Chapter 6: Layouts

What is a layout?

In case you were wondering, layouts have nothing to do with large estates in Florida. Layouts are logback components responsible for transforming an incoming event into a String. The doLayout() method in the Layout interface takes an object that represents an event (of any type) and returns a String. A synopsis of the Layout interface is shown below.

public interface Layout<E> extends ContextAware, LifeCycle {

  String doLayout(E event);
  String getFileHeader();
  String getPresentationHeader();
  String getFileFooter();
  String getPresentationFooter();
  String getContentType();
}

This interface is rather simple and yet is sufficient for many formatting needs. The Texan developer from Texas, whom you might know from Joseph Heller's Catch-22, might exclaim: it just takes five methods to implement a layout!!?

Logback-classic

Logback-classic is wired to process only events of type ch.qos.logback.classic.​spi.ILoggingEvent. This fact will be apparent throughout this section.

Writing your own custom Layout

Let us implement a simple yet functional layout for the logback-classic module that prints the time elapsed since the start of the application, the level of the logging event, the caller thread between brackets, its logger name, a dash followed by the event message and a new line.

Sample output might look like:

Here is a possible implementation, authored by the Texan developer:

package chapters.layouts;

import ch.qos.logback.classic.​spi.ILoggingEvent;
import ch.qos.logback.core.​LayoutBase;

public class MySampleLayout extends LayoutBase<ILoggingEvent> {

  public String doLayout(ILoggingEvent event) {
    StringBuffer sbuf = new StringBuffer(128);
    sbuf.append(event.​getTimeStamp() - event.getLoggingContextVO.​getBirthTime());
    sbuf.append(" ");
    sbuf.append(event.​getLevel());
    sbuf.append(" [");
    sbuf.append(event.​getThreadName());
    sbuf.append("] ");
    sbuf.append(event.​getLoggerName();
    sbuf.append(" - ");
    sbuf.append(event.​getFormattedMessage(​));
    sbuf.append(CoreConstants.​LINE_SEP);
    return sbuf.toString();
  }
}

Note that MySampleLayout extends LayoutBase. This class manages state common to all layout instances, such as whether the layout is started or stopped, header, footer and content type data. It allows the developer to concentrate on the formatting expected from his/her Layout. Note that the LayoutBase class is generic. In its class declaration, MySampleLayout extends LayoutBase<ILoggingEvent>.

The doLayout(ILoggingEvent event) method, i.e. the only method in MySampleLayout, begins by instantiating a StringBuffer. It proceeds by adding various fields of the event parameter. The Texan from Texas was careful to print the formatted form of the message. This is significant if one or more parameters were passed along with the logging request.

After adding these various characters to the string buffer, the doLayout() method converts the buffer into a String and returns the resulting value.

In the above example, the doLayout method ignores any eventual exceptions contained in the event. In a real world layout implementation, you would most probably want to print the contents of exceptions as well.

Configuring your custom layout

Custom layouts are configured as any other component. As mentioned earlier, FileAppender and its subclasses expect an encoder. In order to fulfill this requirement, we pass to FileAppender an instance of LayoutWrappingEncoder which wraps our MySampleLayout. Here is the configuration file:

<configuration>

  <appender name="STDOUT" class="ch.qos.logback.​core.ConsoleAppender">
    <encoder class="ch.qos.logback.​core.encoder.LayoutWrappingEncoder">
      <layout class="chapters.​layouts.MySampleLayout" />
    </encoder>
  </appender>

  <root level="DEBUG">
    <appender-ref ref="STDOUT" />
  </root>
</configuration>

The sample application chapters.layouts.​SampleLogging configures logback with the configuration script passed as its first argument and then logs a debug message, followed by an error message.

To run this example issue the following command from within the logback-examples directory.

java chapters.layouts.​SampleLogging src/main/java/chapters/​layouts/sampleLayoutConfig.​xml

This will produce:

0 DEBUG [main] chapters.layouts.​SampleLogging - Everything's going well
0 ERROR [main] chapters.layouts.​SampleLogging - maybe not quite...

That was simple enough. The skeptic Pyrrho of Elea, who insists that nothing is certain except perhaps uncertainty itself, which is by no means certain either, might ask: how about a layout with options? The reader shall find a slightly modified version of our custom layout in MySampleLayout2.java. As mentioned throughout this manual, adding a property to a layout or any other logback component is as simple as declaring a setter method for the property.

The MySampleLayout2 class contains two properties. The first one is a prefix that can be added to the output. The second property is used to choose whether to display the name of the thread from which the logging request was sent.

