::Go back to Oozie Documentation Index::

Building Oozie

System Requirements

JDK commands (java, javac) must be in the command path.

The Maven command (mvn) must be in the command path.

Oozie Documentation Generation

To generate the documentation, Oozie uses a patched Doxia plugin for Maven with improved twiki support.

The source of the modified plugin is available in the Oozie GitHub repository, in the ydoxia branch.

To build and install it locally run the following command in the ydoxia branch:

$ mvn install

Passphare-less SSH Setup

NOTE: SSH actions are deprecated in Oozie 2.

To run SSH Testcases and for easier Hadoop start/stop configure SSH to localhost to be passphrase-less.

Create your SSH keys without a passphrase and add the public key to the authorized file:

$ ssh-keygen -t dsa
$ cat ~/.ssh/id_dsa.pub >> ~/.ssh/authorized_keys2

Test that you can ssh without password:

$ ssh localhost

Building with different Java Versions

Oozie requires a minimum Java version of 1.6. Any newer version can be used but by default bytecode will be generated which is compatible with 1.6. This can be changed by specifying the build property targetJavaVersion .

Building and Testing Oozie

The JARs for the specified Hadoop and Pig versions must be available in one of the Maven repositories defined in Oozie main 'pom.xml' file. Or they must be installed in the local Maven cache.

Examples Running Oozie Testcases with Different Configurations

Using embedded Hadoop minicluster with 'simple' authentication:

$ mvn clean test

Using a Hadoop cluster with 'simple' authentication:

$ mvn clean test -Doozie.test.hadoop.minicluster=false

Using embedded Hadoop minicluster with 'simple' authentication and Derby database:

$ mvn clean test -Doozie.test.hadoop.minicluster=false -Doozie.test.db=derby

Using a Hadoop cluster with 'kerberos' authentication:

$ mvn clean test -Doozie.test.hadoop.minicluster=false -Doozie.test.hadoop.security=kerberos

NOTE: The embedded minicluster cannot be used when testing with 'kerberos' authentication.

Using a custom Oozie configuration for testcases:

$ mvn clean test -Doozie.test.config.file=/home/tucu/custom-oozie-sitel.xml

Running the testcases with different databases:

$ mvn clean test -Doozie.test.db=[hsqldb*|derby|mysql|postgres|oracle]

Using mysql and oracle enables profiles that will include their JARs files in the build. If using oracle , the Oracle JDBC JAR file must be manually installed in the local Maven cache (the JAR is not available in public Maven repos).

Build Options Reference

All these options can be set using -D .

Except for the options marked with (*) , the options can be specified in the test.properties in the root of the Oozie project. The options marked with (*) are used in Maven POMs, thus they don't take effect if specified in the test.properties file (which is loaded by the XTestCase class at class initialization time).

includeHadoopJars (*): includes Hadoop JARs and its transitive dependencies in the Oozie WAR file, default is undefined (Hadoop JARs are not included).

hadoop.version (*) : indicates the Hadoop version(Hadoop-1 or Hadoop-2) you wish to build Oozie against specifically. It will substitute this value in the Oozie POM properties and pull the corresponding Hadoop artifacts from Maven. Default version is 1.0.1 for Hadoop-1 (the most common case). For Hadoop-2, the version you can pass is 2.0.0-alpha .

generateSite (*): generates Oozie documentation, default is undefined (no documentation is generated)

skipTests (*): skips the execution of all testcases, no value required, default is undefined

test = (*): runs a single test case, to run a test give the test class name without package and extension, no default

oozie.test.db (*): indicates the database to use for running the testcases, supported values are 'hsqldb', 'derby', 'mysql', 'postgres' and 'oracle'; default value is 'hsqldb'. For each database there is =core/src/test/resources/DATABASE-oozie-site.xml file preconfigured.

oozie.test.properties (*): indicates the file to load the test properties from, by default is test.properties . Having this option allows having different test properties sets, for example: minicluster, simple & kerberos.

oozie.test.waitfor.ratio : multiplication factor for testcases using waitfor, the ratio is used to adjust the effective time out. For slow machines the ratio should be increased. The default value is =1 .

oozie.test.config.file = : indicates a custom Oozie configuration file for running the testcases. The specified file must be an absolute path. For example, it can be useful to specify different database than HSQL for running the testcases.

