Host and service check specific timeouts

NEWS

@@ -1,3 +1,10 @@
+Next:
+Breaking changes:
+* Due to object structure changes, it is recommended to recompile any NEB modules
+
+Features:
+* new attribute 'check_timeout' on `host` and `service` structs, overrides the global host / service check timeouts. (#525)
+
 1.5.0 - Feb 03 2026
 ===================
 Breaking changes:

src/naemon/checks.h

@@ -63,6 +63,7 @@ typedef struct check_result {
 	double latency;
 	struct timeval start_time;			/* time the service check was initiated */
 	struct timeval finish_time;			/* time the service check was completed */
+	int timeout;						/* timeout used in this check */
 	int early_timeout;                              /* did the service check timeout? */
 	int exited_ok;					/* did the plugin check return okay? */
 	int return_code;				/* plugin return code */

src/naemon/checks_host.c

@@ -281,7 +281,7 @@ static int run_async_host_check(host *hst, int check_options, double latency)
 	end_time.tv_usec = 0L;
 	tv_set(&hst->last_update);
 
-	neb_result = broker_host_check(NEBTYPE_HOSTCHECK_ASYNC_PRECHECK, NEBFLAG_NONE, NEBATTR_NONE, hst, CHECK_TYPE_ACTIVE, hst->current_state, hst->state_type, start_time, end_time, hst->check_command, hst->latency, 0.0, host_check_timeout, FALSE, 0, NULL, NULL, NULL, NULL, NULL);
+	neb_result = broker_host_check(NEBTYPE_HOSTCHECK_ASYNC_PRECHECK, NEBFLAG_NONE, NEBATTR_NONE, hst, CHECK_TYPE_ACTIVE, hst->current_state, hst->state_type, start_time, end_time, hst->check_command, hst->latency, 0.0, hst->check_timeout, FALSE, 0, NULL, NULL, NULL, NULL, NULL);
 
 	if (neb_result == NEBERROR_CALLBACKCANCEL || neb_result == NEBERROR_CALLBACKOVERRIDE) {
 		log_debug_info(DEBUGL_CHECKS, 0, "Check of host '%s' (id=%u) was %s by a module\n",
@@ -338,12 +338,13 @@ static int run_async_host_check(host *hst, int check_options, double latency)
 	cr->latency = latency;
 	cr->start_time = start_time;
 	cr->finish_time = start_time;
+	cr->timeout = hst->check_timeout;
 	cr->early_timeout = FALSE;
 	cr->exited_ok = TRUE;
 	cr->return_code = STATE_OK;
 	cr->output = NULL;
 
-	neb_result = broker_host_check(NEBTYPE_HOSTCHECK_INITIATE, NEBFLAG_NONE, NEBATTR_NONE, hst, CHECK_TYPE_ACTIVE, hst->current_state, hst->state_type, start_time, end_time, hst->check_command, hst->latency, 0.0, host_check_timeout, FALSE, 0, processed_command, NULL, NULL, NULL, cr);
+	neb_result = broker_host_check(NEBTYPE_HOSTCHECK_INITIATE, NEBFLAG_NONE, NEBATTR_NONE, hst, CHECK_TYPE_ACTIVE, hst->current_state, hst->state_type, start_time, end_time, hst->check_command, hst->latency, 0.0, hst->check_timeout, FALSE, 0, processed_command, NULL, NULL, NULL, cr);
 
 	/* neb module wants to override the service check - perhaps it will check the service itself */
 	if (neb_result == NEBERROR_CALLBACKOVERRIDE || neb_result == NEBERROR_CALLBACKCANCEL) {
@@ -354,7 +355,7 @@ static int run_async_host_check(host *hst, int check_options, double latency)
 		return neb_result == NEBERROR_CALLBACKOVERRIDE ? OK : ERROR;
 	}
 
-	runchk_result = wproc_run_callback(processed_command, host_check_timeout, handle_worker_host_check, (void *)cr, &mac);
+	runchk_result = wproc_run_callback(processed_command, hst->check_timeout, handle_worker_host_check, (void *)cr, &mac);
 	if (runchk_result == ERROR) {
 		nm_log(NSLOG_RUNTIME_ERROR,
 		       "Unable to send check for host '%s' to worker (ret=%d)\n", hst->name, runchk_result);
@@ -623,7 +624,7 @@ int handle_async_host_check_result(host *temp_host, check_result *cr)
 	    temp_host->check_command,
 	    temp_host->latency,
 	    temp_host->execution_time,
-	    host_check_timeout,
+	    cr->timeout,
 	    cr->early_timeout,
 	    cr->return_code,
 	    NULL,
@@ -1178,7 +1179,7 @@ static void check_for_orphaned_hosts_eventhandler(struct nm_event_execution_prop
 				continue;
 
