- Registrar DNS vs Cloud DNS: when the default becomes the outageThe DNS that came with the domain name is a convenience, not an architecture. It will answer until it will not: a high TTL during a migration, no health checks during an origin failure, no Anycast during a flood.
- Vanity nameservers: branded NS on Anycast DNSResellers and brand-sensitive products want ns1.theirname.com. Operating anycast DNS for that name is the expensive part. Vanity nameservers keep the brand; Cloud DNS keeps the network.
- IPv6 DNS: publishing AAAA without lyingIPv6 is a record and a transport. Publishing AAAA for an origin that only speaks IPv4 breaks dual-stack clients. Not publishing AAAA when Cloud CDN can deliver over IPv6 leaves performance on the table.
- Lower DNS TTL before a migration — a checklistThe most expensive DNS mistake is changing a record that still has yesterday's 24-hour TTL. Lower first, wait, change, watch, then raise.
- Weighted DNS load balancing and canariesApplication load balancers only see connections that already arrived. DNS is the first chance to send 10% of answers to a new pool. It will not be perfectly even. It will be good enough to canary a region.
- DNS health checks: connect endpoint health to the recordStatic DNS will keep handing out an address after the application behind it is gone. Health checks make eligibility a measured fact, then a one-second TTL makes that fact usable.
- DNSSEC explained without the folkloreUnsigned zones can be spoofed between authority and resolver. Teams delay DNSSEC because they think it forbids geo, failover or low TTLs. On Cloud DNS those features still sit on a signed zone.
- CNAME flattening: how the apex can follow a hostnameRFC1034 forbids a CNAME at the zone apex beside other records. Sites still need example.com to follow a CDN hostname. Flattening is how the authority returns A/AAAA for that target without dropping mail.
- DNS TTL explained, and when to use a 1-second TTLTTL is a promise to recursive resolvers: you may remember this answer for this long. A 24-hour TTL on a record you might move is a 24-hour outage you pre-authorised.
- DNS failover: stop handing out dead endpointsIf the origin dies and DNS still returns it, every client retries the outage. Failover that waits on a 300-second TTL is not failover. It is a delayed post-mortem.
- GeoDNS explained: steering by country, ASN and subnetA single A record sends everyone to the same endpoint. GeoDNS lets the answer depend on where the lookup appears to come from. On Cloud DNS that can be coordinates, continent, country, ASN, IP or subnet.
- What is Anycast DNS?Unicast DNS has a home address. Anycast DNS has a home announcement: the same nameserver addresses are reachable from many places, and the network carries the resolver to a nearby available path.