Logging

Logging: A Quick Overview

For logging, we use logback which implements the slf4j API. This means you can use any slf4j compliant logging framework, with a default configuration which just works out of the box and bindings to the other common libraries (java.util.logging, log4j, …) if you prefer one of those.

To use:

  • Users: If using a brooklyn binary installation, simply edit the logback.xml or logback-custom.xml supplied in the archive, sometimes in a conf/ directory.

  • Developers: When setting up a new project, if you want logging it is recommended to include the brooklyn-logback-xml project as an optional and provided maven dependency, and then to put custom logging configuration in either logback-custom.xml or logback-main.xml, as described below.

Customizing Your Logging

The project brooklyn-logback-xml supplies a logback.xml configuration, with a mechanism which allows it to be easily customized, consumed, and overridden. You may wish to include this as an optional dependency so that it is not forced upon downstream projects. This logback.xml file supplied contains just one instruction, to include logback-main.xml, and that file in turn includes:

  • logback-custom.xml
  • brooklyn/logback-appender-file.xml
  • brooklyn/logback-appender-stdout.xml
  • brooklyn/logback-logger-excludes.xml
  • brooklyn/logback-debug.xml

For the most common customizations, simply create a logback-custom.xml on your classpath (ensuring it is loaded before brooklyn classes in classpath ordering in the pom) and supply your customizations there:

<included>
    <!-- filename to log to -->           
    <property name="logging.basename" scope="context" value="acme-app" />
    
    <!-- additional loggers -->
    <logger name="com.acme.app" level="DEBUG"/>
</included>

For other configuration, you can override individual files listed above. For example:

  • To remove debug logging, create a trivial brooklyn/logback-debug.xml, containing simply <included/>.
  • To customise stdout logging, perhaps to give it a threshhold WARN instead of INFO, create a brooklyn/logback-appender-stdout.xml which defines an appender STDOUT.
  • To discard all brooklyn’s default logging, create a logback-main.xml which contains your configuration. This should look like a standard logback configuration file, except it should be wrapped in <included> XML tags rather than <configuration> XML tags (because it is included from the logback.xml which comes with brooklyn-logback-xml.)
  • To redirect all jclouds logging to a separate file include brooklyn/logback-logger-debug-jclouds.xml. This redirects all logging from org.jclouds and jclouds to one of two files: anything logged from Brooklyn’s persistence thread will end up in a persistence.log, everything else will end up in jclouds.log.

You should not supply your own logback.xml if you are using brooklyn-logback-xml. If you do, logback will detect multiple files with that name and will scream at you. If you wish to supply your own logback.xml, do not include brooklyn-logback-xml. (Alternatively you can include a logback.groovy which causes logback to ignore logback.xml.)

You can set a specific logback config file to use with:

-Dlogback.configurationFile=/path/to/config.xml

Assemblies

When building an assembly, it is recommended to create a conf/logback.xml which simply includes logback-main.xml (which comes from the classpath). Users of the assembly can then edit the logback.xml file in the usual way, or they can plug in to the configuration mechanisms described above, by creating files such as logback-custom.xml under conf/.

Including brooklyn-logback-xml as an optional and provided dependency means everything should work correctly in IDE’s but it will not include the extra logback.xml file in the assembly. (Alternatively if you include the conf/ dir in your IDE build, you should exclude this dependency.)

With this mechanism, you can include logback-custom.xml and/or other files underneath src/main/resources/ of a project, as described above (for instance to include custom logging categories and define the log file name) and it should get picked up, both in the IDE and in the assembly.

Tests

Brooklyn projects test scope includes the brooklyn-utils-test-support project which supplies a logback-test.xml. logback uses this file in preference to logback.xml when available (ie when running tests). However the logback-test.xml Brooklyn uses includes the same logback-main.xml call path above, so your configurations should still work.

The only differences of the logback-test.xml configuration is that:

  • Debug logging is included for all Brooklyn packages
  • The log file is called brooklyn-tests.log

Caveats

  • logback uses SLF4J version 1.6 which is not compatible with 1.5.x. If you have dependent projects using 1.5.x (such as older Grails) things may break.

  • If you’re not getting the logging you expect in the IDE, make sure src/main/resources is included in the classpath. (In eclipse, right-click the project, the Build Path -> Configure, then make sure all dirs are included (All) and excluded (None) – mvn clean install should do this for you.)

  • You may find that your IDE logs to a file brooklyn-tests.log if it doesn’t distinguish between test build classpaths and normal classpaths.

  • Logging configuration using file overrides such as this is very sensitive to classpath order. To get a separate brooklyn-tests.log file during testing, for example, the brooklyn-test-support project with scope test must be declared as a dependency before brooklyn-logback-includes, due to the way both files declare logback-appender-file.xml.

  • Similarly note that the logback-custom.xml file is included after logging categories and levels are declared, but before appenders are declared, so that logging levels declared in that file dominate, and that properties from that file apply to appenders.

  • Finally remember this is open to improvement. It’s the best system we’ve found so far but we welcome advice. In particular if it could be possible to include files from the classpath with wildcards in alphabetical order, we’d be able to remove some of the quirks listed above (though at a cost of some complexity!).