Amazon Application Recovery Controller (ARC) separates recovery control from the application data plane. Its routing controls, Region switch, zonal shift, and related recovery features let operators change where traffic goes even when the application environment itself is unhealthy or partly unavailable. Recovery design in Amazon AWS SAA-C03 therefore separates health observation from the guarded authority to shift traffic and requires readiness evidence, routing controls, and a rehearsed operator sequence.
That distinction matters because ordinary DNS health checks answer only part of a recovery problem. A system also needs guarded authority to declare a Region or Availability Zone unavailable, a rehearsed sequence for shifting traffic, and evidence that the secondary environment is actually ready to accept production load. ARC controls need pre-authorized operators, independent recovery access, tested traffic-shift runbooks, and telemetry from the destination environment so a routing change is not mistaken for a completed recovery.
ARC also changed in 2026: readiness checks are no longer open to new customers, while routing controls, Region switch, zonal shift, and zonal autoshift remain supported. New designs should therefore use current recovery mechanisms rather than assuming every historical ARC feature is still available for onboarding.
Failover is an operating decision, not only a DNS rule
Route 53 routing policies can react to health signals, but major incidents often require an explicit declaration that one environment should stop receiving traffic. ARC provides controls intended for that operational decision, separating recovery authority from the failing workload.
Regional failover is a runbook, control-plane, capacity, and data-consistency problem. DNS is one mechanism inside that system, not the entire recovery strategy, so the recovery design must connect traffic movement with application and data readiness.
Routing controls create a guarded switch
ARC routing controls are designed to alter traffic routing through highly available control-plane infrastructure. They are useful when operators need a deliberate, auditable switch rather than an application-local script that might fail for the same reason as the workload.
Protect routing-control changes with permissions, approvals, and clear ownership. A recovery control is intentionally powerful, so the architecture should make accidental or conflicting changes harder while keeping the emergency path fast enough for the required recovery time objective.
Recovery controls also need a safe abort path. If the destination Region fails validation after traffic starts moving, operators must know whether to continue, pause, or reverse the change without creating duplicate writes or oscillating between Regions. Define those decision points before the exercise and collect the telemetry required to make them quickly.
Region switch adds orchestration around recovery
Region-switch planning should name the dependencies that must move with traffic: application endpoints, capacity, data stores, secrets, queues, and any external allowlists. A routing control can redirect users quickly while the destination Region is still unable to serve them. Treat the traffic switch as one step in a recovery sequence whose preconditions are measurable before an operator is allowed to execute it.
ARC Region switch is the current direction for managed multi-Region recovery orchestration. It can coordinate a recovery plan and evaluate plan state, which moves the design beyond a binary DNS flip toward a sequence of checks and actions.
Use orchestration where recovery requires more than traffic movement: database promotion, capacity activation, cache warm-up, dependency checks, or controlled rollback. The sequence should reflect actual application dependencies rather than a generic list of cloud resources.
Readiness checks have a new availability boundary
ARC readiness checks are no longer open to new customers, although existing customers can continue using them. AWS explicitly recommends Region switch for similar multi-Region readiness and recovery-orchestration needs.
New documentation and new designs should state that transition directly. Continuing to present readiness checks as the default onboarding path would teach an architecture that a new customer cannot adopt, and it would obscure the newer plan-evaluation model.
DNS TTL and client behavior still shape cutover
Even when a routing decision is made quickly, clients and recursive resolvers can cache DNS answers until TTLs expire. Long-lived TCP connections may also continue until they fail or are closed, so traffic movement is not the same as instantaneous application migration.
Route 53 routing policies interact with client caching and connection lifetimes, so a DNS change alone does not define recovery time. Exercises should measure the real tail of old-Region traffic rather than assuming the configured TTL is the complete answer.
Treat quotas as part of recovery readiness. A standby Region can contain the correct templates yet still be unable to scale because of service quotas, subnet address space, instance availability, or dependency limits. Periodic capacity tests should prove that the failover target can support the load expected during the first recovery window, not merely that resources can be created.
