记录一次没有收集直方图优化器选择全表扫描导致CPU耗尽

场景：数据库升级第二天，操作系统CPU使用率接近100%。

查看ash报告：

再看TOP SQL

具体SQL：

select count(1) as chipinCount, sum(bets) as sumBets from t_chipin_temp where status in (0) and lottype='gp'

看到这个sql的执行计划走全表扫描。再查看相关列上是否有索引，结果是有索引的。那么问题来了，既然有索引，而且升级之后对该表格还进行过基本的统计信息收集，那么为什么优化器没有走索引。

收集该sql的10053事件

SQL> select child_number from v$sql where sql_id='57tdrj6a4ync4';

CHILD_NUMBER
  ------------
        0

SQL> execute DBMS_SQLDIAG.DUMP_TRACE(p_sql_id=>'57tdrj6a4ync4', p_child_number=>0, p_component=>'Optimizer', p_file_id=>'SQL_TRACE_10053');

PL/SQL procedure successfully completed.

SQL> SELECT value FROM v$diag_info WHERE name='Default Trace File';

VALUE
--------------------------------------------------------------------------------
/u01/app/oracle/diag/rdbms/orcl11g/ORCL/trace/ORCL_ora_32070_SQL_TRACE_10053.trc

vi  /u01/app/oracle/diag/rdbms/orcl11g/ORCL/trace/ORCL_ora_32070_SQL_TRACE_10053.trc

Access path analysis for T_CHIPIN_TEMP
***************************************
SINGLE TABLE ACCESS PATH
Single Table Cardinality Estimation for T_CHIPIN_TEMP[T_CHIPIN_TEMP]
Column (#8): STATUS(
AvgLen: 4 NDV: 4 Nulls: 0 Density: 0.250000 Min: -2 Max: 4
Column (#44): lottype(
AvgLen: 3 NDV: 2 Nulls: 0 Density: 0.500000
ColGroup (#4, Index) TX_CHIPIN_TMP_TIMCODE2
Col#: 5 11 31 33 CorStregth: -1.00
ColGroup (#2, Index) IDX_T_CHIPIN_TEMP
Col#: 13 34 44 CorStregth: -1.00
ColGroup (#3, Index) TX_CHIPIN_TMP_TIME2
Col#: 8 20 44 CorStregth: -1.00
ColGroup (#5, Index) TX_CHIPIN_TMP_AWARD2
Col#: 15 35 36 CorStregth: -1.00
ColGroup (#6, Index) IX_CHIPIN_TEMP_STALOC2
Col#: 8 44 CorStregth: 2.67
ColGroup (#1, Index) IDX_S_P
Col#: 8 31 CorStregth: -1.00
ColGroup Usage:: PredCnt: 2 Matches Full: #6 Partial: Sel: 0.3420
Table: T_CHIPIN_TEMP Alias: T_CHIPIN_TEMP
Card: Original: 5671105.00 Rounded: 1939622 Computed: 1939621.71 Non Adjusted: 1939621.71
           原始行数                    近似值                       精确值                            非修正值

Access Path: TableScan
Cost: 62160.58 Resp: 62160.58 Degree: 0
Cost_io: 61966.00 Cost_cpu: 7711574370
Resp_io: 61966.00 Resp_cpu: 7711574370
kkofmx: index filter:"T_CHIPIN_TEMP"."lottype"=:B1

kkofmx: index filter:"T_CHIPIN_TEMP"."STATUS"=:B1

kkofmx: index filter:"T_CHIPIN_TEMP"."lottype"=:B1

Access Path: index (RangeScan)
Index: IDX_S_P
resc_io: 918156.00 resc_cpu: 8328403927
ix_sel: 0.250000 ix_sel_with_filters: 0.250000
Cost: 918366.14 Resp: 918366.14 Degree: 1

Access Path: index (skip-scan)
SS scan sel: 0.500000 SS filter sel: 0.500000 ANDV (#skips): 564.000000
SS io: 14684.000000 vs. table scan io: 61966.000000
Skip Scan chosen
Access Path: index (SkipScan)
Index: IDX_T_CHIPIN_TEMP
resc_io: 226050.00 resc_cpu: 5276879027
ix_sel: 0.500000 ix_sel_with_filters: 0.500000
Cost: 226183.15 Resp: 226183.15 Degree: 1
ColGroup Usage:: PredCnt: 2 Matches Full: #6 Partial: Sel: 0.3420
ColGroup Usage:: PredCnt: 2 Matches Full: #6 Partial: Sel: 0.3420

