add problem timestamps and duration

lib/nsutils.c

@@ -82,7 +82,19 @@ const char *mkstr(const char *fmt, ...)
 	return ret;
 }
 
-
+/* format duration seconds into human readable string */
+const char* duration_string(unsigned long duration) {
+	int days, hours, minutes, seconds;
+
+	days = duration / 86400;
+	duration -= (days * 86400);
+	hours = duration / 3600;
+	duration -= (hours * 3600);
+	minutes = duration / 60;
+	duration -= (minutes * 60);
+	seconds = duration;
+	return (char *)mkstr("%dd %dh %dm %ds", days, hours, minutes, seconds);
+}
 
 /* close and reopen stdin, stdout and stderr to /dev/null */
 void close_standard_fds(void)

lib/nsutils.h

@@ -98,6 +98,14 @@ extern int online_cpus(void);
 extern const char *mkstr(const char *fmt, ...)
 	__attribute__((__format__(__printf__, 1, 2)));
 
+/**
+ * format duration seconds into human readable string.
+ * @note The returned string must *not* be free()'d!
+ * @param[in] duration The duration in seconds
+ * @return A pointer to the formatted string on success. Undefined on errors
+ */
+extern const char *duration_string(unsigned long);
+
 /**
  * Calculate the millisecond delta between two timeval structs
  * @param[in] start The start time

src/naemon/checks_host.c

@@ -1008,14 +1008,18 @@ static int handle_host_state(host *hst, int *alert_recorded)
 		/* update the problem id when transitioning to a problem state */
 		if (hst->last_state == STATE_UP) {
 			/* don't reset last problem id, or it will be zero the next time a problem is encountered */
-			hst->current_problem_id = next_problem_id;
-			next_problem_id++;
+			nm_free(hst->current_problem_id);
+			hst->current_problem_id = (char*)g_uuid_string_random();
+			hst->problem_start = current_time;
+			hst->problem_end = 0L;
 		}
 
 		/* clear the problem id when transitioning from a problem state to an UP state */
 		if (hst->current_state == STATE_UP) {
 			hst->last_problem_id = hst->current_problem_id;
-			hst->current_problem_id = 0L;
+			hst->current_problem_id = NULL;
+			if(hst->problem_start > 0)
+				hst->problem_end = current_time;
 		}
 
 		/* write the host state change to the main log file */

src/naemon/checks_service.c

@@ -686,14 +686,18 @@ int handle_async_service_check_result(service *temp_service, check_result *queue
 		/* update the problem id when transitioning to a problem state */
 		if (temp_service->last_state == STATE_OK) {
 			/* don't reset last problem id, or it will be zero the next time a problem is encountered */
-			temp_service->current_problem_id = next_problem_id;
-			next_problem_id++;
+			nm_free(temp_service->current_problem_id);
+			temp_service->current_problem_id = (char*)g_uuid_string_random();
+			temp_service->problem_start = current_time;
+			temp_service->problem_end = 0L;
 		}
 
 		/* clear the problem id when transitioning from a problem state to an OK state */
 		if (temp_service->current_state == STATE_OK) {
 			temp_service->last_problem_id = temp_service->current_problem_id;
-			temp_service->current_problem_id = 0L;
+			temp_service->current_problem_id = NULL;
+			if(temp_service->problem_start > 0)
+				temp_service->problem_end = current_time;
 		}
 	}
 

src/naemon/globals.h

@@ -158,9 +158,7 @@ extern int currently_running_service_checks;
 extern int currently_running_host_checks;
 
 extern unsigned long next_event_id;
-extern unsigned long next_problem_id;
 extern unsigned long next_comment_id;
-extern unsigned long next_notification_id;
 
 extern unsigned long modified_process_attributes;
 extern unsigned long modified_host_process_attributes;

