[
https://issues.apache.org/jira/browse/SOLR-1612?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12784970#action_12784970
]
Matt Inger commented on SOLR-1612:
----------------------------------
Thanks for the suggestion, and I'm in the process of doing that, but I have
noticed a few things wrong with the general custom transformer implementation:
1. If I package my custom transformer up in it's own jar file, it seems the
ClassLoader is unable to find the Transformer class from dataimporter jar file,
so I have
to add that to my example/example-DIH/lib directory as well in order to
get it to find the Transformer class.
2. The implementation of the loading of the Transformer class has a bug in it
in regards to how it deals with Simple transformers. It seems if you extend
the Transformer class, you still have to implement the simpler version of the
transformRow method:
{code}
public Object transformRow(Map<String,Object>);
{code}
This slice of code, which is in EntityProcessorWrapper, around line 100, is the
root of the problem.
{code}
try {
Class clazz = DocBuilder.loadClass(trans, context.getSolrCore());
if (clazz.newInstance() instanceof Transformer) {
transformers.add((Transformer) clazz.newInstance());
} else {
final Method meth = clazz.getMethod(TRANSFORM_ROW, Map.class);
if (meth == null) {
String msg = "Transformer :"
+ trans
+ "does not implement Transformer interface or does not
have a transformRow(Map m)method";
log.error(msg);
throw new DataImportHandlerException(
SEVERE, msg);
}
transformers.add(new ReflectionTransformer(meth, clazz, trans));
}
{code}
In particular, the class to "clazz.getMethod" DOES NOT return a null value if
the method does not exist (at least in JDK 1.5). It in fact, throws a
NoSuchMethodException, which means that I have to have the simpler method no
matter whether i'm using it or not.
> Use java.sql.Array to populate Multivalue field in DataImportHandler
> --------------------------------------------------------------------
>
> Key: SOLR-1612
> URL: https://issues.apache.org/jira/browse/SOLR-1612
> Project: Solr
> Issue Type: Improvement
> Components: contrib - DataImportHandler
> Affects Versions: 1.4
> Reporter: Matt Inger
> Fix For: 1.5
>
> Attachments: SOLR-1612.patch
>
>
> I am attaching a patch to the JdbcDataSource class which allows the it to
> populate a multiValued field from a query which returns a java.sql.Array type
> for a column value.
> In order to make this work, you need to do the following:
> 1. In your schema.xml file, set the "multiValued" attribute to the value
> "true"
> <field name=foo" type="string" ... multiValued="true" />
> 2. Have your sql query return a sql type of java.sql.Types.ARRAY. How to do
> this is very database dependent, but in oracle, you first create a type:
> create or replace type FOO_TYPE IS TABLE OF VARCHAR2(255);
> and then when you select the values, you have to use MULTISET and cast
> to the created type (assuming here that FOO_PARENT_ID is the condition to
> limit which foos belong to the current record):
> CAST(MULTISET(SELECT FOO FROM FOOS WHERE
> FOO_PARENT=FOO_PARENT_ID) AS FOO_TYPE ) MY_FOOS
> The main advantage this has is it can save you a ton of queries over the
> current method which will fire off an individual query for each document in
> order to retrieve the values for each multiValued field. This method
> executes much faster, and saves alot of database resources.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
