Statistics Gathering and Histograms - DOs and DON'Tscoug.ab.ca/.../Statistics_Gathering_and_Histograms... · Statistics Gathering and Histograms ... To save histogram you should set

Statistics Gatheringand Histograms

DOs ✔ and DON’Ts ✘

Wolfgang Breitlingwww.centrexcc.com

Copyright 2009 Centrex Consulting Corporation. Personal use of this material is permitted. However, permission to reprint/republish

this material for advertising and promotional purposes or for creating new collective works for resale or redistribution to servers

or lists, or to reuse any copyrighted component of this work in other works must be obtained from Centrex Consulting.

1

© Wolfgang Breitling, Centrex Consulting Corporation Calgary Oracle Users Group October 18, 2012

Who Am I

Independent Consultant since 1996

Specializing in Oracle and Peoplesoft Setup,Administration, and Performance Tuning

Member of the Oaktable Network

30+ Years in Database ManagementDL/1, IMS, ADABAS, SQL/DS, DB2, Oracle

Oracle since 1993 (7.0.12)

OCP Certified DBA - 7, 8, 8i, 9i

2


DOn’ts

✘ Do not use ANALYZE ANYMORE

✘ Do not use DBMS_UTILITY.ANALYZE_ {DATABASE | SCHEMA |PART_OBJECT}

✘ Do not use DBMS_DDL.ANALYZE_OBJECT ( pre 10g ) )

3


DO

✔ use DBMS_STATS.GATHER_xxx_STATS

✔ from 11g on use auto_sample_size✔ in 10g or less ✔ if you can’t use estimate_percent 100, gather table stats with cascade=>false

✔ then gather index stats separately with estimate_percent=>100.

4


Effect Of Samplesize

OBJ# SAMPLESIZE ROWCNT BLKCNT ANALYZETIME SAVE_TIME ------- ----------- --------- -------- ------------------- ------------------- 81504 39573017 39573017 1308033 2011-07-09 10:52:17 2011-07-16 10:54:15 81504 39698473 39698473 1316217 2011-07-16 10:54:15 2011-07-16 16:29:06 81504 7940776 39703880 1316217 2011-07-16 16:29:06 2011-07-23 10:53:36 81504 39815392 39815392 1316217 2011-07-23 10:53:36 2011-07-30 08:16:13 81504 7990753 39953765 1324401 2011-07-30 08:16:13 2011-07-30 10:58:37 OBJ# SAMPLESIZE ROWCNT LVL LEAFCNT DISTKEY LBLKKEY DBLKKEY CLUFAC ANALYZETIME SAVE_TIME ------- ---------- --------- ---- -------- --------- ------- ------- --------- ------------------- ------------------- 177197 432525 38212054 2 90202 16294 5 153 2507361 2011-07-09 10:52:18 2011-07-16 10:54:17 177197 451419 39881272 2 94266 16337 5 162 2660112 2011-07-16 10:54:17 2011-07-16 16:29:48 177197 7788338 38941690 2 92280 15622 5 162 2542045 2011-07-16 16:29:48 2011-07-23 10:53:37 177197 476255 42075445 2 100185 16369 6 173 2837600 2011-07-23 10:53:37 2011-07-30 08:16:40 177197 8132718 40663590 2 96885 15660 6 165 2598410 2011-07-30 08:16:40 2011-07-30 10:58:39

177225 199779 37600712 3 198563 1009304 1 15 15845153 2011-07-09 10:52:19 2011-07-16 10:54:18 177225 210652 39647136 3 210609 1041176 1 16 16733511 2011-07-16 10:54:18 2011-07-16 16:30:17 177225 8055319 40276595 3 211595 225587 1 73 16565200 2011-07-16 16:30:17 2011-07-23 10:53:38 177225 215853 40626024 3 216067 1021255 1 16 16907983 2011-07-23 10:53:38 2011-07-30 08:16:59 177225 7981072 39905360 3 213145 216277 1 76 16595720 2011-07-30 08:16:59 2011-07-30 10:58:39

5


Column Statistics Gathering

✘ Do not use“for all indexed columns ...”

✔ Do use“for all columns size 1”then✔ “for columns size 254 col1[,col2,...]”✘ not “for columns col1[,col2,...] size 254” *

for columns requiring histograms * see www.centrexcc.com/Active%20Statistics.ppt.pdf

6


When To Use Histograms

Histograms are stored in the dictionary.There is a maintenance and space cost for using histograms.Only create histograms on columns that are used in WHERE clauses of queriesand have a highly-skewed data distribution.method_opt “size auto” is supposed to do just that but does not always get the second part right.

