Bind Peeking - The Endless Tuning Nightmare
0 likes348 views
The document discusses the challenges and solutions related to Oracle's bind peeking and query optimization, particularly regarding varying data distributions and performance impacts. It highlights best practices for handling SQL statements with bind variables, including the use of histograms and adaptive cursors, and suggests strategies for improving execution plans. Additionally, it emphasizes the importance of regularly gathering statistics and the implications of data patterns on SQL performance.
Bind Peeking - The Endless Tuning Nightmare
- 1. SAGE Computing Services Customised Oracle Training Workshops and ConsultingCustomised Oracle Training Workshops and Consulting Bind Peeking – The Endless Tuning Nightmare Penny Cookson Managing Director and Principal Consultant Working with Oracle products since 1987 Oracle Magazine Educator of the Year 2004 Oracle ACE www.sagecomputing.com.au penny@sagecomputing.com.au
- 2. Characteristics I haven’t changed anything Its really slow this morning I did the same thing yesterday and it was fine Actually its OK nowy No its not Thank you so much you’ve fixed it (I haven’t done anything)y y ( y g)
- 3. Oracle Version < 9 SELECT COUNT(l quantity) FROM Shared Pool SELECT COUNT(l.quantity) SELECT COUNT(l.quantity) FROM bookings_skew l WHERE resource_code = :v1; FROM bookings_skew_large l WHERE resource_code = :v1; FULL SCAN How many rows do I expect? ‘BRLG’ FULL SCAN ‘PC1’‘PC1’ <Version 9 database N bi d kiNo bind peeking
- 4. Oracle Version < 9 SELECT COUNT(l quantity) FROM Shared Pool SELECT COUNT(l.quantity) SELECT COUNT(l.quantity) FROM bookings_skew l WHERE resource_code = :v1; FROM bookings_skew l WHERE resource_code = :v1; FULL SCAN How many rows do I expect? ‘PC1’ FULL SCAN ‘BRLG’‘BRLG’ <Version 9 database N bi d kiNo bind peeking
- 5. What is Bind Peeking?g SELECT COUNT(l quantity) FROM Shared Pool SELECT COUNT(l.quantity) SELECT COUNT(l.quantity) FROM bookings_skew l WHERE resource_code = :v1; FROM bookings_skew l WHERE resource_code = :v1; FULL SCAN How many rows do I expect? ‘BRLG’ FULL SCAN ‘PC1’‘PC1’ >=Version 9 database Bi d kiBind peeking
- 6. What is Bind Peeking?g SELECT COUNT(l quantity) FROM Shared Pool SELECT COUNT(l.quantity) SELECT COUNT(l.quantity) FROM bookings_skew l WHERE resource_code = :v1; FROM bookings_skew l WHERE resource_code = :v1; INDEXED ACCESS How many rows do I expect? ‘PC1’ ACCESS ‘ G‘BRLG’ >=Version 9 database Bi d kiBind peeking
- 7. No Histogramg *************************************** SINGLE TABLE ACCESS PATHSINGLE TABLE ACCESS PATH Column (#3): RESOURCE_CODE(VARCHAR2) AvgLen: 5.00 NDV: 9 Nulls: 0 Density: 0.11111 Table: BOOKINGS SKEW Alias: L 5527238/9 = 614138 Table: BOOKINGS_SKEW Alias: L Card: Original: 5527238 Rounded: 614138 Computed: 614137.56 Non Adjusted: 614137.56 Access Path: TableScanAccess Path: TableScan Cost: 7701.80 Resp: 7701.80 Degree: 0 Cost_io: 7426.00 Cost_cpu: 1605072831 Resp io: 7426.00 Resp cpu: 1605072831p_ p_ p Access Path: index (AllEqRange) Index: BK_RESSKEW resc_io: 32580.00 resc_cpu: 509053165 ix_sel: 0.11111 ix_sel_with_filters: 0.11111 Cost: 32667.47 Resp: 32667.47 Degree: 1 Best:: AccessPath: TableScan Cost: 7701.80 Degree: 1 Resp: 7701.80 Card: 614137.56 Bytes: 0 ***************************************
- 8. Histogram – Majority Firstg j y *************************************** SINGLE TABLE ACCESS PATH Column (#3): RESOURCE_CODE(VARCHAR2) AvgLen: 5.00 NDV: 9 Nulls: 0 Density: 9.0892e-008 Histogram: Freq #Bkts: 9 UncompBkts: 5966 EndPtVals: 9 Table: BOOKINGS_SKEW Alias: L Card: Original: 5464800 Rounded: 1727558 Computed: 1727558.30 Non Adjusted: 1727558.30 A P th T bl SAccess Path: TableScan Cost: 7710.70 Resp: 7710.70 Degree: 0 Cost_io: 7426.00 Cost_cpu: 1656892911 Resp io: 7426 00 Resp cpu: 1656892911 RESO COUNT(*) ---- ---------- VCR1 495711 CONF 495720 LNCH 743576Resp_io: 7426.00 Resp_cpu: 1656892911 Access Path: index (AllEqRange) Index: BK_RESSKEW resc io: 94139 00 resc cpu: 1468544140 LNCH 743576 BRSM 743583 PC1 47858 FLPC 495720 BRLG 1739277 resc_io: 94139.00 resc_cpu: 1468544140 ix_sel: 0.31612 ix_sel_with_filters: 0.31612 Cost: 94391.34 Resp: 94391.34 Degree: 1 Best:: AccessPath: TableScan TAP1 247864 VCR2 495715 Best:: AccessPath: TableScan Cost: 7710.70 Degree: 1 Resp: 7710.70 Card: 1727558.30 Bytes: 0 ***************************************
