server/sstate.c, NEWS.adoc: wait a second before declaring UPS data stale

NEWS.adoc

@@ -161,6 +161,13 @@ https://github.com/networkupstools/nut/milestone/12
      of the existing mappings. Suggest how user can help improve the driver
      if too few data points were seen. [#3082, #3095]
 
+ - `upsd` data server updates:
+   * Sometimes "Data for UPS [X] is stale" and "UPS [X] data is no longer
+     stale" messages were logged in the same second, especially no busy
+     systems. Now we allow one more second on top of `MAXAGE` setting to
+     declare the device dead, just in case fractional/whole second rounding
+     comes into play and breaks things. [issue #661]
+
  - `upsdrvctl` tool updates:
    * Make use of `setproctag()` and `getproctag()` to report parent/child
      process names. [#3084]

server/sstate.c

@@ -478,12 +478,24 @@ int sstate_dead(upstype_t *ups, int arg_maxage)
 	if ((elapsed > (arg_maxage / 3)) && (difftime(now, ups->last_ping) > (arg_maxage / 3)))
 		sendping(ups);
 
-	if (elapsed > arg_maxage) {
+	if (elapsed > arg_maxage + 1) {
 		upsdebugx(3, "%s: didn't hear from driver for UPS [%s] for %g seconds (max %d)",
 			__func__, ups->name, elapsed, arg_maxage);
 		return 1;	/* dead */
 	}
 
+	if (elapsed > arg_maxage) {
+		/* Per https://github.com/networkupstools/nut/issues/661 we do
+		 * often see "is stale" and "is alive" messages in the same
+		 * second, especially on busy systems that can not dedicate
+		 * every scheduled time slot to NUT daemons and data pipes.
+		 * So data exchange frequency settings may almost race...
+		 */
+		upsdebugx(3, "%s: didn't hear from driver for UPS [%s] for %g seconds (max %d); will declare it dead next second (delaying just in case of whole-second rounding issues)",
+			__func__, ups->name, elapsed, arg_maxage);
+		return 0;	/* probably dead */
+	}
+
 	/* ignore DATAOK/DATASTALE unless the dump is done */
 	if ((ups->dumpdone) && (!ups->data_ok)) {
 		upsdebugx(3, "%s: driver for UPS [%s] says data is stale", __func__, ups->name);