Design and Implementation of Business‑Driven Monitoring Systems at JD Cloud

Monitoring is the lifeline of operations; its goal is to quickly discover, locate, and resolve issues to reduce MTTR. JD Cloud defines a unified monitoring standard covering four layers—basic, liveliness, performance, and business—to ensure comprehensive data collection and avoid gaps.

The article answers why operations rely on monitoring, typical system designs, and how business‑driven high‑availability monitoring differs, based on a JD Cloud and Intel joint online class presented by architect Yan Zhijie.

Key capabilities required for an effective monitoring system include data collection, data processing (aggregation, enrichment), anomaly detection with alerting, dashboard‑based diagnosis, rapid mitigation via a playbook platform, and high availability of the monitoring infrastructure itself.

Functionally, a typical monitoring system consists of modules for collection, computation, storage, alerting, algorithms, and business interfaces. Data flows from collection (agents, APIs, probes) into three processing streams: computation (e.g., aggregating Nginx PV), storage (for dashboards), and alerting (for anomaly detection). Advanced analytics provide root‑cause recommendations.

From a data‑centric perspective, monitoring is treated as a data‑processing pipeline: abstraction (CMDB), collection, aggregation, storage, and alerting. Standardizing data (metrics, logs, traces) enables reusable libraries (F(x)) across modules, simplifying design and enhancing reuse.

JD Cloud’s monitoring standards emphasize four layers: basic (CPU, memory, host health), liveliness (process, port status), performance (PV, latency, errors, capacity), and business (user‑side experience). Implementation includes scoring mechanisms for managers, one‑click configuration recommendations for operators, change‑visualization integration, and a standardized playbook platform.

A practical case study—alarm‑convergence project—shows how JD Cloud quantified alarm volume, introduced responsible‑owner dashboards, merged duplicate alerts, and provided mobile‑friendly views, dramatically improving alert relevance.

Looking ahead, the speaker envisions moving from rule‑based processing to statistical and machine‑learning methods, leveraging cloud‑native observability, semantic log analysis, multi‑metric correlation, and AI‑driven automation to close the monitoring loop.