How JD’s Data Platforms Scaled for the 618 Mega‑Sale: Operations, Stress‑Testing, and Dual‑Stream Architecture

JD’s Data Product Platform team supports the retail subsidiary’s massive 618 shopping festival by preparing three core platforms—Golden Eye, Shufang, and Business Intelligence—with coordinated data capabilities across procurement, operations, advertising, customer service, finance, and partner domains.

System Pressure Estimation and Capacity Planning

Relying on historical promotion data, the R&D team, together with QA, forecasts traffic volume, order counts, message throughput, and request concurrency. These estimates guide capacity expansion and stress‑testing for data processing pipelines, push services, API interfaces, and containerized services.

Emergency Downgrade Strategies

Because predictions cannot be 100% accurate, the team designs tiered degradation plans. They prioritize real‑time over offline data, sales over traffic, and internal metrics over industry benchmarks. Unified downgrade switches control refresh intervals, and critical external APIs have specific fallback mechanisms. Small‑traffic windows are used for live drills to validate the plans without impacting users.

Dual‑Stream Architecture and High‑Fidelity Testing

For the first time, the promotion employed a dual‑stream core data pipeline, achieving physical isolation and double backup across two data centers. This eliminated single‑point‑of‑failure risks and boosted resilience. The promotion also featured a full‑link high‑fidelity stress test to guarantee end‑to‑end data availability.

Rapid Development for Live Business Needs

During the promotion, the Golden Eye team faced simultaneous development and preparation. They quickly delivered data for live streaming scenarios, integrated 7FRESH data into the retail dashboard, and supported Beijing consumption‑coupon analysis by coordinating with central platform and government affairs teams. The team also handled numerous real‑time management dashboards and PR screens.

Comprehensive Monitoring and On‑Call Practices

To ensure stability, the Shufang team added detailed logs, micro‑service performance metrics, and call‑chain tracing. Grafana dashboards provide real‑time visibility of system health, enabling early detection of issues and swift activation of emergency procedures. A daily on‑call rotation and over 400 contingency plans keep the team ready for any incident.

Shufang: User Growth Platform Scaling

Shufang, a user‑operation platform, experiences traffic several times higher than the previous year’s 618 peak. The team upgraded big‑data scheduling, enabling dynamic Yarn queue scaling for Spark jobs to match varying compute and storage demands across promotion phases.

In May, Shufang merged with Jiushù, unifying permissions and functions, and extending capabilities such as user asset balance sheets, global user tags, and 4A lifecycle segmentation. The integration also added richer outreach channels and smarter投放 strategies.

Continuous monitoring via Grafana, extensive logging, and a robust alert system ensure that any performance degradation is promptly addressed.

Conclusion

The preparation efforts are means to an end: delivering reliable data services that enable merchants and partners to navigate the promotion’s rhythm and grow their business. JD’s R&D teams remain agile, ready to tackle upcoming challenges, and committed to supporting the success of future large‑scale events.