One product. Three doors.
Built for the day one of them won't open.
Two real backends — AWS and Scaleway — behind three independent domains. A fourth, on a third cloud, is on the way.
Fastly-routed. Health-checked failover between the AWS and Scaleway boxes.
EU-direct path to the Scaleway (fr-par) box. No Fastly in front.
US-direct path to the AWS (us-east-1) box.
Planned third box on Microsoft Azure — a third independent cloud.
Hover or focus any cube to see the provider + region that layer terminates at.
One product, reached through three doors today — glassbreak.io (Fastly-fronted failover), glassbreak.cloud (EU-direct), and glass-break.com (US-direct) — backed by two independent cloud boxes (AWS + Scaleway), with a Microsoft Azure box planned as the third. Each box is a full end-to-end stack — DNS, edge/TLS, static web, Node API, and in-box database — so a failure at any one provider leaves the others operational.
Two backends, three doors
Glassbreak runs the same product on two always-on boxes — one on AWS in us-east-1 and one on Scaleway in fr-par — each a Docker Compose stack running Caddy, the same Node API, and its own in-box PostgreSQL. The two boxes are kept in step by native Postgres streaming replication.
Those two backends are exposed through three different domains, because the domain you visit determines the path you take. Same product, different infrastructure underneath.
glassbreak.io
The front door. Fastly's edge picks between AWS and Scaleway in real time using health checks. You don't notice failover — that's the point.
glassbreak.cloud
A side door bypassing the front. Direct line to Scaleway in the EU, no Fastly in the path. Same data, simpler compliance story for EU-residency-conscious buyers.
glass-break.com
The other side door. Direct line to the AWS box in the US, no Fastly in the path. Keeps the AWS-end-to-end story whole when the front door is down.
How traffic flows
- You visit one of the three domains. Each has two independent DNS authorities serving the same records, so losing one DNS provider does not unresolve the name.
- On
glassbreak.ioyour request lands at Fastly, which health-checks both the AWS and Scaleway boxes and routes you to whichever is healthy. Onglassbreak.cloudandglass-break.comyou connect directly — no CDN in between. - Each box runs its own in-box PostgreSQL. The two are kept in sync by native Postgres streaming replication managed by Patroni + etcd — one box is the primary/writer (currently AWS), the other a hot standby — over a Rosenpass-secured post-quantum WireGuard mesh. No shared message queue, no shared region.
What survives
- AWS outage —
.iofails over to Scaleway via Fastly;.cloudis unaffected. - Scaleway outage —
.iostays on AWS via Fastly;.comis unaffected. - Fastly outage —
.cloudand.comboth bypass Fastly entirely. - DNS provider outage — every domain has a second authority serving identical records.
A fourth door, for the rare day
A planned third box on Microsoft Azurewill be a fully independent path on a third cloud — its own DNS, its own in-box Postgres, and zero shared infrastructure with either AWS or Scaleway. It exists for the day when both production clouds are simultaneously degraded. We're not committing to a date, but the architecture is locked.
Want to see this live? The dots on the cards above are real — each card polls its own domain's /api/public/status endpoint directly from your browser every 30 seconds. Open DevTools and watch the network panel: you'll see requests going to four different hostnames. We don't aggregate the result on a single backend — that would defeat the whole point.