AWS Global Accelerator handles network outages through several integrated mechanisms designed to maintain high availability, fault tolerance, and seamless failover across AWS Regions and network zones.
**Fault Isolation with Network Zones
When you create an accelerator, AWS Global Accelerator allocates two static IPv4 addresses serviced by independent network zones. These network zones are isolated physical infrastructures with unique IP subnets, similar to Availability Zones. If one static IP address becomes unavailable due to network disruptions or IP address blocking, client applications can automatically retry using the healthy static IP address from the other network zone. This fault-isolating design ensures that even if one network zone experiences an outage, the other can continue to serve traffic without interruption[1][8].
**Global Performance-Based Routing and Instant Failover
Global Accelerator continuously monitors the health of all application endpoints (such as load balancers or EC2 instances) across multiple AWS Regions. If an endpoint or an entire region becomes unhealthy or unavailable, the service instantly reroutes traffic to the next closest healthy endpoint in another AWS Region. This failover happens automatically and typically within less than one minute, minimizing downtime and maintaining application availability during regional outages[3][6][7][9].
**TCP Termination at the Edge for Faster Recovery
Global Accelerator terminates TCP connections at the AWS edge locations closest to the client, then establishes a separate optimized TCP connection to the application endpoint in the AWS Region. This architecture reduces connection setup time and improves responsiveness. In case of endpoint failure, new connections are quickly routed to healthy endpoints without waiting for long TCP timeouts, aiding in faster recovery from network issues[1][5].
**Failover Behavior and Traffic Management
If no healthy endpoints with a positive weight are found in the primary endpoint group, Global Accelerator attempts failover to healthy endpoints in other endpoint groups, ignoring traffic dial settings to ensure availability. If no healthy endpoints are found in the closest three endpoint groups, it "fails open" by routing traffic to a random endpoint in the closest group to the client. Once the original endpoints recover, Global Accelerator resumes normal routing to healthy endpoints with configured traffic weights[6].
**Health Checks and Continuous Monitoring
Global Accelerator performs continuous health checks on endpoints to detect failures promptly. It directs new connections only to healthy endpoints, while existing connections continue until idle timeout or reset. This approach ensures minimal disruption to active sessions during failover events[5][6].
**Protection and Resilience
By masking application origins behind static IP addresses and leveraging AWS Shield for DDoS protection, Global Accelerator enhances security and resilience against network attacks that could cause outages. It also uses the AWS global network, which is congestion-free and highly redundant, to route traffic optimally and avoid network bottlenecks or failures[1][8].
In summary, AWS Global Accelerator handles network outages by isolating faults at the network zone level, continuously monitoring endpoint health, and instantly rerouting traffic to healthy endpoints across multiple AWS Regions. Its design ensures minimal downtime, session continuity, and improved performance even during regional or network disruptions. This makes it a robust solution for disaster recovery, multi-region resiliency, and high availability of applications on AWS[1][3][6][7][8][9].
Citations:
[1] https://aws.amazon.com/global-accelerator/features/
[2] https://docs.aws.amazon.com/global-accelerator/latest/dg/about-endpoints.avoid-connection-collisions.html
[3] https://docs.aws.amazon.com/global-accelerator/latest/dg/introduction-benefits-of-migrating.html
[4] https://repost.aws/questions/QUiAtCMMEpSTC4w8u8mU_B9Q/does-aws-global-accelerator-maintain-the-connection-integrity
[5] https://docs.aws.amazon.com/global-accelerator/latest/dg/introduction-how-it-works.html
[6] https://docs.aws.amazon.com/global-accelerator/latest/dg/about-endpoints-endpoint-weights.unhealthy-endpoints.html
[7] https://dev.to/farrukhkhalid/crafting-a-zero-downtime-multi-region-architecture-on-aws-1df9
[8] https://docs.aws.amazon.com/global-accelerator/latest/dg/disaster-recovery-resiliency.html
[9] https://aws.amazon.com/global-accelerator/faqs/
[10] https://aws.amazon.com/blogs/networking-and-content-delivery/creating-disaster-recovery-mechanisms-using-amazon-route-53/