Tips Preventing Suboptimal Execution Plans

1
Copyright 2011, Oracle and/or its affiliates. All rights reserved.
Oracle Optimizer: Top Tips For Preventing Suboptimal Execution plans

2 Copyright 2011, Oracle and/or its affiliates. All rights reserved.
What to expect
Explanation of common mistakes that cause sub-optimal execution plans and how to avoid them The assumption is that all statistics are accurate and the cause of the sub-optimal plan lies elsewhere
What is assumed of you

A very basic understand of an execution plan A common definition of an access predicate
Where clause component used for data retrieval
The start and stop keys for an index If rowids are passed to a table scan
A common definition of a filter predicate

Where clause component that is not used for data retrieval but to eliminate uninteresting row once the data is found
Agenda
Using the right tools Functions, friends or foes? Data type dilemmas ANSI versus Oracle syntax Influencing the execution plan without adding hints
Using the right tools

Expected index range scan but got full table scan? Query How many customers do we have in a specific zipcode
SELECT count(cust_first_name) FROM customers2 WHERE zipcode = :n;
Customers2 table has b-tree index on zipcode There is a histogram on the zipcode column due to data skew

Expected index range scan but got full table scan?
Set bind variable :n to 94065

Using literal value gets the expected plan
Why is the bind variable version getting the wrong plan?


Why is simple equality predicate being applied as filter and no as access?
Why is there a TO_NUMBER function on the bind variable n after it was defined as a number?

Bad plan is actually caused by using autotrace Autotrace is not aware of bind at all
Hence TO_NUMBER on n No bind peeking takes place
Cardinality estimate doesnt use histogram it is calculated using

ROW_NUM NDV 20,000 2
NOTE cursor generated by autotrace will be shared by other sessions


Solution use DBMS_XPLAN.DISPLAY_CURSOR
Execute the statement with the bind then run DBMS_XPLAN command
11

Solution use DBMS_XPLAN.DISPLAY_CURSOR
Can add additional parameters to display the bind variable values along with the plan
12
Agenda
Data type dilemmas ANSI versus Oracle syntax Influencing the execution plan without adding hints
13
Functions friends or foe?

Expected index range scan but got fast full index scan Query How many packages of bounce did we sell?
SELECT /*+ gather_plan_statistics */ count(*) FROM sales2 WHERE to_char(prod_id)=139;
Sales 2 has a b-tree index on the prod_id column
14

Expected index range scan but got fast full index scan
Use hint to capture execution statistics Use format parameter of DBMS_XPLAN to show estimated and execution statistics in plan Additional column added to the plan to show Actual Rows produce by each operation
15

Cardinality estimate is in the right ballpark so not a problem with statistics
Why is an equality predicate being evaluated as a filter and not an access predicate? Could it have something to do with the TO_CHAR function

Expected index range scan but got fast full index scan What data type is the prod_id column ?
But literal value is a character string 139 Better to apply inverse function on other side of predicate

Solution - Use inverse function on other side of predicate Query rewrite
SELECT /*+ gather_plan_statistics */ count(*) FROM sales2 WHERE prod_id=to_number(139);
18

Expected query to access only 4 partitions but its accesses all
Query calculate total amount sold for one year

SELECT sum(amount_sold) FROM sh.sales s, sh.times t WHERE s.time_id = t.time_id AND TO_CHAR(s.time_id,YYYYMMDD) between 19990101 and 20000101;
Sales table is partitioned on the time_id column Sales table has quarterly partitions for historical data

Why has an additional INTERNAL_FUNCTION been added to our predicate?
20

INTERNAL_FUNCTION typically means a data type conversion has occurred Predicate is TO_CHAR(s.time_id,YYYYMMDD) Optimizer has no idea how function will effect the values in the time_id column Optimizer cant determine which partitions will be accessed now
21

Solution Use inverse function on other side of predicate
Only 4 partitions accessed
22

Using inverse function on other side of predicate
Keep the following in mind when deciding where to place the function
Try to place functions on top of constants (literals, binds) rather than on columns Avoid using a function on index columns or partition keys as it prevents index use or partition pruning For function-based index to be considered, use that exact function as specified in index If multiple predicates involve the same columns, write predicates such that they share common expressions For example, WHERE f(a) = b Should be AND a = c rewritten as WHERE a = inv_f(b) This will allow transitive predicate c=inv_f(b) to be AND a=c added by the optimizer
23
Agenda
24
Data type dilemmas

