另一个选项是模拟Appender并验证消息是否已记录到此Appender。Log4j 1.2的示例。X和mockito:

import static org.junit.Assert.assertEquals;
import static org.mockito.Mockito.mock;
import static org.mockito.Mockito.verify;

import org.apache.log4j.Appender;
import org.apache.log4j.Level;
import org.apache.log4j.Logger;
import org.apache.log4j.spi.LoggingEvent;
import org.junit.After;
import org.junit.Before;
import org.junit.Test;
import org.mockito.ArgumentCaptor;

public class MyTest {

    private final Appender appender = mock(Appender.class);
    private final Logger logger = Logger.getRootLogger();

    @Before
    public void setup() {
        logger.addAppender(appender);
    }

    @Test
    public void test() {
        // when
        Logger.getLogger(MyTest.class).info("Test");

        // then
        ArgumentCaptor<LoggingEvent> argument = ArgumentCaptor.forClass(LoggingEvent.class);
        verify(appender).doAppend(argument.capture());
        assertEquals(Level.INFO, argument.getValue().getLevel());
        assertEquals("Test", argument.getValue().getMessage());
        assertEquals("MyTest", argument.getValue().getLoggerName());
    }

    @After
    public void cleanup() {
        logger.removeAppender(appender);
    }
}

对于log4j2，解决方案略有不同，因为AppenderSkeleton不再可用。此外，使用Mockito或类似的库来创建带有ArgumentCaptor的Appender将无法工作，因为MutableLogEvent在多个日志消息上被重用。我为log4j2找到的最佳解决方案是:

import org.apache.logging.log4j.LogManager;
import org.apache.logging.log4j.core.LogEvent;
import org.apache.logging.log4j.core.Logger;
import org.apache.logging.log4j.core.appender.AbstractAppender;

private static MockedAppender mockedAppender;
private static Logger logger;

@Before
public void setup() {
    mockedAppender.message.clear();
}

/**
 * For some reason mvn test will not work if this is @Before, but in eclipse it works! As a
 * result, we use @BeforeClass.
 */
@BeforeClass
public static void setupClass() {
    mockedAppender = new MockedAppender();
    logger = (Logger)LogManager.getLogger(MatchingMetricsLogger.class);
    logger.addAppender(mockedAppender);
    logger.setLevel(Level.INFO);
}

@AfterClass
public static void teardown() {
    logger.removeAppender(mockedAppender);
}

@Test
public void test() {
    // do something that causes logs
    for (String e : mockedAppender.message) {
        // add asserts for the log messages
    }
}

private static class MockedAppender extends AbstractAppender {

    List<String> message = new ArrayList<>();

    protected MockedAppender() {
        super("MockedAppender", null, null);
    }

    @Override
    public void append(LogEvent event) {
        message.add(event.getMessage().getFormattedMessage());
    }
}

正如前面提到的，您可以使用mock框架。为此，您必须在类中公开记录器(尽管我可能更倾向于使其包私有而不是创建公共setter)。

另一种解决方案是手工创建一个假记录器。您必须编写伪记录器(更多的fixture代码)，但在这种情况下，我更喜欢使用模拟框架中保存的代码来增强测试的可读性。

我会这样做:

class FakeLogger implements ILogger {
    public List<String> infos = new ArrayList<String>();
    public List<String> errors = new ArrayList<String>();

    public void info(String message) {
        infos.add(message);
    }

    public void error(String message) {
        errors.add(message);
    }
}

class TestMyClass {
    private MyClass myClass;        
    private FakeLogger logger;        

    @Before
    public void setUp() throws Exception {
        myClass = new MyClass();
        logger = new FakeLogger();
        myClass.logger = logger;
    }

    @Test
    public void testMyMethod() {
        myClass.myMethod(true);

        assertEquals(1, logger.infos.size());
    }
}

这里有一个简单有效的Logback解决方案。 它不需要添加/创建任何新类。 它依赖于ListAppender:一个白盒回logback appender，其中将日志条目添加到公共List字段中，我们可以使用该字段来进行断言。

这里有一个简单的例子。

Foo类:

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

public class Foo {

    static final Logger LOGGER = LoggerFactory.getLogger(Foo .class);

    public void doThat() {
        LOGGER.info("start");
        //...
        LOGGER.info("finish");
    }
}

FooTest类:

import org.slf4j.LoggerFactory;
import ch.qos.logback.classic.Level;
import ch.qos.logback.classic.Logger;
import ch.qos.logback.classic.spi.ILoggingEvent;
import ch.qos.logback.core.read.ListAppender;

public class FooTest {

    @Test
    void doThat() throws Exception {
        // get Logback Logger 
        Logger fooLogger = (Logger) LoggerFactory.getLogger(Foo.class);

        // create and start a ListAppender
        ListAppender<ILoggingEvent> listAppender = new ListAppender<>();
        listAppender.start();

        // add the appender to the logger
        // addAppender is outdated now
        fooLogger.addAppender(listAppender);

        // call method under test
        Foo foo = new Foo();
        foo.doThat();

        // JUnit assertions
        List<ILoggingEvent> logsList = listAppender.list;
        assertEquals("start", logsList.get(0)
                                      .getMessage());
        assertEquals(Level.INFO, logsList.get(0)
                                         .getLevel());

        assertEquals("finish", logsList.get(1)
                                       .getMessage());
        assertEquals(Level.INFO, logsList.get(1)
                                         .getLevel());
    }
}

JUnit断言听起来不太适合断言列表元素的某些特定属性。 像AssertJ或Hamcrest这样的Matcher/断言库似乎更好:

