Implementing a Real-Time Pre-Alert Monitoring System to Improve Fund Trading System Stability

The fund trading system of JD Finance processes tens of billions of yuan daily, making system stability crucial; however, the existing alarm mechanism—thousands of instant alerts and a daily end‑of‑day check—creates noise, delays fault handling, and leads to reconciliation problems and customer complaints.

Because instant alerts are overwhelming and end‑of‑day checks are often too late, the team sought a simpler, more proactive method to discover anomalies early, aiming to reduce ticket volume, shorten reconciliation time, and free resources spent on handling complaints.

The proposed solution adds a supplemental pre‑alert layer that monitors key business points (order status, MQ backlog, payment completion, etc.) and triggers an alert when a point does not change within a predefined time window (e.g., 15 minutes). Alerts are displayed on a mobile interface, allowing engineers to notice suspicious situations within minutes.

The monitoring strategy defines “suspicious” as any critical point that remains unchanged beyond the expected time. The severity is inferred from the number of such points, enabling quick prioritisation of the most impactful issues.

Implementation integrates the monitoring service with the existing MCube template‑based rendering pipeline, using expression and event engines to evaluate the time‑based rules and generate alerts that are pushed to a dedicated UI.

After deployment, ticket volume dropped from 334 to 207 (a reduction of 127 tickets), daily reconciliation consistently finished before 16:30, and abnormal orders became rare occurrences. A notable case during the 2023 Spring Festival showed that early detection of non‑trading‑time redemption settings prevented a potential batch of thousands of erroneous orders and associated customer complaints.

The authors name this approach “先知预警” (Prophetic Pre‑Alert), a time‑series‑driven, business‑aware method that can be applied to other scenarios such as promotional activity monitoring, demonstrating that a modest technical effort combined with clear business logic can significantly improve system reliability.