Mobile APM Performance Monitoring Application and Practice – Summary of 58 Group Technical Salon

The fourth 58 Group technical salon, organized by the Technology Engineering Platform and the HR Magic Academy on November 28, 2018 at the Beijing headquarters, focused on "Mobile APM Performance Monitoring Application and Practice." Speakers included Cheng Mengyang, product lead of Listening Cloud App, and Xu Gang, architect from 58 TEG Wireless Architecture.

Key Takeaways

1. Listening Cloud Mobile APM Performance Monitoring Application and Practice

Mobile Performance Monitoring Principles and Methods

Listening Cloud provides full‑stack visualization of mobile app performance, monitoring crashes, errors, and jank during real user sessions. The monitoring architecture includes both active and passive approaches.

Active Monitoring (Automation‑Based)

Advantages: no device resource consumption, no code changes, supports various protocols, can capture network performance data.

Disadvantages: synthetic data only, limited test points (region, carrier, signal), cannot reach code‑level details, cannot pinpoint crash, ANR, or slow interaction causes.

Passive Monitoring (Code‑Embedding)

Advantages: real‑user data, minimal detection point constraints, code‑level insight to identify crashes, ANR, and performance bottlenecks, helping developers understand execution efficiency.

Disadvantages: requires code modifications, adds some device resource overhead.

Example of code embedding:

add_library(wblog-lib SHARED IMPORTED)
set_target_properties(wblog-lib
         PROPERTIES IMPORTED_LOCATION
         ${CMAKE_SOURCE_DIR}/src/main/third/${ANDROID_ABI}/libwblog-lib.so)

Startup Experience Optimization

Analyzing startup performance reveals issues such as slow startup, crashes, and frequent request errors (HTTP, network). By collecting multi‑dimensional metrics on the app side, teams can locate performance bottlenecks and improve conversion rates.

Slow Startup Diagnosis

Problems often stem from heavy multi‑dex initialization (e.g., attachBaseContext) and Dalvik linearAlloc bugs on Android 4.0 and below. Recommendations include using ProGuard to shrink dex size and thorough testing across devices.

2. 58 Group Mobile SDK Quality Assurance Practice

Traditional SDK issue detection relied on passive feedback (WeChat groups, tickets, phone calls), leading to low timeliness and high effort. The new strategy collects SDK exceptions (crash, performance, functional anomalies) via WMDALite, stores events locally, and uploads them for backend analysis, enabling proactive quality monitoring.

Collected exception statistics are visualized in dashboards (e.g., message send error distribution, internal exception distribution).

Log Retrieval Service

The service captures SDK logs locally and, upon user‑reported issues, retrieves logs using phone number or user ID. It supports cross‑platform SDKs (C/C++, iOS, Android) with a layered design: C/C++ log writer, Lib‑layer retrieval, and secure upload using asymmetric encryption of symmetric keys.

C++ Hybrid Integration

For iOS, a static library with header inclusion suffices; for Android, CMake IMPORTED handling ensures a single SO file in the APK. Example configuration is shown above.

Case Studies

• An app experienced intermittent message‑send failures despite normal network conditions; analysis of error‑code distribution led to a retry strategy that resolved the issue.

• Another app suffered frequent OpenSSL‑related crashes; crash signal analysis and log tracing identified the root cause, and a code fix eliminated the crashes.

3. Conclusions

Mobile app performance monitoring and quality assurance are complex but critical. Both Listening Cloud and 58 Group have made substantial practical advances. The salon facilitated deep discussion on pain points and solutions, and participants look forward to further innovations in quality assurance and performance optimization.