Question

A financial services company runs a customer-facing API on Amazon ECS behind an Application Load Balancer. The API calls a downstream third-party payment provider that occasionally becomes slow or unresponsive during peak periods. When the payment provider degrades, threads in the API tier become blocked waiting for responses, cascading failures consume all available connections, and the entire API becomes unavailable — even for endpoints that do not depend on payments. The architecture team must improve resilience so that a slow downstream dependency cannot take down the whole service. Which approach BEST addresses this requirement?

Accepted Answer

Implement a circuit breaker pattern with bulkhead isolation, using separate connection/thread pools per downstream dependency and timeouts, so failures to the payment provider fail fast and are isolated from unrelated endpoints.

Answer

Increase the ECS task count and enable target-tracking auto scaling on CPU utilization so there are always enough tasks to absorb blocked threads during payment provider slowdowns.

Answer

Place an Amazon SQS queue in front of every API endpoint so that all requests are buffered and retried asynchronously until the payment provider recovers.

Answer

Add an Amazon ElastiCache for Redis cluster to cache payment provider responses so that subsequent calls avoid contacting the slow downstream dependency.