Xianyu Android Log Governance and Performance Monitoring Solution

Current situation: Xianyu relies on Alibaba infrastructure for log collection, with capabilities such as crash aggregation, local TLog, online event and user‑behavior logs. However many issues remain: missing crash/ANR logs for crashes, difficulty locating business issues, incomplete log content, limited local log capability, low command‑push success, and low feedback coverage.

Overall governance design: a new system integrates local console logs, TLog, online logs, and a retrieval strategy to close the gaps.

Enhancing local log capability: capture Android logcat logs (LOG_ID_MAIN, LOG_ID_EVENTS, LOG_ID_CRASH) via adb commands and package them with AUS to OSS. Example commands:

adb logcat -d -v threadtime -t 20000
adb logcat -d -b events -v threadtime -t 6666
adb logcat -d -b crash -v threadtime -t 6666

Local log retrieval: upload via AUS/TLog, improve success rate, and provide a batch retrieval platform for difficult‑to‑obtain logs.

Online lag/ANR detection: existing BlockCanary and adb bugreport have limitations in production. Permission issues prevent reading /data/anr/traces.txt on Android 6+. Sample code:

File mSystemTraceFile;
this.mSystemTraceFilePath = "/data/anr/traces.txt";
this.mSystemTraceFile = new File(this.mSystemTraceFilePath);
if (!this.mSystemTraceFile.exists()) {
    String propSystemTraceFilePath = SystemPropertiesUtils.get("dalvik.vm.stack-trace-file");
    this.mSystemTraceFile = new File(propSystemTraceFilePath);
}

BlockCanary principle: sets Looper.mLogging to monitor UI thread tasks. Sample:

public void start() {
    if (!mMonitorStarted) {
        mMonitorStarted = true;
        Looper.getMainLooper().setMessageLogging(mBlockCanaryCore.monitor);
    }
}

New solution: replace Looper task monitoring with Android frame callbacks. Record timestamps on each frame; if the interval exceeds 500 ms, treat it as a lag, and if continuous lag >5 s with unchanged stack, treat it as an ANR. No extra delayed tasks are created, resulting in minimal overhead.

Detection effect: injecting 500 ms and 5 s sleeps in a CardView click handler produces the expected lag reports, confirming the approach works.

Active problem discovery: build monitoring dashboards for key metrics (5 s lag, request failures, error toasts, etc.) and a self‑built opinion‑tracking platform for real‑time logs.

Log retrieval platform: supports user‑ID query by issue name, batch retrieval, and integrates with TLog and OSS for comprehensive log access.

Summary and outlook: technical opinion ratio dropped from 10.5 % to 4.7 %, upload success improved, and future work includes log visualization, semantic description, and intelligent parsing.