oozie.test.hadoop.minicluster = : indicates if Hadoop minicluster should be started for testcases, default value 'true'

oozie.test.job.tracker = : indicates the URI of the JobTracker when using a Hadoop cluster for testing, default value 'localhost:8021'

oozie.test.name.node = : indicates the URI of the NameNode when using a Hadoop cluster for testing, default value 'hdfs://localhost:8020'

oozie.test.hadoop.security = : indicates the type of Hadoop authentication for testing, valid values are 'simple' or 'kerberos, default value 'simple'

oozie.test.kerberos.keytab.file = : indicates the location of the keytab file, default value '${user.home}/oozie.keytab'

oozie.test.kerberos.realm = : indicates the Kerberos real, default value 'LOCALHOST'

oozie.test.kerberos.oozie.principal = : indicates the Kerberos principal for oozie, default value '${user.name}/localhost'

oozie.test.kerberos.jobtracker.principal = : indicates the Kerberos principal for the JobTracker, default value 'mapred/localhost'

oozie.test.kerberos.namenode.principal = : indicates the Kerberos principal for the NameNode, default value 'hdfs/localhost'

oozie.test.user.oozie : specifies the user ID used to start Oozie server in testcases, default value is =${user.name} .

oozie.test.user.test : specifies primary user ID used as the user submitting jobs to Oozie Server in testcases, default value is =test .

oozie.test.user.test2 : specifies secondary user ID used as the user submitting jobs to Oozie Server in testcases, default value is =test2 .

oozie.test.user.test3 : specifies secondary user ID used as the user submitting jobs to Oozie Server in testcases, default value is =test3 .

oozie.test.group : specifies group ID used as group when submitting jobs to Oozie Server in testcases, default value is =testg .

NOTE: The users/group specified in oozie.test.user.test2 , oozie.test.user.test3 = and oozie.test.user.group = are used for the authorization testcases only.

oozie.test.dir : specifies the directory where the =oozietests directory will be created, default value is /tmp . The oozietests directory is used by testcases when they need a local filesystem directory.

hadoop.log.dir : specifies the directory where Hadoop minicluster will write its logs during testcases, default value is =/tmp .

test.exclude : specifies a testcase class (just the class name) to exclude for the tests run, for example =TestSubmitCommand .

test.exclude.pattern : specifies one or more patterns for testcases to exclude, for example =**/Test*Command.java .

Testing Map Reduce Pipes Action

Pipes testcases require Hadoop's wordcount-simple pipes binary example to run. The wordcount-simple pipes binary should be compiled for the build platform and copied into Oozie's core/src/test/resources/ directory. The binary file must be named wordcount-simple .

If the wordcount-simple pipes binary file is not available the testcase will do a NOP and it will print to its output file the following message 'SKIPPING TEST: TestPipesMain, binary 'wordcount-simple' not available in the classpath'.

There are 2 testcases that use the wordcount-simple pipes binary, TestPipesMain and TestMapReduceActionExecutor , the 'SKIPPING TEST..." message would appear in the testcase log file of both testcases.

Building an Oozie Distribution

An Oozie distribution bundles an embedded Tomcat server. The Oozie distro module downloads Tomcat TAR.GZ from Apache once (in the distro/downloads/ directory) and uses it when creating the distro.

The following Maven invocation builds an Oozie distribution:

$ mvn clean package assembly:single

Build instructions with Hadoop Yarn (Hadoop-2.x)

Following values can be passed as maven properties

  • -Dhadoop.version : 2.0.0-alpha
  • -Dpig.version (optional) : either 0.9.2 or 0.10.2

For e.g.

$ mvn clean package assembly:single -Dhadoop.version=2.0.0-alpha

If none of these parameters are specified, the defaults are

  • -Dhadoop.version : 1.0.1
  • -Dhadoop.auth.version : 2.0.0-alpha
  • -Dpig.version : 0.9.0

IMPORTANT: It is strongly recommend to use a Bigtop distribution if using Hadoop 2.x because the Oozie sharelibs built from the tarball distribution will not work with it.

The following properties should be specified when building a release:

  • -DgenerateDocs : forces the generation of Oozie documentation
  • -Dbuild.time= : timestamps the distribution
  • -Dvc.revision= : specifies the source control revision number of the distribution
  • -Dvc.url= : specifies the source control URL of the distribution

The provided bin/mkdistro.sh script runs the above Maven invocation setting all these properties to the right values (the 'vc.*' properties are obtained from the local git repository).

IDE Setup

Eclipse and IntelliJ can use directly Oozie Maven project files.

The only special consideration is that the following source directories from the client module must be added to the core module source path:

  • client/src/main/java : as source directory
  • client/src/main/resources : as source directory
  • client/src/test/java : as test-source directory
  • client/src/test/resources : as test-source directory

::Go back to Oozie Documentation Index::