Question: Is connection pooling needed for DynamoDB?

Answer

DynamoDB does not need a pool of database sessions, but reusing the client's HTTP connections improves performance. Requests go to an HTTPS API rather than a long-lived SQL connection with transaction state.

The AWS SDK's HTTP transport manages sockets and connection reuse. In Node.js, AWS SDK for JavaScript v3 enables TCP keep-alive by default. Reusing one client allows it to reuse those connections instead of repeating connection and TLS setup for every request.

For a busy application, inspect request concurrency, timeouts and the HTTP agent's socket limit. If the application submits more work than the transport can send, requests can wait in a local queue. Raising concurrency indiscriminately can instead create DynamoDB throttling, so measure both client latency and service metrics.

Configure separate clients for different endpoints or credential contexts, rather than creating and destroying one per operation. In Lambda, initialize clients outside the handler. Avoid disabling keep-alive as a routine fix for intermittent failures; first examine timeouts, retries and networking.

DynamoDB's managed infrastructure handles server-side scaling, but it does not manage your application's sockets. See client reuse and latency troubleshooting.

Source: AWS Node.js keep-alive.