About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
Split-horizon DNS (sometimes called split-brain DNS) means one fully qualified domain name resolves to different addresses depending on who asks and from which network. After a line-of-business application moves to hosting, the same branded hostname might need to point at a private address for office VLANs, a gateway address for remote staff, and a public address for partners — all at once. If you treat DNS as "change the A record once", you get intermittent login failures that look like application bugs.
Why it matters
DNS is cached at laptops, branch resolvers, ISP servers and cloud load balancers. A cutover that flips a name globally without horizon planning breaks some users while others succeed. Australian businesses with hybrid periods — old server still read-only, new server live — amplify the pain when both environments answer similar names.
Remote staff on VPN may need the same answer as the office (private path), while staff on split-tunnelling VPN may need a different path than office users depending on how DNS is assigned in the VPN profile. Home users without VPN need a controlled public entry. None of those audiences reads the same resolver chain.
Material difference from published coverage: Port forward works from the internet but not inside: NAT hairpin or split DNS explains split DNS when NAT hairpin fails. Internal DNS names as a production dependency between hosted applications covers steady-state internal naming. This article focuses on multiple staff audiences (office, branch, VPN, home) during and after hosting migration.
How it works
On Windows Server 2016 and later, Microsoft documents split-brain DNS using DNS policy to serve different answers to internal and external clients (Use DNS Policy for split-brain DNS deployment). The DNS overview notes forwarding and conditional forwarding at a high level; conditional forwarders that send only hosted suffixes to a hoster resolver are an example scenario, not a Microsoft-prescribed migration pattern.
Core concepts (RFC 1034 / RFC 1035): resolvers cache answers for the record TTL. Lowering TTL before migration reduces stale answers only if you lower them earlier than the old TTL — it does not fix wrong horizon design.
Illustrative resolver behaviour (not sourced — validate with your VPN and DNS design):
| Location | Typical resolver | Desired behaviour for app.example.com |
|---|---|---|
| Office LAN | Local AD DNS | Private hosted address or on-prem bridge |
| Branch via site-to-site | Branch forwarder to HQ DNS or local stub | Match office private view |
| VPN full tunnel | Corporate DNS | Same as office |
| VPN split / home | Depends on VPN DNS rules and split tunnel | Often public or gateway answer; confirm NRPT/suffix rules |
Steady-state hybrid design differs from one-shot cutover articles such as DNS changed but users still reach the old application.
What to choose and why
Choose split zones with DNS policy when you must serve different answers for the same name internally and externally and you can maintain two authoritative views responsibly (Windows Server 2016+).
Choose conditional forwarders when only hosted suffixes differ and the rest of AD DNS stays unchanged.
Choose separate names (app vs app-legacy) when politics block split horizons — uglier for users and TLS certificates, but clearer for operators (editorial opinion).
Before cutover, run four resolution checks from each audience using nslookup or Resolve-DnsName from an office PC, branch PC, VPN laptop, and home PC without VPN. These checks are untested templates in this article — record answers in your migration runbook.
For scavenging risk during hybrid periods, see Clean up stale AD DNS records: scavenging without deleting live ones instead of relying on a generic scavenging page.
Read next
Related articles you may want next.
- HostingInternal DNS names as a production dependency between hosted applicationsWhy hard-coded hostnames, short TTLs and split zones couple routine DNS changes to hosted application outages.Read article
- TroubleshootingPort Forward Works From the Internet but Not Inside: NAT Hairpin or Split DNSPublic name works outside but not on the LAN? Prove the NAT hairpin fault with captures, then choose hairpin NAT (DNAT plus SNAT) or split DNS with the TLS name intact.Read article
- TroubleshootingDNS Changed but Users Still Reach the Old ApplicationDistinguish authoritative DNS, resolver caches and existing connections so an application migration has a measurable cutover and return plan.Read article
Risks and limits
CNAME chains that expose private addresses. CNAME chains that accidentally point home users at RFC1918 addresses leak confusion into helpdesk queues (illustrative risk).
Certificate name mismatch. For URLs that use a DNS hostname, TLS clients validate the originally requested hostname against the certificate presented at the endpoint (RFC 9110 section 4.3.4; RFC 9525 on service identity in TLS). Clients do not treat the resolved IP address as the identity to match when the URI host is a DNS name. Split horizons that make the name resolve differently for some audiences still need a certificate valid for that hostname on the endpoint users actually reach.
Provider scope. Who publishes public DNS for customer-facing names and who operates office resolvers depends on the contract.
Related: VPC networking. Plan branch private connectivity and site-to-site routing with your provider in the same workshop as horizon design.
Sources
- Microsoft Learn — Use DNS Policy for split-brain DNS deployment — Windows Server 2016+.
- Microsoft Learn — DNS overview.
- RFC 1034 / RFC 1035 — name resolution and TTL behaviour.
- RFC 9110 — HTTP Semantics section 4.3.4 — client must verify service identity against the URI’s origin host (DNS-ID when the host is a name).
- RFC 9525 — Service Identity in TLS — reference identifiers for TLS; DNS resolution intermediates are not reference identifiers.
Learning notes
- One hostname can need multiple correct answers depending on where the user sits.
- Test DNS from office, branch, VPN and home before you call cutover done.
- TTL lowering helps migration; horizon design prevents steady-state surprises.
- Editorial: split-brain DNS is maintenance-heavy — separate names may be simpler.
- Hosting provider DNS and office DNS ownership depends on the contract.
This article was originally published on .