Here is a copy of the MySampleLayout2 class :

package chapters.layouts;

import ch.qos.logback.classic.;spi.ILoggingEvent;
import ch.qos.logback.core.;LayoutBase;

public class MySampleLayout2 extends LayoutBase<ILoggingEvent> {

  String prefix = null;
  boolean printThreadName = true;

  public void setPrefix(String prefix) {
    this.prefix = prefix;
  }

  public void setPrintThreadName(boolean printThreadName) {
    this.printThreadName = printThreadName;
  }

  public String doLayout(ILoggingEvent event) {
    StringBuffer sbuf = new StringBuffer(128);
    if (prefix != null) {
      sbuf.append(prefix + ": ");
    }
    sbuf.append(event.;getTimeStamp() - event.getLoggerContextVO().getBirthTime());
    sbuf.append(" ");
    sbuf.append(event.;getLevel());
    if (printThreadName) {
      sbuf.append(" [");
      sbuf.append(event.;getThreadName());
      sbuf.append("] ");
    } else {
      sbuf.append(" ");
    }
    sbuf.append(event.;getLoggerName());
    sbuf.append(" - ");
    sbuf.append(event.;getFormattedMessage());
    sbuf.append(LINE_SEP);
    return sbuf.toString();
  }
}

The addition of the corresponding setter method is all that is needed to enable the configuration of a property. Note that the PrintThreadName property is a boolean and not a String. Configuration of logback components was covered in detail in the chapter on configuration. The chapter on Joran provides further detail. Here is the configuration file tailor made for MySampleLayout2.

<configuration>

  <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
    <encoder class="ch.qos.logback.core.encoder.LayoutWrappingEncoder">
      <layout class="chapters.layouts.MySampleLayout2">
        <prefix>MyPrefix</prefix>
        <printThreadName>false</printThreadName>
      </layout>
    </encoder>
  </appender>

  <root level="DEBUG">
    <appender-ref ref="STDOUT" />
  </root>
</configuration>

PatternLayout

Logback classic ships with a flexible layout called PatternLayout. As all layouts, PatternLayout takes a logging event and returns a String. However, this String can be customized by tweaking PatternLayout's conversion pattern.

The conversion pattern of PatternLayout is closely related to the conversion pattern of the printf() function in the C programming language. A conversion pattern is composed of literal text and format control expressions called conversion specifiers. You are free to insert any literal text within the conversion pattern. Each conversion specifier starts with a percent sign '%' and is followed by optional format modifiers, a conversion word and optional parameters between braces. The conversion word controls the data field to convert, e.g. logger name, level, date or thread name. The format modifiers control field width, padding, and left or right justification.

As already mentioned on several occasions, FileAppender and subclasses expect an encoder. Consequently, when used in conjunction with FileAppender or its subclasses a PatternLayout must be wrapped within an encoder. Given that the FileAppender/PatternLayout combination is so common, logback ships with an encoder named PatternLayoutEncoder, designed solely for the purpose of wrapping a PatternLayout instance so that it can be seen as encoder. Below is an example which programmatically configures a ConsoleAppender with a PatternLayoutEncoder:

package chapters.layouts;

import org.slf4j.LoggerFactory;

import ch.qos.logback.classic.;Logger;
import ch.qos.logback.classic.;LoggerContext;
import ch.qos.logback.classic.;encoder.PatternLayoutEncoder;
import ch.qos.logback.classic.;spi.ILoggingEvent;
import ch.qos.logback.core.;ConsoleAppender;

public class PatternSample {

  static public void main(String[] args) throws Exception {
    Logger rootLogger = (Logger)LoggerFactory.;getLogger(Logger.;ROOT_LOGGER_NAME);
    LoggerContext loggerContext = rootLogger.getLoggerContext();
    // we are not interested in auto-configuration
    loggerContext.reset();

    PatternLayoutEncoder encoder = new PatternLayoutEncoder();
    encoder.setContext(loggerContext);
    encoder.setPattern("%-5level [%thread]: %message%n");
    encoder.start();

    ConsoleAppender<ILoggingEvent> appender = new ConsoleAppender<ILoggingEvent>();
    appender.setContext(loggerContext);
    appender.setEncoder(encoder);
    appender.start();

    rootLogger.addAppender(appender);

    rootLogger.debug("Message 1");
    rootLogger.warn("Message 2");
  }
}

In the above example, the conversion pattern is set to be "%-5level [%thread]: %message%n". A synopsis of conversion word included in logback will be given shortly. Running PatternSample application as:

java java chapters.layouts.​PatternSample

will yield the following output on the console.

