# Failover

Failover is the transfer of traffic or processing from a failed or unhealthy resource to a healthy alternative so a service can continue operating.

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

> Failover uses redundant resources, health detection, and a mechanism for transferring work to reduce interruptions when components fail. In cloud architectures, distributing those resources across multiple Availability Zones helps maintain availability when an instance or an entire zone becomes unavailable.

Practice exams covering Failover: [ANS-C01 AWS Certified Advanced Networking - Specialty](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty), [AIB-C01 (beta) AWS Certified AI Business Strategist](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist), [CLF-C02 AWS Certified Cloud Practitioner](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner), [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), [MLA-C02 (beta) AWS Certified Machine Learning Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate), [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 Failover

Effective failover combines independent redundant resources, reliable health detection, sufficient capacity, consistent data handling, and tested recovery procedures.

1. **Failure detection and transfer**: Health checks or monitoring identify when a resource cannot serve its workload, and routing or orchestration moves work to an alternative. Configure thresholds to avoid both slow recovery and unnecessary failovers.
2. **Independent failure domains**: Place redundant resources across multiple Availability Zones to reduce dependence on one zone's infrastructure. Verify that databases and other critical dependencies have equivalent protection.
3. **Ready capacity**: Surviving resources must have enough capacity to accept transferred work. Reserve headroom or maintain suitable standby capacity rather than relying entirely on resources that must be provisioned after failure.
4. **State and data continuity**: Replicate required data and avoid storing essential session state only on a single application instance. Understand replication lag, write ownership, and the possibility of data loss during promotion.
5. **Application recovery behavior**: Applications must reconnect after endpoint or connection changes and use bounded retries with backoff. Make retried writes idempotent where possible so recovery does not create duplicate transactions.
6. **Testing and controlled failback**: Exercise failover under realistic load and measure recovery time and data loss against objectives. Return traffic to recovered resources only after validating their health and data consistency.

## Cloud services for Failover

Elastic Load Balancing directs application traffic to healthy targets distributed across Availability Zones, while Amazon EC2 Auto Scaling maintains and replenishes the underlying instance capacity. Amazon RDS Multi-AZ deployments provide database redundancy and managed failover, but applications still need connection recovery and safe retry logic. Amazon Route 53 can provide DNS-level failover between separate endpoints, including deployments in different Regions when the architecture supports it. Together, these services address different layers of availability; they still require compatible data replication, sufficient surviving capacity, and tested end-to-end recovery procedures.

- **[Amazon RDS](https://blowstack.com/tools/cloud-services-hub/amazon-rds)**: Amazon RDS supports managed database failover through Multi-AZ deployments. In a Multi-AZ DB instance deployment, RDS synchronously replicates to a standby in another Availability Zone and can automatically fail over to it; that standby does not serve read traffic.
- **[Amazon EC2 Auto Scaling](https://blowstack.com/tools/cloud-services-hub/amazon-ec2-auto-scaling)**: Amazon EC2 Auto Scaling maintains desired instance capacity and replaces unhealthy instances. An Auto Scaling group spanning multiple Availability Zones complements traffic failover by maintaining and replenishing compute capacity, although instance launch and application startup take time.
- **[Amazon Route 53](https://blowstack.com/tools/cloud-services-hub/amazon-route-53)**: Amazon Route 53 supports DNS failover routing that can return a secondary endpoint when the primary is considered unhealthy. Recovery is affected by health evaluation, DNS caching, and client behavior, so it does not immediately move every client or preserve existing connections.

## Failover exam questions

Real practice exam questions on Failover, with the correct answer and its explanation.

### Question 1 (AIB-C01 (beta))

**Question:**

A document-processing company is replacing manual document classification with an AI-based process. Contracts require classification to resume within two minutes if the active AI deployment stops working. The company has a secondary deployment in a separate failure location, with approved data access and tested capacity for the full workload. Application routing can switch between deployments automatically.

During a disruption exercise, the active deployment's infrastructure health checks remain successful even though its model cannot produce classifications. Traffic continues going to that deployment, and processing stops for 12 minutes until an operator intervenes.

What should the AI business strategist require before approving the transition?

**Correct answer:** Require end-to-end classification health checks that trigger automatic routing to the healthy secondary deployment, and demonstrate recovery within two minutes in disruption tests.

