Optimizer Hints

  • View

  • Download

Embed Size (px)

Text of Optimizer Hints

1. Indian PUG 2015 11th April Kumar Rajeev Rastogi Prasanna Optimizer Hint 2. KUMAR RAJEEV RASTOGI Senior Technical Leader at Huawei Technology for almost 7 years Have worked to develop various features on PostgreSQL (for internal projects) as well as on In-House DB. Active PostgreSQL community members, have contributed many patches. Holds around 10 patents in my name in various DB technologies. This is my second talk in Indian PUG, first one was last year. Prior to this, worked at Aricent Technology for 3 years. Blog - rajeevrastogi.blogspot.in LinkedIn - http://in.linkedin.com/in/kumarrajeevrastogi Who am I? 3. Introduction Why Optimizer Hint Query Hint Statistics Hint Data Hint Drawback Reference Agenda 4. Introduction The overall query execution is described below: 5. Introduction Contd 1. Parser: Parses the query submitted by the client to do the syntactical validation. Output of this step is parsed tree. 2. Analyzer: It validates the parsed tree semantically. 3. Utility Commands: All of the DDL and other utility commands then gets executed by this sub-module. 4. Optimizer: This is almost like brain of complete SQL execution engine. It find the best possible plan for its execution. 5. Executor: The output from optimizer is Plan Tree, which then gets converted to Execution Tree, where each node of the tree denotes the kind of operation it has to do like IndexScan, SeqScan, Agg, Join etc. Then each node gets executed to yield the final result. 6. Optimizer Zoomed In It takes the validated query tree from analyzer, looks for the all the possible plan and then selects the best plan for execution. There are multiple possible plan because there are several different methods of doing the same operation, some of them are: a. Two scan algorithms (index scan, sequential scan) b. Three core join algorithms (nested loops, hash join, merge join) c. Join Order 7. Factors Used to Decide Best Plan The planner chooses between plans based on their estimated cost Some of the parameters used to decide cost are sequential I/O cost, random I/O cost and CPU cost. Estimated I/O required is calculated based on the number of pages to be scanned. CPU cost is based on estimated number of records and qualification. Random IO is (much) more expensive than sequential IO on modern hardware 8. Optimizer Accuracy It is believed that Optimizer makes best effort to produce the best plan for execution using the parameters mentioned and actually in majority of cases it is being accurate to give the correct and optimal plan for execution. Why Optimizer Hint? 9. Why Optimizer Hint What is Optimizer Hint: As the name suggest, it is a Hint to optimizer to change the resultant plan as per the hint. As it is just hint, so resultant plan may or may not be as per the hint given. Why Optimizer Hint: As a DBA you might know information about your data but the optimizer may not know, in this case DBA should be able to instruct the optimizer to choose a certain query execution plan based on some criteria. 10. Why Optimizer Hint: Scenario-1 Consider for a query, the possible plans for joining two tables were merge join and nested loop join but based on the total cost estimation merge join were selected as winner plan. Now consider scenario where we might require getting the first row result at the earliest, then in that case the plan with smallest start-up cost will be more useful although total cost for this plan is more. 11. Why Optimizer Hint: Scenario-2 Consider the following query SELECT * FROM TBL1, TBL3, TBL2 WHERE . Now lets us say the selectivity for each tables are as below: TBL1: 0.5 TBL2: 0.5 TBL3: 0.3 Since currently planner does not maintain the selectivity for join tables or columns, then in order to check selectivity for join tables, it simply multiply the selectivity of individual tables. So now in our example case, selectivity will be as below: TBL1, TBL2: 0.5*0.5 = 0.25 TBL1, TBL3: 0.5*0.3 = 0.15 But DBA knows that the selectivity of join of TBL1 AND TBL2 is not correct as it will result in maximum of one record where as selectivity of join of TBL1 and TBL3 is correct. 12. Hint Types 1. Query Hints 2. Statistics Hints 3. Data Hints 13. Query Hint This kind of hints force optimizer to choose desired plan on specific relation or join of relations. E.g. DBA can specify to select Sequence scan path for scanning a relation. Index scan path for scanning a relation. Merge join path for joining two relations. Order to evaluate join of relations. 14. Query Hint Contd Following are ways to provide query hint for a particular query Hints along with query in a commented format with syntax as: SELECT/UPDATE/DELETE/INSERT /*hint*/ ; Some DB provide it as separate command instead of embedding into query. Sometime it is also given as property along with table name in query. 