The Day the DNS Died

on YouTube ↗34m

" summary

SparkPost runs 8,000 DNS queries per second and hits undocumented cloud limits that throttle at roughly 1,024 qps per network interface, which causes 90%+ query loss and seven-hour mail delays. The response is a ground-up DNS tier: dedicated VPCs, per-tier resolver clusters, multiple ENIs/IPs, local caches (dnsmasq) per instance, tuned Unbound with serve expired, and network ACLs to avoid stateful security-group tracking. The redesign restores availability and survives subsequent DNS incidents.

AI-written summary · about

" chapters (6)

1. 0:00Start
2. 0:50Introduction
3. 2:20Version 1. Centralized Internal Resolver Cluster
4. 5:00Version 2, AWS VPC Resolver
5. 6:19Version 1.5. Centralized Internal Resolver Cluster
6. 20:27New DNS Design

" related talks

• 29mDNS: Old Solution for Modern ProblemsThomas Jackson, Rauf Guliyev · SREcon16 Europe
• 25mDark Side of the DNS ForceErik Wu · Black Hat USA 2016
• 34mIt’s Always DNS: So Let’s Break It On PurposeJoel Studler, Georgios Daskalopoulos · ContainerDays Hamburg 2026
• 28mDarwin & ProtocolsPaul Mockapetris · dotScale 2013
• 15mYou Depend on DNS, This Is How It Works and You Won't...Philip Rowlands · SREcon23 Europe/Middle East/Africa
• 09mDNS is the entry point for every application on the internetKristopher Beevers · O'Reilly Velocity Conference 2017 - New York, New York
• 01h 27mDNS at 30Jim Hague · ACCU 2017
• 26mMisuse of DNS, the Second Most Used ProtocolEmilio Couto · Black Hat Asia 2020