**Explanation:** The secondary deployment and routing mechanism are already available, but failure detection does not reflect useful business operation. Functional checks, automatic transfer, and timed disruption tests address the demonstrated gap and provide evidence that the continuity requirement can be met.

### Question 2 (MLA-C02 (beta))

**Question:**

A company runs a stateless ML inference API on Amazon EC2 instances in an Auto Scaling group across two Availability Zones. An Application Load Balancer (ALB) distributes requests to the instances. Each Availability Zone has enough running capacity to handle the full production workload.

Occasionally, an instance's model-serving process becomes unresponsive, but its web proxy continues to return HTTP 200 for the ALB's static /health page. EC2 status checks also continue to pass. Requests routed to the affected instance time out. Loading the model onto a replacement instance takes 8 minutes.

The company needs to automatically redirect new requests away from affected instances to healthy, already-running instances and replace the affected instances. Which configuration meets these requirements?

**Correct answer:** Configure the ALB to check an application health endpoint that succeeds only when the model-serving process can complete a lightweight inference within a timeout. Enable Elastic Load Balancing health checks for the Auto Scaling group.

**Explanation:** The application-aware check detects instances that cannot perform inference. After the unhealthy threshold is reached, the ALB stops routing new requests to those instances and uses the healthy instances with sufficient capacity in the other zone. Enabling Elastic Load Balancing health checks allows the Auto Scaling group to replace instances that the load balancer reports as unhealthy.

### Question 3 (SAP-C03)

**Question:**

A company operates a document-processing API in two AWS Regions. Amazon Route 53 failover records direct clients to the active Region or a fully provisioned standby Region. Each record has an associated HTTPS health check with response-body string matching. The company is preparing a failover exercise to measure recovery time through the public API hostname.

During baseline testing, Route 53 reports the active endpoint as unhealthy even though document-processing requests succeed. The health check requests /diagnostics and searches for READY_TO_SERVE. Packet captures show the following behavior consistently across all health checker locations:

• The endpoint returns HTTP 200 within 100 milliseconds.
• The response body arrives 2.4 seconds after the HTTP status code.
• The search string starts at byte 5,115 and ends at byte 5,128 of the response body.

The endpoint includes READY_TO_SERVE only when the application and its required dependencies can process documents. It returns HTTP 200 even when that string is absent. The standby endpoint has the same implementation.

The company must eliminate these baseline health check failures without weakening application-readiness detection. The existing DNS TTL, health check interval, and failure threshold must remain unchanged so that recovery-time measurements are comparable with previous exercises.

Which change should a solutions architect recommend before repeating the exercise?

**Correct answer:** Modify both Regional endpoints to return a compact readiness response within one second after the HTTP status code. Include READY_TO_SERVE entirely within the first 1,024 bytes only when the application and its required dependencies can process documents. Keep string matching enabled.

**Explanation:** This fixes both causes of the baseline failures without weakening readiness detection. Route 53 requires the response body within two seconds after the status code, and the entire search string must appear within the first 5,120 bytes. A compact, promptly delivered response satisfies both constraints while preserving the meaning of the readiness marker.

### Question 4 (CLF-C02)

**Question:**

A company runs a stateless web application on two Amazon EC2 instances in different Availability Zones within one AWS Region. Either instance has enough capacity to handle all requests. Currently, users connect directly to one instance, and an administrator changes the application's DNS record if that instance fails. The company wants to distribute requests across both instances and automatically stop sending new requests to an unhealthy instance. Which solution best meets these requirements?

**Correct answer:** Use Elastic Load Balancing with an Application Load Balancer enabled in both Availability Zones, register both instances, configure health checks, and direct users to the load balancer.

**Explanation:** The load balancer distributes requests across healthy registered instances in both Availability Zones. Health checks allow it to stop routing new requests to an unhealthy instance and use the healthy instance instead.

### Question 5 (SOA-C03)

**Question:**

A company runs an order-management database on an Amazon RDS for PostgreSQL Multi-AZ DB instance deployment. Reporting queries are increasing the load on the primary DB instance.

A CloudOps engineer must redesign the database deployment to provide two readable replicas in separate Availability Zones and automatic failover if the writer's Availability Zone becomes unavailable. The same replicas that serve reporting queries must be eligible targets for automatic writer failover. The application must use an AWS-managed writer endpoint and an AWS-managed reader endpoint. The company must retain RDS for PostgreSQL, but migration to a new deployment is permitted. The required deployment types are supported in the company's Region and PostgreSQL version.

