- It was decided to deviate from rfc7871, as full
implementation would be potentially very cache-hungry
(there are very many meaningful IP prefixes).
Instead, a geo DB will be used to partition IPs into discrete
categories. That's what the authoritative NSs use AFAIK.
- For now the granularity would be a country, as used by
https://www.maxmind.com/en/geoip2-databases
- Cache deduplicates equal records for different locations.
That's implemented by splitting the mapping into two:
usual key + location -> hash, and usual key + hash -> RRdata.
Timestamp and TTL of the RRset is stored in the first part,
and the stored RRs have TTLs zeroed.
(All this forced changes in lib/cache.h API.)