An Overview of Meituan-Dianping's Open‑Source Distributed Monitoring System CAT

Project Background

CAT (Central Application Tracking) is an open‑source distributed real‑time monitoring system developed by Meituan‑Dianping using Java. It provides a unified solution for storage, high‑performance communication, massive online access, service governance, real‑time monitoring, containerization, and intelligent cluster scheduling. CAT serves as a unified monitoring component at the application layer and is widely used in middleware such as RPC, databases, caches, and MQ.

The article introduces CAT’s client, performance, and architecture. A previous article detailed the design of CAT’s client and server; readers are invited to refer to that piece for deeper technical insight.

Product Value

Reduces fault discovery time.

Lowers fault localization cost.

Assists application optimization.

Technical Advantages

Real‑time processing to capture time‑sensitive information.

Full‑volume data collection for in‑depth fault analysis.

High availability to support fault recovery and diagnosis.

Fault tolerance so that failures do not affect business operation.

High throughput to handle massive monitoring data.

Scalable architecture supporting distributed, cross‑IDC deployment.

Usage Status

CAT now covers core business lines of Meituan‑Dianping such as food delivery, travel, mobility, and finance, and is deployed in almost all core applications. Since early 2016, the number of integrated applications has grown by 400%, the number of machines by 900%, processing up to 320 billion messages per day, storing nearly 400 TB of data, and reaching a peak QPS of 6.5 million per second.

The rapid traffic growth posed challenges in communication, computation, and storage, leading to multiple architectural upgrades (message sampling aggregation, storage redesign, multi‑dimensional metric monitoring, unified alerting, etc.). After seven years of continuous development, CAT is being further iterated as an open‑source project.

New Features

Multi‑language Clients – In addition to the Java client, CAT now provides C/C++, Python, Node.js, and Golang SDKs. The core C client offers the fundamental API, which other language SDKs wrap.

Performance Improvements

Message Sampling Aggregation – Reduces client message volume and network overhead by aggregating sampled messages locally before reporting.

Communication Protocol Optimization – Switch from a custom text protocol to a binary protocol, significantly boosting performance while maintaining backward compatibility.

Message File Storage Redesign – Introduces a two‑level index structure to reduce random I/O and improve read/write performance. The design uses 4 KB buckets, with the first bucket as a primary index header and the remaining buckets as secondary indexes.

Future Plans

Technology Stack Upgrade – Adopt mainstream open‑source tools (Spring, MyBatis, etc.) to lower learning and development costs.

Product Experience – Redesign UI/UX for better user experience.

Open‑Source Community Building – Enhance the official website and organize technical exchanges.

More Language SDKs – Continue expanding language support.

About Open‑Source

Since its open‑source launch in 2011, CAT has earned over 5,900 Stars and 2,400 Forks on GitHub and is used by more than 100 companies, including Ctrip, Lufax, Liepin, and Ping An. It is regularly presented at global conferences such as QCon and architecture/operations summits.

The project is hosted at https://github.com/dianping/cat . The team welcomes contributions via GitHub issues.

Conclusion

CAT is a long‑term, high‑availability, high‑reliability distributed monitoring middleware. The team invites developers to join, contribute, and help shape the future of the project.

Recruitment Information

Meituan‑Dianping’s infrastructure team is hiring senior Java engineers in Beijing and Shanghai to work on core infrastructure components such as distributed monitoring, service governance, high‑performance communication, messaging middleware, storage, containerization, and cluster scheduling. Interested candidates can send resumes to yong.you@… (link omitted).