Oracle PL/SQL for DBAs

Distributing Records

One really cool thing about parallel table functions is that you can tell Oracle how to distribute records among the parallel function instances. Two different aspects of distribution can be specified. The first is partitioning — how Oracle should decide which records to send to which function instance. The second is streaming— how each function instance should order the records partitioned to it. The options for each are shown below.

Partitioning

Records can be partitioned based on ranges of one or more columns, hash values based on one or more columns, or simply as Oracle sees fit. The PARTITION BY clause specifies the specific partitioning option for the table function, as shown in the following examples.

Streaming

Records can be ordered or clustered by specific columns. The ORDER or CLUSTER keyword specifies the specific streaming option for the table function, as shown in the following examples.

I’ll explain each of these components and how they interact while working through development of my account transaction summary function.

Random partitioning (PARTITION BY ANY)

First, I’ll set up the function and tell Oracle to run it in parallel, but simply partition the records as it sees fit. The PARALLEL_ENABLE clause in the function header tells Oracle that this function has been coded with parallelism in mind and that it would really like to take advantage of this feature. The PARTITION BY ANY parameter supplied in this example states that the records returned ...