Which solution meets these requirements?

**Correct answer:** Migrate to an RDS for PostgreSQL Multi-AZ DB cluster with one writer and two readable instances distributed across three Availability Zones. Use the cluster endpoint for writes and the reader endpoint for reporting.

**Explanation:** An RDS Multi-AZ DB cluster provides a writer and two readable instances across three Availability Zones. The readable instances serve read traffic and are eligible for automatic promotion if the writer fails. The cluster provides managed writer and reader endpoints.

### Question 6 (DOP-C02)

**Question:**

A company runs a transaction-processing application on an Amazon Aurora PostgreSQL global database with one primary DB cluster and one secondary DB cluster in another AWS Region. Both clusters have provisioned DB instances and use an engine version that supports managed global database failover and recovery point objective enforcement.

During a network impairment, the primary cluster continues acknowledging transactions while cross-Region replication falls several minutes behind. The company requires the database to enforce a maximum cross-Region recovery point objective (RPO) of 60 seconds. If replication cannot maintain this objective, delaying transaction commits in the primary Region is acceptable. Regional recovery will use a managed failover to the existing secondary cluster. The company does not want custom application write-gating logic.

Which change should a DevOps engineer implement?

**Correct answer:** Configure the primary cluster's rds.global_db_rpo parameter to 60 seconds in its DB cluster parameter group.

**Explanation:** Aurora PostgreSQL global databases support RPO enforcement through rds.global_db_rpo. With one secondary Region, Aurora can block primary transaction commits when necessary until that secondary catches up sufficiently to satisfy the configured RPO. This provides database-level enforcement and explicitly trades primary write availability for the required data protection.

### Question 7 (SAA-C03)

**Question:**

A company runs private IPv4-only EC2 workers in three Availability Zones in one VPC. Availability Zones A and B each have a zonal public NAT gateway that must remain for other workloads. Workers in Availability Zone C currently share the NAT gateway in Availability Zone A. Both NAT gateways have sufficient capacity for all workers, and their public subnets have routes to an internet gateway.

If Availability Zone A fails, workers in Availability Zone C must automatically recover outbound internet access. A brief interruption and reconnection of existing sessions are acceptable. The company has an existing Regional automation platform that can detect egress failures and call Amazon VPC APIs from outside the affected Availability Zone. Its incremental cost is negligible. Worker traffic is low enough that cross-AZ data transfer costs are much lower than the fixed cost of another NAT gateway.

Which solution meets the recovery requirement with the LOWEST additional recurring network cost?

**Correct answer:** Configure the existing automation to detect loss of egress through Availability Zone A and replace the default route in the Availability Zone C workers' route table with a route targeting the existing NAT gateway in Availability Zone B.

**Explanation:** The automation supplies health detection and redirects new connections to a healthy NAT gateway outside the failed Availability Zone. It reuses existing capacity without adding NAT gateway hourly charges. The brief interruption and loss of existing NAT sessions are permitted, and the stem establishes that cross-AZ traffic costs are lower than another gateway's fixed cost.

### Question 8 (ANS-C01)

**Question:**

A company operates an application on three public IPv4 endpoints in different AWS Regions. Each endpoint can handle the entire workload. The company uses Amazon Route 53 and wants approximately 80% of DNS selections to return endpoint A and 20% to return endpoint B while both are healthy. Endpoint C must receive no selections while A or B is healthy, but must receive the selections when both are unhealthy and C is healthy. Operations must also receive Route 53 health-check status for every endpoint, including C while it is on standby, so that loss of standby readiness is detected before an active endpoint fails. Each endpoint's HTTPS /ready path returns HTTP 200 when the application can serve requests and HTTP 503 otherwise, although its TCP port remains open. These routing requirements apply when at least one endpoint is healthy, after health checks have converged and cached DNS answers have expired. Which configuration meets these requirements?

**Correct answer:** Create weighted A records for the application name with weights of 80, 20, and 0 for A, B, and C, respectively. Associate each record with an HTTPS health check of its endpoint's /ready path.

