JUnit4 多线程执行测试用例

转载地址：http://testerhome.com/topics/2502

前言：
之前发过类似的文章，现重新调整了部分格式，部分内容稍作调整和添加，便于阅读。
评论中，有人说直接使用TestNG，就可以实现多线程，是的，但是方式不一样；我们是按照自己的需求对JUnit4自定义多线程Runner，直接在某个类加上相应的注解即可，运行该类就行，支持类和方法级别；TestNG只在方法上有注解 @Test(threadPoolSize = m, invocationCount = n, timeOut = i)实现了对这个方法进行多线程重复跑，threadPoolSize多少个线程执行该方法，invocationCount被执行次数，timeOut每次执行该方法的超时时间，这仅是用多线程重复执行这一个方法，而不是类下面的所有方法同时并发执行，并不是所谓的方法级别并发；TestNG是在xml指定并发的类，方法，组件，具体参照TestNG Executing Parallel Tests Example。
这里不讨论TestNG与JUnit4谁好谁坏，JUnit 4 vs TestNG，只要能满足自己的业务需要即可。
本文仅针对JUnit4进行二次开发。

JUnit4本身是支持多线程，但没有提供多线程的注解；本文将介绍JUnit4自身的多线程实现，自定义对单个类进行多线程执行的Runner和自定义聚合多个类进行多线程执行的Runner。

（一）JUnit4自身的多线程实现

JUnit4提供了ParallerComputer类来使用多线程执行测试用例。
java.lang.Object
extended by org.junit.runner.Computer
extended by org.junit.experimental.ParallelComputer
源码如下：

001    package org.junit.experimental;
002    
003    import java.util.concurrent.ExecutorService;
004    import java.util.concurrent.Executors;
005    import java.util.concurrent.TimeUnit;
006    
007    import org.junit.runner.Computer;
008    import org.junit.runner.Runner;
009    import org.junit.runners.ParentRunner;
010    import org.junit.runners.model.InitializationError;
011    import org.junit.runners.model.RunnerBuilder;
012    import org.junit.runners.model.RunnerScheduler;
013    
014    public class ParallelComputer extends Computer {
015        private final boolean classes;
016    
017        private final boolean methods;
018    
019        public ParallelComputer(boolean classes, boolean methods) {
020            this.classes = classes;
021            this.methods = methods;
022        }
023    
024        public static Computer classes() {
025            return new ParallelComputer(true, false);
026        }
027    
028        public static Computer methods() {
029            return new ParallelComputer(false, true);
030        }
031    
032        private static Runner parallelize(Runner runner) {
033            if (runner instanceof ParentRunner) {
034                ((ParentRunner<?>) runner).setScheduler(new RunnerScheduler() {
035                    private final ExecutorService fService = Executors.newCachedThreadPool();
036    
037                    public void schedule(Runnable childStatement) {
038                        fService.submit(childStatement);
039                    }
040    
041                    public void finished() {
042                        try {
043                            fService.shutdown();
044                            fService.awaitTermination(Long.MAX_VALUE, TimeUnit.NANOSECONDS);
045                        } catch (InterruptedException e) {
046                            e.printStackTrace(System.err);
047                        }
048                    }
049                });
050            }
051            return runner;
052        }
053        //  类的维度
054        @Override
055        public Runner getSuite(RunnerBuilder builder, java.lang.Class<?>[] classes)
056                throws InitializationError {
057            Runner suite = super.getSuite(builder, classes);
058            return this.classes ? parallelize(suite) : suite;
059        }
060        // 方法的维度
061        @Override
062        protected Runner getRunner(RunnerBuilder builder, Class<?> testClass)
063                throws Throwable {
064            Runner runner = super.getRunner(builder, testClass);
065            return methods ? parallelize(runner) : runner;
066        }
067    }

ParallelComputer类中parallelize(Runner runner)方法重写了
ParentRunner类的方法runner.setScheduler(RunnerSchedulerscheduler) ，重新定义了调度顺序，定义了一个线程池 private final ExecutorService fService = Executors.newCachedThreadPool()来多线程执行，运行结束后finished()，关闭线程池fService.shutdown()，并返回该runner。
其中ParallelComputer类重写了父类 Computer的getSuite()和getRunner:

        @Override
        public Runner getSuite(RunnerBuilder builder, java.lang.Class<?>[] classes)
                throws InitializationError {
            Runner suite = super.getSuite(builder, classes);
            return this.classes ? parallelize(suite) : suite;
        }

        @Override
        protected Runner getRunner(RunnerBuilder builder, Class<?> testClass)
               throws Throwable {
           Runner runner = super.getRunner(builder, testClass);
            return methods ? parallelize(runner) : runner;
        }

getSuite()和getRunner()根据ParallelComputer类的全局final变量classes和methods的值去决定是否多线程执行；
classes为true时，并发以类为维度，如下：
return this.classes ? parallelize(suite) : suite;
methods为true时，并发以方法为维度，如下：
return methods ? parallelize(runner) : runner;

