add problem timestamps and duration

this PR makes hosts / services save the start and end timestamp of the current
problem. Those values can then be used as macros, ex. in notification scripts.
For this, there are several new macros available:

- $HOSTPROBLEMSTART$ start timestamp of problem
- $HOSTPROBLEMEND$ end timestamp of problem (or zero if problem still persists)
- $HOSTPROBLEMDURATIONSEC$ duration of problem
- $HOSTPROBLEMDURATION$ duration as human readable text

the same macros exist for services:
- $SERVICEPROBLEMSTART$
- $SERVICEPROBLEMEND$
- $SERVICEPROBLEMDURATIONSEC$
- $SERVICEPROBLEMDURATION$

While there is a currently ongoing problem, the values point to this current
problem and the end timestamp is zero. Once the problem is resolved, the values
can still be used and won't change until a new problem starts.

This makes it possible to use the problem duration in recovery notifications which
otherwise would not be possible.

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

@@ -1010,12 +1010,15 @@ static int handle_host_state(host *hst, int *alert_recorded)
 			/* 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++;
+			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->problem_end = current_time;
 		}
 
 		/* write the host state change to the main log file */

src/naemon/checks_service.c

@@ -688,12 +688,16 @@ int handle_async_service_check_result(service *temp_service, check_result *queue
 			/* 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++;
+			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 = 0L;
+			temp_service->problem_end = current_time;
 		}
 	}
 

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:
@@ -762,6 +749,26 @@ static int grab_standard_host_macro_r(nagios_macros *mac, int macro_type, host *
 	case MACRO_LASTHOSTPROBLEMID:
 		*output = (char *)mkstr("%lu", 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)
 			*output = temp_host->action_url;
@@ -950,10 +957,6 @@ 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;
 
@@ -1044,24 +1047,12 @@ 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)
+		if (macro_type == MACRO_HOSTDURATIONSEC)
 			*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:
@@ -1082,6 +1073,26 @@ static int grab_standard_service_macro_r(nagios_macros *mac, int macro_type, ser
 	case MACRO_LASTSERVICEPROBLEMID:
 		*output = (char *)mkstr("%lu", 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)
 			*output = temp_service->action_url;
@@ -1569,6 +1580,11 @@ 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 +1704,10 @@ 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 +2711,14 @@ 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/objects_host.h

@@ -91,6 +91,8 @@ struct host {
 	unsigned long last_event_id;
 	unsigned long current_problem_id;
 	unsigned long last_problem_id;
+	time_t  problem_start;
+	time_t  problem_end;
 	double  latency;
 	double  execution_time;
 	int     is_executing;

src/naemon/objects_service.h

@@ -85,6 +85,8 @@ struct service {
 	unsigned long last_event_id;
 	unsigned long current_problem_id;
 	unsigned long last_problem_id;
+	time_t  problem_start;
+	time_t  problem_end;
 	time_t	last_notification;
 	time_t  next_notification;
 	int     no_more_notifications;

