Microsoft’s July 23, 2026 West US incident shows why a multi-region label is not enough. Resilience depends on application routing, data replication, identity, network paths, dependencies, observability, and a tested failover decision. A second region that cannot accept real traffic is capacity, not recovery.
Current as of 2026-08-15
Microsoft’s post-incident review reports impact from 14:44 UTC to 19:41 UTC, with a longer recovery tail for some services, following an error in a route-conversion process. Microsoft recommends multi-region design for critical workloads.
Decision summary
- Resolve every critical dependency, not only compute.
- Verify failover capacity, data behavior, DNS/routing, identity, and operational authority.
- Use availability zones for datacenter-level resilience and regions for regional failure scenarios.
- Test failover and failback with measurable recovery objectives.
Map the complete dependency path
- Ingress, DNS, CDN, load balancers, and certificates.
- Application compute, queues, caches, and secrets.
- Databases, storage, replication, and recovery-point behavior.
- Identity, key management, monitoring, and alert delivery.
- ExpressRoute, VPN, internet, and third-party API dependencies.
Choose the correct regional pattern
Microsoft’s region and availability-zone guidance distinguishes zonal and regional risks. Active-active can reduce recovery time but increases data and operational complexity. Active-passive may be simpler but requires proven activation, capacity, and data recovery.
Remove hidden network single points
Review Microsoft’s ExpressRoute resilience guidance when private connectivity is required. Independent circuits are useful only when routing, provider diversity, on-premises equipment, and application paths are also resilient.
Exercise the decision
- Declare a realistic regional scenario.
- Measure detection and decision time.
- Fail traffic and validate user journeys.
- Verify data consistency and business reconciliation.
- Operate long enough to expose hidden dependencies.
- Fail back under a separate controlled plan.
- Record observed RTO, RPO, defects, owners, and due dates.
For related guidance from ITECS, see ITECS managed Azure cloud services.
Sources and update trigger
- Microsoft Azure — July 23, 2026 incident history
- Microsoft Learn — Regions and availability zones
- Microsoft Learn — ExpressRoute resiliency
Review trigger: Review after every significant Azure incident, architecture change, dependency change, or resilience exercise.