DEBUG [main]: Message 1 WARN [main]: Message 2

Note that in the conversion pattern "%-5level [%thread]: %message%n" there is no explicit separator between literal text and conversion specifiers. When parsing a conversion pattern, PatternLayout is capable of differentiating between literal text (space characters, the brackets, colon character) and conversion specifiers. In the example above, the conversion specifier %-5level means the level of the logging event should be left justified to a width of five characters. Format specifiers will be explained below.

In PatternLayout, parenthesis can be used to group conversion patterns. It follows that the '(' and ')' carry special meaning and need to be escaped if intended to be used as literals. The special nature of parenthesis is further explained below.

As mentioned previously, certain conversion specifiers may include optional parameters passed between braces. A sample conversion specifier with options could be %logger{10}. Here "logger" is the conversion word, and 10 is the option. Options are further discussed below.

The recognized conversions words along with their options are described in the table below. When multiple conversion words are listed in the same table cell, they are considered as aliases.

mainPackage.foo.​bar.TestException: Houston we have a problem
  at mainPackage.foo.​bar.TestThrower.​fire(TestThrower.​java:22)
  at mainPackage.foo.​bar.TestThrower.​readyToLaunch(TestThrower.​java:17)
  at mainPackage.ExceptionLauncher.​main(ExceptionLauncher.​java:38)

mainPackage.foo.​bar.TestException: Houston we have a problem
  at mainPackage.foo.​bar.TestThrower.​fire(TestThrower.​java:22)

mainPackage.foo.​bar.TestException: Houston we have a problem
  at mainPackage.foo.​bar.TestThrower.​fire(TestThrower.​java:22)
  at mainPackage.foo.​bar.TestThrower.​readyToLaunch(TestThrower.​java:17)
  at mainPackage.ExceptionLauncher.​main(ExceptionLauncher.​java:38)

mainPackage.foo.​bar.TestException: Houston we have a problem
  at mainPackage.foo.​bar.TestThrower.​fire(TestThrower.​java:22)
  at mainPackage.foo.​bar.TestThrower.​readyToLaunch(TestThrower.​java:17)

0    [main] DEBUG - logging statement
Caller+0   at mainPackage.sub.;sample.Bar.sampleMethodName(Bar.java:22)
Caller+1   at mainPackage.sub.;sample.Bar.createLoggingRequest(Bar.java:17)

16   [main] DEBUG - logging statement
Caller+0   at mainPackage.sub.;sample.Bar.sampleMethodName(Bar.java:22)
Caller+1   at mainPackage.sub.;sample.Bar.createLoggingRequest(Bar.java:17)
Caller+2   at mainPackage.ConfigTester.;main(ConfigTester.;java:38)

0    [main] DEBUG - logging statement
Caller+0   at mainPackage.sub.;sample.Bar.createLoggingRequest(Bar.java:17)

%d thread=%thread level=%level logger=%logger user=%X{user} %message

%d %prefix(%thread %level %logger %X{user}) %message

java.lang.NullPointerException
  at com.xyz.Wombat(Wombat.;java:57) ~[wombat-1.3.jar:1.3]
  at com.xyz.Wombat(Wombat.;java:76) ~[wombat-1.3.jar:1.3]
Wrapped by: org.springframework.;BeanCreationException: Error creating bean with name 'wombat':
  at org.springframework.;AbstractBeanFactory.;getBean(AbstractBeanFactory.;java:248) [spring-2.0.jar:2.0]
  at org.springframework.;AbstractBeanFactory.;getBean(AbstractBeanFactory.;java:170) [spring-2.0.jar:2.0]
  at org.apache.catalina.;StandardContext.;listenerStart(StandardContext.;java:3934) [tomcat-6.0.26.jar:6.0.26]
                    

% character has special meaning

Given that in the context of conversion patterns the percent sign carries special meaning, in order to include it as a literal, it needs to be escaped with a backslash, e.g. "%d %p \% %m%n".

Restrictions on literals immediately following conversion words

In most cases literals naturally contain spaces or other delimiting characters so that they are not confused with conversion words. For example, the pattern "%level [%thread] - %message%n" contains the string literals " [" and "] - ". However, if a character which can be part of a java identifier immediately follows a conversion word, logback"s pattern parser will be fooled into thinking that the literal is part of the conversion word. For example, the pattern "%date%nHello" will be interpreted as two conversion words %date and %nHello and since %nHello is not a known conversion word, logback will output %PARSER_ERROR[nHello] for %nHello. If you wish the string literal "Hello" to immediately separate %n and Hello, pass an empty argument list to %n. For example, "%date%n{}Hello" will be interpreted as %date followed by %n followed by the literal "Hello".