Expected index range scan but got fast full index scan Query Simple IAS part of an ETL process
INSERT /*+ APPEND gather_plan_statistics */ INTO t1 FROM (row_id, modification_num, operation, last_upd) t2 SELECT row_id, 1 , 'I', last_upd WHERE t2.last_upd > systimestamp;
T2 has a b-tree index on the last_upd column

Data type dilemmas

Only one non-equality access predicate
So why is our access predicate applied as a filter? What does INTERNAL_FUNCTION mean?
26
Data type dilemmas

Expected index range scan but got fast full index scan INTERNAL_FUNCTION typically means a data type conversion has occurred Predicate is t2.last_upd > systimestamp What data type is the last_upd column
27
Data type dilemmas

Why is the cardinality estimate wrong? Presence of the INTERNAL_FUNCTION cause the Optimizer to guess the cardinality estimate
Optimizer has no way of knowing how function effects data in LAST_UPD column Without a function based index or extended statistics the Optimizer has to guess Guess is 5% of the rows in the table
5% of 823296 is 41,164.8 or 41,165

Data type dilemmas

Solution correct data type mismatch Changing the predicate from SYSTIMESTAMP to SYSDATE
INSERT /*+ APPEND gather_plan_statistics */ INTO t1 (row_id, modification_num, operation, last_upd) SELECT row_id, 1 , 'I', last_upd FROM t2 WHERE t2.last_upd > sysdate;
;
29
Data type dilemmas

Expected to get partition pruning via a join but didnt
Query calculate total amount sold that was return same day

SELECT sum(amount_sold) FROM sh.sales s, sh.sales_returns sr WHERE s.time_id = sr.time_id AND sr.time_id=31-DEC-01;
Sales table is range partitioned on time_id Sales table has 7 years of data in quarterly partitions
Data type dilemmas

Expected to get partition pruning via a join but didnt
Getting transitive predicate but Internal_function on partitioned column prevents pruning Function needed because the join columns have different data types
Data type dilemmas

Solution ensure join columns have the same data type
KEY means dynamic pruning at execution time AP means And Pruning, caused by bloom filter Now get transitive predicate without data type conversion hence pruning
Agenda
33
ANSI versus Oracle Syntax

Expected Nest Loop Outer Join but got Sort Merge Join While migrating an application to Oracle the following ANSI SQL was translated into Oracle syntax
SELECT d.dname, e.empno FROM dept d LEFT OUTER JOIN emp e ON (d.deptno = e.deptno AND (d.loc = 'NEW YORK' or e.ename like 'S%') ); SELECT d.dname, e.empno FROM dept d, emp e WHERE d.deptno = e.deptno (+) AND (d.loc = 'NEW YORK' or e.ename like 'S%');
34

Expected Nest Loop Outer Join but got Sort Merge Join ANSI Syntax
SELECT d.dname, e.empno FROM dept d LEFT OUTER JOIN emp e ON (d.deptno = e.deptno AND (d.loc = 'NEW YORK' or e.ename like 'S%'));
35

Expected Nest Loop Outer Join but got Sort Merge Join Oracle Syntax
SELECT d.dname, e.empno FROM dept d, emp e WHERE d.deptno = e.deptno (+) AND (d.loc = 'NEW YORK' or e.ename like 'S%');
Why does the predicate section of the Oracle Syntax plan look so different from the Ansi SQL version?
36

Oracle syntax didnt return the same results as ANSI
ANSI syntax returns7 rows Oracle syntax returns only 5 rows
37

Expected Nest Loop Outer Join but got Sort Merge Join Oracle Syntax is not equivalent to Ansi SQL In Ansi SQL filter predicates are evaluated before join Predicates are evaluated by creating a lateral view on View definition would be EMP
SELECT e.empno FROM emp e WHERE (d.loc = 'NEW YORK' or e.ename like 'S%') AND d.deptno = e.deptno;
38

Expected Nest Loop Outer Join but got Sort Merge Join Correct Oracle translation of the Ansi SQL
SELECT d.dname, v.empno FROM dept d, lateral(SELECT e.empno FROM emp e WHERE (d.loc = 'NEW YORK' or e.ename like 'S%') AND d.deptno = e.deptno)(+) v;
In Oracle Database11g lateral keyword only active if events 22829 is set


