Sei sulla pagina 1di 3

Oracle PL/SQL performance tuning using BULK COLLECT and FORALL

Published October 10, 2012 | By admin Here Ill show you, in Oracle database, how to do performance tuning of your pl/sql code using BULK COLLECT and FORALL. Using BULK COLLECT and FORALL instead of normal FOR LOOP I have achieved significant performance benefit. In some cases I was able to tune the performance from 14 minutes to just 10 seconds. You may already know that in Oracle database the PL/SQL code is executed by PL/SQL engine and SQL is executed by SQL engine. Also you know that SQL is embedded with PL/SQL. When PL/SQL engine encounters SQL code, it passes control to SQL engine for execution. This is called context switching between SQL and PL/SQL. For example see the code below: FOR i in 1..1000 LOOP insert into emp values (); END LOOP; In the above code when PL/SQL engine is executing the for loop, it needs to execute the INSERT statement 1000 times. This also means there were 1000 context switching. This generally degrades performance and causes longer execution time. Instead of context switching 1000 times, by using BULK COLLECT and FORALL method, the same operation can be achieved with just a single context switch. So that is where you can achieve significant performance benefit. The test case here copies 1,000,000 records from a simple source table to a destination table. The source and destination tables have same structure. The only difference is that the source table has a primary key. Here is the code that I used to create the source table called t_source and the destination table called t_dest: create table t_source( empno number, ename varchar2(10), joindate date); alter table t_source add constraint pk_src primary key (empno); create table t_dest as select * from t_source; Then I created 1,000,000 records in t_source using this simple code declare begin for i in 1..1000000 loop insert into t_source values (i, emp||i, sysdate); end loop; commit; end; / After inserting source data, I analyzed the table so that the optimizer has the latest statistics about the source table. That is to help Oracle as much as possible to find out the best way to get the data from t_source table. This may be not necessary for a simple table such as this. But depending on the number of rows and number of columns, it may help enormously.

Anyway here is the code to gather table stats. EXEC DBMS_STATS.gather_table_stats(JOE, T_SOURCE); Then lets set the SQL environment so that we can see SQL execution time. Run this from SQL prompt: SET TIMING ON; Now I am going to populate t_dest using the usual FOR LOOP which is a very common method used by programmers. The code basically gets data into a cursor and then read each record from the cursor and inserts into t_dest table. declare cursor c1 is select * from t_source; begin for src_rec in c1 loop insert into t_dest values (src_rec.empno, src_rec.ename, src_rec.joindate); end loop; commit; end; Once the execution is complete you will get the output something like this. PL/SQL procedure successfully completed. Elapsed: 00:02:40.12 SQL> Depending on the speed of your machine and database parameters, your code may take more time than you can see here or it may take less time. We are going to need this timing to compare against code that uses BULK COLLECT and FORALL. Now run this code from your SQL*PLUS session truncate table t_dest; declare cursor c1 is select * from t_source; TYPE src_tab IS TABLE OF t_source%ROWTYPE INDEX BY BINARY_INTEGER; rec_tab src_tab; begin open c1; fetch c1 BULK COLLECT INTO rec_tab limit 10000; WHILE rec_tab.COUNT > 0 LOOP FORALL i IN 1..rec_tab.COUNT INSERT INTO t_dest (empno, ename, joindate) VALUES (rec_tab(i).empno,rec_tab(i).ename,rec_tab(i).joindate); fetch c1 BULK COLLECT INTO rec_tab limit 10000; END LOOP; CLOSE c1; end; The output should shoething like this: PL/SQL procedure successfully completed. Elapsed: 00:00:16.13 SQL>

As you can see the time to insert 1,000,000 records from source to destination has decreased from 00:02:40.12 to just 00:00:16.13. This is a huge performance gain. When dealing with millions of records the performance benefit may be tremendous.

Potrebbero piacerti anche