 			/* determine the time at which the check results should have come in (allow 10 minutes slack time) */
-			expected_time = (time_t)(temp_host->next_check + temp_host->latency + host_check_timeout + check_reaper_interval + 600);
+			expected_time = (time_t)(temp_host->next_check + temp_host->latency + temp_host->check_timeout + check_reaper_interval + 600);
 
 			/* this host was supposed to have executed a while ago, but for some reason the results haven't come back in... */
 			if (expected_time < current_time.tv_sec) {

src/naemon/checks_service.c

@@ -402,14 +402,15 @@ static int run_scheduled_service_check(service *svc, int check_options, double l
 	cr->latency = latency;
 	cr->start_time = start_time;
 	cr->finish_time = start_time;
+	cr->timeout = svc->check_timeout;
 	cr->early_timeout = FALSE;
 	cr->exited_ok = TRUE;
 	cr->return_code = STATE_OK;
 	cr->output = NULL;
 	cr->host_name = nm_strdup(svc->host_name);
 	cr->service_description = nm_strdup(svc->description);
 
-	neb_result = broker_service_check(NEBTYPE_SERVICECHECK_INITIATE, NEBFLAG_NONE, NEBATTR_NONE, svc, CHECK_TYPE_ACTIVE, start_time, end_time, svc->check_command, svc->latency, 0.0, service_check_timeout, FALSE, 0, processed_command, cr);
+	neb_result = broker_service_check(NEBTYPE_SERVICECHECK_INITIATE, NEBFLAG_NONE, NEBATTR_NONE, svc, CHECK_TYPE_ACTIVE, start_time, end_time, svc->check_command, svc->latency, 0.0, svc->check_timeout, FALSE, 0, processed_command, cr);
 
 	/* neb module wants to override the service check - perhaps it will check the service itself */
 	if (neb_result == NEBERROR_CALLBACKOVERRIDE || neb_result == NEBERROR_CALLBACKCANCEL) {
@@ -421,7 +422,7 @@ static int run_scheduled_service_check(service *svc, int check_options, double l
 	}
 
 	/* paw off the check to a worker to run */
-	runchk_result = wproc_run_callback(processed_command, service_check_timeout, handle_worker_service_check, (void *)cr, &mac);
+	runchk_result = wproc_run_callback(processed_command, svc->check_timeout, handle_worker_service_check, (void *)cr, &mac);
 	if (runchk_result == ERROR) {
 		nm_log(NSLOG_RUNTIME_ERROR,
 		       "Unable to send check for service '%s' on host '%s' to worker (ret=%d)\n", svc->description, svc->host_name, runchk_result);
@@ -1125,7 +1126,7 @@ int handle_async_service_check_result(service *temp_service, check_result *queue
 	    temp_service->check_command,
 	    temp_service->latency,
 	    temp_service->execution_time,
-	    service_check_timeout,
+	    queued_check_result->timeout,
 	    queued_check_result->early_timeout,
 	    queued_check_result->return_code,
 	    NULL,
@@ -1186,7 +1187,7 @@ static void check_for_orphaned_services_eventhandler(struct nm_event_execution_p
 				continue;
 
 			/* determine the time at which the check results should have come in (allow 10 minutes slack time) */
-			expected_time = (time_t)(temp_service->next_check + temp_service->latency + service_check_timeout + check_reaper_interval + 600);
+			expected_time = (time_t)(temp_service->next_check + temp_service->latency + temp_service->check_timeout + check_reaper_interval + 600);
 
 			/* this service was supposed to have executed a while ago, but for some reason the results haven't come back in... */
 			if (expected_time < current_time.tv_sec) {

