# Active-Active Architecture

Active-active architecture runs multiple production environments simultaneously, distributing live traffic across them so healthy environments can continue serving users when another fails.

Cloud concept. Updated 2026-10-29. Page: https://blowstack.com/tools/cloud-concepts-hub/active-active-architecture

> Active-active architecture combines availability and scalability by keeping two or more environments actively serving production requests rather than reserving one solely for recovery. Across cloud Regions, it can reduce user latency and support regional disaster recovery, but it requires deliberate traffic management, data consistency, and capacity planning.

Practice exams covering Active-Active Architecture: [ANS-C01 AWS Certified Advanced Networking - Specialty](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty), [SOA-C03 AWS Certified CloudOps Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate), [DOP-C02 AWS Certified DevOps Engineer - Professional](https://blowstack.com/practice-exams/aws-certified-devops-engineer-professional), [SAA-C03 AWS Certified Solutions Architect - Associate](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate), [SAP-C03 AWS Certified Solutions Architect - Professional](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional).

## 6 key aspects of Active-Active Architecture

Active-active combines concurrent serving, isolated environments, health-aware routing, deliberate data semantics, and tested capacity and recovery procedures.

1. **Concurrent production serving**: Two or more environments handle real requests during normal operation. Traffic shares can differ, provided the environments are genuinely active rather than waiting solely for a failure.
2. **Failure isolation and independent operation**: Each environment should retain the dependencies needed to serve its intended workload when another environment becomes unavailable. Use separate Regions when regional disaster recovery is a requirement.
3. **Health-aware traffic distribution**: Route requests according to endpoint health and application requirements such as latency or geographic placement. Include application-level readiness checks, and account for client behavior and DNS caching when estimating recovery time.
4. **Explicit data consistency**: Choose write ownership, replication, and conflict-resolution rules that match business requirements. Multi-writer operation is optional, and asynchronous replication can leave a data-loss window during failures.
5. **Survivor capacity**: Plan for healthy environments to absorb displaced requests without overload. Validate spare capacity, scaling speed, service quotas, and downstream limits through failure testing.
6. **Tested recovery and restoration**: Maintain procedures for removing failed endpoints, verifying service continuity, reconciling data, and safely restoring traffic. Keep independent backups because an active replica is not a historical recovery point.

## Cloud services for Active-Active Architecture

Route 53 or Global Accelerator provides the global traffic-management layer, directing requests to application stacks operating concurrently in multiple Regions. Elastic Load Balancing distributes requests within each Region, while EC2 Auto Scaling adjusts local compute capacity. DynamoDB global tables can supply a replicated, writable regional data layer when their consistency semantics fit the workload. These services address different parts of the architecture; none alone establishes end-to-end active-active resilience. Engineers must also configure independent dependencies, application-aware health checks, survivor capacity, replication monitoring, backups, and tested recovery procedures.

- **[Amazon Route 53](https://blowstack.com/tools/cloud-services-hub/amazon-route-53)**: Latency-based or weighted DNS routing can distribute requests across active regional endpoints, with health checks or supported target-health evaluation guiding endpoint selection. DNS caching means routing changes do not immediately redirect every client or existing connection.
- **[AWS Global Accelerator](https://blowstack.com/tools/cloud-services-hub/aws-global-accelerator)**: For standard accelerators, static anycast IP addresses provide an entry point to healthy application endpoints across Regions. Endpoint groups and traffic controls support regional distribution, but they do not replicate application state or supply recovery capacity.
- **[Amazon DynamoDB](https://blowstack.com/tools/cloud-services-hub/amazon-dynamodb)**: Global tables provide multi-Region, multi-active replication with writes accepted in each replica Region. With multi-Region eventual consistency, replication is asynchronous and concurrent item updates use last-writer-wins reconciliation, so applications must account for stale reads and conflicting writes.
- **[Amazon EC2 Auto Scaling](https://blowstack.com/tools/cloud-services-hub/amazon-ec2-auto-scaling)**: Separate regional Auto Scaling groups can adjust application capacity as normal demand changes or traffic shifts after a failure. Scaling policies must be supported by sufficient quotas, available capacity, and startup times that meet recovery objectives.

## Active-Active Architecture exam questions

Real practice exam questions on Active-Active Architecture, with the correct answer and its explanation.

### Question 1 (SOA-C03)

**Question:**

A company runs a stateless API on Amazon EC2 Auto Scaling groups in two AWS Regions. Both Regions serve production traffic concurrently. At the global peak of 1,600 requests per second, traffic is divided equally between the Regions. Each Region normally has eight running instances. Load tests show that each instance can handle at most 100 requests per second while meeting the required response time.

During a regional failure exercise, traffic shifts to the surviving Region within 30 seconds. However, response times exceed the requirement until additional instances become ready 7 minutes later. Data stores and other dependencies remain available independently in each Region.

The company requires acceptable response times to resume within 2 minutes of either Region failing, including at global peak traffic. Both Regions must continue serving production traffic during normal operation. Which change meets these requirements?

