Publishing · Multi-site
Ferrolane Media
312 title sites, one network, and a cutover that had to finish before the morning edition published.
-71%
median time to first byte across all 312 sites, measured ninety days after the move
Why they moved
Ferrolane runs local news and trade titles from a single multi-site install. Every title shared one database, and one badly written plugin on one title could slow all 312. Editors published from six countries, so there was no quiet hour: whenever a deploy went wrong, someone was mid-edition.
The requirement was blunt — separate the blast radius, keep one editorial login, and do it without a maintenance window anyone would notice.
Cutover night, hour by hour
21:40
Final sync of 312 databases begins from the rehearsal snapshot taken that morning.
23:15
Editorial moved to read-only. Publishing queue held, nothing dropped.
01:05
DNS switched title by title in batches of twenty, with rollback armed at each batch.
04:30
Writes reopened, caches warmed, and the Tokyo morning edition published on time.
One night, 312 titles, no delayed editions
Ferrolane came off a single shared cluster. The rebuild isolated every region, so a bad plugin now stops at one title instead of all of them.
312
title sites moved in a single night
0
editions delayed or articles lost
1 of 312
titles affected the next time a plugin
misbehaved
The setup they run now
Architecture
Four isolated clusters by region
Databases
One primary per cluster, per-title schemas
Media
Shared object storage, edge cached
Deploys
Staged per cluster, 20-title canary
“We had rehearsed it twice, so the real night was mostly waiting. That is exactly what you want from a migration of this size.”
Guy Hawkins
Head of Engineering, Ferrolane Media