The connection pool is an object in the Physical layer that describes access to the data source.
The connection pool contains information about the connection between the Oracle BI Server and that data source.
The Physical layer in the BI Server contains at least one connection pool for each database. When you create the Physical layer by importing a schema for a data source, the connection pool is created automatically. You can configure multiple connection pools for a database. Connection pools allow multiple concurrent data source requests (queries) to share a single database connection, reducing the overhead of connecting to a database.
It is recommended that you create a dedicated connection pool for initialization blocks. See About Connection Pools for Initialization Blocks for more information.
For each connection pool, you must specify the maximum number of concurrent connections allowed. After this limit is reached, the connection request waits until a connection becomes available.
Increasing the allowed number of concurrent connections can potentially increase the load on the underlying database accessed by the connection pool. Test and consult with your DBA to make sure the data source can handle the number of connections specified in the connection pool. Also, if the data sources have a charge back system based on the number of connections, you might want to limit the number of concurrent connections to keep the charge-back costs down.
In addition to the potential load and costs associated with the database resources, the Oracle BI Server allocates shared memory for each connection upon server startup. This raises the number of connections and increases Oracle BI Server memory usage.
You should create a dedicated connection pool for initialization blocks.
This connection pool should not be used for queries.
Additionally, it is recommended that you isolate the connections pools for different types of initialization blocks. This also makes sure that authentication and login-specific initialization blocks do not slow down the login process. The following types should have separate connection pools:
All authentication and login-specific initialization blocks such as language, externalized strings, and group assignments.
All initialization blocks that set session variables.
All initialization blocks that set repository variables. These initialization blocks should always be run using credentials with administrator privileges.
Be aware of the number of these initialization blocks, their scheduled refresh rate, and when they are scheduled to run. Typically, it would take an extreme case for this scenario to affect resources. For example, refresh rates set in minutes, greater than 15 initialization blocks that refresh concurrently, and a situation in which either of these scenarios could occur during prime user access time frames.
Initialization blocks should be designed so that the maximum number of Oracle BI Server variables may be assigned by each block. For example, if you have five variables, it is more efficient and less resource intensive to construct a single initialization block containing all five variables. When using one initialization block, the values are resolved with one call to the back end tables using the initialization string. Constructing five initialization blocks, one for each variable, would result in five calls to the back end tables for assignment.
If an initialization block fails for a particular connection pool during Oracle BI Server start-up, no more initialization blocks using that connection pool are processed. Instead, the connection pool is blacklisted and subsequent initialization blocks for that connection pool are skipped. This behavior ensures that the Oracle BI Server starts in a timely manner, even when a connection pool has a large number of associated initialization blocks or variables.
If this occurs, a message similar to the following appears in the server log:
[OracleBIServerComponent] [ERROR:1]  Blacklisted connection pool name_of_connection_pool
If you see this error, check the initialization blocks for the given connection pool to ensure they are correct.