DNS — generic / self-hosted
The vendor-neutral default. When no cloud is named, this is the answer.
What to run
- BIND — the reference authoritative + recursive server; ubiquitous, flexible, zone-file driven. Heaviest to operate; large attack surface if misconfigured.
- PowerDNS — authoritative server with pluggable backends (SQL, etc.) plus a separate recursor; good when zones live in a database. Some geo/load features via backends (e.g. geoip).
- Knot DNS / NSD — modern, fast authoritative-only servers; pair with Unbound as the recursive resolver. Common for high-performance authoritative tiers.
- CoreDNS — plugin-based, common inside Kubernetes for service discovery.
Topology
- Authoritative tier holds your zones; run at least two name servers, on separate networks/sites. Standard DNS supports primary→secondary zone transfer (AXFR/IXFR) so secondaries stay in sync — use it for redundancy.
- Anycast the authoritative IPs across sites via BGP for proximity and DDoS resilience (you operate the routing).
- Recursive resolvers are usually the client's/ISP's; you run recursors only for internal resolution.
Routing policies without a managed service
Plain BIND/NSD give you simple and round-robin (multiple A records) out of the
box. Weighted, latency, geo, and failover policies are features layered on
top — PowerDNS geoip/lua backends, dnsdist, or a GSLB appliance — and health-
checked failover means running your own prober that rewrites zone records on
failure. This operational burden is exactly why most teams use a managed provider
for traffic steering.
Limits / things that bite
- Self-run health-check failover is DIY: you build the prober, the thresholds, and the record-rewrite path — and it's still TTL-bound for recovery.
- Zone transfers must be secured (TSIG) or you leak/expose your zone.
- A misconfigured open recursive resolver becomes a DDoS amplifier — lock it down.
- Geo/latency steering needs an IP-geolocation database you keep current.
Pitfalls
- One physical site for all authoritative servers → SPOF; spread them.
- Forgetting DNSSEC key rollover if signing zones (broken validation = outage).
- Treating round-robin
Arecords as load balancing — clients cache one answer and there's no health awareness.
Maximum control and zero lock-in, but you own anycast/BGP, health checking, and geo data. Use when running your own infra or when a managed provider can't meet a constraint; otherwise a managed zone removes most of this toil.