使用DBMS_STATS来收集统计信息

Index Statistics

Oracle collects the following statistics for an index. Statistics marked with an asterisk are always computed exactly. For conventional indexes, the statistics appear in the data dictionary views USER_INDEXES, ALL_INDEXES, and DBA_INDEXES in the columns in parentheses.（带*为准确值）

oracle收集如下的索引统计信息。对于常规索引，可以在视图USER_INDEXES, ALL_INDEXES, and DBA_INDEXES中查看到如下的统计信息。

Where are the statistics stored?

These views are prefixed with DBA_ or ALL_ or USER_. For ease of reading, we will use DBA_% views, but ALL_% views or USER_% views could be used as well.

统计信息存储在数据字典中，在sys用户下的表内。通过视图我们可以非常方便的从这些表中获取信息。视图通常以DBA_ USER_ ALL_开始。为了简便，我们以DBA_开头的视图为例。

Conventions Used

- Statistics available only since 8.0.X rdbms release : (*)
- Statistics available only since 8.1.X rdbms release : (**)
- Statistics not available at partition or subpartition level : (G)
- Statistics not available at subpartition level : (GP)

Table level statistics can be retrieved from:

Columns to look at are:

 NUM_ROWS : Number of rows (always exact even when computed 
 	 with ESTIMATE method) 
 BLOCKS : Number of blocks which have been used even 
 if they are empty due to delete statements 
 EMPTY_BLOCKS : Number of empty blocks (these blocks have 
 never been used) 
 AVG_SPACE : Average amount of FREE space in bytes in blocks 
 allocated to the table : Blocks + Empty Blocks 
 CHAIN_CNT : Number of chained or migrated rows 
 AVG_ROW_LEN : Average length of rows in bytes 
 AVG_SPACE_FREELIST_BLOCKS (*)(G) : Average free space of blocks in the freelist 
 NUM_FREELIST_BLOCKS (*)(G) : Number of blocks in the freelist 
 SAMPLE_SIZE : Sample defined in ESTIMATE method (0 if COMPUTE) 
 LAST_ANALYZED : Timestamp of last analysis 
 GLOBAL_STATS (**) : For partitioned tables, YES means statistics 
 are collected for the TABLE as a whole 
 NO means statistics are estimated from statistics 
 on underlying table partitions or subpartitions 
 USER_STATS (**) : YES if statistics entered directly by the user

Index level statistics can be retrieved from:

Columns to look at are:

 BLEVEL : B*Tree level : depth of the index from its root 
 block to its leaf blocks (从0开始)
 LEAF_BLOCKS : Number of leaf blocks 
 DISTINCT_KEYS : Number of distinct keys 
 AVG_LEAF_BLOCKS_PER_KEY : Average number of leaf blocks in which each 
 distinct key appears (1 for a UNIQUE index) 
 AVG_DATA_BLOCKS_PER_KEY : Average number of data blocks in the table that 
 are pointed to by a distinct key 
 CLUSTERING_FACTOR : - if near the number of blocks, then the table is 
 ordered : index entries in a single leaf block 
 tend to point to rows in same data block 
 - if near the number of rows, the table is 
 randomly ordered : index entries in a single 
 leaf block are unlikely to point to rows in 
 same data block 
 SAMPLE_SIZE : Sample defined in ESTIMATE method (0 if COMPUTE) 
 LAST_ANALYZED : Timestamp of last analysis 
 GLOBAL_STATS (**) : For partitioned indexes, YES means statistics 
 are collected for the INDEX as a whole 
 NO means statistics are estimated from statistics 
 on underlying index partitions or subpartitions 
 USER_STATS (**) : YES if statistics entered directly by the user 
 PCT_DIRECT_ACCESS (**)(GP) : For secondary indexes on IOTs, percentage of 
 rows with VALID guess（可以通过alter index index_name update block references来更新）

Column level statistics can be retrieved from:

The last three views extract statistics data from DBA_TAB_COLUMNS.（后三个视图是从DBA_TAB_COLUMNS获取数据）

Columns to look at are:

 NUM_DISTINCT : Number of distinct values 
 LOW_VALUE : Lowest value 
 LOW_VALUE : Highest value 
 DENSITY : Density 
 NUM_NULLS : Number of columns having a NULL value 
 AVG_COL_LEN : Average length in bytes 
 NUM_BUCKETS : Number of buckets in histogram for the column 
 SAMPLE_SIZE : Sample defined in ESTIMATE method (0 if COMPUTE) 
 LAST_ANALYZED : Timestamp of last analysis 
 (**)GLOBAL_STATS : For partitioned tables, YES means statistics 
 are collected for the TABLE as a whole 
 NO means statistics are estimated from statistics 
 on underlying table partitions or subpartitions 
 (**)USER_STATS : YES if statistics entered directly by the user

Compute statistics vs. Estimate statistics

Both computed and estimated statistics are used by the Oracle optimizer to choose the execution plan for SQL statements that access analyzed objects. These statistics may also be useful to application developers who write such statements.

