1. Author: DuyTran Minh Doan 10,April 2012 4/3/20141 Advanced PL/SQL

2. Objective 4/3/20142 You should have basic knowledge about PL/SQL and database management before reading this. This slide will guide you through some advanced concepts of PL/SQL like hints, execution plan, bulk processing, You should use this slide as reference but remember to do testing everything before apply to your code.

3. Agenda 4/3/20143 A. Flow Control B. Bulk Processing C. Oracle Hints D. Resources E. Q/A

4. A.1. Code optimization LOOP & IF 4/3/2014A. Flow Control4 Given LOOP and IF statement for count1 in 1..2000 loop for count2 in 1..2000 loop mod1 := mod(count1, 10); mod2 := mod(count2, 10); sqrt1 := sqrt(count1); sqrt2 := sqrt(count2); if (mod1 = 0) then if (mod2 = 0) then sum1 := sum1 + sqrt1 + sqrt2; end if; end if; end loop; end loop; Executed in 8.297 seconds Bad

5. A.1. Code optimization LOOP & IF(cont) 4/3/2014A. Flow Control5 Enhanced version for count1 in 1..2000 loop mod1 := mod(count1, 10); sqrt1 := sqrt(count1); for count2 in 1..2000 loop mod2 := mod(count2, 10); sqrt2 := sqrt(count2); if (mod1 = 0) then if (mod2 = 0) then sum1 := sum1 + sqrt1 + sqrt2; end if; end if; end loop; end loop; Executed in 4.359 seconds Not bad, but

6. A.1. Code optimization LOOP & IF(cont) 4/3/2014A. Flow Control6 More fine grained for count1 in 1..2000 loop mod1 := mod(count1, 10); if (mod1 = 0) then sqrt1 := sqrt(count1); for count2 in 1..2000 loop mod2 := mod(count2, 10); sqrt2 := sqrt(count2); if (mod2 = 0) then sum1 := sum1 + sqrt1 + sqrt2; end if; end loop; end if; end loop; Executed in 0.453 seconds Good 20 times faster than the original one !!!

7. A.1. Code optimization LOOP & IF(cont) 4/3/2014A. Flow Control7 Conclusion Minimize the number of iterations: use EXIT to cease looping where necessary. Remove any statements within a loop that could be processed outside the loop, especially when you have nested loop. Specify the most probable condition first in a compound IF statement: try to reduce the number of evaluations.

8. A.2. Code optimization - Recursion 4/3/2014A. Flow Control8 Fibonacci calculating function rec_proc(n_limit integer) return integer is begin if (n_limit > 1) then return n_limit + rec_proc(n_limit-1); else return n_limit; end if; end; function non_rec_proc(n_limit integer) return integer is n_sum integer := 0; begin for i in 0.. n_limit loop n_sum := n_sum + n_limit; end loop; return n_sum; end; Executed in 2 seconds Executed in 0.234 seconds

9. A.2. Code optimization - Recursion(cont) 4/3/2014A. Flow Control9 Conclusion Elegant from coding perspective, but consume memory and are usually slower than iterative alternative. Almost all recursive algorithms have a non-recursive equivalent.

10. B.1. Bulk processing in PL/SQL 4/3/2014B. Bulk Processing10 Row by row processing in PL/SQL SQL statement executor SQL Engine Procedural statement executor PL/SQL Runtime Engine PL/SQL Block FOR rec IN emp_cur LOOP UPDATE emp SET salary = WHERE emp_id = rec.emp_id; END LOOP; Oracle Server

11. B.1. Bulk processing in PL/SQL(cont) 4/3/2014B. Bulk Processing11 Bulk processing with FORALL SQL statement executor SQL Engine Procedural statement executor PL/SQL Runtime Engine PL/SQL Block FOR rec IN emp_cur LOOP UPDATE emp SET salary = WHERE emp_id = rec.emp_id; END LOOP; Oracle Server

12. B.1. Bulk processing in PL/SQL(cont) 4/3/2014B. Bulk Processing12 Bulk processing with FORALL (cont.) Fewer context switches. Same SQL behavior. Use with inserts, updates, deletes and merges. Move data from collections to tables.

13. B.1. Bulk processing in PL/SQL(cont) 4/3/2014B. Bulk Processing13 Warning BEFORE &AFTER statement-level triggers fired only once per FORALL INSERT statement. Make code run faster but user session will consume more PGA memory. whats PGA memory ???

14. B.2. Oracle DB memory essential 4/3/2014B. Bulk Processing14 Oracle memory structure Program GlobalArea

15. B.2. Oracle DB memory essential(cont) 4/3/2014B. Bulk Processing15 Oracle memory structure more details

16. B.2. Oracle DB memory essential(cont) 4/3/2014B. Bulk Processing16 Memory solution Bulk collection with LIMIT. CAUTION !!! Do not checking %NOTFOUND right after fetch. Do one of the following instead: Check %NOTFOUND at the end of the loop. Exit when collection.COUNT = 0 right after fetch. Exit when collection.COUNT < limit at the end of the loop.