The secondary Region must be a real production target
Changes to the recovery design should be rehearsed against the secondary Region, not only reviewed in the primary. Compare replication lag, available capacity, endpoint health, and the behavior of dependencies that remain regional. A modification that looks harmless during normal operation can lengthen recovery if the standby environment no longer has the data, permissions, or scale needed when the switch is invoked.
Failover controls cannot create database capacity, quotas, secrets, certificates, dependencies, or warmed application fleets in the destination Region. A recovery plan is only as strong as the least-prepared dependency required after the switch.
Capacity reservations, replication lag, identity dependencies, third-party endpoints, and Region-specific quotas should be tested under production-like load. Cross-Region resilience depends on those operational dependencies surviving the same failure as the routing control, not merely on a second Region appearing in the architecture diagram. Recovery is credible only when the alternate Region has the capacity, data, identity, and external connectivity required to serve real traffic.
Data consistency sets the recovery boundary
An application can be reachable in the secondary Region and still be unsafe if its data is stale or split-brain writes are possible. Recovery controls need a clear relationship with replication mode, recovery point objective, write ownership, and promotion rules.
Document which data stores can fail over automatically, which require operator confirmation, and which can tolerate temporary read-only service. Traffic recovery should never outrun the data-recovery semantics that make the application correct.
Exercises should test people and controls together
During an incident, the runbook should separate three questions: is the primary Region actually impaired, is the recovery destination ready, and is the routing control in the intended state. Mixing those questions encourages premature failover. Operators need a short evidence chain that shows the health signal, the readiness signal, and the control action so the team can explain why traffic moved and what condition will allow failback.
A failover rehearsal must include decision authority, communication, change execution, application validation, and the path back to steady state. A scripted routing switch that no human has practiced is not a complete recovery capability.
Record how long each stage takes and which preconditions fail most often. The purpose of an exercise is to find uncertainty while the primary Region is healthy, then remove that uncertainty from the runbook before a real incident.
Failback deserves the same rigor as failover. Data written during the secondary-Region period must be reconciled, replication direction may need to change, and traffic should return only after the former primary is healthy and synchronized. A design that can move away from a Region but cannot safely return is only half a recovery system.
Exercises should include a partial failure rather than only a clean Region outage. Test a degraded dependency, stale replication, and a control action performed under pressure, then measure how much traffic remains on the old path because of DNS caching or persistent connections. That produces a realistic recovery-time picture and exposes whether the operator sequence assumes a cleaner failure than production is likely to provide.
Authority to move traffic should be tightly defined. The recovery team needs named operators who can change routing controls, application owners who can declare the destination usable, and data owners who can judge consistency risk. Separating those approvals prevents a single health alarm from becoming an automatic decision to expose users to a standby Region whose application or data state is not ready.
Keep a known-good recovery snapshot that records routing-control state, expected health signals, replication status, and the capacity available in the secondary Region. That baseline helps responders distinguish a real service failure from drift in the standby environment. It should be refreshed after significant architecture changes so the recovery plan does not rely on assumptions from an older topology.
Recovery cost is mainly a question of how much capacity and data readiness the secondary Region must maintain before an incident. A warm environment shortens recovery but consumes more continuously; a thinner standby costs less but needs more actions before it can take traffic. Choose the posture from the recovery objective and test that the chosen amount of pre-provisioned capacity can actually absorb the expected load.
Failback deserves its own criteria. Do not return traffic simply because the primary Region is reachable again; first confirm data convergence, dependency health, capacity, and the impact of another routing change on connected clients. Record the sequence for restoring the original traffic path so the team does not improvise while both Regions may contain recent production activity.
Emergency access to recovery controls should be auditable and temporary. Log who changed routing state, the evidence used to authorize the action, and any bypass of normal approval steps. After the event, remove temporary permissions and reconcile the final control state with the documented architecture so an incident workaround does not silently become the new production design.
Recovery architecture should minimize improvisation
The strongest ARC design gives operators a small number of well-understood actions backed by current capacity, replication, and validation evidence. It avoids a long emergency checklist of console changes that each depend on tribal knowledge.
Resilience combines routing, control-plane independence, secondary capacity, data protection, and tested operations. ARC can improve the control surface, but successful failover still depends on a system that was designed to run somewhere else.