Amazon Application Recovery Controller (ARC) supports recovery through zonal shifts and regional routing controls. Zonal shifts avoid one Availability Zone; routing controls change Route 53 health checks to switch traffic between application replicas.
Zonal shift
ARC zonal shiftA zonal shift temporarily moves a supported resource traffic away from one Availability Zone. temporarily moves a supported resource’s traffic away from one Availability ZoneAn Availability Zone is an isolated location within an AWS Region., an isolated location within a Region. It lets the application use the other zones while a zonal problem is investigated.
How it works
- Enable zonal shift on the ALB, where it is disabled by default. Check that the resource is active, fully provisioned, opted in, and in the Region and zone being shifted away from.
- Prescale the remaining zones, meaning provision capacity before the shift rather than waiting for demand-driven scaling.
- Start a shift for one zone and set an expiration. For ALB with cross-zone balancing enabled, the zonal addresses leave DNSDNS resolves application names into addresses that clients can connect to., the system that resolves names to addresses, and targets in the shifted zone stop receiving new traffic.
- ALB records an AdministrativeOverrideAn administrative override records deliberate target exclusion separately from target health. separately from TargetHealth. A healthy target can show zonal_shift_active because it is deliberately excluded. Existing connections stay intact until they close, so DNS caching and connection reuse affect how quickly all traffic leaves.
- Extend, cancel, or let the shift expire. Canceling or expiration makes the zone available again and traffic starts returning.
- For AWS-initiated shifts, configure zonal autoshiftZonal autoshift authorizes AWS to move configured resource traffic when AWS telemetry detects a potential zonal impairment. and its required practice runsA practice run tests whether an application can operate with resource traffic shifted away from one zone.. AWS uses internal telemetryTelemetry is measurement data AWS uses to detect a potential zonal impairment., measurements of AWS infrastructure, rather than testing individual resource health.
| Mode | Trigger | End condition |
|---|---|---|
| Manual zonal shift | Operator starts a shift | Expiration or cancellation |
| Zonal autoshift | AWS detects a potential zonal impairment | AWS determines the zone has recovered |
| Practice run | Required weekly autoshift test | About 30 minutes for one resource |
ALB added zonal shift and autoshift support with cross-zone balancing enabled in November 2024. Older launch guidance requiring cross-zone balancing to be off predates this change.
Exam cues
- If the question says one ALB zone is impaired, think zonal shift, because it moves new traffic to other zones in the same Region.
- If the question says ALB cross-zone balancing is enabled, think zonal shift is still supported, because ALB can administratively exclude the affected zone’s targets.
- If the question says AWS should trigger evacuation, think zonal autoshift, because AWS telemetry starts the shift.
- If the question says all load balancer targets are unhealthy and a shift has no effect, think fail openFail open is a load balancer state in which lack of healthy targets prevents zonal shift from moving traffic away., because zonal shift cannot force a move in that state.
Remember
- ALB and NLB support zonal shift with cross-zone balancing on or off.
- One zone per load balancer shift.
- Initial manual expiration: one minute to three days, or 72 hours; extendable.
- Autoshift requires weekly practice runs lasting about 30 minutes per resource.
- Autoshift and practice runs do not wait for auto scaling.
Trap: Administrative exclusion and health are separate. A healthy ALB target can be excluded by zonal shift, while existing connections remain until they close.
Enable the resource and start a shift away from one zone.
With remaining healthy capacity, ALB sends new traffic to targets outside the shifted zone.
From the AWS docs (19)
- ARC zonal shift moves a supported resource traffic away from one impaired Availability Zone to healthy zones in the same Region.
- ALB zonal shift is disabled by default and must be enabled on each load balancer before use.
- ARC supports zonal shift for ALB and NLB with cross-zone load balancing enabled or disabled.
- The shifted resource must be active, fully provisioned, opted in, and deployed in the Region and zone being shifted away from.
- Before a zonal shift, prescale capacity so the remaining zones can handle the application traffic.
- A zonal shift for a specific load balancer can affect only one Availability Zone at a time.
- A manual ARC zonal shift requires an expiration initially from one minute to three days, or 72 hours; it can be extended or canceled.
- When a zonal shift expires or is canceled, the zone becomes available to the resource and traffic begins returning.
- For an ALB with cross-zone balancing enabled, zonal shift removes zonal IP addresses from DNS and stops new traffic to targets in the affected zone.
- ALB zonal shift sets AdministrativeOverride independently from TargetHealth; zonal_shift_active identifies targets in the shifted zone.