Format modifiers

By default the relevant information is output as-is. However, with the aid of format modifiers it is possible to change the minimum and maximum width and the justifications of each data field.

The optional format modifier is placed between the percent sign and the conversion character or word.

The first optional format modifier is the left justification flag which is just the minus (-) character. Then comes the optional minimum field width modifier. This is a decimal constant that represents the minimum number of characters to output. If the data item contains fewer characters, it is padded on either the left or the right until the minimum width is reached. The default is to pad on the left (right justify) but you can specify right padding with the left justification flag. The padding character is space. If the data item is larger than the minimum field width, the field is expanded to accommodate the data. The value is never truncated.

This behavior can be changed using the maximum field width modifier which is designated by a period followed by a decimal constant. If the data item is longer than the maximum field, then the extra characters are removed from the beginning of the data item. For example, if the maximum field width is eight and the data item is ten characters long, then the first two characters of the data item are dropped. This behavior deviates from the printf function in C where truncation is done from the end.

Truncation from the end is possible by appending a minus character right after the period. In that case, if the maximum field width is eight and the data item is ten characters long, then the last two characters of the data item are dropped.

Below are various format modifier examples for the logger conversion specifier.

The table below list examples for format modifier truncation. Please note that the square brackets, i.e the pair of "[]" characters, are not part of the output. They are used to delimit the width of output.

[           main.Name]

[main.Name           ]

[o.bar.Name]

[main.foo.f]

Output just one letter for the level

Instead of printing TRACE, DEBUG, WARN, INFO or ERROR for the level, you may want to print just T, D, W, I and E. You could write a custom converter for this purpose, or simply make use of format modifiers (just discussed) to shorten the level value to a single character. The appropriate conversion specifier would be "%.-1level".

Conversion word options

A conversion specifier can be followed by options. They are always declared between braces. We have already seen some of the possibilities offered by options, for instance in conjunction with the MDC conversion specifier, as in: %mdc{someKey}.

A conversion specifier might have more than one option. For example, a conversion specifier that makes use of evaluators, which will be covered soon, may add evaluator names to the option list, as shown below:

<pattern>%-4relative [%thread] %-5level - %msg%n \
  %caller{2, DISP_CALLER_EVAL, OTHER_EVAL_NAME, THIRD_EVAL_NAME}</pattern>

If the option includes special characters such as a braces, spaces or commas, you can enclose it between single or double quotes. For example, consider the next pattern.

<pattern>%-5level - %replace(%msg){'\d{14,16}', 'XXXX'}%n</pattern>

We pass the options \d{16} and XXXX to the replace conversion word. It replaces any sequence of 14, 15 or 16 digits contained in the message with XXXX effectively obfuscating credit card numbers. Note that "\d" which is a shorthand for a single digit in regular expressions. The "{14,16\}" is interpreted as "{14, 16}", that is, repeat the previous item at least 14 but at most 16 times.

Grouping with parentheses

In logback, parentheses within the pattern string are treated as grouping tokens. Thus, it is possible to group a sub-pattern and apply formatting directives on that sub-pattern.

For example, the pattern

%-30(%d{HH:mm:ss.SSS} [%thread]) %-5level %logger{32} - %msg%n

will group the output generated by the sub-pattern "%d{HH:mm:ss.SSS} [%thread]" so that it is right-padded if less than 30 characters.

Assuming without the grouping the output was

13:09:30 [main] DEBUG c.q.logback.demo.;ContextListener - Classload hashcode is 13995234 13:09:30 [main] DEBUG c.q.logback.demo.;ContextListener - Initializing for ServletContext 13:09:30 [main] DEBUG c.q.logback.demo.;ContextListener - Trying platform Mbean server 13:09:30 [pool-1-thread-1] INFO ch.qos.logback.demo.;LoggingTask - Howdydy-diddly-ho - 0 13:09:38 [btpool0-7] INFO c.q.l.demo.lottery.;LotteryAction - Number: 50 was tried. 13:09:40 [btpool0-7] INFO c.q.l.d.prime.NumberCruncherImpl - Beginning to factor. 13:09:40 [btpool0-7] DEBUG c.q.l.d.prime.NumberCruncherImpl - Trying 2 as a factor. 13:09:40 [btpool0-7] INFO c.q.l.d.prime.NumberCruncherImpl - Found factor 2

