目的
arthas 是一个非常优秀的生产诊断工具,可以热部署.查看jvm信息,强制gc等等. 这里查看源码的目的有两点:
- 通常的jvm是不支持对已载入的class进行重新部署的,想知道它是如何实现的(
javassist可以实现修改一次,但没办法反复修改.) - trace命令 官方只支持展开一层,想自己修改源码实现展开3层.或者指定层数.
准备工作
arthas 源码: 源码下载地址
zulu jdk: zulu jdk 下载地址
源码解读
入口函数
找到项目arthas-boot的pom文件,即可以看到arthas的入口类为com.taobao.arthas.boot.Bootstrap
对应的函数为该类的main方法:
<manifest>
<mainClass>com.taobao.arthas.boot.Bootstrap</mainClass>
</manifest>
我们先来看看main方法,看一个大概即可.
String javaHome = System.getProperty("java.home");
if (javaHome != null) {
AnsiLog.info("JAVA_HOME: " + javaHome);
}
Package bootstrapPackage = Bootstrap.class.getPackage();
if (bootstrapPackage != null) {
String arthasBootVersion = bootstrapPackage.getImplementationVersion();
if (arthasBootVersion != null) {
AnsiLog.info("arthas-boot version: " + arthasBootVersion);
}
}
try {
String javaToolOptions = System.getenv("JAVA_TOOL_OPTIONS");
if (javaToolOptions != null && !javaToolOptions.trim().isEmpty()) {
AnsiLog.info("JAVA_TOOL_OPTIONS: " + javaToolOptions);
}
} catch (Throwable e) {
// ignore
}
这部分获取环境变量,设置arthas的home目录.
Bootstrap bootstrap = new Bootstrap();
CLI cli = CLIConfigurator.define(Bootstrap.class);
CommandLine commandLine = cli.parse(Arrays.asList(args));
try {
CLIConfigurator.inject(commandLine, bootstrap);
} catch (Throwable e) {
e.printStackTrace();
System.out.println(usage(cli));
System.exit(1);
}
这里配置命令行交互,也可以忽略后面一大段都是在校验pid,home目录等等,全部跳过.主要看这里,启动core jar
AnsiLog.info("Try to attach process " + pid);
AnsiLog.debug("Start arthas-core.jar args: " + attachArgs);
ProcessUtils.startArthasCore(pid, attachArgs);
进入该方法,我们可以看到如下代码:
command.addAll(attachArgs);
// "${JAVA_HOME}"/bin/java \
// ${opts} \
// -jar "${arthas_lib_dir}/arthas-core.jar" \
// -pid ${TARGET_PID} \
// -target-ip ${TARGET_IP} \
// -telnet-port ${TELNET_PORT} \
// -http-port ${HTTP_PORT} \
// -core "${arthas_lib_dir}/arthas-core.jar" \
// -agent "${arthas_lib_dir}/arthas-agent.jar"
ProcessBuilder pb = new ProcessBuilder(command);
// https://github.com/alibaba/arthas/issues/2166
pb.environment().put("JAVA_TOOL_OPTIONS", "");
这里相当于通过命令行执行java -jar 来启动arthas-core.jar了
启动完core之后,继续往下看.对arthas-client进行了启动
URLClassLoader classLoader = new URLClassLoader(
new URL[] { new File(arthasHomeDir, "arthas-client.jar").toURI().toURL() });
Class<?> telnetConsoleClas = classLoader.loadClass("com.taobao.arthas.client.TelnetConsole");
Method mainMethod = telnetConsoleClas.getMethod("main", String[].class);
即调用com.taobao.arthas.client.TelnetConsole的main方法.
到这里,arthas-boot 流程走完了. 这里大胆猜测一下,arthas-core模块负责的是对目标进程的attach,并开启一个telnet 来监听,arthas-client 则是当前arthas 输入转发,以及结果的输出展示.
现在,接着我们继续查看arthas-core的代码,之前看源码得知,我们是通过java -jar来运行的arthas-core,所以先看core的pom文件,找到入口类.
