Async Request Cancellation in DynamoDB SDK v2 Does Not Abort In-Flight HTTP Calls

Async Cancellation Behavior in AWS SDK v2

When using the asynchronous DynamoDB client in AWS SDK v2, developers can cancel a CompletableFuture returned by operations like scan() or query(). However, this cancellation does not translate to aborting the underlying HTTP request sent to DynamoDB. The Netty-based HTTP client continues processing the request until completion, consuming read capacity and generating response data that is ultimately discarded.

This behavior has operational implications for applications that rely on cancellation for resource management or user-initiated request termination. Connection pool exhaustion can occur if many requests are cancelled without actual network-level abort, and billing for consumed capacity continues regardless of client-side cancellation.

The SDK provides apiCallTimeout and apiCallAttemptTimeout in ClientOverrideConfiguration to enforce client-side timeouts, but these mechanisms do not provide a way to cancel the underlying HTTP request either. DynamoDB has no server-side API to cancel an in-flight request, and the SDK does not currently propagate cancellation signals to the Netty channel.

Specific Questions

• Does AWS SDK v2 (or v3) provide a way to cancel the underlying HTTP request when cancelling a CompletableFuture, or is this behavior by design?
• How can applications reliably detect and handle request abandonment to avoid connection pool exhaustion and unexpected capacity consumption?
• What monitoring or metrics should be configured to track the impact of cancelled requests on DynamoDB usage and costs?