ParallelComputer类提供了带参的构造函数：public ParallelComputer(boolean classes, boolean methods)
可以在类初始化时，直接定义多线程执行（不同维度）的对象。
JUnitCore类中的方法runClasses():public static Result runClasses(Computer computer,Class<?>... classes)，可以在main()函数里直接运行测试用例，参数Computer是ParallelComputer的父类，可以直接new ParallelComputer（boolean classes, boolean methods）对象作为第一个形参。

实例1：

public class A {

    @Test
    public void a() {
        assertThat(3, is(1));
    }

    @Test
    public void b() {
        assertThat(3, not(1));
    }

}

public class B {

    @Test
    public void c() {
        assertThat(3, greaterThan(1));
    }

    @Test
    public void d() {
        assertThat(3, lessThan(1));
    }

}

public class ParallelTest {

    public static void main(String[] args) {
        Class[] cls = { A.class, B.class };

        Result rt;

        // 并发以类为维度
        // rt = JUnitCore.runClasses(ParallelComputer.classes(), cls);

        // 并发以方法为维度
        // rt = JUnitCore.runClasses(ParallelComputer.methods(), cls);

        // 并发以类和方法为维度
        rt = JUnitCore.runClasses(new ParallelComputer(true, true), cls);

        System.out.println(rt.getRunCount() + " " + rt.getFailures()  + " " + rt.getRunTime());
    }

}

// A,B两个类并发执行，但类的方法还是串行执行；
JUnitCore.runClasses(ParallelComputer.classes(), cls);
// A,B两个类串行执行，但类的方法并发执行
JUnitCore.runClasses(ParallelComputer.methods(), cls);
// A,B两个类并发执行，其方法也并发执行
JUnitCore.runClasses(new ParallelComputer(true, true), cls);

（二）自定义对单个类进行多线程执行的Runner

package com.weibo.concurrent;

import java.util.concurrent.atomic.AtomicInteger;

import org.junit.runner.notification.RunNotifier;
import org.junit.runners.BlockJUnit4ClassRunner;
import org.junit.runners.model.FrameworkMethod;
import org.junit.runners.model.InitializationError;
import org.junit.runners.model.Statement;

/**
 * Runs all tests in parallel and waits for them to complete. 
 * 
 */
public class MultiThreadedRunner extends BlockJUnit4ClassRunner {

    private AtomicInteger numThreads;

    public static int maxThreads = 10;

    public MultiThreadedRunner (Class<?> klass) throws InitializationError {
        super (klass);
        numThreads = new AtomicInteger(0);
    }

    // Runs the test corresponding to child，which can be assumed to be an element of the list returned by getChildren()
    @Override
    protected void runChild(final FrameworkMethod method, final RunNotifier notifier) {
        while (numThreads.get() > maxThreads) {
            try {
                Thread.sleep(1000);
            } catch (InterruptedException e) {
                System.err.println ("Interrupted: " + method.getName());
                e.printStackTrace();
                return; // The user may have interrupted us; this won't happen normally
            }
        }
        numThreads.incrementAndGet();
        // 用线程执行父类runChild（method, notifier）
        new Thread (new Test(method, notifier)).start();
    }

    // childrenInvoker() call runChild(Object, RunNotifier) on each object returned by getChildren()
    // evaluate() run the action, 调用父类BlockJUnit4ClassRunner的evaluate()
    @Override
    protected Statement childrenInvoker(final RunNotifier notifier) {
        return new Statement() {
            @Override
            public void evaluate() throws Throwable {
                MultiThreadedRunner.super.childrenInvoker(notifier).evaluate();
                // wait for all child threads (tests) to complete
                while (numThreads.get() > 0) {
                    Thread.sleep(1000);
                }
            }
        };
    }

    class Test implements Runnable {
        private final FrameworkMethod method;
        private final RunNotifier notifier;

        public Test (final FrameworkMethod method, final RunNotifier notifier) {
            this.method = method;
            this.notifier = notifier;
        }

        @Override
        public void run () {
            System.err.println (method.getName());
            MultiThreadedRunner.super.runChild(method, notifier);
            numThreads.decrementAndGet();
        }
    }

}

只要在单个测试类前，加上注解：@RunWith(MultiThreadRunner.class),就可以并发的执行用例。
如下图：

(三)自定义聚合多个类进行多线程执行的Runner

有时我们需要聚合同一个模块的测试类，如果使用@RunWith(Suite.class)@SuiteClasses({A.class,B.class})，当类较多时，需要一一列举，效率不高；可以使用ClasspathSuite，支持过滤，将类名符合一定规则的类聚合，官方文档。