src/naemon/macros.c

@@ -627,10 +627,6 @@ static int grab_standard_host_macro_r(nagios_macros *mac, int macro_type, host *
 	objectlist *temp_objectlist = NULL;
 	time_t current_time = 0L;
 	unsigned long duration = 0L;
-	int days = 0;
-	int hours = 0;
-	int minutes = 0;
-	int seconds = 0;
 	char *buf1 = NULL;
 	char *buf2 = NULL;
 	int total_host_services = 0;
@@ -708,19 +704,10 @@ static int grab_standard_host_macro_r(nagios_macros *mac, int macro_type, host *
 	case MACRO_HOSTDURATION:
 		time(&current_time);
 		duration = (unsigned long)(current_time - temp_host->last_state_change);
-
 		if (macro_type == MACRO_HOSTDURATIONSEC)
 			*output = (char *)mkstr("%lu", duration);
 		else {
-
-			days = duration / 86400;
-			duration -= (days * 86400);
-			hours = duration / 3600;
-			duration -= (hours * 3600);
-			minutes = duration / 60;
-			duration -= (minutes * 60);
-			seconds = duration;
-			*output = (char *)mkstr("%dd %dh %dm %ds", days, hours, minutes, seconds);
+			*output = (char *)mkstr("%s", duration_string(duration));
 		}
 		break;
 	case MACRO_HOSTEXECUTIONTIME:
@@ -748,7 +735,7 @@ static int grab_standard_host_macro_r(nagios_macros *mac, int macro_type, host *
 		*output = (char *)mkstr("%d", temp_host->current_notification_number);
 		break;
 	case MACRO_HOSTNOTIFICATIONID:
-		*output = (char *)mkstr("%lu", temp_host->current_notification_id);
+		*output = temp_host->current_notification_id;
 		break;
 	case MACRO_HOSTEVENTID:
 		*output = (char *)mkstr("%lu", temp_host->current_event_id);
@@ -757,10 +744,32 @@ static int grab_standard_host_macro_r(nagios_macros *mac, int macro_type, host *
 		*output = (char *)mkstr("%lu", temp_host->last_event_id);
 		break;
 	case MACRO_HOSTPROBLEMID:
-		*output = (char *)mkstr("%lu", temp_host->current_problem_id);
+		if(temp_host->current_problem_id != NULL)
+			*output = temp_host->current_problem_id;
 		break;
 	case MACRO_LASTHOSTPROBLEMID:
-		*output = (char *)mkstr("%lu", temp_host->last_problem_id);
+		if(temp_host->last_problem_id != NULL)
+			*output = temp_host->last_problem_id;
+		break;
+	case MACRO_HOSTPROBLEMSTART:
+		*output = (char *)mkstr("%lu", (unsigned long)temp_host->problem_start);
+		break;
+	case MACRO_HOSTPROBLEMEND:
+		*output = (char *)mkstr("%lu", (unsigned long)temp_host->problem_end);
+		break;
+	case MACRO_HOSTPROBLEMDURATIONSEC:
+	case MACRO_HOSTPROBLEMDURATION:
+		if(temp_host->problem_end > 0) {
+			duration = (unsigned long)(temp_host->problem_end - temp_host->problem_start);
+		} else if(temp_host->problem_start > 0) {
+			time(&current_time);
+			duration = (unsigned long)(current_time - temp_host->problem_start);
+		}
+		if (macro_type == MACRO_HOSTPROBLEMDURATIONSEC)
+			*output = (char *)mkstr("%lu", duration);
+		else {
+			*output = (char *)mkstr("%s", duration_string(duration));
+		}
 		break;
 	case MACRO_HOSTACTIONURL:
 		if (temp_host->action_url)
@@ -950,10 +959,6 @@ static int grab_standard_service_macro_r(nagios_macros *mac, int macro_type, ser
 	objectlist *temp_objectlist = NULL;
 	time_t current_time = 0L;
 	unsigned long duration = 0L;
-	int days = 0;
-	int hours = 0;
-	int minutes = 0;
-	int seconds = 0;
 	char *buf1 = NULL;
 	char *buf2 = NULL;
 