15. Query Hint By Other DBs Many databases are using query hint as optimization hint, some of them are: 1. Oracle 2. Sybase 3. MySQL 4. EnterpriseDB PostgreSQL Plus 5. Recently we have also implemented query hint in Huawei for internal Database. 16. Query Hint By Oracle: Example Some of the hint used by Oracle in query are: 1. /*+ FULL(tbl)*/ Hint to optimizer to choose sequence scan for table tbl; 2. /*+ORDERED */ Hint to optimizer to choose the same join ordering as the table names are given in FROM clause. 3. /*+ USE_NL(tbl1 tbl2) */ Hint to optimizer to choose the nested loop join between tables tbl1 and tbl2; 17. Query Hint By Sybase: Example Some of the hint used by Sybase in query are: 1. set forceplan [on|off] Hint to optimizer to choose the same join ordering as the table names are given in FROM clause, if it is on. 2. We can specify the index to use for a query using the (index index_name) clause in select, update, and delete statements 18. Query Hint By MySQL: Example Some of the hint used by MySQL in query are: 1. /*! STRAIGHT_JOIN */ This hint tells optimizer to join the tables in the order that they are specified in the FROM clause. (MySQL hint is similar to oracle except it uses ! instead of +. 2. USE {INDEX|KEY} (index_list)] Provide hints to give the optimizer information about how to choose indexes during query processing 19. Query Hint By EDB: Example Hint used in EDB are similar to used in Oracle: 1. /*+ FULL(tbl)*/ Hint to optimizer to choose sequence scan for table tbl; 2. /*+ORDERED */ Hint to optimizer to choose the same join ordering as the table names are given in FROM clause. 3. /*+ USE_NL(tbl1 tbl2) */ Hint to optimizer to choose the nested loop join between tables tbl1 and tbl2; 20. Does PostgreSQL has query Hint? NO (or indirectly YES) PostgreSQL does not support hint directly but there are many GUC configuration parameter, using which we can force to disable particular plan (Notice here that we can configure to disable a particular plan unlike other DB, where Hint provided to choose a particular plan). e.g. enable_indexscan to off: Forces optimizer to skip index scan enable_mergejoin to off: Forces optimizer to skip merge join Also this setting is session-wise not query-wise. Similar to hint, this setting can also be ignored if not possible to process e.g. even if we make enable_seqscan to off still it can select sequence scan to scan a table if there is no index on that particular table. These setting are very useful for a developer or DBA working on tuning the planner or their application respectively. 21. Statistics Hint Statistics Hint is used to provide any kind of possible statistics related to query, which can be used by optimizer to yield the even better plan compare to what it would have done otherwise. Since most of the databases stores statistics for a particular column or relation but doesnt store statistics related to join of column or relation. Rather these databases just multiply the statistics of individual column/relation to get the statistics of join, which may not be always correct. So for such case statistics based hints can be used to provides direct statistics of join of relation or columns. 22. Statistics Hint: Example Lets say there is query as SELECT * FROM EMPLOYEE WHERE GRADE>5 AND SALARY > 10000; If we calculate independent stats for a and b. suppose sel(GRADE) = .2 and sel(SALARY) = .2; then sel (GRADE and SALARY) = sel(GRADE) * sel (SALARY) = .04. In all practical cases if we see, these two components will be highly be dependent i.e. first column satisfy, 2nd column will also satisfy. Then in that case sel (GRADE and SALARY) should be .2 not .04. But current optimizer will be incorrect in this case and may give wrong plan. 23. Statistics Hint: Example Contd The query with statistics hint will look like: SELECT /*+ SEL (GRADE and SALARY) AS 0.2* FROM EMPLOYEE WHERE GRADE>5 AND SALARY > 10000; 24. Data Hint This kind of hints provides the information about the relationship/ dependency among relations or column to influence the plan instead of directly hinting to provide desired plan or direct selectivity value. Optimizer can consider dependency information to derive the actual selectivity. 25. Data Hint: Example Lets say there is a query as SELECT * FROM TBL WHERE ID1 = 5 AND ID2=NULL; SELECT * FROM TBL WHERE ID1 = 5 AND ID2!=NULL; Now here if we specify that the dependency as If TBL.ID1 = 5 then TBL.ID2 is NULL then the optimizer will always consider this dependency pattern and accordingly combined statistics for these two columns can be choosen. 26. Drawback 1. Required periodic maintenance to verify that hint supplied is still giving positive result 2. Interference with upgrades: today's helpful hints become anti- performance after an upgrade. 3. Encouraging bad DBA habits slap a hint on instead of figuring out the real issue. 27. Reference http://docs.oracle.com/cd/B19306_01/server.102/b14211/hintsref.h tm http://manuals.sybase.com/onlinebooks/group- as/asg1250e/ptallbk/@Generic__BookTe