src/naemon/objects_host.c

@@ -81,7 +81,7 @@ host *create_host(const char *name)
 	return new_host;
 }
 
-int setup_host_variables(host *new_host, const char *display_name, const char *alias, const char *address, const char *check_period, int initial_state, double check_interval, double retry_interval, int max_attempts, int notification_options, double notification_interval, double first_notification_delay, const char *notification_period, int notifications_enabled, const char *check_command, int checks_enabled, int accept_passive_checks, const char *event_handler, int event_handler_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, const char *vrml_image, const char *statusmap_image, int x_2d, int y_2d, int have_2d_coords, double x_3d, double y_3d, double z_3d, int have_3d_coords, int retain_status_information, int retain_nonstatus_information, int obsess, unsigned int hourly_value)
+int setup_host_variables(host *new_host, const char *display_name, const char *alias, const char *address, const char *check_period, int initial_state, int check_timeout, double check_interval, double retry_interval, int max_attempts, int notification_options, double notification_interval, double first_notification_delay, const char *notification_period, int notifications_enabled, const char *check_command, int checks_enabled, int accept_passive_checks, const char *event_handler, int event_handler_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, const char *vrml_image, const char *statusmap_image, int x_2d, int y_2d, int have_2d_coords, double x_3d, double y_3d, double z_3d, int have_3d_coords, int retain_status_information, int retain_nonstatus_information, int obsess, unsigned int hourly_value)
 {
 	timeperiod *check_tp = NULL, *notify_tp = NULL;
 
@@ -104,6 +104,10 @@ int setup_host_variables(host *new_host, const char *display_name, const char *a
 		nm_log(NSLOG_CONFIG_ERROR, "Error: Invalid check_interval value for host '%s'\n", new_host->name);
 		return -1;
 	}
+	if (check_timeout <= 0) {
+		nm_log(NSLOG_CONFIG_ERROR, "Error: check_timeout must be a positive integer host '%s'\n", new_host->name);
+		return -1;
+	}
 	if (notification_interval < 0) {
 		nm_log(NSLOG_CONFIG_ERROR, "Error: Invalid notification_interval value for host '%s'\n", new_host->name);
 		return -1;
@@ -157,6 +161,7 @@ int setup_host_variables(host *new_host, const char *display_name, const char *a
 	/* duplicate non-string vars */
 	new_host->hourly_value = hourly_value;
 	new_host->max_attempts = max_attempts;
+	new_host->check_timeout = check_timeout;
 	new_host->check_interval = check_interval;
 	new_host->retry_interval = retry_interval;
 	new_host->notification_interval = notification_interval;
@@ -545,6 +550,7 @@ void fcache_host(FILE *fp, const host *temp_host)
 		fprintf(fp, "u\n");
 	else
 		fprintf(fp, "o\n");
+	fprintf(fp,"\tcheck_timeout\t%d\n", temp_host->check_timeout);
 	fprintf(fp, "\thourly_value\t%u\n", temp_host->hourly_value);
 	fprintf(fp, "\tcheck_interval\t%f\n", temp_host->check_interval);
 	fprintf(fp, "\tretry_interval\t%f\n", temp_host->retry_interval);

src/naemon/objects_host.h

@@ -33,6 +33,7 @@ struct host {
 	struct servicesmember *services;
 	char    *check_command;
 	int     initial_state;
+	int		check_timeout;
 	double  check_interval;
 	double  retry_interval;
 	int     max_attempts;
@@ -153,7 +154,7 @@ void destroy_objects_host(void);
 
 host *create_host(const char *name);
 /** This is an evil legacy function which you should never, ever use */
-int setup_host_variables(host *new_host, const char *display_name, const char *alias, const char *address, const char *check_period, int initial_state, double check_interval, double retry_interval, int max_attempts, int notification_options, double notification_interval, double first_notification_delay, const char *notification_period, int notifications_enabled, const char *check_command, int checks_enabled, int accept_passive_checks, const char *event_handler, int event_handler_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, const char *vrml_image, const char *statusmap_image, int x_2d, int y_2d, int have_2d_coords, double x_3d, double y_3d, double z_3d, int have_3d_coords, int retain_status_information, int retain_nonstatus_information, int obsess_over_host, unsigned int hourly_value);
+int setup_host_variables(host *new_host, const char *display_name, const char *alias, const char *address, const char *check_period, int initial_state, int check_timeout, double check_interval, double retry_interval, int max_attempts, int notification_options, double notification_interval, double first_notification_delay, const char *notification_period, int notifications_enabled, const char *check_command, int checks_enabled, int accept_passive_checks, const char *event_handler, int event_handler_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, const char *vrml_image, const char *statusmap_image, int x_2d, int y_2d, int have_2d_coords, double x_3d, double y_3d, double z_3d, int have_3d_coords, int retain_status_information, int retain_nonstatus_information, int obsess_over_host, unsigned int hourly_value);
 int register_host(host *new_host);
 void destroy_host(host *this_host);
 

src/naemon/objects_service.c

@@ -88,7 +88,7 @@ service *create_service(host *hst, const char *description)
 	return new_service;
 }
 
