Design and Implementation of a Distributed Monitoring System at Autohome

Autohome originally used Zabbix for monitoring, progressing from a simple dual‑node hot‑cold backup to a proxy‑based distributed model and finally to cross‑data‑center disaster recovery, reflecting a typical growth path for enterprise monitoring systems.

The new system was designed to overcome Zabbix's limitations, such as heavy database dependence and low modularity, by adopting a more scalable, distributed architecture.

Key design goals include precise alerting, automatic fault localization, and self‑healing capabilities, with the system serving both operations teams and business units for configuration and alert consumption.

The proposed architecture consists of agents for data collection, transfer components to forward data to back‑ends (analyzers, storage, etc.), storage for historical metrics, dashboards for configuration and visualization, detectors, analyzers, senders, and processors for automated remediation.

Implementation leveraged existing open‑source building blocks like collectd, statsd, MySQL, HBase, and later adopted the Open‑Falcon project as a foundation for further development.

Product differentiators include a custom service tree aligned with corporate organization, an enhanced dashboard supporting multi‑level grouping, advanced alarm expression functions (e.g., daydiff, daypdiff), and sophisticated alarm escalation and de‑duplication strategies to mitigate alert storms.

A Windows‑based agent was developed in Python to monitor Windows services, IIS, and SQL Server, running as a Windows Service and providing an HTTP proxy interface, distinguishing it from existing Linux‑oriented agents.

The roadmap outlines future work such as fault localization using temporal and semantic event correlation, dynamic threshold analysis, behavior‑based monitoring, and multi‑site data synchronization for core components.

In conclusion, building a monitoring system is an ongoing effort driven by evolving requirements, and Autohome commits to open‑sourcing parts of the solution and collaborating with the community.