<mainClass>com.taobao.arthas.core.Arthas</mainClass>
打开对应的类com.taobao.arthas.core.Arthas我们注意看这一行代码:
virtualMachine.loadAgent(arthasAgentPath,
configure.getArthasCore() + ";" + configure.toString());
这里是通过虚拟机的loadAgent 接口,来进行对目标进程的attach一个agent. 附加的agent是arthas-agent 通过查询虚拟机的相关API,我们得知,virtualMachine.loadAgent 执行启动agent后,会执行对应agent的premain 或者agentmain方法.这里我们打开arthas-agent源码,在pom文件可以看到如下代码:
<Agent-Class>com.taobao.arthas.agent334.AgentBootstrap</Agent-Class>
打开该类,找到入口函数:
public static void premain(String args, Instrumentation inst) {
main(args, inst);
}
public static void agentmain(String args, Instrumentation inst) {
main(args, inst);
}
继续展开,往下面看
private static void bind(Instrumentation inst, ClassLoader agentLoader, String args) throws Throwable {
/**
* <pre>
* ArthasBootstrap bootstrap = ArthasBootstrap.getInstance(inst);
* </pre>
*/
Class<?> bootstrapClass = agentLoader.loadClass(ARTHAS_BOOTSTRAP);
Object bootstrap = bootstrapClass.getMethod(GET_INSTANCE, Instrumentation.class, String.class).invoke(null, inst, args);
boolean isBind = (Boolean) bootstrapClass.getMethod(IS_BIND).invoke(bootstrap);
if (!isBind) {
String errorMsg = "Arthas server port binding failed! Please check $HOME/logs/arthas/arthas.log for more details.";
ps.println(errorMsg);
throw new RuntimeException(errorMsg);
}
ps.println("Arthas server already bind.");
}
这段代码的意思是,从arthas-core.jar中,加载com.taobao.arthas.core.server.ArthasBootstrap类,并调用该类的静态方法getInstance
public synchronized static ArthasBootstrap getInstance(Instrumentation instrumentation, String args) throws Throwable {
if (arthasBootstrap != null) {
return arthasBootstrap;
}
Map<String, String> argsMap = FeatureCodec.DEFAULT_COMMANDLINE_CODEC.toMap(args);
// 给配置全加上前缀
Map<String, String> mapWithPrefix = new HashMap<String, String>(argsMap.size());
for (Entry<String, String> entry : argsMap.entrySet()) {
mapWithPrefix.put("arthas." + entry.getKey(), entry.getValue());
}
return getInstance(instrumentation, mapWithPrefix);
}
Instrumentation 这个对象是进行class重载的关键对象,这里我们继续看该类的作用:
this.instrumentation = instrumentation;
initFastjson();
// 1. initSpy()
initSpy();
// 2. ArthasEnvironment
initArthasEnvironment(args);
String outputPathStr = configure.getOutputPath();
if (outputPathStr == null) {
outputPathStr = ArthasConstants.ARTHAS_OUTPUT;
}
outputPath = new File(outputPathStr);
outputPath.mkdirs();
// 3. init logger
loggerContext = LogUtil.initLogger(arthasEnvironment);
// 4. 增强ClassLoader
enhanceClassLoader();
// 5. init beans
initBeans();
// 6. start agent server
bind(configure);
executorService = Executors.newScheduledThreadPool(1, new ThreadFactory() {
@Override
public Thread newThread(Runnable r) {
final Thread t = new Thread(r, "arthas-command-execute");
t.setDaemon(true);
return t;
}
});
shutdown = new Thread("as-shutdown-hooker") {
@Override
public void run() {
ArthasBootstrap.this.destroy();
}
};
transformerManager = new TransformerManager(instrumentation);
Runtime.getRuntime().addShutdownHook(shutdown);
从这里我们可以看到,目标进程进行attach来启动agent后,arthas-agent会加载arthas-core来进行 agent service 的启动,这样相当于目标进程运行了一个服务端,然后等待客户端连接,然后通过命令行交互,来执行对应命令的事项.注册关闭的钩子,做一些必要的清理退出等等.
到这里,整体流程差不多就很清晰了.我们画一个图.
后面我们再来看,命令交互与实现.
