spawned);
 		if (init_workers(num_check_workers) < 0) {
@@ -572,31 +599,14 @@ int main(int argc, char **argv)
 		}
 		timing_point("Connected %u workers\n", wproc_num_workers_online);
 
-		/* read in all object config data */
-		if (result == OK) {
-			nm_log(NSLOG_INFO_MESSAGE, "Reading all config object data\n");
-			timing_point("Reading all object data\n");
-			result = read_all_object_data(config_file);
-			timing_point("Read all object data\n");
-		}
-
-		/*
-		 * the queue has to be initialized before loading the neb modules
-		 * to give them the chance to register user events.
-		 * (initializing event queue requires number of objects, so do
-		 * this after parsing the objects)
-		 */
-		timing_point("Initializing Event queue\n");
-		init_event_queue();
-		timing_point("Initialized Event queue\n");
-
 		/* load modules */
 		nm_log(NSLOG_INFO_MESSAGE, "Loading neb modules\n");
 		timing_point("Loading modules\n");
 		if (neb_load_all_modules() != OK) {
 			nm_log(NSLOG_CONFIG_ERROR, "Error: Module loading failed. Aborting.\n");
 			/* give already loaded modules a chance to deinitialize */
 			neb_unload_all_modules(NEBMODULE_FORCE_UNLOAD, NEBMODULE_NEB_SHUTDOWN);
+			shutdown_command_file_worker();
 			exit(EXIT_FAILURE);
 		}
 		timing_point("Loaded modules\n");
@@ -634,6 +644,7 @@ int main(int argc, char **argv)
 			broker_program_state(NEBTYPE_PROCESS_SHUTDOWN, NEBFLAG_PROCESS_INITIATED, NEBATTR_SHUTDOWN_ABNORMAL);
 
 			cleanup();
+			shutdown_command_file_worker();
 			exit(ERROR);
 		}
 
@@ -695,13 +706,6 @@ int main(int argc, char **argv)
 		log_service_states(INITIAL_STATES, NULL);
 		timing_point("Logged initial states\n");
 
-		registered_commands_init(200);
-		register_core_commands();
-		/* fire up command file worker */
-		timing_point("Launching command file worker\n");
-		launch_command_file_worker();
-		timing_point("Launched command file worker\n");
-
 		broker_program_state(NEBTYPE_PROCESS_EVENTLOOPSTART, NEBFLAG_NONE, NEBATTR_NONE);
 
 		/* get event start time and save as macro */
@@ -718,6 +722,8 @@ int main(int argc, char **argv)
 			}
 		}
 
+		nm_log(NSLOG_INFO_MESSAGE, "Naemon successfully initialized (PID=%d)\n", (int)getpid());
+
 		timing_point("Entering event execution loop\n");
 		/***** start monitoring all services *****/
 		/* (doesn't return until a restart or shutdown signal is encountered) */

src/naemon/query-handler.c

@@ -394,7 +394,7 @@ int qh_init(const char *path)
 	result = iobroker_register(nagios_iobs, qh_listen_sock, NULL, qh_registration_input);
 	if (result < 0) {
 		g_hash_table_destroy(qh_table);
-		close(qh_listen_sock);
+		qh_close_socket();
 		nm_log(NSLOG_RUNTIME_ERROR, "qh: Failed to register socket with io broker: %s\n", iobroker_strerror(result));
 		return ERROR;
 	}
@@ -408,3 +408,9 @@ int qh_init(const char *path)
 
 	return 0;
 }
+
+void qh_close_socket() {
+	if( qh_listen_sock > 0 )
+		close(qh_listen_sock);
+	qh_listen_sock = -1;
+}

src/naemon/query-handler.h

@@ -20,6 +20,7 @@ int qh_init(const char *path);
 void qh_deinit(const char *path);
 int qh_register_handler(const char *name, const char *description, unsigned int options, qh_handler handler);
 const char *qh_strerror(int code);
+void qh_close_socket(void);
 
 NAGIOS_END_DECL
 

src/naemon/utils.c

@@ -178,6 +178,8 @@ int host_skip_check_dependency_status = DEFAULT_SKIP_CHECK_STATUS;
 
 static long long check_file_size(char *, unsigned long, struct rlimit);
 
+static int lock_file_fd = -1; /* the file handle of the lockfile */
+
 time_t max_check_result_file_age = DEFAULT_MAX_CHECK_RESULT_AGE;
 
 check_stats     check_statistics[MAX_CHECK_STATS_TYPES];
@@ -504,7 +506,6 @@ int signal_parent(int sig)
 int daemon_init(void)
 {
 	int pid = 0;
-	int lockfile = 0;
 	int val = 0;
 	char buf[256];
 	struct flock lock;
@@ -515,16 +516,16 @@ int daemon_init(void)
 
 	umask(S_IWGRP | S_IWOTH);
 