17. B.3. Exception handling 4/3/2014B. Bulk Processing17 Got exception in FORALL statement ? Add SAVE EXCEPTIONS to suppress errors at statement level. Exception will be handled in Exception block.

18. B.3. Exception handling(cont) 4/3/2014B. Bulk Processing18 Typical errors with bulk processing Rollback segment too small Cause: too many uncommitted changes, the rollback segment can not handle all. Solution: may still need to use incremental commit. Snapshot is too old ... Cause: a cursor is held open too long, Oracle can no longer maintain the snapshot information. Solution: open-close cursor.

19. B.3. Bulk processing conclusion 4/3/2014B. Bulk Processing19 Conclusion Should always try to execute multi-row SQL operations in PL/SQL. Watch out for impact on PGA memory.

20. C.1. SQL processing 4/3/2014C. Oracle Hints20 SQL processing diagram Read more: Understand explain plan

21. C.1. SQL processing(cont) 4/3/2014C. Oracle Hints21 SQL processing Oracle execute SQL in different ways: Full table scans. Index scans. Nested loops. Hash joins. Execution plan. We have more information on context than Oracle.

22. C.1. SQL processing(cont) 4/3/2014C. Oracle Hints22 Access paths for the Cost Based Optimizer Full table scans SELECT /*+ FULL(e) */ FROM Employee e. Rowid scans. Index scans. Cluster scans. Hash scans. Sample table scans.

23. C.1. SQL processing(cont) 4/3/2014C. Oracle Hints23 Join methods Nested loop joins. Hash joins. Sort merge joins. Cartesian joins. Outer joins.

24. C.1. SQL processing(cont) 4/3/2014C. Oracle Hints24 Why hints are needed ? Incorrect/incomplete statistics. Keywords that slow performance down. Highly volatile data changes. Configurations outside Optimizer knowledge scope.

25. C.2. Oracle hints introduction 4/3/2014C. Oracle Hints25 What is Oracle hint? A code snippet that is embedded into a SQL statement. Suggest to Oracle how the statement should be executed. Note: Hints should only be used as a last-resort if statistics were gathered and the query is still following a sub-optimal execution plan. Hint syntax --+RULE /*+RULE */ Read more: http://psoug.org/reference/hints.html

26. C.2. Oracle hints introduction(cont) 4/3/2014C. Oracle Hints26 Some of the most useful hints Global hints: rule, first_rows, first_rows_n all_rows, driving_site. Table join hints: use_nl, use_hash, ordered. Table access hints: parallel, full, cardinality. Index hints: index, no_index, index_combine.

27. C.3. Example hints 4/3/2014C. Oracle Hints27 Example Example suggesting that a FULLTABLE SCAN method be used: SELECT /*+ FULL(x) */ FROM tab1 xWHERE col1 = 10; Suggest that Oracle uses a specific index: SELECT /*+ INDEX(x emp_idx1) */ ... FROM scott.emp x... Suggest that Oracle DOES NOT USE a specific index: SELECT /*+ NO_INDEX(x emp_idx1) */ ... FROM scott.emp x...

28. C.4. Oracle Parallel Query 4/3/2014C. Oracle Hints28 Oracle Parallel Query PQO allows one to break-up a given SQL statement Its parts can run simultaneously on different processors in a multi- processor machine Typical operations that can run parallel are: full table scans, sorts, sub- queries, data loading , Advantage: improve performance of certain types of operations dramatically on a multi-CPU system (usually be paired with full table scans)

29. C.4. Oracle Parallel Query(cont) 4/3/2014C. Oracle Hints29 How to invoke Parallel query? Alter the table (or index) to indicate that Oracle should try to parallelize operations performed against it ALTERTABLE table_name PARALLEL (DEGREE 8); Put hints in SQL statements to indicate that Oracle should try to execute them in parallel SELECT /*+PARALLEL(table_alias, degree, nodes)*/ * FROM table_name ... Note:You must set all the INIT.ORA parameters necessary for Parallel Query to work

30. C.4. Oracle Parallel Query(cont) 4/3/2014C. Oracle Hints30 Using OPQ, Oracle partitions the table into logical chunks. Oracle fires off parallel query slaves (sometimes called factotum processes), and each slave simultaneously reads a piece of the large table. Upon completion of all slave processes, Oracle passes the results back to a parallel query coordinator, which will reassemble the data, perform a sort if required, and return the results back to the end user.

31. Resources 4/3/201431 Doing SQL from PL/SQL: Best andWorst Practices Oracle. Tuning Oracle Stored Procedures Guy Harrison, Quest Software. Turbo-charge PL/SQL Performance with Bulk Processing Features Steven Feuerstein,ToadWorld.com. The Top 10 Dumbest PL/SQLThings I Have Seen or Done Steven Feuerstein, Quest Software. Oracle Documentation about MemoryArchitecture Understanding explain plan When NOT EXIST should NOT EXIST D. Resources

32. 4/3/201432 Q/A E. Q/A