DNS — GCP
Service mapping
GCP also splits zone hosting from global steering — most traffic steering happens in the load balancer, not in Cloud DNS.
- Cloud DNS — managed authoritative hosting on Google's anycast network. Public and private zones, standard record types, DNSSEC, and routing policies: Weighted round-robin and Geolocation (answer by client region), plus health-checked failover for internal/private routing. Apex support is via the policy/record set rather than a separate alias type — point records at a global LB IP.
- Cloud Load Balancing (global external) — the primary "latency/geo" steerer.
A single anycast global IP front-ends all regions; Google's edge routes users
to the nearest healthy backend and fails over without DNS TTL waits. Prefer
this over DNS-level geo for speed and fast failover. Edge caching (Cloud CDN) is
content-delivery. - Cloud DNS routing policies cover DNS-level weighted/geo when you specifically need answers to differ per client rather than a single anycast IP.
When to pick which
Default: one global external Application LB on an anycast IP, with Cloud DNS holding a simple record pointing at it — proximity and failover handled at the LB, no TTL dependency. Reach for Cloud DNS geolocation/weighted policies when you need the answer itself to differ (data residency, canary across distinct IPs).
Limits / things that bite (verify against current docs)
- DNS-level geo/weighted policies are TTL-bound; the global LB's anycast steering is not.
- Cloud DNS geolocation maps by client region (resolver-influenced), not exact user.
- Per-zone record-set and per-policy item quotas apply.
- Health-checked DNS failover applies mainly to private/internal zones — public fast failover is the LB's job.
Pitfalls
- Building DNS-level geo routing when a single global anycast LB IP would be simpler and faster.
- Forgetting DNSSEC key management once enabled (validation breaks = outage).
- Assuming Cloud DNS does latency routing like a managed competitor — favor the LB instead.