无论是采用compute还是采用estimat的方式计算统计信息，优化器都会根据这些信息来选择执行计划。程序员也可以根据这些统计信息来编写sql语句。

COMPUTE STATISTICS

When computing statistics, an entire object is scanned to gather data about the object. This data is used by Oracle to compute exact statistics about the object. Slight variances throughout the object are accounted for in these computed statistics. Because an entire object is scanned to gather information for computed statistics, the larger the size of an object, the more work that is required to gather the necessary information.

To perform an exact computation, Oracle requires enough space to perform a scan and sort of the table. If there is not enough space in memory, then temporary space may be required. For estimations, Oracle requires enough space to perform a scan and sort of only the rows in the requested sample of the table. For indexes, computation does not take up as much time or space, so it is best to perform a full computation.

Some statistics are always computed exactly, such as the number of data blocks currently containing data in a table or the depth of an index from its root block to its leaf blocks.

Use estimation for tables and clusters rather than computation, unless you need exact values. Because estimation rarely sorts, it is often much faster than computation, especially for large tables.

当COMPUTE STATISTICS时，oracle会精确计算被分析对象的统计信息，并将其存储在数据字典中。oracle会扫描整个对象来获取数据，并根据这些数据计算统计信息。对于这种方式，基本是轻微的变化也会被计算在内。因为整个对象都会被扫描，因此对象越大就会需要越多的工作量来完成统计。

为了完成精确统计，oracle需要足够的空间来执行扫描和排序作业。如果在内存中不存在足够的空间，就会占用磁盘的临时空间。对于estimation方式，oracle仅仅需要扫描和排序所采样的内容。如果我们统计的对象是索引，computation方式不会占用太多的时间和空间，因此对于索引我们最好采用compute方式。

某些统计信息总是精确计算的，例如表所占用的数据块数量和索引的深度。

对于表和聚簇，我们建议使用estimation的方式，除非真的需要精确的统计信息。因此estatimation方式通常不会发生排序，速度更快，尤其在分析大表时。

ESTIMATE STATISTICS

When estimating statistics, Oracle gathers representative information from portions of an object. This subset of information provides reasonable, estimated statistics about the object. The accuracy of estimated statistics depends upon how representative the sampling used by Oracle is. Only parts of an object are scanned to gather information for estimated statistics, so an object can be analyzed quickly. You can optionally specify the number or percentage of rows that Oracle should use in making the estimate.

estimate statistics 使得oracle评估待分析对象的统计信息并将它们存储在数据字典中。当评估统计信息时，oracle在待分析对象的部分区间内收集信息。这部分信息为分析对象提供了足够的内容。estimate方式的准确程度主要依赖于oracle是如何采样的。由于只有部分内容被扫描，因此速度更快。我们可以指定oracle采样的百分比。

To estimate statistics, Oracle selects a random sample of data. You can specify the sampling percentage and whether sampling should be based on rows or blocks.

对于estimate方式，oracle会随机采样数据。我们可以指定采样的百分比，也可以指定是根据记录还是根据块来采样。

Notes on estimating statistics

DBMS_STATS functions and variable definitions

Most of the DBMS_STATS procedures include the three parameters statown, stattab, and statid. These parameters allow you to store statistics in your own tables (outside of the dictionary), which does not affect the optimizer. Therefore, you can maintain and experiment with sets of statistics.

大部分DBMS_STAT过程包含三个参数STATOWN,STATTAB和statid。这些参数允许我们将统计信息存放到自己的表中，这些统计信息不回影响优化器。因此，我们可以维护和测试统计信息。

The stattab parameter specifies the name of a table in which to hold statistics, and it is assumed that it resides in the same schema as the object for which statistics are collected (unless the statown parameter is specified). Users may create multiple tables with different stattabidentifiers to hold separate sets of statistics.stattab参数规定了保存统计信息的表明，通常情况下，如果没有指定statown参数，oracle以被统计对象所在的模式用户为stattab的拥有者。我们可以使用不同的stattab来分别存储不同的统计信息。

Additionally, users can maintain different sets of statistics within a single stattab by using the statid parameter, which can help avoid cluttering the user's schema.

灵位，我们也可以指定statid参数，从而在相同的stattab中存储不同的统计信息，这样可以使用户模式显得井井有条。

For all of the SET or GET procedures, if stattab is not provided (i.e., NULL), then the operation works directly on the dictionary statistics; therefore, users do not need to create these statistics tables if they only plan to modify the dictionary directly. However, if stattab is not NULL, then the SET or GET operation works on the specified user statistics table, and not the dictionary.

对于所有的set和get过程，如果我们没有指定stattab，oracle会将统计信息写入数据字典，如果指定了stattab，orcle只会将统计信息写入用户自定义表，而不会更新数据字典。

Create Stats Table

DBMS_STATS.CREATE_STAT_TABLE (
 ownname VARCHAR2, 
 stattab VARCHAR2,
 tblspace VARCHAR2 DEFAULT NULL);

stattab : Name of the table to create. This value should be passed as the stattab parameter to other procedures when the user does not want to modify the dictionary statistics directly. tblspace : Tablespace in which to create the stat tables. If none is specified, then they are created in the user's default tablespace.