**Correct answer:** Maintain at least 16 healthy, running instances in each Region and retain automatic traffic redirection away from a failed Region.

**Explanation:** Each Region needs 16 instances to handle the full peak of 1,600 requests per second independently. Keeping that capacity running in both Regions allows the surviving deployment to accept the entire workload after the tested 30-second traffic shift without waiting for instance launches.

### Question 2 (SAP-C03)

**Question:**

A financial services company runs a stateless application on independent Amazon EKS clusters in two AWS Regions. Both clusters serve production traffic through regional load balancers. Global traffic routing, regional compute capacity, container images, and application data already support operation after the loss of either Region.

Both clusters use a Kubernetes validating admission webhook to enforce mandatory security policies when application Pods are created. The webhook runs only in the primary Region, and both clusters call its HTTPS endpoint with failurePolicy set to Fail. During a disaster recovery exercise, existing Pods in the secondary Region continued serving requests, but replacement Pods could not be created when connectivity to the primary Region was interrupted.

The company requires the surviving cluster to create replacement Pods throughout a regional outage. Every new Pod must pass the mandatory security checks; bypassing validation is prohibited. Approved policy data can be replicated and evaluated independently in either Region.

Which solution meets these requirements?

**Correct answer:** Deploy a highly available webhook service in each cluster and configure each cluster's webhook registration to reference its local service. Replicate approved policy data to each Region so validation requires no calls to the other Region. Retain failurePolicy set to Fail.

**Explanation:** Each cluster can independently validate and admit replacement Pods because its webhook and required policy data are available locally. Regional replication of approved policies removes the cross-Region runtime dependency, while failurePolicy set to Fail preserves the requirement that Pods cannot bypass validation.

### Question 3 (SAA-C03)

**Question:**

A company provides an HTTPS route-planning API to customers in North America and Europe. The API calculates routes from a read-only dataset that is copied to both AWS Regions before each release. Requests are stateless, and no request requires access to another customer's data.

The company wants to use compute capacity in both Regions for production traffic during normal operation and direct users to the Region that provides lower network latency. If either Region becomes unavailable, the other Region must serve the full peak workload without waiting for additional instances to launch. The company accepts DNS caching delays during traffic redirection.

Which architecture meets these requirements?

**Correct answer:** Deploy a complete application stack and a local dataset copy in each Region, with each stack provisioned for the full peak workload. Configure Amazon Route 53 latency alias records for the Regional load balancers and enable target health evaluation.

**Explanation:** Both Regions serve production requests, and latency routing selects the healthy Regional endpoint with lower latency for users. Each Region has its own application dependencies and sufficient provisioned capacity to handle the full workload if the other Region fails. Target health evaluation allows Route 53 to exclude an unhealthy Regional load balancer from new DNS responses.

### Question 4 (DOP-C02)

**Question:**

A company uses AWS Organizations with all features enabled. It is expanding a stateless application so that deployments in three AWS Regions will serve production traffic concurrently. Each member account in a production organizational unit (OU) requires the same Regional infrastructure.

A DevOps engineer must implement centralized infrastructure deployment with the following requirements:
• Automatically deploy the infrastructure to accounts that subsequently join the production OU.
• Avoid creating and maintaining CloudFormation execution roles separately in each member account.
• Deploy updates in the order us-east-1, eu-west-1, and ap-southeast-2. A Region's deployment must finish successfully before deployment begins in the next Region.
• Update only one account at a time within each Region. If an account deployment fails, stop the operation without starting deployments in subsequent Regions.

Which solution meets these requirements?

**Correct answer:** Enable trusted access for CloudFormation StackSets in AWS Organizations. Create a service-managed StackSet targeting the production OU, and enable automatic deployment. Use sequential Region concurrency with the specified Region order, MaxConcurrentCount of 1, and FailureToleranceCount of 0.

**Explanation:** Service-managed StackSets use Organizations integration to manage the required permissions and can automatically deploy stack instances when accounts join a targeted OU. Sequential Region execution honors the specified order. A maximum concurrency of one updates one account at a time, and zero failure tolerance stops the operation after the first account failure, preventing deployment in subsequent Regions.

### Question 5 (ANS-C01)

**Question:**

A company operates an internet-facing application in us-east-1 and eu-west-1. Each Region has separate stable-release and canary-release endpoints with fixed public IPv4 addresses. All four endpoints actively serve production traffic. Each endpoint exposes an HTTPS readiness URL that reports whether the application and its required dependencies can serve requests.

A network engineer must configure Amazon Route 53 so that DNS resolution first selects the healthy Region that provides the lowest latency according to Route 53 latency measurements. Within that Region, approximately 90% of weighted DNS selections must return the stable endpoint and 10% must return the canary endpoint when both are healthy. An unhealthy endpoint must be excluded while its regional peer remains healthy. If both endpoints in one Region become unhealthy, new DNS resolutions must select the other Region, assuming its endpoints remain healthy. These requirements apply after health-check detection and DNS cache expiration.

