DNS TTL Tradeoffs During Zero-Downtime Infrastructure Migrations
Most "instant" DNS cutovers still leak traffic to the old origin for hours because recursive resolvers honor the previous Time-to-Live (TTL) value, not the new one. If your A or CNAME record carried an 86400-second (24-hour) TTL, flipping the IP five minutes before maintenance leaves half the internet hitting the old box.
The Three-Step Cutover Checklist
A predictable migration adjusts TTLs ahead of time:
1. T-48h : Lower A / AAAA / CNAME TTL from 86400 -> 120 seconds 2. T-0 : Keep both old & new origins warm; switch DNS target 3. T+24h : Verify 0 hits on old origin, then restore TTL -> 3600
Note that when switching authoritative nameservers (NS records) at the registrar level, parent TLD zones typically cache NS delegations for 24–48 hours regardless of your apex record TTL. Keeping identical TLS certificates and static routes active on both sides during that overlap window prevents spurious 5xx spikes while resolvers converge.