Expected Nest Loop Outer Join but got Sort Merge Join Complete Oracle translation of the Ansi SQL
Alter session set events '22829 trace name context forever, level 1'; SELECT d.dname, v.empno FROM dept d, lateral(SELECT e.empno FROM emp e WHERE (d.loc = 'NEW YORK' or e.ename like 'S%') AND d.deptno = e.deptno)(+) v;
40
Agenda
41
How to influence the execution plan without adding hints

Even if statistics are accurate and up to date you may need to influence the optimizer with a hint Most third party application dont provide a mechanism to add hints to generate SQL Solution is to use SQL Plan Management (SPM) SPM available in EE, no additional options required
42

1. In a SQL*Plus session run the non-hinted SQL statement to begin the SQL plan baseline capture
43

2. Find the SQL_ID for the non-hinted statement in V$SQL
44

3. Create a SQL plan baseline for the non-hinted SQL statement
45

4. The plan that was captured is not our desired plan so it should be disabled using DBMS_SPM.ALTER_SQL_PLAN_BASELINE
46

5. Modify the SQL statement to use the hint(s) & execute it
47

6. Find new SQL_ID and PLAN_HASH_VALUE for the hinted SQL stmt in V$SQL
48

7. Associate hinted plan with original SQL statements SQL HANDLE
Sql_id & plan_hash_value belong to hinted statement
sql_handle is for the non-hinted statement
49

8. In DBA_SQL_PLAN_BASELINE we nowhave two plans for this stmt, one enabled and one disables
Hinted plan only accepted plan for non-hinted SQL stmt

Confirm that non-hinted SQL statement will now use hinted

Non-hinted SQL text but it is using the plan hash value for the hinted stmt Note section also confirms SQL plan baseline used for stmt
51
Summary
Choose your plan displaying tool wisely
They dont all show the correct plan and can effect other sessions
Place functions carefully

Avoid adding them to table columns as it can limit access paths
Data types
If INTERNAL_FUNCTION appears in the predicate section check data types used in where clause predicates and across joins
Translate ANSI SQL with caution

Some times it better to leave it as ANSI SQL
Q&A
53
Oracle Optimizer Schedule for Oracle Open World!

Date Title Loca,on Speaker
Monday Oct 3rd 12:30 PM"
Oracle Optimizer: Prevent Suboptimal " Execution Plans" " Hands-on-Lab! Oracle Optimizer: Best Practices for Managing Optimizer Statistics"
Marriott Marquis - Salon 12/13"
Maria Colgan"
Senior Principal Member of Technical Staff" Oracle"
Wednesday Oct 5th 10:15 AM "
Moscone South - 103"
Maria Colgan"
Thursday Oct 6th 12:00 PM"
Oracle Database Optimizer: Tips for " Preventing Suboptimal Execution Plans"
Moscone South - 104"
Maria Colgan"
Mohamed Zait"
Architect" Oracle"
54
55

Tips Preventing Suboptimal Execution Plans

Загружено:

Сведения о документе

Исходное описание:

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Tips Preventing Suboptimal Execution Plans

Загружено:

Авторское право:

Доступные форматы

1

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Oracle Optimizer: Top Tips For Preventing Suboptimal Execution plans

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

What is assumed of you

A common definition of a filter predicate

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Using the right tools

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Using the right tools

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Using the right tools

Why is the bind variable version getting the wrong plan?

Using the right tools

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Using the right tools

Cardinality estimate doesnt use histogram it is calculated using

NOTE cursor generated by autotrace will be shared by other sessions

Using the right tools

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Using the right tools

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Functions friends or foe?

Sales 2 has a b-tree index on the prod_id column

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Functions friends or foe?

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Functions friends or foe?

Cardinality estimate is in the right ballpark so not a problem with statistics

Functions friends or foe?

Functions friends or foe?

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Functions friends or foe?

Query calculate total amount sold for one year

Functions friends or foe?

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Functions friends or foe?

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Functions friends or foe?

Only 4 partitions accessed

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Functions friends or foe?

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Data type dilemmas

T2 has a b-tree index on the last_upd column

Data type dilemmas

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Data type dilemmas

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Data type dilemmas

5% of 823296 is 41,164.8 or 41,165

Data type dilemmas

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

Data type dilemmas

Data type dilemmas

Data type dilemmas

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

ANSI versus Oracle Syntax

Copyright 2011, Oracle and/or its affiliates. All rights reserved.

ANSI versus Oracle Syntax