为什么80%的码农都做不了架构师？>>> 
已经有一篇文章，经测试是可以的，不过好像不太正宗，可以到这里参考原文。
官方的连接repository的示例代码如下：
KettleEnvironment.init(); RepositoriesMeta repositoriesMeta = new RepositoriesMeta(); repositoriesMeta.readData(); RepositoryMeta repositoryMeta = repositoriesMeta.findRepository( repositoryName ); PluginRegistry registry = PluginRegistry.getInstance(); Repository repository = registry.loadClass( RepositoryPluginType.class, repositoryMeta, Repository.class ); repository.init(repositoryMeta); repository.connect(username, password); RepositoryDirectoryInterface tree=repository.loadRepositoryDirectoryTree(); TransMeta transMeta=repository.loadTransformation("first",tree,listener,false,null); Trans Trans = new Trans(transMeta); trans.execute(arguments); trans.waitUntilFinished(); Result result = trans.getResult(); //这样就获取到了数据库形式的资源库，可以加载转换和job并运行了。repositoryName就是在spoon中创建的资源库名称， 这里会读取用户目录下.kettle/repositories.xml文件，并根据repositoryName获取资源库的具体信息。
添加 parameters, variables 和 arguments
try {
for (String key : parameterMap.keySet()) {
transMeta.setParameterValue(key, parameterMap.get(key));
}
for (String key : variableMap.keySet()) {
transMeta.setVariable(key, variableMap.get(key));
}
} catch (UnknownParamException e) {
error(e.getMessage());
}
transMeta.setArguments(arguments);
获取转换或job中某一步骤的输入和输出的数据，需要添加一个RowListener（RowAdapter）：
Trans Trans = new Trans(transMeta);
// prepare the execution of the transformation (instead of simply execute)
trans.prepareExecution(arguments);
// Find a step thread (ready to run but not yet started)
// You can also use method Trans.findBaseStep which gives back a list of all the step copies
//先获取感兴趣的那个步骤，也可以使用Trans.findBaseStep方法获取所有步骤的一个副本。
StepInterface step = trans.findRunThread("Your Step Name");
// Attach a row listener to a step copy
step.addRowListener(new RowAdapter() {
public void rowReadEvent(RowMetaInterface rowMeta, Object[] row) throws KettleStepException {
// Here you get the rows as they are read by the step
//该步骤读取的数据，即上一步骤传入的数据
}
public void rowWrittenEvent(RowMetaInterface rowMeta, Object[] row) throws KettleStepException {
// Here you get the rows as they are written by the step
//该步骤输出的数据，即处理后传给下一步骤的数据
}
}
);
// Now start the transformation threads...
trans.startThreads();
// If you want to wait until the transformation is finished...
trans.waitUntilFinished();
// If you want to know about the execution result.
Result result = trans.getResult();
还可以手动向转换中添加数据，不需要从文件或数据库中读取，详情这里。
Options
| Option | Definition |
|---|---|
| Step name | The name of this step as it appears in the transformation workspace. |
| Fieldname | Specify the field name of the rows to inject. |
| Type | Specify the type of data. |
| Length | For Number: Total number of significant figures in a number; For String: total length of string; For Date: length of printed output of the string. |
| Precision | For Number: Number of floating point digits; For String, Date, Boolean: unused. |
Example
Here is some information on how to do it:
- You can ask a Trans object for a RowProducer object
- Also see the unit test case: org.pentaho.di.trans.RowProducerTest
Use this type of code:
Trans trans = new Trans(... TransMeta ...); trans.prepareExecution(args); RowProcuder rp = trans.addRowProducer(String stepname, int stepCopy);
After that you start the threads in the transformation. Then you can inject the rows while the transformation is running:
trans.startThreads(); ... rp.putRow(some row you need to inject); ...
You can also specify the rows you are expecting to be injected. This makes it easier to build transformations because you have the meta-data at design time.
嵌入到其它项目中使用，参考这里文档
javadoc
开发者中心
未完待续，如有更好的相关博文，可以留言贴地址，继续参考丰富一下。