How JD.com’s Shopping Cart Gained 30% Faster Performance with Full Asynchronous Refactoring

In the context of increasing business complexity, JD.com’s retail shopping‑cart team applied a full asynchronous refactor to boost system performance and user experience.

Background

The cart now supports many new services, deeper integration of front‑end calls, pre‑checkout features such as coupons and JD beans, and larger item capacity, leading to a rapid growth in RPC interfaces and pagination calls.

Full Asynchronous Refactor

The solution consists of three main parts:

Parallelizing different RPCs : Analyze dependencies, split code into asynchronous requests and result handling, and execute RPCs in parallel to reduce waiting time.

Batch interfaces with multi‑page parallelism : Wrap pagination logic in an async tool that automatically splits data exceeding interface limits into parallel pages, improving single‑interface response speed.

Underlying JSF async calls : Use JD’s RPC framework JSF (version 1.7.5+), which supports CompletableFuture.

Issues and Solutions

Fine‑grained retry : After async conversion, retries must be triggered on timeout during result processing, and only the failed page should be retried.

Complex async RPC monitoring : Record start time at pagination call and end time when async result arrives; separate metrics for normal and retry calls, and also monitor Get‑wait time.

Preserving async results per page : Results must be returned unchanged to retain Provider information for timeout diagnostics.

Per‑page timeout control : Each page’s timeout is calculated as

RPC_TIMEOUT > (now - asyncStart) ? RPC_TIMEOUT - (now - asyncStart) : 0

.

Page balancing : Evenly distribute data across pages to avoid skew and maximize performance.

Benefits

After refactoring, core cart interface latency dropped by about 30%, improving user experience while saving substantial server resources. New RPCs on non‑critical paths have minimal impact, and capacity growth now has little performance penalty.