@@ -1044,31 +1049,19 @@ static int grab_standard_service_macro_r(nagios_macros *mac, int macro_type, ser
 		break;
 	case MACRO_SERVICEDURATIONSEC:
 	case MACRO_SERVICEDURATION:
-
 		time(&current_time);
 		duration = (unsigned long)(current_time - temp_service->last_state_change);
-
-		/* get the state duration in seconds */
 		if (macro_type == MACRO_SERVICEDURATIONSEC)
 			*output = (char *)mkstr("%lu", duration);
-
-		/* get the state duration */
 		else {
-			days = duration / 86400;
-			duration -= (days * 86400);
-			hours = duration / 3600;
-			duration -= (hours * 3600);
-			minutes = duration / 60;
-			duration -= (minutes * 60);
-			seconds = duration;
-			*output = (char *)mkstr("%dd %dh %dm %ds", days, hours, minutes, seconds);
+			*output = (char *)mkstr("%s", duration_string(duration));
 		}
 		break;
 	case MACRO_SERVICENOTIFICATIONNUMBER:
 		*output = (char *)mkstr("%d", temp_service->current_notification_number);
 		break;
 	case MACRO_SERVICENOTIFICATIONID:
-		*output = (char *)mkstr("%lu", temp_service->current_notification_id);
+		*output = temp_service->current_notification_id;
 		break;
 	case MACRO_SERVICEEVENTID:
 		*output = (char *)mkstr("%lu", temp_service->current_event_id);
@@ -1077,10 +1070,30 @@ static int grab_standard_service_macro_r(nagios_macros *mac, int macro_type, ser
 		*output = (char *)mkstr("%lu", temp_service->last_event_id);
 		break;
 	case MACRO_SERVICEPROBLEMID:
-		*output = (char *)mkstr("%lu", temp_service->current_problem_id);
+		*output = temp_service->current_problem_id;
 		break;
 	case MACRO_LASTSERVICEPROBLEMID:
-		*output = (char *)mkstr("%lu", temp_service->last_problem_id);
+		*output = temp_service->last_problem_id;
+		break;
+	case MACRO_SERVICEPROBLEMSTART:
+		*output = (char *)mkstr("%lu", (unsigned long)temp_service->problem_start);
+		break;
+	case MACRO_SERVICEPROBLEMEND:
+		*output = (char *)mkstr("%lu", (unsigned long)temp_service->problem_end);
+		break;
+	case MACRO_SERVICEPROBLEMDURATIONSEC:
+	case MACRO_SERVICEPROBLEMDURATION:
+		if(temp_service->problem_end > 0) {
+			duration = (unsigned long)(temp_service->problem_end - temp_service->problem_start);
+		} else if(temp_service->problem_start > 0) {
+			time(&current_time);
+			duration = (unsigned long)(current_time - temp_service->problem_start);
+		}
+		if (macro_type == MACRO_SERVICEPROBLEMDURATIONSEC)
+			*output = (char *)mkstr("%lu", duration);
+		else {
+			*output = (char *)mkstr("%s", duration_string(duration));
+		}
 		break;
 	case MACRO_SERVICEACTIONURL:
 		if (temp_service->action_url)
@@ -1569,6 +1582,11 @@ static int grab_macrox_value_r(nagios_macros *mac, int macro_type, char *arg1, c
 	case MACRO_LASTHOSTPROBLEMID:
 	case MACRO_LASTHOSTSTATE:
 	case MACRO_LASTHOSTSTATEID:
+	case MACRO_HOSTPROBLEMSTART:
+	case MACRO_HOSTPROBLEMEND:
+	case MACRO_HOSTPROBLEMDURATIONSEC:
+	case MACRO_HOSTPROBLEMDURATION:
+
 
