- [多线程优化 booster-transform-thread](#多线程优化-booster-transform-threadhttpsgithubcomdidiboostertreemasterbooster-transform-thread)
- 性能瓶颈检测 booster-transform-lint
本文继续分析booster的实现原理。
多线程优化 booster-transform-thread
对于这个组件官方是这样描述的: 对于开发者来说，线程管理一直是个头疼的问题，特别是第三方 SDK 中的线程，过多的线程可能会导致内存不足，然而幸运的是，这些问题都能通过 Booster 来解决。
那booster是如何解决的呢? 其实实现思路类似上一篇文章booster分析-修复系统bug。即通过gradle transform在编译时做字节码插桩,来实现代码的动态替换(整个booster框架大部分功能的实现都是这个原理)。
那么来看一下booster到底如何优化了多线程管理吧。
一般一个比较成熟的应用在运行时后台都会有很多线程在跑，为了便于线程的管理。 booster在编译时强制给应用程序运行时的线程(不论是自己开启的还是三方库开启的)都起了一个名字。这样无论是bug的定位，还是线程的调试都方便了不少。
booster 在编译时通过修改字节码文件动态给HandlerThread、Timer、AsyncTask、Thread、ThreadPoolExecutor等开启的线程都起了名字。
比如，所有的Thread在start()之前:
ThreadTransformer.java
private fun MethodInsnNode.transformInvokeVirtual(context: TransformContext, klass: ClassNode, method: MethodNode) {
if (context.klassPool.get("java/lang/Thread").isAssignableFrom(this.owner)) {
when ("${this.name}${this.desc}") {
"start()V" -> {
method.instructions.insertBefore(this, LdcInsnNode(makeThreadName(klass.className)))
method.instructions.insertBefore(this, MethodInsnNode(Opcodes.INVOKESTATIC, SHADOW_THREAD, "setThreadName", "(Ljava/lang/Thread;Ljava/lang/String;)Ljava/lang/Thread;", false))
}
}
}
}
即在Thread.start()之前调用了ShadowThread.setThreadName():
public class ShadowThread {
public static Thread setThreadName(final Thread t, final String prefix) {
t.setName(makeThreadName(t.getName(), prefix));
return t;
}
public static String makeThreadName(final String name, final String prefix) {
return name == null ? prefix : (name.startsWith(MARK) ? name : (prefix + "#" + name));
}
}
举个例子:
class ThreadTest {
private val TAG = "Booster-Thread"
fun test(){
Thread({
Log.d(TAG, "thread name : ${Thread.currentThread().name}")
}).start()
}
}
上面代码经过booster优化后log输入的线程名为:
D/Booster-Thread: thread name : com.susion.boostertest.thread.ThreadTest
即线程的名字改为 : 包+类名, 而没有优化前名称是:
D/Booster-Thread: thread name : Thread-2
经过booster优化后确实更容易识别线程了。 我们写多线程代码时还是给线程起个规范点的名字比较好
先来回顾一下线程池的构造函数:
public ThreadPoolExecutor(int corePoolSize,
int maximumPoolSize,
long keepAliveTime,
TimeUnit unit,
BlockingQueue<Runnable> workQueue)
keepAliveTime参数是用来设置线程的超时时长，当线程的闲置时长超过这个时间时线程就会被回收，但是这个策略并不会应用在核心线程上。
但当对线程池设置allowCoreThreadTimeOut(true)时，核心线程也会被回收退出。
booster通过对字节码的动态替换，对应用中所有的线程池都设置了allowCoreThreadTimeOut(true), 这样可以保证线程资源及时回收, 比如:
ThreadTransformer.java
private fun MethodInsnNode.transformInvokeStatic(context: TransformContext, klass: ClassNode, method: MethodNode) {
when (this.owner) {
"java/util/concurrent/Executors" -> {
when (this.name) {
"newCachedThreadPool",
"newFixedThreadPool",
"newSingleThreadExecutor",
"newSingleThreadScheduledExecutor",
"newScheduledThreadPool" -> {
val r = this.desc.lastIndexOf(')')
val name = this.name.replace("new", "newOptimized")
val desc = "${this.desc.substring(0, r)}Ljava/lang/String;${this.desc.substring(r)}"
this.owner = “com/didiglobal/booster/instrument/ShadowExecutors”
this.name = name
this.desc = desc
method.instructions.insertBefore(this, LdcInsnNode(makeThreadName(klass.className)))
}
}
}
}
}
即将Executors.newCachedThreadPool()等方法都替换为ShadowExecutors.newOptimizedXXX方法:
ShadowExecutors.java
public static ExecutorService newOptimizedCachedThreadPool() {
final ThreadPoolExecutor executor = new ThreadPoolExecutor(0, Integer.MAX_VALUE, 60L, TimeUnit.SECONDS, new SynchronousQueue<Runnable>(), Executors.defaultThreadFactory());
executor.allowCoreThreadTimeOut(true);
return executor;
}
这个组件通过编译时的lint检查来找出一些在应用中调用可能阻塞UI线程的API,如：I/O API 等。主要有两个特点:
- 支持自定义检查API列表。
- 对于查找到的有问题的API可以产生一张调用图便于查看与定位。
首先指定一些受检查的方法:
LintTransformer.java
class LintTransformer : ClassTransformer {
override fun onPreTransform(context: TransformContext) {
...
