性能测试脚本的编写和调试

以前，我曾说过， 您总是想保留一些调试器断点作为例外 。 这有助于防止代码在不注意的情况下腐烂掉，有时掩盖了另一个问题。

如果您认真对待这一点，那么最好将此想法扩展到自动化测试中。 但是想出一个全面的解决方案并不简单。 您可以仅从try / catch开始，但不会捕获其他线程上的异常。 您也可以使用AOP进行操作； 但是，根据框架的不同，您不能保证完全捕获所有内容，这确实意味着您正在使用稍微不同的代码进行测试，这将使您有些担心。

几天前，我遇到了有关如何编写自己的调试器的博客文章，我想知道java进程是否有可能自行调试。 事实证明，可以的，这是我作为这个小小的思想实验的一部分提出的代码。

该类的第一部分仅包含一些相当hacky的代码，用于根据启动参数来猜测连接回同一VM所需的端口。 可能可以使用Attach机制启动调试器。 但是我没有看到一种明显的方法来使其工作。 然后只有几个工厂方法带有要查找的异常列表。

package com.kingsfleet.debug;import com.sun.jdi.Bootstrap;
import com.sun.jdi.ReferenceType;
import com.sun.jdi.VirtualMachine;
import com.sun.jdi.connect.AttachingConnector;
import com.sun.jdi.connect.Connector;
import com.sun.jdi.connect.IllegalConnectorArgumentsException;
import com.sun.jdi.event.ClassPrepareEvent;
import com.sun.jdi.event.Event;
import com.sun.jdi.event.EventQueue;
import com.sun.jdi.event.EventSet;
import com.sun.jdi.event.ExceptionEvent;
import com.sun.jdi.event.VMDeathEvent;
import com.sun.jdi.event.VMDisconnectEvent;
import com.sun.jdi.request.ClassPrepareRequest;
import com.sun.jdi.request.EventRequest;
import com.sun.jdi.request.ExceptionRequest;import java.io.IOException;import java.lang.management.ManagementFactory;
import java.lang.management.RuntimeMXBean;import java.util.Collections;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.TimeUnit;public class ExceptionDebugger implements AutoCloseable {public static int getDebuggerPort() {// Try to work out what port we need to connect toRuntimeMXBean runtime = ManagementFactory.getRuntimeMXBean();List<String> inputArguments = runtime.getInputArguments();int port = -1;boolean isjdwp = false;for (String next : inputArguments) {if (next.startsWith("-agentlib:jdwp=")) {isjdwp = true;String parameterString = next.substring("-agentlib:jdwp=".length());String[] parameters = parameterString.split(",");for (String parameter : parameters) {if (parameter.startsWith("address")) {int portDelimeter = parameter.lastIndexOf(":");if (portDelimeter != -1) {port = Integer.parseInt(parameter.substring(portDelimeter + 1));} else {port = Integer.parseInt(parameter.split("=")[1]);}}}}}return port;}public static ExceptionDebugger connect(final String... exceptions) throws InterruptedException {return connect(getDebuggerPort(),exceptions);}public static ExceptionDebugger connect(final int port, final String... exceptions) throws InterruptedException {ExceptionDebugger ed = new ExceptionDebugger(port, exceptions);return ed;}

构造函数创建一个简单的守护程序线程，以启动与虚拟机的连接。 非常重要的是，这是一个单独的线程，否则很明显，当我们遇到断点时，VM会停止运行。 确保该线程中的代码不会引发异常是一个好主意-目前，我只是希望做到最好。

最后，代码仅维护了一个禁止的异常列表，如果您有更多的时间，应该可以将堆栈跟踪存储在发生异常的位置。

// // Instance variablesprivate final CountDownLatch startupLatch = new CountDownLatch(1);private final CountDownLatch shutdownLatch = new CountDownLatch(1);private final Set<String> set = Collections.synchronizedSet(new HashSet<String>());private final int port;private final String exceptions[];private Thread debugger;private volatile boolean shutdown = false;//// Object construction and methods//private ExceptionDebugger(final int port, final String... exceptions) throws InterruptedException {this.port = port;this.exceptions = exceptions;debugger = new Thread(new Runnable() {@Overridepublic void run() {try {connect();} catch (Exception ex) {ex.printStackTrace();}}}, "Self debugging");debugger.setDaemon(true); // Don't hold the VM opendebugger.start();// Make sure the debugger has connectedif (!startupLatch.await(1, TimeUnit.MINUTES)) {throw new IllegalStateException("Didn't connect before timeout");}}@Overridepublic void close() throws InterruptedException {shutdown = true;// Somewhere in JDI the interrupt was being eaten, hence the volatile flag debugger.interrupt();shutdownLatch.await();}/*** @return A list of exceptions that were thrown*/public Set<String> getExceptionsViolated() {return new HashSet<String>(set);}/*** Clear the list of exceptions violated*/public void clearExceptionsViolated() {set.clear();}

主要的连接方法是一个相当简单的代码块，可确保连接并配置任何初始断点。

//// Implementation details//private void connect() throws java.io.IOException {try {// Create a virtual machine connectionVirtualMachine attach = connectToVM();try{// Add prepare and any already loaded exception breakpointscreateInitialBreakpoints(attach);// We can now allow the rest of the work to go on as we have created the breakpoints// we requiredstartupLatch.countDown();// Process the eventsprocessEvents(attach);}finally {// Disconnect the debuggerattach.dispose();// Give the debugger time to really disconnect// before we might reconnect, couldn't find another// way to do thistry {TimeUnit.SECONDS.sleep(1);} catch (InterruptedException e) {Thread.currentThread().interrupt();}}} finally {// Notify watchers that we have shutdownshutdownLatch.countDown();}}