src/naemon/xrddefault.c

@@ -180,6 +180,8 @@ int xrddefault_save_state_information(void)
 		fprintf(fp, "current_event_id=%lu\n", temp_host->current_event_id);
 		fprintf(fp, "current_problem_id=%lu\n", temp_host->current_problem_id);
 		fprintf(fp, "last_problem_id=%lu\n", temp_host->last_problem_id);
+		fprintf(fp, "problem_start=%lu\n", temp_host->problem_start);
+		fprintf(fp, "problem_end=%lu\n", temp_host->problem_end);
 		fprintf(fp, "plugin_output=%s\n", (temp_host->plugin_output == NULL) ? "" : temp_host->plugin_output);
 		fprintf(fp, "long_plugin_output=%s\n", (temp_host->long_plugin_output == NULL) ? "" : temp_host->long_plugin_output);
 		fprintf(fp, "performance_data=%s\n", (temp_host->perf_data == NULL) ? "" : temp_host->perf_data);
@@ -275,6 +277,8 @@ int xrddefault_save_state_information(void)
 		fprintf(fp, "current_event_id=%lu\n", temp_service->current_event_id);
 		fprintf(fp, "current_problem_id=%lu\n", temp_service->current_problem_id);
 		fprintf(fp, "last_problem_id=%lu\n", temp_service->last_problem_id);
+		fprintf(fp, "problem_start=%lu\n", temp_service->problem_start);
+		fprintf(fp, "problem_end=%lu\n", temp_service->problem_end);
 		fprintf(fp, "current_attempt=%d\n", temp_service->current_attempt);
 		fprintf(fp, "max_attempts=%d\n", temp_service->max_attempts);
 		fprintf(fp, "normal_check_interval=%f\n", temp_service->check_interval);
@@ -1064,6 +1068,10 @@ int xrddefault_read_state_information(void)
 							temp_host->current_problem_id = strtoul(val, NULL, 10);
 						else if (!strcmp(var, "last_problem_id"))
 							temp_host->last_problem_id = strtoul(val, NULL, 10);
+						else if (!strcmp(var, "problem_start"))
+							temp_host->problem_start = strtoul(val, NULL, 10);
+						else if (!strcmp(var, "problem_end"))
+							temp_host->problem_end = strtoul(val, NULL, 10);
 						else if (!strcmp(var, "state_type"))
 							temp_host->state_type = atoi(val);
 						else if (!strcmp(var, "last_state_change"))
@@ -1310,6 +1318,10 @@ int xrddefault_read_state_information(void)
 							temp_service->current_problem_id = strtoul(val, NULL, 10);
 						else if (!strcmp(var, "last_problem_id"))
 							temp_service->last_problem_id = strtoul(val, NULL, 10);
+						else if (!strcmp(var, "problem_start"))
+							temp_service->problem_start = strtoul(val, NULL, 10);
+						else if (!strcmp(var, "problem_end"))
+							temp_service->problem_end = strtoul(val, NULL, 10);
 						else if (!strcmp(var, "state_type"))
 							temp_service->state_type = atoi(val);
 						else if (!strcmp(var, "last_state_change"))

src/naemon/xsddefault.c

@@ -202,6 +202,8 @@ int xsddefault_save_status_data(void)
 		fprintf(fp, "\tcurrent_event_id=%lu\n", temp_host->current_event_id);
 		fprintf(fp, "\tcurrent_problem_id=%lu\n", temp_host->current_problem_id);
 		fprintf(fp, "\tlast_problem_id=%lu\n", temp_host->last_problem_id);
+		fprintf(fp, "\tproblem_start=%lu\n", temp_host->problem_start);
+		fprintf(fp, "\tproblem_end=%lu\n", temp_host->problem_end);
 		fprintf(fp, "\tplugin_output=%s\n", (temp_host->plugin_output == NULL) ? "" : temp_host->plugin_output);
 		fprintf(fp, "\tlong_plugin_output=%s\n", (temp_host->long_plugin_output == NULL) ? "" : temp_host->long_plugin_output);
 		fprintf(fp, "\tperformance_data=%s\n", (temp_host->perf_data == NULL) ? "" : temp_host->perf_data);
@@ -269,6 +271,8 @@ int xsddefault_save_status_data(void)
 		fprintf(fp, "\tcurrent_event_id=%lu\n", temp_service->current_event_id);
 		fprintf(fp, "\tcurrent_problem_id=%lu\n", temp_service->current_problem_id);
 		fprintf(fp, "\tlast_problem_id=%lu\n", temp_service->last_problem_id);
+		fprintf(fp, "\tproblem_start=%lu\n", temp_service->problem_start);
+		fprintf(fp, "\tproblem_end=%lu\n", temp_service->problem_end);
 		fprintf(fp, "\tcurrent_attempt=%d\n", temp_service->current_attempt);
 		fprintf(fp, "\tmax_attempts=%d\n", temp_service->max_attempts);
 		fprintf(fp, "\tstate_type=%d\n", temp_service->state_type);