val parser = SAXParserFactory.newInstance().newSAXParser()
context.artifacts.get(ArtifactManager.MERGED_MANIFESTS).forEach { manifest ->
val handler = ComponentHandler()
parser.parse(manifest, handler)
// Attach component entry points to graph ROOT
mapOf(
Pair(handler.applications, APPLICATION_ENTRY_POINTS),
Pair(handler.activities, ACTIVITY_ENTRY_POINTS),
Pair(handler.services, SERVICE_ENTRY_POINTS),
Pair(handler.receivers, RECEIVER_ENTRY_POINTS),
Pair(handler.providers, PROVIDER_ENTRY_POINTS)
).forEach { components, entryPoints ->
components.forEach { component ->
entryPoints.map {
CallGraph.Node(component.replace('.', '/'), it.name, it.desc)
}.forEach {
globalBuilder.addEdge(CallGraph.ROOT, it) // 所有的分析点
}
}
}
}
}
}
它检查的方法有:
constant.kt
/**
* Main/UI thread entry point of Application
*/
internal val APPLICATION_ENTRY_POINTS = arrayOf(
"onConfigurationChanged(Landroid/content/res/Configuration;)V",
"onCreate()V",
"onLowMemory()V",
"onTerminate()V",
"onTrimMemory(I)V"
).map(EntryPoint.Companion::valueOf).toSet()
/**
* Main/UI thread entry point of Activity
*/
internal val ACTIVITY_ENTRY_POINTS = arrayOf(
// <editor-fold desc="public methods of Activity">
"onActionModeFinished(Landroid/view/ActionMode;)V",
"onActionModeStarted(Landroid/view/ActionMode;)V",
"onActivityReenter(ILandroid/content/Intent;)V",
"onAttachFragment(Landroid/app/Fragment;)V",
"onAttachedToWindow()V",
....
即Aplication/Activity/Service/BroadcastReceiver/ContentProvider等相关API。
override fun onPostTransform(context: TransformContext) {
val graph = globalBuilder.build()
...
// Analyse global call graph and separate each chain to individual graph
graph[CallGraph.ROOT].forEach { node ->
graph.analyse(context, node, listOf(CallGraph.ROOT, node), lints) { chain ->
val builder = graphBuilders.getOrPut(node.type) {
CallGraph.Builder().setTitle(node.type.replace('/', '.'))
}
chain.toEdges().forEach { edge ->
builder.addEdges(edge)
}
}
}
}
analyse()用来检查当前API是否命中了Lint检查列表。如果命中就会加入到dot文件绘图节点中。
booster默认定义的敏感API有:
/**
* Sensitive APIs
*/
internal val LINT_APIS = setOf(
...
"android/graphics/BitmapFactory.decodeFile(Ljava/lang/String;)Landroid/graphics/Bitmap;",
"android/graphics/BitmapFactory.decodeFileDescriptor(Ljava/io/FileDescriptor;)Landroid/graphics/Bitmap;",
"android/graphics/BitmapFactory.decodeFileDescriptor(Ljava/io/FileDescriptor;Landroid/graphics/Rect;Landroid/graphics/BitmapFactory\$Options;)Landroid/graphics/Bitmap;",
"android/graphics/BitmapFactory.decodeResource(Landroid/content/res/Resources;I)Landroid/graphics/Bitmap;",
"android/graphics/BitmapFactory.decodeResource(Landroid/content/res/Resources;ILandroid/graphics/BitmapFactory\$Options;)Landroid/graphics/Bitmap;",
"android/graphics/BitmapFactory.decodeResourceStream(Landroid/content/res/Resources;Landroid/util/TypedValue;Ljava/io/InputStream;Landroid/graphics/Rect;Landroid/graphics/BitmapFactory\$Options;)Landroid/graphics/Bitmap;",
"android/graphics/BitmapFactory.decodeStream(Ljava/io/InputStream;)Landroid/graphics/Bitmap;",
"android/graphics/BitmapFactory.decodeStream(Ljava/io/InputStream;Landroid/graphics/Rect;Landroid/graphics/BitmapFactory\$Options;)Landroid/graphics/Bitmap;"
// </editor-fold>
).map(Node.Companion::valueOf).toSet()
即主要是一些可能导致主线程卡顿的API。
比如有下面这个类:
class LabThreadTestActivity : AppCompatActivity() {
override fun onCreate(savedInstanceState: Bundle?) {
super.onCreate(savedInstanceState)
setContentView(R.layout.activity_lab_thread_test)
val test1 = getSharedPreferences("default", Context.MODE_PRIVATE).edit().putBoolean("test",true).commit()
// val test2 = getSharedPreferences("default", Context.MODE_PRIVATE).edit().putBoolean("test",true).apply()
}
}
经过booster-transform-lint检查后会生成com.susion.boostertest.thread.LabThreadTestActivity.dot文件。这个文件转为png图片如下:
更多分享见AdvancedAndroid