with the "%-30()" grouping it would be

13:09:30 [main] DEBUG c.q.logback.demo.;ContextListener - Classload hashcode is 13995234 13:09:30 [main] DEBUG c.q.logback.demo.;ContextListener - Initializing for ServletContext 13:09:30 [main] DEBUG c.q.logback.demo.;ContextListener - Trying platform Mbean server 13:09:30 [pool-1-thread-1] INFO ch.qos.logback.demo.;LoggingTask - Howdydy-diddly-ho - 0 13:09:38 [btpool0-7] INFO c.q.l.demo.lottery.;LotteryAction - Number: 50 was tried. 13:09:40 [btpool0-7] INFO c.q.l.d.prime.NumberCruncherImpl - Beginning to factor. 13:09:40 [btpool0-7] DEBUG c.q.l.d.prime.NumberCruncherImpl - Trying 2 as a factor. 13:09:40 [btpool0-7] INFO c.q.l.d.prime.NumberCruncherImpl - Found factor 2

The latter form is more comfortable to read.

If you need to treat the parenthesis character as a literal, it needs to be escaped by preceding each parenthesis with a backslash. As in, \(%d{HH:mm:ss.SSS} [%thread]\).

Coloring

Grouping by parentheses as explained above allows coloring of sub-patterns. As of version 1.0.5, PatternLayout recognizes "%black", "%red", "%green","%yellow","%blue", "%magenta","%cyan", "%white", "%gray", "%boldRed","%boldGreen", "%boldYellow", "%boldBlue", "%boldMagenta""%boldCyan", "%boldWhite" and "%highlight" as conversion words. These conversion words are intended to contain a sub-pattern. Any sub-pattern enclosed by a coloring word will be output in the specified color.

Below is a configuration file illustrating coloring. Note the %cyan conversion specifier enclosing "%logger{15}". This will output the logger name abbreviated to 15 characters in cyan. The %highlight conversion specifier prints its sub-pattern in bold-red for events of level ERROR, in red for WARN, in BLUE for INFO, and in the default color for other levels.

<configuration debug="true">
  <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
    <withJansi>true</withJansi>
    <encoder>
      <pattern>[%thread] %highlight(%-5level) %cyan(%logger{15}) -%kvp -%msg %n</pattern>
    </encoder>
  </appender>
  <root level="DEBUG">
    <appender-ref ref="STDOUT" />
  </root>
</configuration>

Setting withJansi to true enables ANSI color code interpretation by the Jansi library, which transparently filters out ANSI escape sequences if the underlying terminal is not compatible. This is the safest choice for cross-platform deployments, but requires org.fusesource.jansi:jansi:1.​17 or higher on the class path. Note that Unix-based operating systems such as Linux and Mac OS X support ANSI color codes natively and usually do not require enabling the Jansi library, but doing so is harmless. On Windows however, enabling Jansi is recommended to benefit from color code interpretation on DOS command prompts, which otherwise risk being sent ANSI escape sequences that they cannot interpret.

Here is the corresponding output:

[main] WARN  c.l.TrivialMain - a warning message 0
[main] DEBUG c.l.TrivialMain - hello world number1
[main] DEBUG c.l.TrivialMain - hello world number2
[main] INFO  c.l.TrivialMain - hello world number3
[main] DEBUG c.l.TrivialMain - hello world number4
[main] WARN  c.l.TrivialMain - a warning message 5
[main] ERROR c.l.TrivialMain - Finish off with fireworks

It takes very few lines of code to create a coloring conversion word. The section entitled creating a custom conversion specifier discusses the steps necessary for registering a conversion word in your configuration file.

Evaluators

As mentioned above, option lists come in handy when a conversion specifier is required to behave dynamically based on one or more EventEvaluator objects. EventEvaluator objects have the responsibility to determine whether a given logging event matches the criteria of the evaluator.

Let us review an example involving a EventEvaluator. The next configuration file outputs the logging events to the console, displaying date, thread, level, message and caller data. Given that extracting the caller data of a logging event is on the expensive side, we will do so only when the logging request originates from a specific logger, and when the message contains a certain string. Thus, we make sure that only specific logging requests will have their caller information generated and displayed. In other cases, where the caller data is superfluous, we will not penalize application performance.

Evaluators and in particular evaluation expressions are presented in a dedicated section of the chapter on filters which you MUST read if you want to use evaluators in any meaningful way. Also note that the examples below are implicitly based on JaninoEventEvaluator which requires the Janino library. Please see the corresponding section of the setup document.

