Practical Lessons from Upgrading JD's Phone Recharge System Architecture

JD's phone‑recharge service saw a 40% increase in active users, driving a surge in order volume that exceeded the capacity of its legacy architecture, prompting a comprehensive system upgrade.

Application‑layer improvements : a Redis‑based cache (JimDB) was added for hot data such as feature flags and white‑lists, reducing database load; concurrent processing programs were written to parallelize independent tasks like order settlement; and the overall system structure was refactored to improve cohesion and reduce coupling.

Micro‑service transformation : the monolithic application was split into five services (PC order, mobile order, MQ processor, admin console, worker). Each service communicates via JD's RPC framework (JSF) and message framework (JMQ), enabling independent development, testing, and deployment while maintaining overall functionality.

Database‑layer strategies : a vertical‑plus‑horizontal sharding scheme separates enterprise and consumer orders, using hash‑based routing on account PIN or order ID; read‑write separation directs low‑latency reads to replicas; low‑frequency pages are statically generated and served via Nginx; and data aggregation is handled by syncing shards to Elasticsearch.

Deployment and operations : resource isolation is achieved with Hystrix for semaphore and thread‑pool isolation, independent containers, and multi‑datacenter traffic distribution; gray‑release and version‑rollback mechanisms use data‑flow switches and cache flags to ensure seamless upgrades without user impact.

After these upgrades, the system achieved optimal throughput and stability, successfully handling peak traffic during major sales events such as 618 and Double‑11.