-int setup_service_variables(service *new_service, const char *display_name, const char *check_command, const char *check_period, int initial_state, int max_attempts, int accept_passive_checks, double check_interval, double retry_interval, double notification_interval, double first_notification_delay, char *notification_period, int notification_options, int notifications_enabled, int is_volatile, const char *event_handler, int event_handler_enabled, int checks_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, int retain_status_information, int retain_nonstatus_information, int obsess, unsigned int hourly_value)
+int setup_service_variables(service *new_service, const char *display_name, const char *check_command, const char *check_period, int initial_state, int check_timeout, int max_attempts, int accept_passive_checks, double check_interval, double retry_interval, double notification_interval, double first_notification_delay, char *notification_period, int notification_options, int notifications_enabled, int is_volatile, const char *event_handler, int event_handler_enabled, int checks_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, int retain_status_information, int retain_nonstatus_information, int obsess, unsigned int hourly_value)
 {
 	timeperiod *cp = NULL, *np = NULL;
 	command *cmd;
@@ -123,6 +123,10 @@ int setup_service_variables(service *new_service, const char *display_name, cons
 		nm_log(NSLOG_CONFIG_ERROR, "Error: check_interval must be a non-negative integer for service '%s' on host '%s'\n", new_service->description, new_service->host_name);
 		return -1;
 	}
+	if (check_timeout <= 0) {
+		nm_log(NSLOG_CONFIG_ERROR, "Error: check_timeout must be a positive integer for service '%s' on host '%s'\n", new_service->description, new_service->host_name);
+		return -1;
+	}
 	if (retry_interval <= 0) {
 		nm_log(NSLOG_CONFIG_ERROR, "Error: retry_interval must be a positive integer for service '%s' on host '%s'\n", new_service->description, new_service->host_name);
 		return -1;
@@ -171,6 +175,7 @@ int setup_service_variables(service *new_service, const char *display_name, cons
 	}
 
 	new_service->hourly_value = hourly_value;
+	new_service->check_timeout = check_timeout;
 	new_service->check_interval = check_interval;
 	new_service->retry_interval = retry_interval;
 	new_service->max_attempts = max_attempts;
@@ -484,6 +489,7 @@ void fcache_service(FILE *fp, const service *temp_service)
 		fprintf(fp, "c\n");
 	else
 		fprintf(fp, "o\n");
+	fprintf(fp,"\tcheck_timeout\t%d\n", temp_service->check_timeout);
 	fprintf(fp, "\thourly_value\t%u\n", temp_service->hourly_value);
 	fprintf(fp, "\tcheck_interval\t%f\n", temp_service->check_interval);
 	fprintf(fp, "\tretry_interval\t%f\n", temp_service->retry_interval);

src/naemon/objects_service.h

@@ -32,6 +32,7 @@ struct service {
 	char    *check_command;
 	char    *event_handler;
 	int     initial_state;
+	int     check_timeout;
 	double	check_interval;
 	double  retry_interval;
 	int	max_attempts;
@@ -152,7 +153,7 @@ int init_objects_service(int elems);
 void destroy_objects_service(int truncate_lists);
 
 service *create_service(host *hst, const char *description);
-int setup_service_variables(service *svc, const char *display_name, const char *check_period, const char *check_command, int initial_state, int max_attempts, int accept_passive_checks, double check_interval, double retry_interval, double notification_interval, double first_notification_delay, char *notification_period, int notification_options, int notifications_enabled, int is_volatile, const char *event_handler, int event_handler_enabled, int checks_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, int retain_status_information, int retain_nonstatus_information, int obsess, unsigned int hourly_value);
+int setup_service_variables(service *svc, const char *display_name, const char *check_period, const char *check_command, int initial_state, int check_timeout, int max_attempts, int accept_passive_checks, double check_interval, double retry_interval, double notification_interval, double first_notification_delay, char *notification_period, int notification_options, int notifications_enabled, int is_volatile, const char *event_handler, int event_handler_enabled, int checks_enabled, int flap_detection_enabled, double low_flap_threshold, double high_flap_threshold, int flap_detection_options, int stalking_options, int process_perfdata, int check_freshness, int freshness_threshold, const char *notes, const char *notes_url, const char *action_url, const char *icon_image, const char *icon_image_alt, int retain_status_information, int retain_nonstatus_information, int obsess, unsigned int hourly_value);
 int register_service(service *new_service);
 void destroy_service(service *svc, int truncate_lists);