重新连接到自我只是找到合适的连接器的过程，在本例中是Socket，尽管我猜想如果您稍稍修改一下代码，便可以在某些平台上使用共享内存传输。

private VirtualMachine connectToVM() throws java.io.IOException {List<AttachingConnector> attachingConnectors = Bootstrap.virtualMachineManager().attachingConnectors();AttachingConnector ac = null;found:for (AttachingConnector next : attachingConnectors) {if (next.name().contains("SocketAttach")) {ac = next;break;}}Map<String, Connector.Argument> arguments = ac.defaultArguments();arguments.get("hostname").setValue("localhost");arguments.get("port").setValue(Integer.toString(port));arguments.get("timeout").setValue("4000");try {return ac.attach(arguments);} catch (IllegalConnectorArgumentsException e) {throw new IOException("Problem connecting to debugger",e);}}

连接调试器时，您不知道是否已加载您感兴趣的异常，因此您需要为准备类的点和已经加载的点注册断点。

请注意，设置的断点仅用于中断一个线程的策略–否则，出于显而易见的原因，如果调试器线程也进入睡眠状态，则当前VM将停止运行。

private void createInitialBreakpoints(VirtualMachine attach) {// Our first exception is for class loadingfor (String exception : exceptions) {ClassPrepareRequest cpr = attach.eventRequestManager().createClassPrepareRequest();cpr.addClassFilter(exception);cpr.setSuspendPolicy(EventRequest.SUSPEND_EVENT_THREAD);cpr.setEnabled(true);}// Then we can check each in turn to see if it have already been loaded as we might// be late to the game, remember classes can be loaded more than once//for (String exception : exceptions) {List<ReferenceType> types = attach.classesByName(exception);for (ReferenceType type : types) {createExceptionRequest(attach, type);}}}private static void createExceptionRequest(VirtualMachine attach, ReferenceType refType) {ExceptionRequest er = attach.eventRequestManager().createExceptionRequest(refType, true, true);er.setSuspendPolicy(EventRequest.SUSPEND_EVENT_THREAD);er.setEnabled(true);}

事件处理循环轮询包含一个或多个事件实例的EventSet实例。 尽管并非所有这些事件都属于断点请求，所以您必须注意不要总是在事件集上调用简历。 这是因为您可能连续有两个事件集，而代码甚至在阅读第二个事件集之前就调用了resume。 随着代码的赶上，这会导致错过断点。

由于某种原因， JDI似乎正在吃掉中断的标志，因此该布尔属性使用以前的close方法停止循环。

private void processEvents(VirtualMachine attach) {// Listen for eventsEventQueue eq = attach.eventQueue();eventLoop: while (!Thread.interrupted() && !shutdown) {// Poll for event sets, with a short timeout so that we can// be interrupted if requiredEventSet eventSet = null;try {eventSet = eq.remove(500);}catch (InterruptedException ex) {Thread.currentThread().interrupt();continue eventLoop;  }// Just loop again if we have no eventsif (eventSet == null) {continue eventLoop;}//boolean resume = false;for (Event event : eventSet) {EventRequest request = event.request();if (request != null) {int eventPolicy = request.suspendPolicy();resume |= eventPolicy != EventRequest.SUSPEND_NONE;}if (event instanceof VMDeathEvent || event instanceof VMDisconnectEvent) {// This should never happen as the VM will exit before this is called} else if (event instanceof ClassPrepareEvent) {// When an instance of the exception class is loaded attach an exception breakpointClassPrepareEvent cpe = (ClassPrepareEvent) event;ReferenceType refType = cpe.referenceType();createExceptionRequest(attach, refType);} else if (event instanceof ExceptionEvent) {String name = ((ExceptionRequest)event.request()).exception().name();set.add(name);}}// Dangerous to call resume always because not all event suspend the VM// and events happen asynchornously.if (resume)eventSet.resume();}}}

因此，剩下的只是一个简单的测试示例，因为这是JDK 7，而ExceptionDebugger是AutoCloseable，所以我们可以使用try-with-resources构造进行此操作，如下所示。 显然，如果要进行自动化测试，请使用您选择的测试框架固定装置。

public class Target {public static void main(String[] args) throws InterruptedException {try (ExceptionDebugger ex = ExceptionDebugger.connect(NoClassDefFoundError.class.getName())) {doSomeWorkThatQuietlyThrowsAnException();System.out.println(ex.getExceptionsViolated());}System.exit(0);}private static void doSomeWorkThatQuietlyThrowsAnException() {// Check to see that break point gets firedtry {Thread t = new Thread(new Runnable() {public void run() {try{throw new NoClassDefFoundError();}catch (Throwable ex) {}}});t.start();t.join();} catch (Throwable th) {// Eat this and don't tell anybody}}
}

因此，如果使用以下VM参数运行此类，请注意suspend = n，否则代码将不会开始运行，您会发现它可以重新连接到自身并开始运行。

-agentlib:jdwp=transport=dt_socket,address=localhost:5656,server=y,suspend=n

这将为您提供以下输出，请注意来自VM的额外调试行：

Listening for transport dt_socket at address: 5656java.lang.NoClassDefFoundError 
Listening for transport dt_socket at address: 5656

每个人都想读一下这是否对人们有用并有助于消除任何明显的错误。

翻译自: https://www.javacodegeeks.com/2013/10/write-an-auto-debugger-to-catch-exceptions-during-test-execution.html

性能测试脚本的编写和调试