- 9. Histogram – Minority Firstg y *************************************** SINGLE TABLE ACCESS PATH Column (#3): RESOURCE_CODE(VARCHAR2) AvgLen: 5.00 NDV: 9 Nulls: 0 Density: 9.0892e-008 Histogram: Freq #Bkts: 9 UncompBkts: 5966 EndPtVals: 9 Table: BOOKINGS_SKEW Alias: L Card: Original: 5464800 Rounded: 43968 Computed: 43967.55 Non Adjusted: 43967.55 A P th T bl SAccess Path: TableScan Cost: 7693.35 Resp: 7693.35 Degree: 0 Cost_io: 7426.00 Cost_cpu: 1555877511 Resp io: 7426 00 Resp cpu: 1555877511 RESO COUNT(*) ---- ---------- VCR1 495711 CONF 495720 Resp_io: 7426.00 Resp_cpu: 1555877511 Access Path: index (AllEqRange) Index: BK_RESSKEW resc io: 2399 00 resc cpu: 37397785 LNCH 743576 BRSM 743583 PC1 47858 FLPC 495720 BRLG 1739277resc_io: 2399.00 resc_cpu: 37397785 ix_sel: 0.0080456 ix_sel_with_filters: 0.0080456 Cost: 2405.43 Resp: 2405.43 Degree: 1 Best:: AccessPath: IndexRange Index: BK RESSKEW BRLG 1739277 TAP1 247864 VCR2 495715 Best:: AccessPath: IndexRange Index: BK_RESSKEW Cost: 2405.43 Degree: 1 Resp: 2405.43 Card: 43967.55 Bytes: 0 ***************************************
- 10. What is the CBO OK at Data Condition Literal/Bind Var Histogramg Even Distribution Equality Literal N/A Even Distribution Equality Bind N/A Skewed Equality Literal NO Skewed Equality Literal YESSkewed Equality Literal YES Skewed Equality Bind NO Skewed Equality Bind YES
- 11. Histogramsg V i < 9 Hi t ith bi d i blVersions < 9: Histograms are no use with bind variables Versions >= 9: Histograms are worse than no useVersions > 9: Histograms are worse than no use with bind variables Each distinct SQL uses only either minority or majority Unless… values but not both Or you don’t keep/use the statements in the shared pool (in which case you might as well use literals)( y g )
- 12. So if I have no Skewed Data I am OK? Data Condition Literal/Bind Var Histogramg Even Distribution Equality Literal N/A Even Distribution Equality Bind N/A Skewed Equality Literal NO Skewed Equality Literal YESSkewed Equality Literal YES Skewed Equality Bind NO Skewed Equality Bind YES
- 13. Minority Range Firsty g ******************************************************************************** SINGLE TABLE ACCESS PATHSINGLE TABLE ACCESS PATH Column (#2): EVENT_NO(NUMBER) AvgLen: 5.00 NDV: 99178 Nulls: 0 Density: 1.0083e-005 Min: 211 Max: 100175 Table: BOOKINGS EVEN Alias: LTable: BOOKINGS_EVEN Alias: L Card: Original: 5522697 Rounded: 4973 Computed: 4972.66 Non Adjusted: 4972.66 Access Path: TableScanAccess Path: TableScan Cost: 7742.79 Resp: 7742.79 Degree: 0 Cost_io: 7492.00 Cost_cpu: 1459499283 Resp io: 7492.00 Resp cpu: 1459499283 MIN MAX ------ ------ 100 100200p_ p_ p Access Path: index (RangeScan) Index: BK_EVTEVEN resc_io: 4992.00 resc_cpu: 37790328 ix_sel: 9.0040e-004 ix_sel_with_filters: 9.0040e-004 Cost: 4998.49 Resp: 4998.49 Degree: 1 Best:: AccessPath: IndexRange Index: BK_EVTEVEN Cost: 4998.49 Degree: 1 Resp: 4998.49 Card: 4972.66 Bytes: 0 ******************************************************************************
- 14. So if I have no Skewed Data I am OK? Data Condition Literal/Bind Var Histogramg Even Distribution Equality Literal N/A Even Distribution Equality Bind N/A Skewed Equality Literal NO Skewed Equality Literal YESSkewed Equality Literal YES Skewed Equality Bind NO Skewed Equality Bind YES Even Distribution Range Bind N/A
- 15. Partitions? Which statistics shall we use? Resource code Resource_code has 1 value Partition empty _ has approx 5600 values has 1 valueempty <100 <200 <300 <400 <500 Global statistics resource_code has 6116 values
