[dnsdist] dnsdist latency bucket metric still broken in 1.8.0?

[Remi Gacogne]

Hi Christoph,

On 14/04/2023 17:04, Christoph via dnsdist wrote:
> here is our dnsdist.conf,
> maybe it helps to reproduce the issue.

If I'm not mistaken, you are mostly dealing with DoT and DoH queries, 
not UDP ones?
I'm asking because since 1.8 these latency metrics are now only updated 
for UDP queries, and we have new ones that are broken down by protocol, 
as implemented in [1]. We have more information at [2] and a sample 
prometheus output at [3].
We do know it will require updating a few dashboards, but exactly how 
the latency was measured when several protocols were in use has been 
very hard to understand since dnsdist 1.0, with some advocating for the 
UDP latency to be only one meaningful and others wanting a global 
overview regardless of the protocol. Both use-cases can now be done by 
aggregating the metrics as needed.

Note that I still don't understand how we can end up with all recorded 
queries in the "slow" bucket ("dnsdist_latency_bucket{le="+Inf"}") and 
still get 0 as the sum ("dnsdist_latency_sum"), but I have not been able 
to reproduce this behaviour so I don't know what is going on.

[1]: https://github.com/PowerDNS/pdns/pull/11716
[2]: 
https://dnsdist.org/statistics.html?highlight=latency#latency-doh-avg100
[3]: 
https://dnsdist.org/guides/webserver.html?highlight=latency#get--metrics

Hope that helps,

Remi

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
URL: <http://mailman.powerdns.com/pipermail/dnsdist/attachments/20230503/d7563624/attachment.sig>