Start with failure domains
Paris, New York and Toronto each have a role, dependencies and a recovery path. This separation makes node-loss and site-loss scenarios explicit.
Observe before disruption
Prometheus targets and 19 Grafana panels establish a baseline. Galera, GlusterFS and the application must be observed together.
Preserve evidence
Thirty-seven test files document connectivity, replication, storage and failovers. The archive makes results reviewable and verifiable.