7


When Not To Use Histograms

Histograms are not useful ifall predicates on the column use bind variables the column data is uniformly distributed the column is not used in WHERE clauses of queries the column is unique and is used only with equality predicates

8


method_opt “size auto”

Histograms are not gathered on columns for which there is no predicate information in the col_usage$ table.

col_usage$ is only populatedwhen queries are hard parsedwith columns referenced in predicates

9


Bind Peeking

A cursor will be moved out of the Shared Pool and therefore subsequently requires a hard parse in a number of circumstances:Database shutdown/restartCursor not in use by any session, and aged out by LRU algorithmIt will not get moved out by flushing the Shared Pool if it is pinned (i.e. in use)

Change to the structure of any referenced object (e.g. alter table)

Granting/revoking privileges on a referenced object.Change to the stats associated with any referenced object(e.g. following a gather stats job). However, that may not be immediate if no_invalidate is AUTO

10


export_table_stats

Since 10g whenever optimizer statistics are modified using the DBMS_STATS package, old versions of the statistics are saved automatically for future restoration, so the above is not necessary.

However, after 31 days they DO get purged.See Note 452011.1 "Restoring table statistics in 10G onwards".

✔ Do export your (good) statistics regularly.

11


import_table_stats

✔ DO delete object statistics before importing them. Always.

12


‘gather auto’ vs. ‘gather stale’

gather autoignores all other options and uses

‘for all columns size skewonly’ until 11.1.0.7‘for all columns size auto’ for 11.1.0.7 and up

gather stale other parameters specified will also be taken does not gather statistics on objects which do not have statistics. ( modifications can only be detected relative to existing num_rows )

13


varchar2(n) n>32

ANALYZE command gathers incorrect num_distinct in every release if the length of character column is more than 32 bytes.DBMS_STATS with METHOD_OPT => 'FOR ALL

COLUMNS SIZE 1' gathers correct num_distinct.From 11g on, DBMS_STATS with METHOD_OPT => 'FOR ALL COLUMNS SIZE N' (N >1)generates correct num_distinct.But, it still generates incorrect histograms.

14


Index Rebuild Problem

ALTER INDEX ... REBUILD COMPUTE STATISTICSdeletes histograms of the leading column.

This is due to the Bug 6040988 To save histogram you should set event 8130 at system or session level.

ALTER {session|system} SET EVENTS'8130 TRACE NAME CONTEXT FOREVER';

Reported for 9.2.0.8, appears fixed in 10g

15


Index Rebuild Problem

• Gather statistics of the table• [ wait a few minutes ]• Rebuild the index• Restore the statistics giving a time between gathering the

statistics and the rebuild.

You end up with no statistics at all on the index. Not the ones gathered with the table nor the ones gathered as a result of the rebuild.The statistics are just wiped out.

Fixed in Oracle 11.2.0.1 and up

16


Deleting Histograms

Prior to 11g there is no simple way to delete just the column histogram statistics, leaving the basic column statistics, withoutre-analyzing.Workaround:

Use DBMS_STATS.GET_COLUMN_STATS to get all column stats, change the bucket count in SREC to reflect no histogram and use DBMS_STATS.SET_COLUMN_STATS to re-set them.

Oracle 11g adds a new argument (COL_STAT_TYPE) to DBMS_STATS.DELETE_COLUMN_STATSto provide the ability to delete histogram statistics only. COL_STAT_TYPE can be

ALL is the default behaviour which deletes all column statistics HISTOGRAM will only delete the histograms for the column

17


Extended Stats and Histograms

Poor ( overestimated ) table cardinality estimates are possible when using multi-column statistics if there were predicates on columns with histograms that were not part of the column group used to calculate selectivity ( bug 9069046 ).

Affected: 10.2.0.4 <= Oracle version < 11.2.0.2

Fixed in 11.2.0.2

18

Wolfgang [email protected]

Centrex Consulting Corp.www.centrexcc.com

Documents

Statistics Gathering and Histograms - DOs and DON'Tscoug.ab.ca/.../Statistics_Gathering_and_Histograms... · Statistics Gathering and Histograms ... To save histogram you should set