-	lockfile = open(lock_file, O_RDWR | O_CREAT, S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH);
+	lock_file_fd = open(lock_file, O_RDWR | O_CREAT, S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH);
 
-	if (lockfile < 0) {
+	if (lock_file_fd < 0) {
 		nm_log(NSLOG_RUNTIME_ERROR, "Failed to obtain lock on file %s: %s\n", lock_file, strerror(errno));
 		nm_log(NSLOG_PROCESS_INFO | NSLOG_RUNTIME_ERROR, "Bailing out due to errors encountered while attempting to daemonize... (PID=%d)", (int)getpid());
 		return (ERROR);
 	}
 
 	/* see if we can read the contents of the lockfile */
-	if ((val = read(lockfile, buf, (size_t)10)) < 0) {
+	if ((val = read(lock_file_fd, buf, (size_t)10)) < 0) {
 		nm_log(NSLOG_RUNTIME_ERROR, "Lockfile exists but cannot be read");
 		return (ERROR);
 	}
@@ -546,7 +547,7 @@ int daemon_init(void)
 		lock.l_start = 0;
 		lock.l_whence = SEEK_SET;
 		lock.l_len = 0;
-		if (fcntl(lockfile, F_GETLK, &lock) == -1) {
+		if (fcntl(lock_file_fd, F_GETLK, &lock) == -1) {
 			nm_log(NSLOG_RUNTIME_ERROR, "Failed to access lockfile '%s'. %s. Bailing out...", lock_file, strerror(errno));
 			return (ERROR);
 		}
@@ -615,9 +616,9 @@ int daemon_init(void)
 	lock.l_whence = SEEK_SET;
 	lock.l_len = 0;
 	lock.l_pid = getpid();
-	if (fcntl(lockfile, F_SETLK, &lock) == -1) {
+	if (fcntl(lock_file_fd, F_SETLK, &lock) == -1) {
 		if (errno == EACCES || errno == EAGAIN) {
-			fcntl(lockfile, F_GETLK, &lock);
+			fcntl(lock_file_fd, F_GETLK, &lock);
 			nm_log(NSLOG_RUNTIME_ERROR, "Lockfile '%s' looks like its already held by another instance of Naemon (PID %d).  Bailing out, post-fork...", lock_file, (int)lock.l_pid);
 		} else
 			nm_log(NSLOG_RUNTIME_ERROR, "Cannot lock lockfile '%s': %s. Bailing out...", lock_file, strerror(errno));
@@ -626,28 +627,34 @@ int daemon_init(void)
 	}
 
 	/* write PID to lockfile... */
-	lseek(lockfile, 0, SEEK_SET);
-	if (ftruncate(lockfile, 0) != 0) {
+	lseek(lock_file_fd, 0, SEEK_SET);
+	if (ftruncate(lock_file_fd, 0) != 0) {
 		nm_log(NSLOG_RUNTIME_ERROR, "Cannot truncate lockfile '%s': %s. Bailing out...", lock_file, strerror(errno));
 		return (ERROR);
 	}
 	sprintf(buf, "%d\n", (int)getpid());
 
-	if (nsock_write_all(lockfile, buf, strlen(buf)) != 0) {
+	if (nsock_write_all(lock_file_fd, buf, strlen(buf)) != 0) {
 		nm_log(NSLOG_RUNTIME_ERROR, "Cannot write PID to lockfile '%s': %s. Bailing out...", lock_file, strerror(errno));
 		return (ERROR);
 	}
 
 	/* make sure lock file stays open while program is executing... */
-	val = fcntl(lockfile, F_GETFD, 0);
+	val = fcntl(lock_file_fd, F_GETFD, 0);
 	val |= FD_CLOEXEC;
-	fcntl(lockfile, F_SETFD, val);
+	fcntl(lock_file_fd, F_SETFD, val);
 
 	broker_program_state(NEBTYPE_PROCESS_DAEMONIZE, NEBFLAG_NONE, NEBATTR_NONE);
 
 	return OK;
 }
 
+void close_lockfile_fd() {
+	if(lock_file_fd > 0)
+		close(lock_file_fd);
+	lock_file_fd = -1;
+}
+
 /******************************************************************/
 /************************* FILE FUNCTIONS *************************/
 /******************************************************************/

src/naemon/utils.h

@@ -37,6 +37,7 @@ void signal_react(void);				/* General signal reaction routines */
 void handle_sigxfsz(void);				/* handle SIGXFSZ */
 int signal_parent(int);					/* signal parent when daemonizing */
 int daemon_init(void);				     		/* switches to daemon mode */
+void close_lockfile_fd(void);			/* close lock_file file handle */
 
 int init_check_stats(void);
 int update_check_stats(int, time_t);