One dead origin region, zero downtime

One dead origin region, zero downtime

A payments dashboard ran two origins, Virginia and Frankfurt, with the CDN routing each region to its nearest. Clean, until the Frankfurt origin had a database failover and went dark. The CDN kept sending all European misses to a dead box. Europe saw timeouts for nine minutes while Virginia sat healthy and idle.

We configured origin failover with active health checks: the edge probes each origin every few seconds, and on failure it reroutes that region's misses to the surviving origin automatically.

During the next Frankfurt outage, European traffic shifted to Virginia in 6 seconds. Users saw an extra 80 ms of latency on misses, nothing more, instead of a dead site. Cached content, 93% of requests, never noticed the outage at all.

The number that mattered: 6-second failover, down from 9 minutes of timeouts.