实现代码如下：

package com.weibo.concurrent;

import org.junit.experimental.categories.Categories;
import org.junit.extensions.cpsuite.ClasspathSuite;
import org.junit.internal.builders.AllDefaultPossibilitiesBuilder;
import org.junit.runner.Runner;
import org.junit.runners.ParentRunner;
import org.junit.runners.model.InitializationError;
import org.junit.runners.model.RunnerBuilder;
import org.junit.runners.model.RunnerScheduler;

import com.weibo.common.MbLogger;

import java.lang.reflect.Method;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.LinkedList;
import java.util.List;
import java.util.Queue;
import java.util.concurrent.CompletionService;
import java.util.concurrent.ExecutorCompletionService;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.ThreadFactory;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;

/**
 * @author hugang
 *
 **/
public final class ConcurrentSuite extends ClasspathSuite {

    public static Runner MulThread(Runner runner) {
        if (runner instanceof ParentRunner) {
            // setScheduler(RunnerScheduler scheduler):Sets a scheduler that
            // determines the order and parallelization of children
            // RunnerScheduler:Represents a strategy for scheduling when
            // individual test methods should be run (in serial or parallel)
            ((ParentRunner) runner).setScheduler(new RunnerScheduler() {
                private final ExecutorService fService = Executors
                        .newCachedThreadPool();

                // private final ExecutorService fService =
                // Executors.newFixedThreadPool(10);

                // Schedule a child statement to run
                public void schedule(Runnable childStatement) {
                    this.fService.submit(childStatement);
                }

                // Override to implement any behavior that must occur after all
                // children have been scheduled
                public void finished() {
                    try {
                        this.fService.shutdown();
                        this.fService.awaitTermination(9223372036854775807L,
                                TimeUnit.NANOSECONDS);
                    } catch (InterruptedException e) {
                        e.printStackTrace(System.err);
                    }
                }
            });
        }
        return runner;
    }

    public ConcurrentSuite(final Class<?> klass) throws InitializationError {
        // 调用父类ClasspathSuite构造函数
        // AllDefaultPossibilitiesBuilder根据不同的测试类定义（@RunWith的信息）返回Runner,使用职责链模式
        super(klass, new AllDefaultPossibilitiesBuilder(true) {
            @Override
            public Runner runnerForClass(Class<?> testClass) throws Throwable {
                List<RunnerBuilder> builders = Arrays
                        .asList(new RunnerBuilder[] { ignoredBuilder(),
                                annotatedBuilder(), suiteMethodBuilder(),
                                junit3Builder(), junit4Builder() });
                for (RunnerBuilder each : builders) {
                    // 根据不同的测试类定义（@RunWith的信息）返回Runner
                    Runner runner = each.safeRunnerForClass(testClass);
                    if (runner != null)
                        // 方法级别，多线程执行
                        return MulThread(runner);
                }
                return null;
            }
        });

        // 类级别，多线程执行
        setScheduler(new RunnerScheduler() {
            private final ExecutorService fService = Executors
                    .newCachedThreadPool();

            @Override
            public void schedule(Runnable paramRunnable) {
                // TODO Auto-generated method stub
                fService.submit(paramRunnable);
            }

            @Override
            public void finished() {
                // TODO Auto-generated method stub
                try {
                    fService.shutdown();
                    fService.awaitTermination(Long.MAX_VALUE,
                            TimeUnit.NANOSECONDS);
                } catch (InterruptedException e) {
                    e.printStackTrace(System.err);
                }
            }

        });
    }

}

新建一个聚合的IntegrationBeijingOneTests.java文件：

@RunWith(ConcurrentSuite.class)
@ClassnameFilters({"com.weibo.cases.xuelian.*Test", "!.*RemindTest","com.weibo.cases.maincase.*Xuelian"})
@Concurrent
public interface IntegrationBeijingOneTests {

}

再建一个suite文件，XuelianTestSuite.java：

package com.weibo.cases.suite;
import org.junit.experimental.categories.Categories;
import org.junit.runner.RunWith;
import org.junit.runners.Suite.SuiteClasses;

@RunWith(Categories.class)
@SuiteClasses( IntegrationBeijingOneTests.class )
public class XuelianTestSuite {

}

直接运行XuelianTestSuite.java即可，执行过程如下：

写在最后：

设计测试用例时需考虑线程安全。
建议（本组内用例）：
1.账号的使用，同一个测试类中每个测试方法之间需使用不同测试账号（之前未考虑并发，串行执行时方法间使用同样账号，没有影响），咱们组V4的用例共1516个，假设每个用例使用3个账号，则同时执行用例时，则需4548个账号，现库里有1617个账号，可能需要增加用户（空间换时间）； 当然也可以控制并发执行测试方法的数量，来减少用户的使用，比如可以指定同时5个（可调）测试方法并发执行，当然，执行时间上就会相应的增加。
2.非final的全局变量，全改写到测试方法内定义，变成局部变量。