Which configuration meets these requirements?

**Correct answer:** Create latency alias records for the application name, one for each Region, with Evaluate Target Health enabled. Point each alias to a regional name containing weighted A records for the stable and canary endpoints, with weights of 90 and 10. Associate each weighted record with a Route 53 HTTPS health check of its endpoint's readiness URL.

**Explanation:** The latency alias records select the lowest-latency healthy regional branch. The regional weighted records then select between healthy stable and canary endpoints using the configured weights. Endpoint health checks exclude an unhealthy leaf, and Evaluate Target Health propagates the regional branch's health to the latency records. If all endpoints in one branch are unhealthy, Route 53 can select the other healthy regional branch.

## Practice Active-Active Architecture in a full exam

Questions on Active-Active Architecture in each practice exam, with every answer option and an explanation for each:

- [DOP-C02](https://blowstack.com/practice-exams/aws-certified-devops-engineer-professional): 4 questions
- [SOA-C03](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate): 2 questions
- [SAP-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional): 2 questions
- [SAA-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate): 1 question
- [ANS-C01](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty): 1 question

## What is Active-Active Architecture

Active-active architecture means that multiple environments serve production traffic at the same time. These environments can span Availability Zones, data centers, or cloud Regions. The defining characteristic is concurrent production service, not equal traffic distribution: one environment might handle 70 percent of requests while another handles 30 percent. For regional disaster recovery, the architecture must place the required application components and dependencies in separate Regions; distributing instances across Availability Zones alone does not protect against a regional disruption.

A typical cloud implementation deploys independently operable application stacks in two or more Regions. Global traffic management directs users to healthy endpoints using latency, geographic, or weighted routing. Regional load balancers distribute requests across local application instances, while automatic scaling adjusts capacity. Stateless application servers simplify this design because a request can reach another environment without depending on memory or files stored on the original server. Session state must be accessible wherever requests are served, or the application must tolerate session loss.

Data design is usually the hardest part. Active-active application servers do not necessarily require a database that accepts writes in every Region. They can use a single writer, although that writer introduces a dependency whose failure and recovery must be addressed. A multi-writer data layer allows local writes but requires explicit consistency and conflict-handling rules. With asynchronous replication, acknowledged changes might not reach another Region before a failure. Concurrent updates can also produce conflicts. Applications handling inventory, payments, or account balances must evaluate these consequences rather than assuming replication guarantees correctness.

For example, a global product catalog might serve reads from local data replicas while accepting catalog updates through a controlled write path. A shopping-cart service might accept writes in several Regions using a replicated data store, with application rules for merging concurrent changes. Retried checkout requests should carry idempotency keys so traffic rerouting does not create duplicate orders. Each design should document which operations remain available during a network partition and what consistency users can expect.

Compared with backup and restore, pilot light, and warm standby, active-active keeps multiple environments serving live traffic before a disaster. This can reduce recovery time because another production environment is already operating, but it does not guarantee zero recovery time or zero data loss. Health detection, DNS caching, client retries, replication behavior, and available capacity all affect the outcome. Recovery time objective describes the targeted time to restore service; recovery point objective describes the acceptable data-loss window.

Operational procedures should cover failure detection, traffic withdrawal, capacity expansion, validation of critical transactions, data reconciliation, and controlled traffic restoration. If two Regions each normally handle half the workload, the remaining Region must have sufficient headroom or scale quickly enough to absorb the displaced traffic. Test regional loss and dependency failures rather than testing only individual instances. Monitor replication lag, end-to-end transaction success, and saturation in each environment.

The trade-offs include duplicated infrastructure, cross-Region transfer costs, more complex deployments, and harder troubleshooting. Shared identity, configuration, networking, or deployment systems can still cause correlated failures. Independent backups and tested restores remain necessary because replication can propagate accidental deletion or corruption. Active-active is therefore both a serving strategy and an operational discipline, not simply a routing setting.

## Learn Active-Active Architecture in lessons

5 short lessons on Active-Active Architecture in 4 certification courses.

- SOA-C03: Active-Active Architecture: continuity through regional failure (Premium), https://app.blowstack.com/lessons/active-active-architecture-continuity-through-regional-failure-soa-c03
- SAP-C03: Active-Active Architecture: regional independence for container clusters (Premium), https://app.blowstack.com/lessons/active-active-architecture-regional-independence-for-container-clusters-sap-c03
- SAA-C03: Active-Active Architecture: multi-Region serving topology (Premium), https://app.blowstack.com/lessons/active-active-architecture-multi-region-serving-topology-saa-c03
- DOP-C02: Active-Active Architecture: multi-Region serving (Premium), https://app.blowstack.com/lessons/active-active-architecture-multi-region-serving-dop-c02
- SOA-C03: Active-Active Architecture vs Warm Standby (Premium), https://app.blowstack.com/lessons/active-active-architecture-vs-warm-standby-soa-c03