<configuration>
  <evaluator name="DISP_CALLER_EVAL">
    <expression>logger.contains("chapters.layouts") &amp;&amp; \
      message.contains("who calls thee")</expression>
  </evaluator>

  <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
    <encoder>
      <pattern>
        %-4relative [%thread] %-5level -%kvp -%msg%n%caller{2, DISP_CALLER_EVAL}
      </pattern>
    </encoder>
  </appender>

  <root level="DEBUG">
    <appender-ref ref="STDOUT" />
  </root>
</configuration>

The above evaluation expression matches events which emanate from a logger with a name containing the string "chapters.layouts" and the message contains the string "who calls thee". Due to XML encoding rules, the & character cannot be written as is, and needs to be escaped as &.

The following class makes use of some of the characteristics mentioned in above configuration file.

Example: Sample usage of EventEvaluators (logback-examples/​src/main/java/chapters/​layouts/CallerEvaluatorExample.​java)

package chapters.layouts;

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

import ch.qos.logback.classic.LoggerContext;
import ch.qos.logback.classic.joran.JoranConfigurator;
import ch.qos.logback.core.joran.spi.JoranException;
import ch.qos.logback.core.util.StatusPrinter;

public class CallerEvaluatorExample {

  public static void main(String[] args)  {
    Logger logger = LoggerFactory.getLogger(CallerEvaluatorExample.class);
    LoggerContext lc = (LoggerContext) LoggerFactory.getILoggerFactory();

    try {
      JoranConfigurator configurator = new JoranConfigurator();
      configurator.setContext(lc);
      configurator.doConfigure(args[0]);
    } catch (JoranException je) {
      // StatusPrinter will handle this
    }
    StatusPrinter.printInCaseOfErrorsOrWarnings(lc);

    for (int i = 0; i < 5; i++) {
      if (i == 3) {
        logger.debug("who calls thee?");
      } else {
        logger.debug("I know me " + i);
      }
    }
  }
}

The above application does nothing particularly fancy. Five logging requests are issued, the third one emitting the message "who calls thee?"

The command

java chapters.layouts.CallerEvaluatorExample src/main/java/chapters/​layouts/callerEvaluatorConfig.​xml

will yield

0    [main] DEBUG - I know me 0
0    [main] DEBUG - I know me 1
0    [main] DEBUG - I know me 2
0    [main] DEBUG - who calls thee?
Caller+0   at chapters.layouts.CallerEvaluatorExample.​main(CallerEvaluatorExample.​java:28)
0    [main] DEBUG - I know me 4

When a logging request is issued, the corresponding logging event is evaluated. Only the third logging event matches the evaluation criteria, causing its caller data to be displayed. For other logging events, the evaluation criteria do not match and no caller data is printed.

One can change the expression to correspond a real world scenario. For instance, one could combine the logger name and request level. Thus, logging requests of level WARN and up, originating from a sensitive part of an application, e.g. a financial transaction module, would have their caller data displayed.

Important: With the caller conversion word, caller data is output when the expression evaluates to true.

Let us consider at a different situation. When exceptions are included in a logging request, their stack trace is also output. However, one might want to suppress the stack trace for some specific exceptions.

The Java code shown below creates three log requests, each with an exception. The second exception is different from the others: it contains the string "do not display this" and it is of type chapters.layouts.​TestException. As its message commands, let us now prevent the second exception from being printed.

Example: Sample usage of EventEvaluators (logback-examples/​src/main/java/chapters/​layouts/ExceptionEvaluatorExample.​java)

package chapters.layouts;

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

import ch.qos.logback.classic.LoggerContext;
import ch.qos.logback.classic.joran.JoranConfigurator;
import ch.qos.logback.core.joran.spi.JoranException;
import ch.qos.logback.core.util.StatusPrinter;

public class ExceptionEvaluatorExample {

  public static void main(String[] args) {
    Logger logger = LoggerFactory.getLogger(ExceptionEvaluatorExample.class);
    LoggerContext lc = (LoggerContext) LoggerFactory.getILoggerFactory();

    try {
      JoranConfigurator configurator = new JoranConfigurator();
      configurator.setContext(lc);
      lc.reset();
      configurator.doConfigure(args[0]);
    } catch (JoranException je) {
       // StatusPrinter will handle this
    } 
    StatusPrinter.printInCaseOfErrorsOrWarnings(lc);

    for (int i = 0; i < 3; i++) {
      if (i == 1) {
        logger.debug("logging statement " + i, new TestException(
            "do not display this"));
      } else {
        logger.debug("logging statement " + i, new Exception("display"));
      }
    }
  }
}

In the next configuration file, the evaluation expression matches events containing a throwable of type chapters.layouts.​TextException, precisely the type of exceptions we wish to suppress.