**Explanation:** Route 53 distributes selections among healthy nonzero-weight records according to their relative weights. When all nonzero-weight records are unhealthy, it considers zero-weight records, allowing healthy C to serve as the standby. Independent HTTPS /ready health checks detect application readiness for all three endpoints, including C while it receives no application traffic. Route 53 can fail open when all records are unhealthy, so this does not guarantee exclusion of every unhealthy endpoint in that situation; the stem explicitly limits its routing requirements to cases with a healthy endpoint.

## Practice Failover in a full exam

Questions on Failover 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): 24 questions
- [SAP-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional): 23 questions
- [SAA-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate): 13 questions
- [ANS-C01](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty): 13 questions
- [SOA-C03](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate): 12 questions
- [AIB-C01 (beta)](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist): 3 questions
- [CLF-C02](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner): 2 questions
- [MLA-C02 (beta)](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate): 1 question

## What is Failover

Failover is a recovery mechanism that transfers responsibility from an unavailable or unhealthy component to another component capable of doing its work. The component might be an application server, database, network endpoint, or an entire deployment. Failover supports high availability by reducing the time a service remains unavailable after a failure. High availability is the objective; failover is one mechanism used to achieve it.

A practical failover design requires more than a spare server. It needs a way to detect failure, a healthy alternative with sufficient capacity, and a mechanism to redirect traffic or transfer processing responsibility. Health checks should reflect whether a component can perform useful work, not merely whether its operating system responds. Detection thresholds must balance fast recovery against unnecessary transfers caused by temporary network problems or brief performance slowdowns.

In AWS, a Region contains multiple Availability Zones, and each Availability Zone consists of one or more discrete data centers with independent power, networking, and connectivity. Deploying application instances across multiple Availability Zones reduces exposure to a single zone failure. For example, a web application can run instances in two zones behind a load balancer. If instances in one zone become unhealthy, the load balancer can direct new requests to healthy instances in the other zone. Multiple instances located only in one zone do not provide the same protection against a zone-wide disruption.

Application availability also depends on the data tier and other dependencies. A web tier distributed across zones can still become unavailable if it depends on a single database without a recovery alternative. A database failover design typically replicates data to another database resource and promotes or selects an eligible replica when the primary fails. Applications must handle broken connections, reconnect to the appropriate endpoint, and retry operations safely. Keeping user sessions only in an individual application server's memory can also interrupt users when traffic moves elsewhere.

Active-passive designs keep an alternative ready to take over, while active-active designs serve traffic from multiple resources and shift work to surviving resources after a failure. Either approach requires enough surviving capacity to handle the expected workload. Automatic scaling can replenish capacity, but provisioning takes time and should not be treated as instant protection against a sudden zone failure.

Failover involves cost and consistency trade-offs. Redundant capacity, data replication, monitoring, and testing increase expense and operational complexity. Asynchronous replication can leave a promoted replica missing recent writes. Synchronous replication can reduce that risk but adds coordination and latency. Recovery time objective describes the target time to restore service, while recovery point objective describes the acceptable amount of data loss measured in time.

Failover does not necessarily preserve existing connections, guarantee zero downtime, or protect against every failure. Multi-zone deployment addresses zone-level failures within a Region; protection against a regional outage requires a separate cross-Region strategy. Backups support restoration but do not, by themselves, provide a ready failover destination. Test failure scenarios regularly, prevent multiple nodes from incorrectly acting as the sole writer, and plan failback—the controlled return to the preferred resource after recovery.

## Learn Failover in lessons

16 short lessons on Failover in 7 certification courses.

- AIB-C01 (beta): Failover: switching to a healthy alternative (Premium), https://app.blowstack.com/lessons/failover-switching-to-a-healthy-alternative-aib-c01
- MLA-C02 (beta): Failover: capacity for Availability Zone failures (Premium), https://app.blowstack.com/lessons/failover-capacity-for-availability-zone-failures-mla-c02
- SAP-C03: Failover: content origin fallback (Premium), https://app.blowstack.com/lessons/failover-content-origin-fallback-sap-c03
- CLF-C02: Failover: transferring work across Availability Zones (Premium), https://app.blowstack.com/lessons/failover-transferring-work-across-availability-zones-clf-c02
- SOA-C03: Failover: transferring work across Availability Zones (Premium), https://app.blowstack.com/lessons/failover-transferring-work-across-availability-zones-soa-c03
- DOP-C02: Backup and Restore vs Failover (Premium), https://app.blowstack.com/lessons/backup-and-restore-vs-failover-dop-c02