- 16. What Can We do? Own code:- Use Literals – but only for skewed data Write separate code for minority/majority Give user separate menu options Daily report v Annual report Build sql dynamically with hints for various cases User not aware (and keep your histograms)
- 17. What Can We do? Package:- If we have no histogram what is it doing? Full scan Is a full scan acceptable? Yes No histogram Turn off bind peeking _optim_peek_user_binds=false_ p _p _ _ Is a full scan acceptable? Nop Run the majority ones first Different session for majority/minority cases and change session variables
- 18. What Can We do? Package:- If we have no histogram what is it doing? Index scan (So none of these will help No histogram Turn off bind peekingp g _optim_peek_user_binds=false) Remove index? Create an outline and force a full scan Run the majority ones first Different session for majority/minority cases and change session variables
- 19. What Can We do? Package:- Purge the statement from the shared pool each execution (10.2.0.4) : dbms_shared_pool.purge ('&address, &hash_value','c')( _ )
- 20. Oracle 11g - Datag RESO NUM ---- ---------- VCR1 524288 LNCH 786432LNCH 786432 CONF 524288 BRSM 786432BOOKINGS_EVEN PC1 262144 FLPC 524288 BRLG 1310720 TAP1 262144 VCR2 524288VCR2 524288 ---------- sum 5505024
- 21. Oracle 11g - Datag RESO NUM ---- ---------- RESO NUM ---- ---------- PC2 99 PC3 99 PC4 99 PC1 47865 TAP1 247853 FLPC 495713 PC5 99 PC6 99 PC7 99 PC8 99 BOOKINGS_SKEW CONF 495714 VCR2 495718 VCR1 495720 PC8 99 PC9 99 PC11 101 PC12 102 BRSM 743571 LNCH 743581 BRLG 1737243 ---------- PC13 103 PC14 104 PC15 199 sum 5505024 PC16 399 PC17 799 PC18 999 PC19 1999PC19 1999 PC20 3999 PC10 199
- 22. Bind Peeking + Adaptive Cursors Summary Statements with bind variables (+histograms) are bind itisensitive The first time you execute a statement with different selectivity it uses the original plan The second time it changes the plan and become bind awareThe second time it changes the plan and become bind aware New values will use a plan for the appropriate selectivity range Be careful when statements become invalidated by:- Gathering statistics Flushing the shared pool Restarting the database Or when they are aged out
- 23. Adaptive Cursors Functionality Adaptive Cursors 11 1 0 6Adaptive Cursors 11.1.0.6 Bind variable with Equality and Histogram Not for range conditionsNot for range conditions Adaptive Cursors 11.1.0.7 Bind variable with Equality and HistogramBind variable with Equality and Histogram Range conditions Do Not Support LIKEDo Not Support LIKE Future? LIKELIKE SELECT /*+ HINT TO MAKE IT BIND AWARE */
- 24. Gathering Statisticsg E l CBO “M k th t ti ti l l ”Early CBO: “Make sure you gather statistics regularly” Later CBO: “Don’t gather statistics unless data patterns change”change BUT If you have new majority values you need to recreate the histogram
- 25. SQL Plan Management Manual capture DBMS_SPM.LOAD_PLANS_FROM_SQLSET DBMS SPM.LOAD PLANS FROM CURSOR CACHE Auto capture of repeatable statements OPTIMIZER_CAPTURE_SQL_PLAN_BASELINE = TRUE DBMS_SPM.LOAD_PLANS_FROM_CURSOR_CACHE B li SQL Management Base M l l d/ t f l Baseline = (Stored and Accepted plans) Manual load/accept of new plan DBMS_SPM.LOAD_PLANS_FROM_SQLSET DBMS_SPM.LOAD_PLANS_FROM_CURSOR_CACHE SQL Tuning Advisor identifies new plan SQL*Profile accepted Auto accept of new plan (if it performs better) N Pl plan – SQL Profile accepted (if it performs better) DBMS_SPM.EVOLVE_SQL_PLAN_BASELINE New Plan identified during execution Stored not accepted
- 26. SQL Plan Management – Binds • Will not automatically handle adaptive cursors • New plan identified on first execution• New plan identified on first execution • New plan recorded as not accepted • Plan will not evolve • All bind variable values use same baseline plan • Plans show as not bind sensitive or aware
- 27. SQL Plan Management – Binds
- 28. SQL Plan Management – Binds
- 29. SQL Plan Management – Binds Force Acceptance of the plan
- 30. SAGE Computing Services Customised Oracle Training Workshops and ConsultingCustomised Oracle Training Workshops and Consulting Questions? tiwww.sagecomputing.com.au penny@sagecomputing.com.au