Access Path: index (AllEqRange)
Index: IX_CHIPIN_TEMP_STALOC2
resc_io: 128169.00 resc_cpu: 3259690463
ix_sel: 0.350930 ix_sel_with_filters: 0.350930
Cost: 128251.25 Resp: 128251.25 Degree: 1
ColGroup Usage:: PredCnt: 2 Matches Full: #6 Partial: Sel: 0.3420
Access Path: index (skip-scan)
SS scan sel: 0.344312 SS filter sel: 0.344312 ANDV (#skips): 1579520.000000
SS io: 20452.000000 vs. table scan io: 61966.000000
Skip Scan chosen
Access Path: index (SkipScan)
Index: TX_CHIPIN_TMP_TIME2
resc_io: 1157882.00 resc_cpu: 10592729810
ix_sel: 0.344312 ix_sel_with_filters: 0.344312
Cost: 1158149.27 Resp: 1158149.27 Degree: 1
****** trying bitmap/domain indexes ******
****** finished trying bitmap/domain indexes ******
******** Begin index join costing ********
****** trying bitmap/domain indexes ******
ColGroup Usage:: PredCnt: 2 Matches Full: #6 Partial: Sel: 0.3420
ColGroup Usage:: PredCnt: 2 Matches Full: #6 Partial: Sel: 0.3420
ColGroup Usage:: PredCnt: 2 Matches Full: #6 Partial: Sel: 0.3420
Access Path: index (AllEqRange)
Index: IX_CHIPIN_TEMP_STALOC2
resc_io: 4610.00 resc_cpu: 420754238
ix_sel: 0.350930 ix_sel_with_filters: 0.350930
Cost: 4620.62 Resp: 4620.62 Degree: 0
Bitmap nodes:
Used IX_CHIPIN_TEMP_STALOC2
Cost = 4627.223361, sel = 0.342018
****** finished trying bitmap/domain indexes ******
******** End index join costing ********
Best:: AccessPath: TableScan
Cost: 62160.58 Degree: 1 Resp: 62160.58 Card: 1939621.71 Bytes: 0

***************************************

10053中看到因为没有直方图存在，所以这里的Column (#8) Density = 0.25 ,Column (#44) Density = 0.5 是从 1/ NDV 算得的,统计信息显示的status in (0)  数据行占总行数的1/4， lottype='gp' 数据行占总行数的1/2, 所以优化器选择做全表扫描.

SQL> select count(1) as chipinCount, sum(bets) as sumBets from t_chipin_temp where status in (0) and lottype='gp' ;

CHIPINCOUNT SUMBETS
-----------         ---------
   3                   20

以上符合status in (0) and lottype='gp'的只有3行，且status, lottype列上建有复合索引，索引是IX_CHIPIN_TEMP_STALOC2，同时也使用了dbms_stats包收集表和索引上的统计信息，照理说CBO应该选择 INDEX RANGE SCAN，而避免全表扫描，但实时上优化器opitimizer仍然全表扫描。

经过一番查阅，发现其原因是在收集统计信息时并没有收集到直方图，只要收集了直方图，那么优化器就会了解到status in (0) and lottype='gp' 条件仅有少量的行，优化器会走索引。

于是重新收集统计信息加上直方图：

execute dbms_stats.gather_table_stats(ownname => 'USER',tabname => 'T_CHIPIN_TEMP' ,method_opt => 'FOR ALL COLUMNS SIZE 2' ,cascade => true);

alter system flush shared_pool;

执行完以上操作后，系统CPU很快降了下来。

再看执行计划，便是 INDEX RANGE SCAN 了。

再看收集直方图之后10053的内容：

Access path analysis for T_CHIPIN_TEMP
***************************************
SINGLE TABLE ACCESS PATH
Single Table Cardinality Estimation for T_CHIPIN_TEMP[T_CHIPIN_TEMP]
Column (#8):
NewDensity:0.000180, OldDensity:0.000000 BktCnt:5558, PopBktCnt:5558, PopValCnt:3, NDV:3
Column (#8): STATUS(
AvgLen: 4 NDV: 3 Nulls: 0 Density: 0.000180 Min: -2 Max: 4
Histogram: Freq #Bkts: 3 UncompBkts: 5558 EndPtVals: 3
Column (#44):
NewDensity:0.000090, OldDensity:0.000000 BktCnt:5559, PopBktCnt:5558, PopValCnt:1, NDV:2
Column (#44): lottype(
AvgLen: 3 NDV: 2 Nulls: 0 Density: 0.000090
Histogram: Freq #Bkts: 2 UncompBkts: 5559 EndPtVals: 2
ColGroup (#5, Index) TX_CHIPIN_TMP_TIMCODE2
Col#: 5 11 31 33 CorStregth: -1.00
ColGroup (#6, Index) TX_CHIPIN_TMP_AWARD2
Col#: 15 35 36 CorStregth: -1.00
ColGroup (#4, Index) TX_CHIPIN_TMP_TIME2
Col#: 8 20 44 CorStregth: -1.00
ColGroup (#3, Index) IDX_T_CHIPIN_TEMP
Col#: 13 34 44 CorStregth: -1.00
ColGroup (#1, Index) IX_CHIPIN_TEMP_STALOC2
Col#: 8 44 CorStregth: 2.00
ColGroup (#2, Index) IDX_S_P
Col#: 8 31 CorStregth: -1.00
ColGroup Usage:: PredCnt: 2 Matches Full: Partial:
Table: T_CHIPIN_TEMP Alias: T_CHIPIN_TEMP
Card: Original: 5682253.000000 Rounded: 1 Computed: 0.18 Non Adjusted: 0.18
Access Path: TableScan
Cost: 62159.09 Resp: 62159.09 Degree: 0
Cost_io: 61966.00 Cost_cpu: 7652553546
Resp_io: 61966.00 Resp_cpu: 7652553546
kkofmx: index filter:"T_CHIPIN_TEMP"."lottype"='gp'

