Logistics R&D Preparation for the 618 Promotion: System Readiness, Stress Testing, and Real‑Time Monitoring

Each major sales promotion is treated like a full‑scale battle; the logistics R&D team prepared from the end of February to May 24, a total of 62 days, with more thorough preparation than in previous years.

Preparation ("Ready"): The team meticulously identified weak points in core systems, introduced stress testing to uncover deep issues, and iterated optimizations to advance the preparation process.

Separate core and auxiliary flows: Keep core processes minimal and stable, handle non‑core work asynchronously or with graceful degradation to avoid rollbacks.

Isolate different business types: Prioritize availability for order services, consistency for inventory, and performance for high‑concurrency features; split memory‑intensive tasks like report import/export to separate resources.

Database sharding: Use vertical sharding (e.g., order DB, product DB) and horizontal expansion (read/write separation, replication), apply partitioning, and separate hot/cold or historical data.

Service dependency principles: Stable services must not depend on volatile ones; avoid circular dependencies; prefer asynchronous weak dependencies across business boundaries; lower‑level services should not depend on higher‑level ones.

Fault transfer: Deploy core systems across multiple data centers, enable multi‑active setups, and provide rapid traffic switching to avoid single‑point failures.

Capacity planning (DID principle): Design for 20× capacity, implement 3×, and deploy 1.5×; evaluate traffic needs without increasing overall resources.

Traffic control: Sequentially manage traffic spikes by diverting load, applying degradation, and finally rate‑limiting with fallback plans.

Return to code: Identify and eliminate "code smells" in the codebase.

Battle ("Fight"): After thorough preparation, the team employed a suite of tools for real‑time monitoring and rapid fault diagnosis.

Customizable real‑time monitoring dashboard per department, with sortable performance metrics.

Comprehensive call‑chain tracing with configurable latency thresholds to locate slow methods.

Alarm statistics portal offering department‑ and application‑level insights for alarm trend analysis.

Slow‑SQL analysis tool that filters by project, IP, and time range.

Configuration scanning covering JVM, Log4j, DB connection pools, etc., plus third‑party library version checks.

Glances – a Linux host performance monitor that replaces the traditional top command with real‑time CPU, memory, load, network, disk, and filesystem metrics.

Arthas – a powerful Java performance diagnostic tool providing full‑stack analysis.

Automatic snapshot generation for medium‑or‑higher severity alarms, preserving fault‑scene data for post‑mortem.

Unified aggregation of alerts from multiple monitoring platforms to give a clear view of application health.

Thanks to all teams for their hard work; we are fully prepared for the 618 promotion!