 		/* a standard host macro */
 		if (arg2 == NULL) {
@@ -1688,6 +1706,10 @@ static int grab_macrox_value_r(nagios_macros *mac, int macro_type, char *arg1, c
 	case MACRO_LASTSERVICEPROBLEMID:
 	case MACRO_LASTSERVICESTATE:
 	case MACRO_LASTSERVICESTATEID:
+	case MACRO_SERVICEPROBLEMSTART:
+	case MACRO_SERVICEPROBLEMEND:
+	case MACRO_SERVICEPROBLEMDURATIONSEC:
+	case MACRO_SERVICEPROBLEMDURATION:
 
 		/* use saved service pointer */
 		if (arg1 == NULL && arg2 == NULL) {
@@ -2691,6 +2713,14 @@ int init_macrox_names(void)
 	add_macrox_name(HOSTVALUE);
 	add_macrox_name(SERVICEVALUE);
 	add_macrox_name(PROBLEMVALUE);
+	add_macrox_name(HOSTPROBLEMSTART);
+	add_macrox_name(HOSTPROBLEMEND);
+	add_macrox_name(HOSTPROBLEMDURATIONSEC);
+	add_macrox_name(HOSTPROBLEMDURATION);
+	add_macrox_name(SERVICEPROBLEMSTART);
+	add_macrox_name(SERVICEPROBLEMEND);
+	add_macrox_name(SERVICEPROBLEMDURATIONSEC);
+	add_macrox_name(SERVICEPROBLEMDURATION);
 
 	return OK;
 }

src/naemon/macros.h

@@ -22,7 +22,7 @@
 /****************** MACRO DEFINITIONS *****************/
 #define MACRO_ENV_VAR_PREFIX			"NAGIOS_"
 #define MAX_USER_MACROS				256	/* max $USERx$ macros */
-#define MACRO_X_COUNT				156	/* size of macro_x[] array */
+#define MACRO_X_COUNT				164	/* size of macro_x[] array */
 
 NAGIOS_BEGIN_DECL
 
@@ -201,7 +201,15 @@ typedef struct nagios_macros nagios_macros;
 #define MACRO_HOSTVALUE                         153
 #define MACRO_SERVICEVALUE                      154
 #define MACRO_PROBLEMVALUE                      155
-
+#define MACRO_HOSTPROBLEMSTART                  156
+#define MACRO_HOSTPROBLEMEND                    157
+#define MACRO_HOSTPROBLEMDURATIONSEC            158
+#define MACRO_HOSTPROBLEMDURATION               159
+#define MACRO_SERVICEPROBLEMSTART               160
+#define MACRO_SERVICEPROBLEMEND                 161
+#define MACRO_SERVICEPROBLEMDURATIONSEC         162
+#define MACRO_SERVICEPROBLEMDURATION            163
+/* NOTE: update MACRO_X_COUNT above to highest macro + 1 */
 
 /************* MACRO CLEANING OPTIONS *****************/
 #define STRIP_ILLEGAL_MACRO_CHARS       1

src/naemon/nebmodules.h

@@ -10,7 +10,7 @@ NAGIOS_BEGIN_DECL
 
 /***** MODULE VERSION INFORMATION *****/
 #define NEB_API_VERSION(x) int __neb_api_version = x;
-#define CURRENT_NEB_API_VERSION    6
+#define CURRENT_NEB_API_VERSION    7
 
 
 /***** MODULE INFORMATION *****/

src/naemon/notifications.c

@@ -400,8 +400,8 @@ int service_notification(service *svc, int type, char *not_author, char *not_dat
 	log_debug_info(DEBUGL_NOTIFICATIONS, 1, "Current notification number: %d (%s)\n", svc->current_notification_number, (increment_notification_number == TRUE) ? "incremented" : "changed");
 
 	/* save and increase the current notification id */
-	svc->current_notification_id = next_notification_id;
-	next_notification_id++;
+	nm_free(svc->current_notification_id);
+	svc->current_notification_id = g_uuid_string_random();
 
 	log_debug_info(DEBUGL_NOTIFICATIONS, 2, "Creating list of contacts to be notified.\n");
 
