ECS Service Connect adds Zone-Aware Routing

ECS Service Connect now routes traffic within the same AZ by default, cutting costs and latency.

What changed

• ECS Service Connect now defaults to Zone-Aware Routing.
• Traffic is preferentially routed to endpoints within the same Availability Zone (AZ) as the originating task.
• Existing services require a one-time redeployment to enable this behavior.

Why it matters

This is the Lambda cold-start fix everyone had stopped waiting for. Prioritizing intra-AZ traffic directly addresses the cost and latency implications of cross-AZ data transfer, a common trade-off for achieving high availability with distributed containerized applications. The honest version: AWS is making it easier to run resilient applications without incurring unexpected data transfer fees. This change should benefit workloads with high inter-service communication, particularly those sensitive to latency, by reducing the effective hop count for requests. It effectively removes the need to manually engineer complex routing rules or accept higher operational costs for resilience.

The catch

While enabled by default for new services, existing services require a redeployment. The honest limit: The announcement does not specify the exact thresholds for when traffic will redistribute across AZs if local endpoints become unhealthy or fall below capacity. This lack of specificity leaves room for unexpected behavior under high load or during partial AZ failures. Furthermore, while the feature itself is free, the underlying ECS tasks and the data transfer itself still incur standard AWS charges.

Ship it

For existing ECS services using Service Connect, plan a rolling redeployment. Pairs with: Amazon VPC Flow Logs can be used with AZ metadata to monitor cross-AZ traffic patterns and validate the effectiveness of the new routing behavior. This is a straightforward change for most services, but monitor your traffic patterns post-redeployment.

Bottom line: ECS Service Connect's new zone-aware routing reduces cross-AZ data transfer costs and latency by default, but existing services need redeployment and specific failure scenarios require monitoring.

— Filed to /aws