The publisher whose CMS outage took the whole site down

The publisher whose CMS outage took the whole site down

A regional publisher purged aggressively — every edit triggered a hard purge that fully evicted the article from the edge. One afternoon their CMS database went down for 25 minutes. With nothing in cache and no origin to fetch from, the edge had only 503s to serve. The entire site went dark mid-traffic-peak.

We switched hard purges to soft purges (mark stale, don't evict) and enabled stale-if-error. Now when origin fails, the edge serves the last-known-good version instead of an error.

During the next origin outage — a 40-minute deploy gone wrong — readers kept seeing fully-rendered articles from the edge. Pageviews stayed flat; bounce rate didn't move. The newsroom fixed origin with zero public-facing downtime.

The number that mattered: 40 minutes of origin downtime, 0 minutes of reader downtime.