How JD.com’s ForceBot Revolutionized 618 Sale Load Testing

JD.com 618 Sales Figures & Black Tech

On June 18, JD.com reported a total order amount of 119.9 billion CNY for its 618 mid‑year shopping festival, comparable to Alibaba’s 120.7 billion CNY during the 2016 Double‑11 event.

During the same day, JD.com launched unmanned delivery robots across several university campuses, marking a step toward fully automated logistics.

Technical Challenges and the Need for Accurate Capacity Planning

Before the 618 event, the technical support team was concerned about bottlenecks in tightly coupled core systems, which could degrade user experience and reduce order volume.

Capacity planning was often based on historical experience, leading to over‑provisioning and imprecise resource estimates.

ForceBot Full‑Link Load‑Testing Initiative

In 2016, JD.com’s infrastructure team launched the ForceBot project to simulate end‑to‑end user behavior across the entire order flow, from homepage browsing to payment.

The system distinguishes simulated traffic from real user traffic to avoid contaminating production metrics.

Design Goals

ForceBot aims to reproduce high‑concurrency user actions such as searching, adding items to the cart, modifying addresses, and completing checkout, both for regular traffic and peak‑sale scenarios.

Architecture Overview

The new platform decouples functional modules to eliminate bottlenecks and support horizontal scaling.

Controller functions are split into single‑task assignment units.

Task Service handles task distribution and horizontal scaling.

Agents register heartbeats, pull tasks, and execute them.

Monitor Service forwards performance data to JMQ.

Dataflow performs streaming calculations on test data and stores results in a database.

Git stores test scripts and libraries, supporting distributed versioning.

Business System Refactoring

Golden‑Path Business Identifies the critical flow from user browsing to successful order placement, covering homepage, search, product detail, cart, and checkout.

Test Traffic Identification Marks users and products to ensure test traffic does not affect production statistics such as PV, UV, and order counts.

Test Data Storage Two approaches are used: tagging data stored in the production database for realistic performance insight, or isolating data in a dedicated test database to avoid impacting production resources.

For payment systems, a mock bank interface was created to handle test transactions without affecting real financial services.

Benefits of ForceBot

ForceBot provides accurate capacity‑planning data, identifies concurrency bottlenecks across all golden‑link systems, and consolidates load‑testing resources, improving overall testing efficiency.

During the 2016 Double‑11 event, ForceBot replaced individual system optimizations, delivering real‑time monitoring of response times, TPS, and latency percentiles.