Vladimír Čunát <vladimir.cunat@nic.cz> schrieb am Sa., 2. Juni 2018 um 08:29 Uhr:

On 05/31/2018 03:32 PM, Ulrich Wisser wrote:

I find the query stats strange. If you count on the answers, why aren't they answer stats? And why aren't the other flags counted. I for one would be interested in AD.

I find that an implementation detail. Usage of EDNS and the DO flag are determined by the query having them set, so it feels more natural (to me) to count them as properties of the queries rather than answers.

I agree that it isn't that important where the flags are counted.

But I would like a count of all flags, not only these two.

I think the prefetch stats need some more definition. Right now I can't see a real use. I would like to see how many queries are considered for prefetch (is this prefetch.queue?) and the total amount of prefetches done.

They are actually documented, but I suppose it's not particularly enlightening unless you know how the prefetch module works. http://knot-resolver.readthedocs.io/en/stable/modules.html#exported-metrics

I have actually read the documentation. Here is my understanding

predict.epoch is no statistic at all, more a marker of the current period

predict.queue is the number of queries that are considered for prefetching

predict.learned is the number of fetched executed in the current period

If that is true, the values are actually unusable for statistics. The graphite modules will write statistics to the database every 5s. What will the predict values tell me?

For the RPZ I think it would be very interesting to see which RPZ and action combination was hit how many times.

Maybe you could look at pihole https://pi-hole.net/ these people have a nice dashboard and good statistics for their DNS service.

Basically all this turned out to require more than a few minutes to fix, so I collected the ideas here: https://gitlab.labs.nic.cz/knot/knot-resolver/issues/364

Of course, I didn't expect a quick fix.

/Ulrich