<configuration>

  <evaluator name="DISPLAY_EX_EVAL">
    <expression>throwable != null &amp;&amp; throwable instanceof  \
      chapters.layouts.TestException</expression>
  </evaluator>

  <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
    <encoder>
      <pattern>%msg%n%ex{full, DISPLAY_EX_EVAL}</pattern>
    </encoder>
  </appender>

  <root level="debug">
    <appender-ref ref="STDOUT" />
  </root>
</configuration>

With this configuration, each time an instance of the chapters.layouts.​TestException is included within a logging request, the stack trace will be suppressed.

Launching the command

java chapters.layouts.​ExceptionEvaluatorExample src/main/java/chapters/​layouts/exceptionEvaluatorConfig.​xml

will yield

logging statement 0 java.lang.Exception: display at chapters.layouts.ExceptionEvaluatorExample.main(ExceptionEvaluatorExample.java:43) [logback-examples-0.9.19.jar:na] logging statement 1 logging statement 2 java.lang.Exception: display at chapters.layouts.ExceptionEvaluatorExample.main(ExceptionEvaluatorExample.java:43) [logback-examples-0.9.19.jar:na]

Notice how the second log statement has no stack trace. We effectively suppressed the stack trace for the TextException. The text between square brackets at the end of each stack trace line is packaging information discussed earlier.

Note With the %ex conversion specifier, the stack trace is displayed when the expression evaluates to false.

Creating a custom conversion specifier

Up to this point we have presented the built-in conversion words in PatternLayout. But it is also possible to add conversion words of your own making.

Building a custom conversion specifier consists of two steps.

Step 1

First, you must extend the ClassicConverter class. ClassicConverter objects are responsible for extracting information out of ILoggingEvent instances and producing a String. For example, LoggerConverter, the converter underlying the %logger conversion word, extracts the name of the logger from ILoggingEvent and returns it as a String. It might abbreviate the logger name in the process.

Here is a customer converter which returns the time elapsed since its creation in nanoseconds:

public class MySampleConverter extends ClassicConverter {

  long start = System.nanoTime();

  @Override
  public String convert(ILoggingEvent event) {
    long nowInNanos = System.nanoTime();
    return Long.toString(nowInNanos-start);
  }
}

This implementation is pretty straightforward. The MySampleConverter class extends ClassicConverter, and implements the convert method which returns the number of nanoseconds elapsed since its creation.

Step 2

In the second step, we must let logback know about the new Converter. For this purpose, we need to declare the new conversion word in the configuration file, as shown below:

<configuration>

  <conversionRule conversionWord="nanos"
                  converterClass="chapters.layouts.MySampleConverter" />

  <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
    <encoder>
      <pattern>%-6nanos [%thread] -%kvp -%msg%n</pattern>
    </encoder>
  </appender>

  <root level="DEBUG">
    <appender-ref ref="STDOUT" />
  </root>
</configuration>

Once the new conversion word has been declared in the configuration file, we can refer to it within PatternLayout pattern, as with any other conversion word.

The command:

should yield output akin to:

4868695 [main] DEBUG - Everything's going well
5758748 [main] ERROR - maybe not quite...

The reader might want to take a look at other Converter implementations such as MDCConverter to learn about more complex behaviours, such as option handling. For creating your own coloring schemes have a look at HighlightingCompositeConverter.

HTMLLayout

HTMLLayout (as included in logback-classic) generates logs in HTML format. HTMLLayout outputs logging events in an HTML table where each row of the table corresponds to a logging event.

Here is a sample output produced by HTMLLayout using its default CSS stylesheet:

The content of table columns are specified with the help of a conversion pattern. See PatternLayout for documentation on conversion patterns. As such, you have full control over the contents and format of the table. You can select and display any combination of converters PatternLayout knows about.

One notable exception about the use of PatternLayout with HTMLLayout is that conversion specifiers should not be separated by space characters or more generally by literal text. Each specifier found in the pattern will result in a separate column. Likewise a separate column will be generated for each block of literal text found in the pattern, potentially wasting valuable real-estate on your screen.

Here is simple but functional configuration file illustrating the use of HTMLLayout.

