This week in caching: Varnish under traffic spikes
What full-page caching at the edge of origin actually buys you:
— News site, election night — Varnish in front of WordPress served 96% of requests from RAM; origin saw 350 req/s instead of the ~9,000 hitting the front door. Zero added app servers.
— Ticketing platform — a 2-second grace period (
— Classifieds — VCL that stripped tracking query params before hashing lifted hit rate from 61% to 88% overnight; the cache had been fragmenting on ?utm junk.
The theme: most 'we need more servers' moments are really 'we're not caching the cacheable.'
Bookmark: varnishhist + varnishtop -i ReqURL — find what's punching through to origin and why.
What full-page caching at the edge of origin actually buys you:
— News site, election night — Varnish in front of WordPress served 96% of requests from RAM; origin saw 350 req/s instead of the ~9,000 hitting the front door. Zero added app servers.
— Ticketing platform — a 2-second grace period (
std.healthy + stale-while-revalidate) meant a backend restart during onsale caused zero user-facing errors; users kept getting cached pages.— Classifieds — VCL that stripped tracking query params before hashing lifted hit rate from 61% to 88% overnight; the cache had been fragmenting on ?utm junk.
The theme: most 'we need more servers' moments are really 'we're not caching the cacheable.'
Bookmark: varnishhist + varnishtop -i ReqURL — find what's punching through to origin and why.