Open appliedprivacy opened 1 year ago
I think we have one documentation issue and one feature request:
If this ever gets implemented as different buckets, please please please just add these as tags for the prometheus metrics and don't create them as separately-named metrics. It may even be useful to make the existing metric names be the average of all the protocols, in order for there to be no breaking change and also to give answers in that legacy metric which meet the expectations that people have today of a single value that gives all the protocols. In other words:
The right way: dnsdist_recursor_main_latency0_1 100 dnsdist_recursor_main_latency0_1{protocol="UDP53"} 50 dnsdist_recursor_main_latency0_1{protocol="DOH"} 150 ... etc.
instead of the wrong way: dnsdist_recursor_main_latency0_1 100 dnsdist_recursor_main_DOH_latency0_1 150 dnsdist_recursor_main_UDP53_latency0_1 50 ... etc.
Short description
as pointed out by Remi on the mailing list latency-bucket does not contain DoT/DoH queries and is limited to UDP queries: https://mailman.powerdns.com/pipermail/dnsdist/2023-May/001338.html
would be great to add that information to the latency-bucket documentation: https://dnsdist.org/statistics.html?highlight=latency-bucket#latency-bucket
Does it also apply to the latencyN-M metrics? https://dnsdist.org/statistics.html?highlight=latency#latency0-1