- ALB zonal shift leaves existing connections intact until they close; DNS settings and connection reuse can delay complete traffic movement.
- Zonal shift has no effect when an ALB or NLB is failing open, such as when load balancer target groups have no instances or all instances are unhealthy.
- Zonal autoshift authorizes AWS to shift a configured resource away from a zone when AWS internal telemetry detects a potential zonal impairment.
- Zonal autoshift does not inspect individual resource health and can shift a resource that is not itself experiencing impact.
- Zonal autoshift requires practice runs; they occur weekly and move one resource traffic away from a zone for about 30 minutes.
- Zonal autoshift and practice runs do not wait for auto scaling; preprovision sufficient capacity before enabling them.
- AWS ends a zonal autoshift when the affected Availability Zone recovers.
- An Availability Zone is an isolated location within an AWS Region.
- ALB support for zonal shift and autoshift with cross-zone balancing enabled launched in November 2024.
Routing controls
ARC routing controls let an operator or automation switch DNSDNS resolves application names into addresses that clients can connect to. traffic between application replicasAn application replica is a deployment that can serve the workload in another zone or Region.. A replica is another deployment of the application that can serve the same workload. Each control is an On or Off switch connected to a Route 53 health check; it does not probe the application itself.
How it works
-
Create a routing control cluster, which supplies five redundant regional endpoints. An endpoint is an address used to call the API. The cluster includes a default control panelA control panel groups related routing controls so they can share safety rules., a grouping of related switches.
-
Create routing controls and their Route 53 health checks. Associate the checks with DNS records for the application replicas, such as primary and standby failover records.
-
Add safety rules to the control panel before failover.
Component Purpose Routing control Set the associated health check state through an On or Off switch Assertion ruleAn assertion rule rejects a routing control update when its required state conditions would be violated. Reject changes unless a condition such as at least one control On remains true Gating ruleA gating rule uses a master control to allow or block changes to target routing controls. Permit or block changes to target controls using a master switch -
Use the data planeThe routing control data plane reads and updates the live routing control states used for failover., the API operations that read and update live control states, to set the standby On and primary Off. Multiple state changes can be submitted together. The configuration control planeThe routing control configuration control plane creates or deletes clusters, panels, and routing controls. creates or deletes the resources in advance.
-
Route 53 uses the changed health checks with the DNS records to direct clients to the eligible replica. DNS TTLTTL specifies how long a DNS answer can be cached before it must be refreshed., the time a DNS answer can be cached, and existing connection reuse affect how quickly every client moves.
-
Keep local copies of the five endpoints and routing control ARNs, their AWS resource identifiers. During failover, use the data plane API and retry other endpoints if one is unavailable.
Readiness checksA readiness check audits application replicas for capacity and configuration differences that could affect failover. have a different role: they audit replica capacity, configuration, quotas, and routing policies before an event. Use application monitoring to decide whether to fail over; readiness checks are not the primary disaster trigger.
Exam cues
- If the question says operator-controlled regional DNS failover, think routing controls, because their switches change health checks linked to application DNS records.
- If the question says keep at least one replica enabled, think an assertion rule, because it rejects unsafe state combinations.
- If the question says freeze automated failover changes with one switch, think a gating rule, because it controls permission to update other controls.
- If the question says an endpoint is unavailable during a disaster, think retry another saved cluster endpointA cluster endpoint is a regional API address used to read or update routing control states., because the data plane has five regional endpoints.
Remember
- Routing control state: On or Off.
- Cluster: five redundant regional endpoints.
- Control panel: related routing controls and safety rules.
- Configure before the event; use data plane state updates during recovery.
- Readiness checks audit preparedness; routing controls change traffic eligibility.
Trap: A routing control health check reflects a switch state. It does not independently detect application failure or remove the need for monitoring and a failover decision.
Change primary and standby routing control states through a cluster endpoint.
Associated Route 53 health checks let DNS failover records select the standby replica.
From the AWS docs (17)
- ARC routing controls are On or Off switches that change Route 53 health checks associated with application DNS records.
- ARC routing controls can fail over DNS traffic between application replicas in different Regions.
- A routing control does not probe endpoint health or measure response time; its configured state controls the associated health check.
- An ARC routing control cluster provides five redundant regional endpoints for getting and updating routing control states.
- A control panel groups related routing controls; creating a cluster also creates a default control panel.
