Interview with Lu Pengcheng on Mogu Street’s Monitoring System Architecture and Evolution

At the SDCC 2016 China Software Developers Conference in Beijing, CSDN interviewed Lu Pengcheng, a platform architect at Mogu Street, about his background, the company’s business, and the technical direction of its systems.

Lu introduced himself as a technology enthusiast focusing on monitoring, logging, and storage architectures, with experience across many domains such as distributed systems, smart operations, and both client‑side and server‑side development.

The core web stack of Mogu Street uses mainstream components like LVS, Nginx, and Tomcat for network services, while the main application layer combines PHP, SOA, and Java, supplemented by monitoring, logging, middleware, and DevOps platforms. Open‑source technologies such as HBase, Spark, Storm, Kafka, Redis, and ScyllaDB are heavily employed.

During the company’s transition from a social platform to a large‑scale e‑commerce site, the architecture evolved through several phases: moving from a monolithic PHP codebase to a service‑oriented architecture (SOA) built on the internally developed Tesla framework, shifting monitoring SDKs from PHP to Java, and rebuilding the monitoring system to reduce data overhead and improve performance.

Lu then described the design of Mogu Street’s monitoring platform, highlighting four key principles:

Decentralization – using a gossip‑based SWIM protocol for configuration sync and cluster management.

Role separation – isolating heavy query functions into peripheral systems for better scalability during traffic spikes.

No external dependencies – keeping the core system independent of third‑party services to ensure stability.

Scenario‑based architecture – simplifying configuration by mapping clear usage scenarios for data collection and visualization.

When building a high‑availability monitoring system, Lu emphasized several critical decisions: data type and granularity, sampling and aggregation methods, storage strategies (full vs. rolling, local vs. centralized), and alerting mechanisms (polling vs. real‑time, false‑positive tolerance, intelligent alerts).

The monitoring platform was implemented in C++ to leverage the team’s expertise and to meet the CPU‑intensive requirements of query‑time calculations. Lu argued that embracing C++’s complexity—using modern features like RAII, smart pointers, and lambdas—leads to more stable and efficient code.

Regarding team management, Lu’s monitoring team consists of six engineers who collaborate informally, focusing on mutual discussion, code reviews, and shared passion, which he believes drives productivity more than formal management.

Mogu Street has also contributed open‑source projects such as the IM system TeamTalk and maintains a suite of internally developed technologies (SOA framework, message queues, sharding, monitoring, logging, full‑link tracing, storage). While most are not yet open‑sourced due to business constraints, Lu hopes to release more components in the future.

Finally, Lu announced his upcoming talk at SDCC 2016 Beijing titled “Mogu Street Monitoring System Architecture and Evolution,” inviting attendees to learn more about the platform’s design and roadmap.