How-to: Hadoop Integration with OBIEE 11g

The newest release of Oracle Business Intelligence 11.1.1.7 shows Oracle's continued efforts in trying to integrate its Oracle Business Intelligence Platform with big data technologies such as Hadoop and Hive. Specifically, I'm talking about OBIEE 11g's ability to integrate with a Hadoop Data source.

What is Hadoop?

Hadoop is a framework that enables data to be distributed amongst many servers (nodes), commonly referred to as a 'distributed file system'. The data is not stored in a single database, rather it is spread across multiple clusters.

How does Hadoop process data stored in multiple nodes?

Hadoop uses a programming model called 'MapReduce' for parallel processing across multiple nodes. At a high level this is comprised of two steps:

1. Map step
1. The map step takes the data, divides it into smaller sets of data and distributes the result to worker nodes
2. Reduce step
1. The reduce step collects the data from all of the worker nodes and aggregates it into a single 'output'

What is Hive?

MapReduce functions are generally written in Java and generally require someone with deep knowledge in both Hadoop and MapReduce. The guys over at facebook created a technology called 'Hive' which is a data warehouse infrastructure that sits on top of Hadoop. More simply, Hive does the 'heavy lifting' of creating the MapReduce functions. In order to query a Hadoop distributed file system, instead of having to write MapReduce code, you generate sql-style code in a hive language called 'HQL'

Why does this matter in the Oracle Business Intelligence / Analytics space?

The analytics space is experiencing a shift in both technology and function. Traditional BI projects required a 'data warehouse' to store data in a series of star schemas (denormalized models) for quick query generation and data retrieval. The development and support of the data warehouse is achieved through a team of ETL developers whose main focus is to create the mappings that perform the data transformation from the source to the target.
Unless the functional requirements are clearly understood during this phase, value is usually lost in the data transformation and the potential to eliminate relevant data is certainly possible.

Using OBIEE 11g's Hadoop integration via a Hive ODBC, OBIEE can directly query distributed file systems via Hive. What does this mean? The potential now exists to eliminate or reduce the need for ETL as we now have the ability to directly query gigantic file systems.

The saving grace to ETL developers is that a need still exists for someone to create the HQL functions that populate the 'tables' that OBIEE uses. Ultimately, it could be a change in how ETL is developed.

How do you integrate OBIEE 11g with Hadoop?

Step 1: Download the Hive ODBC Drivers from http://support.oracle.com

You can reference Oracle Note 'Using Oracle Hadoop ODBC Driver with BI Administration Tool [ID 1520733.1]'

Step 2: Create a Hive ODBC Connection via the ODBC Data Source Administrator

Similiar to how you create an ODBC connection to edit the repository online, you're going to create an ODBC connection but this time specify the driver as 'Oracle Apache Hadoop Hive WP Driver'

Once you've created the ODBC Data Source Connection, you can configure the Driver set up under the 'General' tab:



Step 3: Configure Database Connection

Moving into the repository, you're going to create a new database connection like you would for any data source in the physical layer. Note that you need to specify the database type as 'Apache Hadoop' (this is important!).

Step 4: Create Connection Pool

Within the Apache Hadoop database connection you just created in step 3, create a data source with a call interface as type 'ODBC 2.0' or 'ODBC 3.5'. The data source call interface should not be 'Apache Hadoop' (you've already specified the database as type as Apache Hadoop!). If you specify the data source call interface as 'Apache Hadoop' you will receive the following error:

Your connection pool should be similar to the following:

You should now be able to import your tables and columns just like any other connection pool. The BI Server will generate normal SQL statements as if it were querying a traditional Oracle database, but the Hive ODBC driver in turn converts that to HQL which is used to execute mapreduce functions to query the Hadoop distributed file system across multiple nodes.



 

keywords: hadoop, obiee 11g, hive, mapreduce, HQL