AppTrace技术全景：开发者视角下的工具链与实践经验

import os.signpost

let log = OSLog(subsystem: "com.yourapp", category: "performance")
os_signpost(.begin, log: log, name: "Image Processing")
// 执行耗时操作
os_signpost(.end, log: log, name: "Image Processing")

python systrace.py -o mytrace.html -a com.yourapp sched freq idle am wm gfx view binder_driver hal dalvik

// 优化前
fun onBindViewHolder(holder: ViewHolder, position: Int) {
    val bitmap = decodeBitmap(items[position].imagePath) // 主线程解码
    holder.imageView.setImageBitmap(bitmap)
}

// 优化后
private val decodeDispatcher = Executors.newFixedThreadPool(4)

fun onBindViewHolder(holder: ViewHolder, position: Int) {
    val cacheKey = items[position].imagePath
    decodeDispatcher.submit {
        val bitmap = decodeBitmap(cacheKey)
        decodedCache.put(cacheKey, bitmap)
        mainHandler.post { updateImageView(holder, bitmap) }
    }
}

// Gradle配置示例
android {
    buildTypes {
        debug {
            minifyEnabled false
            testCoverageEnabled true
            buildConfigField "boolean", "ENABLE_TRACING", "true"
        }
        release {
            minifyEnabled true
            shrinkResources true
            buildConfigField "boolean", "ENABLE_TRACING", "false"
        }
    }
}

// 基于设备状态的采样
func shouldSample() -> Bool {
    guard ProcessInfo.processInfo.thermalState != .critical else {
        return true // 设备过热时全量采样
    }
    return Int.random(in: 0..<100) < 5 // 5%采样率
}

使用Grafana + Prometheus构建自定义监控看板：
sql
# 示例查询：绘制CPU使用率百分位数
quantile_over_time(0.95, 
  rate(process_cpu_seconds_total{app="mobile-app"}[5m])
) * 100