使用AssertJ，它将是:

import org.assertj.core.api.Assertions;

Assertions.assertThat(listAppender.list)
          .extracting(ILoggingEvent::getMessage, ILoggingEvent::getLevel)
          .containsExactly(Tuple.tuple("start", Level.INFO), Tuple.tuple("finish", Level.INFO));

另一个值得提及的想法是创建一个CDI生成器来注入记录器，这样模拟就变得容易了，尽管这是一个较老的主题。(而且它还提供了不必再声明“整个logger语句”的优势，但这已经跑题了)

例子:

创建要注入的记录器:

public class CdiResources {
  @Produces @LoggerType
  public Logger createLogger(final InjectionPoint ip) {
      return Logger.getLogger(ip.getMember().getDeclaringClass());
  }
}

限定符:

@Qualifier
@Retention(RetentionPolicy.RUNTIME)
@Target({TYPE, METHOD, FIELD, PARAMETER})
public @interface LoggerType {
}

在生产代码中使用记录器:

public class ProductionCode {
    @Inject
    @LoggerType
    private Logger logger;

    public void logSomething() {
        logger.info("something");
    }
}

在测试代码中测试记录器(给出一个easyMock示例):

@TestSubject
private ProductionCode productionCode = new ProductionCode();

@Mock
private Logger logger;

@Test
public void testTheLogger() {
   logger.info("something");
   replayAll();
   productionCode.logSomething();
}

您可能要测试两件事。

当我的程序的操作人员对事件感兴趣时，我的程序是否执行适当的日志记录操作，以通知操作人员该事件。 当我的程序执行日志记录操作时，它产生的日志消息是否具有正确的文本。

这两件事实际上是不同的，所以可以分别测试。然而，测试第二个(消息文本)问题很大，我建议不要这样做。消息文本的测试最终将包括检查一个文本字符串(预期的消息文本)是否与日志记录代码中使用的文本字符串相同，或者可以简单地从文本字符串派生出来。

这些测试根本不测试程序逻辑，它们只测试一个资源(字符串)是否等同于另一个资源。 这些测试是脆弱的;即使是对日志消息格式的微小调整也会破坏您的测试。 测试与日志接口的国际化(翻译)不兼容。测试假设只有一种可能的消息文本，因此只有一种可能的人类语言。

请注意，让您的程序代码(可能实现了一些业务逻辑)直接调用文本日志接口是糟糕的设计(但不幸的是非常常见)。负责业务逻辑的代码还决定一些日志策略和日志消息的文本。它将业务逻辑与用户界面代码混合在一起(是的，日志消息是程序用户界面的一部分)。这些东西应该是分开的。

因此，我建议业务逻辑不要直接生成日志消息的文本。相反，让它委托给一个日志对象。

The class of the logging object should provide a suitable internal API, which your business object can use to express the event that has occurred using objects of your domain model, not text strings. The implementation of your logging class is responsible for producing text representations of those domain objects, and rendering a suitable text description of the event, then forwarding that text message to the low level logging framework (such as JUL, log4j or slf4j). Your business logic is responsible only for calling the correct methods of the internal API of your logger class, passing the correct domain objects, to describe the actual events that occurred. Your concrete logging class implements an interface, which describes the internal API your business logic may use. Your class(es) that implements business logic and must perform logging has a reference to the logging object to delegate to. The class of the reference is the abstract interface. Use dependency injection to set up the reference to the logger.

然后，您可以通过创建一个模拟记录器(它实现了内部日志API)并在测试的设置阶段使用依赖项注入来测试业务逻辑类是否正确地将事件告知日志接口。

是这样的:

 public class MyService {// The class we want to test
    private final MyLogger logger;

    public MyService(MyLogger logger) {
       this.logger = Objects.requireNonNull(logger);
    }

    public void performTwiddleOperation(Foo foo) {// The method we want to test
       ...// The business logic
       logger.performedTwiddleOperation(foo);
    }
 };

 public interface MyLogger {
    public void performedTwiddleOperation(Foo foo);
    ...
 };

 public final class MySl4jLogger: implements MyLogger {
    ...

    @Override
    public void performedTwiddleOperation(Foo foo) {
       logger.info("twiddled foo " + foo.getId());
    }
 }

 public final void MyProgram {
    public static void main(String[] argv) {
       ...
       MyLogger logger = new MySl4jLogger(...);
       MyService service = new MyService(logger);
       startService(service);// or whatever you must do
       ...
    }
 }

 public class MyServiceTest {
    ...

    static final class MyMockLogger: implements MyLogger {
       private Food.id id;
       private int nCallsPerformedTwiddleOperation;
       ...

       @Override
       public void performedTwiddleOperation(Foo foo) {
          id = foo.id;
          ++nCallsPerformedTwiddleOperation;
       }

       void assertCalledPerformedTwiddleOperation(Foo.id id) {
          assertEquals("Called performedTwiddleOperation", 1, nCallsPerformedTwiddleOperation);
          assertEquals("Called performedTwiddleOperation with correct ID", id, this.id);
       }
    };

    @Test
    public void testPerformTwiddleOperation_1() {
       // Setup
       MyMockLogger logger = new MyMockLogger();
       MyService service = new MyService(logger);
       Foo.Id id = new Foo.Id(...);
       Foo foo = new Foo(id, 1);

       // Execute
       service.performedTwiddleOperation(foo);

       // Verify
       ...
       logger.assertCalledPerformedTwiddleOperation(id);
    }
 }