High cpu usage when show_age_threshold = -1.

[fwsmit]

Described by @bakkeby:

As per the comment in the dunst config file the show_age_threshold can be set to -1 to disable the display of the age of a message.

In queues_get_next_datachange if there are no messages that are timing out (i.e. they are permanent until the user interacts with them) then the function will return -1.

dunst/src/queues.c

Line 589 in 464076d

return wakeup_time != G_MAXINT64 ? wakeup_time : -1;

Now in the run function we do not take into account that the return value can be -1 indicating that there is no need to do any further updates.

dunst/src/dunst.c

Lines 95 to 115 in 464076d

	if (active) {
	gint64 timeout_at = queues_get_next_datachange(now);
	// Previous computations may have taken time, update `now`
	// This might mean that `timeout_at` is now before `now`, so
	// we have to make sure that `sleep` is still positive.
	now = time_monotonic_now();
	gint64 sleep = timeout_at - now;
	sleep = MAX(sleep, 1000); // Sleep at least 1ms
	LOG_D("Sleeping for %li ms", sleep/1000);
	if (sleep >= 0) {
	if (reason == 0 || next_timeout < now || timeout_at < next_timeout) {
	if (next_timeout != 0) {
	g_source_remove(next_timeout_id);
	}
	next_timeout_id = g_timeout_add(sleep/1000, run, NULL);
	next_timeout = timeout_at;
	}
	}
	}

The solution can be as simple as bailing if that is the case:

         if (active) {
                 gint64 timeout_at = queues_get_next_datachange(now);
+                if (timeout_at == -1)
+                        return G_SOURCE_REMOVE;
                 // Previous computations may have taken time, update `now`
                 // This might mean that `timeout_at` is now before `now`, so
                 // we have to make sure that `sleep` is still positive.
                 now = time_monotonic_now();
                 gint64 sleep = timeout_at - now;
                 sleep = MAX(sleep, 1000); // Sleep at least 1ms
 
                 LOG_D("Sleeping for %li ms", sleep/1000);
 
                 if (sleep >= 0) {
                         if (reason == 0 || next_timeout < now || timeout_at < next_timeout) {
                                 if (next_timeout != 0) {
                                         g_source_remove(next_timeout_id);
                                 }
                                 next_timeout_id = g_timeout_add(sleep/1000, run, NULL);
                                 next_timeout = timeout_at;
                         }
                 }
         }

(there are other ways to refactor the code to the same effect of course)

Example test scenario is to:

• set show_age_threshold to -1 in the dunstrc configuration file and
• run top -p $(pidof dunst) to monitor CPU usage and
• run notify-send -u critical hello

Expected behaviour without the change above:

• CPU usage increases drastically while the notification is displayed
• the notification may flicker

Expected behaviour with the change above:

• CPU usage remains at 0.0% and
• there are no unwanted side effects (i.e. new notifications come and go as normal, one can interact with the critical notification as normal)

[fwsmit]

@bakkeby Your fix seems to work. Could you submit it or something similar as a PR?

[bakkeby]

Sure, I'll raise one.

[fwsmit]

Fixed with #1164

[myme]

I've been consistently struggling with this issue for a while (and sorry to say without putting effort into solving it). I built from master after #1168 yesterday and have seen no trace of CPU spin ups since. Great work! 👍