kkofmx: index filter:"T_CHIPIN_TEMP"."STATUS"=1

kkofmx: index filter:"T_CHIPIN_TEMP"."lottype"='gp'

Access Path: index (RangeScan)
Index: IDX_S_P
resc_io: 667.00 resc_cpu: 6041476
ix_sel: 0.000180 ix_sel_with_filters: 0.000180
Cost: 667.15 Resp: 667.15 Degree: 1

Access Path: index (skip-scan)
SS scan sel: 0.000180 SS filter sel: 0.000180 ANDV (#skips): 564.000000
SS io: 14379.000000 vs. table scan io: 61966.000000
Skip Scan chosen
Access Path: index (SkipScan)
Index: IDX_T_CHIPIN_TEMP
resc_io: 14450.00 resc_cpu: 104195010
ix_sel: 0.000180 ix_sel_with_filters: 0.000180
Cost: 14452.63 Resp: 14452.63 Degree: 1
ColGroup Usage:: PredCnt: 2 Matches Full: Partial:
ColGroup Usage:: PredCnt: 2 Matches Full: Partial:
Access Path: index (AllEqRange)
Index: IX_CHIPIN_TEMP_STALOC2
resc_io: 4.00 resc_cpu: 29696
ix_sel: 0.000000 ix_sel_with_filters: 0.000000
Cost: 4.00 Resp: 4.00 Degree: 1
ColGroup Usage:: PredCnt: 2 Matches Full: Partial:
Access Path: index (skip-scan)
SS scan sel: 0.000000 SS filter sel: 0.000000 ANDV (#skips): 1582976.000000
SS io: 20631.000000 vs. table scan io: 61966.000000
Skip Scan chosen
Access Path: index (SkipScan)
Index: TX_CHIPIN_TMP_TIME2
resc_io: 20634.00 resc_cpu: 146945003
ix_sel: 0.000000 ix_sel_with_filters: 0.000000
Cost: 20637.71 Resp: 20637.71 Degree: 1
****** trying bitmap/domain indexes ******
****** finished trying bitmap/domain indexes ******
******** Begin index join costing ********
****** trying bitmap/domain indexes ******
ColGroup Usage:: PredCnt: 2 Matches Full: Partial:
ColGroup Usage:: PredCnt: 2 Matches Full: Partial:
ColGroup Usage:: PredCnt: 2 Matches Full: Partial:
Access Path: index (AllEqRange)
Index: IX_CHIPIN_TEMP_STALOC2
resc_io: 3.00 resc_cpu: 21564
ix_sel: 0.000000 ix_sel_with_filters: 0.000000
Cost: 3.00 Resp: 3.00 Degree: 0
Bitmap nodes:
Used IX_CHIPIN_TEMP_STALOC2
Cost = 3.000545, sel = 0.000000
****** finished trying bitmap/domain indexes ******
******** End index join costing ********
Best:: AccessPath: IndexRange
Index: IX_CHIPIN_TEMP_STALOC2
Cost: 4.00 Degree: 1 Resp: 4.00 Card: 0.18 Bytes: 0

从10053可以看出手动指定收集直方图后CBO优化器才能做出正确的选择

注：dbms_stats的自动决定直方图的收集与否及收集的桶数受到col_usage$表影响，如果某张表的列存在于col_usage$中，oracle就认为该列存在收集直方图的必要。

　　Histograms are not gathered on columns for which there is no predicate information captured in the  col_usage$ view. Col_usage$ is populated with the column information only when queries are executed with columns referred in predicate information.