@@ -478,7 +478,7 @@ int service_notification(service *svc, int type, char *not_author, char *not_dat
 		mac.x[MACRO_NOTIFICATIONNUMBER] = nm_strdup(mac.x[MACRO_SERVICENOTIFICATIONNUMBER]);
 
 		/* set the notification id macro */
-		nm_asprintf(&mac.x[MACRO_SERVICENOTIFICATIONID], "%lu", svc->current_notification_id);
+		nm_asprintf(&mac.x[MACRO_SERVICENOTIFICATIONID], "%s", svc->current_notification_id);
 
 		/* notify each contact (duplicates have been removed) */
 		for (temp_notification = notification_list; temp_notification != NULL; temp_notification = temp_notification->next) {
@@ -1283,8 +1283,8 @@ int host_notification(host *hst, int type, char *not_author, char *not_data, int
 	log_debug_info(DEBUGL_NOTIFICATIONS, 1, "Current notification number: %d (%s)\n", hst->current_notification_number, (increment_notification_number == TRUE) ? "incremented" : "unchanged");
 
 	/* save and increase the current notification id */
-	hst->current_notification_id = next_notification_id;
-	next_notification_id++;
+	nm_free(hst->current_notification_id);
+	hst->current_notification_id = g_uuid_string_random();
 
 	log_debug_info(DEBUGL_NOTIFICATIONS, 2, "Creating list of contacts to be notified.\n");
 
@@ -1360,7 +1360,7 @@ int host_notification(host *hst, int type, char *not_author, char *not_data, int
 		mac.x[MACRO_NOTIFICATIONNUMBER] = nm_strdup(mac.x[MACRO_HOSTNOTIFICATIONNUMBER]);
 
 		/* set the notification id macro */
-		nm_asprintf(&mac.x[MACRO_HOSTNOTIFICATIONID], "%lu", hst->current_notification_id);
+		nm_asprintf(&mac.x[MACRO_HOSTNOTIFICATIONID], "%s", hst->current_notification_id);
 
 		/* notify each contact (duplicates have been removed) */
 		for (temp_notification = notification_list; temp_notification != NULL; temp_notification = temp_notification->next) {

src/naemon/objects_host.c

@@ -326,6 +326,9 @@ void destroy_host(host *this_host)
 	nm_free(this_host->icon_image_alt);
 	nm_free(this_host->vrml_image);
 	nm_free(this_host->statusmap_image);
+	nm_free(this_host->current_notification_id);
+	nm_free(this_host->last_problem_id);
+	nm_free(this_host->current_problem_id);
 	nm_free(this_host);
 }
 

src/naemon/objects_host.h

@@ -89,8 +89,10 @@ struct host {
 	int     current_attempt;
 	unsigned long current_event_id;
 	unsigned long last_event_id;
-	unsigned long current_problem_id;
-	unsigned long last_problem_id;
+	char   *current_problem_id;
+	char   *last_problem_id;
+	time_t  problem_start;
+	time_t  problem_end;
 	double  latency;
 	double  execution_time;
 	int     is_executing;
@@ -110,7 +112,7 @@ struct host {
 	int     notified_on;
 	int     current_notification_number;
 	int     no_more_notifications;
-	unsigned long current_notification_id;
+	char   *current_notification_id;
 	int     check_flapping_recovery_notification;
 	int     scheduled_downtime_depth;
 	int     pending_flex_downtime; /* UNUSED */

src/naemon/objects_service.c

@@ -339,6 +339,9 @@ void destroy_service(service *this_service, int truncate_lists)
 	nm_free(this_service->action_url);
 	nm_free(this_service->icon_image);
 	nm_free(this_service->icon_image_alt);
+	nm_free(this_service->current_notification_id);
+	nm_free(this_service->last_problem_id);
+	nm_free(this_service->current_problem_id);
 	nm_free(this_service);
 }