How do I index rich-format documents contained as database BLOBs with Solr 4.0+?

Question

I've found a few related solutions to this problem. The related solutions will not work for me as I'll explain. (I'm using Solr 4.0 and indexing data stored in an Oracle 11g database.)

Jonck van der Kogel's related solution (from 2009) is explained here. He describes creating a custom Transformer, sort of like the ClobTransformer that ships with Solr. This is going down the elegant path but is not using Tika which is now integrated with Solr. (He uses external PDFBox and FontBox.) This creates multiple maintenance / upgrade dependencies. Also, I need to be able to index Word documents in addition to PDF.

Since Kogel's solutions seems to be on the right path, is there a way to use the Tika classes included with Solr in a custom Transformer? That would allow all the Tika functionality with Kogel's elegant database solution.

Another related solution is the ExtractingRequestHandler (ERH) that ships with Solr. However, as the name suggests, this is a request handler, such as to handle HTTP posts of rich-text documents. To extract documents from the database this way has performance and security problems. I would have to make the database BLOBs accessible via HTTP. I've found no discussion of using ERH for direct ingest from a database BLOB. Is it possible to directly ingest from database BLOBs with Solr Cell?

Another related solution is to write a Transformer (like Kogel's above) to convert a byte[] to a string (from DataImportHandler FAQ). With true binary documents this is going to feed junk into the index and not properly extract the text elements like Tika does. Won't work.

A final related solution is UpdateRichDocuments offered by the RichDocumentHandler. This is deprecated and no longer available in Solr. The page refers you to the ExtractingRequestHandler (discussed above).

It seems like the right solution is to use DataImportHandler and a customer Transformer using the Tika class. How does this work?

Answer 1

Many hours later... First, there is a lot of misleading, wrong and useless information on this problem. No page seemed to provide everything in one place. All of the information is well intentioned but between differing versions and some going over my head, it didn't solve the problem. Here is my collection of what I learned and the solution. To reiterate, I'm using Solr 4.0 (on Tomcat) + Oracle 11g.

Solution overview: DataImportHandler + TikaEntityProcessor + FieldStreamDataSource

Step 1, make sure you update your solrconfig.xml so that solr can find the TikaEntityProcessor + DataImportHandler + Solr Cell stuff.

<lib dir="../contrib/dataimporthandler/lib" regex=".*\.jar" />
<!-- will include extras (where TikaEntPro is) and regular DIH -->
<lib dir="../dist/" regex="apache-solr-dataimporthandler-.*\.jar" />
<lib dir="../contrib/extraction/lib" regex=".*\.jar" />
<lib dir="../dist/" regex="apache-solr-cell-\d.*\.jar" />

Step 2, modify your data-config.xml to include your BLOB table. This is where I had the most trouble since the solutions to this problems have changed a lot as versions have changed. Plus, using multiple data sources and plugging them together correctly was not intuitive to me. Very sleek once it's done though. Make sure to replace your IP, SID name, username, password, table names, etc.

<dataConfig>
  <dataSource name="dastream" type="FieldStreamDataSource" />
  <dataSource name="db" type="JdbcDataSource"
    driver="oracle.jdbc.OracleDriver"
    url="jdbc:oracle:thin:@192.1.1.1:1521:sid"
    user="username"
    password="password"/>
  <document>
    <entity 
      name="attachments" 
      query="select * from schema.attachment_table"
      dataSource="db">
      <entity 
        name="attachment" 
        dataSource="dastream"
        processor="TikaEntityProcessor"
        url="blob_column"
        dataField="attachments.BLOB_COLUMN"
        format="text">
        <field column="text" name="body" />
      </entity>
    </entity>
    <entity name="unrelated" query="select * from another_table" dataSource="db">
    </entity>
  </document>
</dataConfig>

Important note here. If you're getting "No field available for name : whatever" errors when you attempt to import, the FieldStreamDataSource is not able to resolve the data field name you gave. For me, I had to have the url attribute with the lower-case column name, and then the dataField attribute with outside_entity_name.UPPERCASE_BLOB_COLUMN. Also, once I had the column name wrong and that will cause the problem as well.

Step 3, you need to modify your schema.xml to add the BLOB-column field (and any other column you need to index/store). Modify according to your needs.

<field name="body" type="text_en" indexed="false" stored="false" />
<field name="attach_desc" type="text_general" indexed="true" stored="true" />

<field name="text" type="text_en" indexed="true" stored="false" multiValued="true" />
<field name="content" type="text_general" indexed="false" stored="true" multiValued="true" />

<copyField source="body" dest="text" />
<copyField source="body" dest="content" />

With that you should be well on your way to saving many hours getting your binary, rich-text documents (aka rich documents) that are stored as BLOBs in a database column indexed with Solr.

Answer 2

The Integration of Tika and DIH is already provided with Solr via TikaEntityProcessor

Integration - SOLR-1358
Blob Handling - SOLR-1737

You need to just find the right combination.