<configuration debug="true">
  <appender name="FILE" class="ch.qos.logback.core.FileAppender">
    <encoder class="ch.qos.logback.core.encoder.LayoutWrappingEncoder">
      <layout class="ch.qos.logback.classic.html.HTMLLayout">
        <pattern>%relative%thread%mdc%level%logger%msg</pattern>
      </layout>
    </encoder>
    <file>test.html</file>
  </appender>

  <root level="DEBUG">
    <appender-ref ref="FILE" />
  </root>
</configuration>

The TrivialMain application logs a few messages finishing with an exception. The command:

java chapters.layouts.​TrivialMain src/main/java/chapters/​layouts/htmlLayoutConfig1.​xml

will create the file test.html in the current folder. The contents of test.html should be similar to:

Stack traces

If you use the %em conversion word to display stack traces, a table column will be created to display stack traces. In most cases the column will be empty, wasting screen real-estate. Moreover, printing a stack trace on a separate column does not yield very readable results. Fortunately, the %ex conversion word is not the only way to display stack traces.

A better solution is available through implementations of IThrowableRenderer interface. Such an implementation can be assigned to HTMLLayout to manage the display data related to exceptions. By default, a DefaultThrowableRenderer is assigned to each HTMLLayout instance. It writes exceptions on a new table row, along with its stack trace, in an easily readable manner, as shown on the figure above.

If for some reason, you still wish to use the %ex pattern, then you can specify NOPThrowableRenderer in the configuration file in order to disable displaying a separate row for the stack trace. We don't have the faintest idea why you would want to do that, but if you wished, you could.

CSS

The presentation of the HTML created by HTMLLayout is controlled through a Cascading Style Sheet (CSS). In the absence of specific instructions, HTMLLayout will default to its internal CSS. However, you can instruct HTMLLayout to use an external CSS file. For this purpose a cssBuilder element can be nested within a <layout> element, as shown below.

<layout class="ch.qos.logback.classic.html.HTMLLayout">
  <pattern>%relative...%msg</;pattern>
  <cssBuilder class="ch.qos.logback.classic.html.UrlCssBuilder">
    <!-- url where the css file is located -->
    <url>http://...</url>
  </cssBuilder>
</layout>

The HTMLLayout is often used in conjunction with SMTPAppender so that outgoing email is pleasantly formatted in HTML.

Log4j XMLLayout

XMLLayout (part of logback-classic) generates output in a log4j.dtd compliant format to interoperate with tools such as Chainsaw and Vigilog capable of processing files generated by log4j's XMLLayout.

As the original XMLLayout in log4j version 1.2.15, XMLLayout in logback-classic takes two boolean properties, locationInfo and properties. Setting locationInfo to true enables the inclusion of location info (caller data) in each event. Setting properties to true enables the inclusion of MDC information. Both options are set to false by default.

Here is a sample configuration

<configuration>
  <appender name="FILE" class="ch.qos.logback.core.FileAppender">
    <file>test.xml</file>
    <encoder class="ch.qos.logback.core.encoder.LayoutWrappingEncoder">
      <layout class="ch.qos.logback.classic.log4j.XMLLayout">
        <locationInfo>true</locationInfo>
      </layout>
    </encoder>
  </appender>

  <root level="DEBUG">
    <appender-ref ref="FILE" />
  </root>
</configuration>

Logback access

Most logback-access layouts are mere adaptations of logback-classic layouts. Logback-classic and logback-access modules address different needs, but in general offer comparable functionality.

Writing your own Layout

Writing a custom Layout for logback access is nearly identical to its sibling Layout in logback-classic.

PatternLayout

PatternLayout in logback-access can be configured in much the same way as its classic counterpart. However it features additional conversion specifiers suited for logging particular bits of information available only in HTTP servlet requests and HTTP servlet responses.

Below is a list of conversion specifiers for PatternLayout in logback-access.

Logback access' PatternLayout also recognizes three keywords, which act like shortcuts.

The common keyword corresponds to the pattern '%h %l %u [%t] "%r" %s %b' which displays client host, remote log name, user, date, requested URL, status code and response's content length

The combined keyword is a shortcut for '%h %l %u [%t] "%r" %s %b "%i{Referer}" "%i{User-Agent}"'. This pattern begins much like the common pattern but also displays two request headers, namely referer, and user-agent.

HTMLLayout

The HTMLLayout class found in logback-access is similar to the HTMLLayout class from logback-classic.

By default, it will create a table containing the following data:

• Remote IP
• Date
• Request URL
• Status code
• Content Length

Here is a sample output produced by HTMLLayout in logback-access:

What can be better than a real world example? Our own log4j properties for logback translator makes use of logback-access to demonstrate live output from RollingFileAppender with HTMLLayout.

On every new user request to our translator web-application, a new entry will be added to the access logs, which you can view by following this link.