- Create an ARC cluster before creating its control panels and routing controls, then associate routing control health checks with Route 53 DNS records.
- ARC configuration control plane operations create or delete routing control resources, while the routing control data plane gets, lists, and updates their states.
- Switching the primary routing control Off and a standby control On changes associated health checks to reroute DNS traffic to the standby application replica.
- ARC supports updating multiple routing control states together so one replica can be turned off while another is turned on.
- An assertion safety rule rejects a routing control state update unless its configured conditions remain true, such as at least one control remaining On.
- A gating safety rule uses control states as a master switch that permits or blocks changes to target routing controls.
- AWS recommends keeping local copies of all five regional cluster endpoints and routing control ARNs before a failure.
- Failover automation should try other ARC regional cluster endpoints when one is unavailable and be able to use all five.
- AWS recommends using the ARC data plane API to get or update routing control states during failover rather than relying on the console.
- ARC routing control failover is affected by DNS settings and existing connection reuse, so traffic might not move immediately.
- Lower DNS TTL values and limited client connection lifetimes help traffic move sooner during routing control failover.
- ARC readiness checks audit replica capacity, configuration, quotas, and routing policies; they should not be the primary failover trigger during a disaster.
Flashcards
Nothing to review here right now. Pick other topics or another filter.
·
All 17 cards as a list
One Availability Zone behind an ALB is impaired. Which ARC feature can move new traffic away while the other zones keep serving?
Start a zonal shift for the opted-in load balancer after checking spare zonal capacity. ALB supports this with cross-zone balancing enabled or disabled. Existing connections can take time to drain, and the shift needs an expiration of up to 72 hours initially.
An ARC zonal shift started for a load balancer, but traffic still reaches the impaired zone and all target groups lack healthy instances. Why?
The load balancer can be failing open. Zonal shift has no effect in that state.
Must an ALB disable cross-zone balancing before an ARC zonal shift?
No. ALB supports zonal shift with cross-zone balancing enabled or disabled. With it enabled, zonal addresses leave DNS and new traffic to targets in the shifted zone stops.
An ALB target is healthy but shows AdministrativeOverride zonal_shift_active. Why does it stop receiving new traffic?
A zonal shift administratively excludes it. AdministrativeOverride is independent of TargetHealth, and existing connections can remain until they close.
What initial expiration can a manual ARC zonal shift use?
From one minute to three days, or 72 hours. Extend the expiration if needed or cancel early to let traffic return.
A team wants AWS to shift traffic when AWS detects a zonal impairment. Which ARC option fits?
Zonal autoshift. It uses AWS telemetry and requires weekly practice runs; it does not inspect individual resource health.
Can a team rely on auto scaling to finish before an ARC autoshift or practice run starts?
No. Autoshift does not wait for scaling. Preprovision capacity so the application can serve traffic without one zone.
How often do ARC zonal autoshift practice runs occur, and how long does each traffic shift last?
Weekly, with traffic for one resource shifted away from a zone for about 30 minutes.
An application has primary and standby stacks in separate Regions and needs an operator-controlled DNS failover switch. Which ARC feature fits?
Routing controls linked to Route 53 health checks and DNS records. Set the primary Off and standby On to change DNS traffic eligibility.
Does an ARC routing control decide its On or Off state by probing application response times?
No. It is a switch controlling an associated Route 53 health check, not an endpoint health monitor.
How many regional endpoints does an ARC routing control cluster provide?
Five redundant regional endpoints for reading and changing routing control states.
An ARC failover script cannot reach its first cluster endpoint. What should it do?
Try another regional endpoint, using locally saved endpoints and ARNs. The retry process should be able to use all five endpoints.
An ARC operator must prevent turning off every application replica. Which safety rule fits?
An assertion rule requiring at least one routing control to remain On. An update that violates the condition is rejected.
An ARC operator wants a master switch that freezes automated changes to several routing controls. Which safety rule fits?
A gating rule. It permits or blocks updates to the target routing controls based on the configured gating control states.
Which ARC API plane should an outage runbook use to update routing control states?
The data plane. Configure resources in advance through the control plane, then use saved cluster endpoints to read and update states during recovery.
An ARC routing control failover succeeded but some clients still use the old Region. What should the team inspect?
DNS TTL and client connection reuse. DNS-based failover and existing connections can delay complete movement.
Should an ARC readiness check be the primary trigger for regional failover during a disaster?
No. It audits preparedness, including capacity and configuration, and complements application monitoring rather than serving as the primary failover trigger.