Drop Stats Table

DBMS_STATS.drop_stat_table (
 ownname VARCHAR2, 
 stattab VARCHAR2);

stattab : User stat table identifier.

Gather Schema Stats (本人在测试过程中，即便指定了stattab，该过程依然更新了数据字典)

DBMS_STATS.gather_schema_stats (
 ownname VARCHAR2,
 estimate_percent NUMBER DEFAULT NULL, 
 block_sample BOOLEAN DEFAULT FALSE,
 method_opt VARCHAR2 DEFAULT 'FOR ALL COLUMNS SIZE 1',（size 1 指在该列上不创建histogram，如果该值大于1，则创建histogram）
 degree NUMBER DEFAULT NULL,
 granularity VARCHAR2 DEFAULT 'DEFAULT', 
 cascade BOOLEAN DEFAULT FALSE,
 stattab VARCHAR2 DEFAULT NULL, 
 statid VARCHAR2 DEFAULT NULL,
 options VARCHAR2 DEFAULT 'GATHER', 
 objlist OUT ObjectTab,
 statown VARCHAR2 DEFAULT NULL);

estimate_percent : Percentage of rows to estimate (NULL means compute): The valid range is [0.000001,100).

block_sample : Whether or not to use random block sampling instead of random row sampling. Random block sampling is more efficient, but if the data is not randomly distributed on disk, then the sample values may be somewhat correlated. Only pertinent when doing an estimate statistics.

method_opt : Method options of the following format (the phrase 'SIZE 1' is required to ensure gathering statistics in parallel and for use with the phrase hidden):

FOR ALL [INDEXED | HIDDEN] COLUMNS [SIZE integer]

This value is passed to all of the individual tables.

degree : Degree of parallelism (NULL means use table default value).

granularity : Granularity of statistics to collect (only pertinent if the table is partitioned).

DEFAULT: Gather global- and partition-level statistics.SUBPARTITION: Gather subpartition-level statistics.PARTITION: Gather partition-level statistics.GLOBAL: Gather global statistics.ALL: Gather all (subpartition, partition, and global) statistics.

cascade : Gather statistics on the indexes as well.

Index statistics gathering is not parallelized. Using this option is equivalent to running the gather_index_stats procedure on each of the indexes in the schema in addition to gathering table and column statistics.

stattab : User stat table identifier describing where to save the current statistics.

statid : Identifier (optional) to associate with these statistics within stattab.

options : Further specification of which objects to gather statistics for:

GATHER: Gather statistics on all objects in the schema.GATHER STALE: Gather statistics on stale objects as determined by looking at the *_tab_modifications views. Also, return a list of objects found to be stale.GATHER EMPTY: Gather statistics on objects which currently have no statistics. also, return a list of objects found to have no statistics.LIST STALE: Return list of stale objects as determined by looking at the *_tab_modifications views.LIST EMPTY: Return list of objects which currently have no statistics.

objlist : List of objects found to be stale or empty.

statown : Schema containing stattab (if different than ownname).

Export Schema Stats(从数据字典导出到用户表）

DBMS_STATS.export_schema_stats (
 ownname VARCHAR2,
 stattab VARCHAR2, 
 statid VARCHAR2 DEFAULT NULL,
 statown VARCHAR2 DEFAULT NULL);

stattab : User stat table identifier describing where to store the statistics.

statid : Identifier (optional) to associate with these statistics within stattab.

statown : Schema containing stattab (if different than ownname).

Import Schema Stats（从用户表导入到数据字典）

DBMS_STATS.import_schema_stats (
 ownname VARCHAR2,
 stattab VARCHAR2, 
 statid VARCHAR2 DEFAULT NULL,
 statown VARCHAR2 DEFAULT NULL);

stattab : User stat table identifier describing from where to retrieve the statistics.

statown : Schema containing stattab (if different than ownname).

Delete Schema Stats

DBMS_STATS.delete_schema_stats (
 ownname VARCHAR2, 
 stattab VARCHAR2 DEFAULT NULL,
 statid VARCHAR2 DEFAULT NULL,
 statown VARCHAR2 DEFAULT NULL);

stattab : User stat table identifier describing from where to delete the statistics. If stattab is NULL, then the statistics are deleted directly in the dictionary.

statid : Identifier (optional) to associate with these statistics within stattab (Only pertinent if stattab is not NULL).

statown : Schema containing stattab (if different than ownname).

Set Table Stats

DBMS_STATS.set_table_stats (
 ownname VARCHAR2, 
 tabname VARCHAR2, 
 partname VARCHAR2 DEFAULT NULL,
 stattab VARCHAR2 DEFAULT NULL, 
 statid VARCHAR2 DEFAULT NULL,
 numrows NUMBER DEFAULT NULL, 
 numblks NUMBER DEFAULT NULL,
 avgrlen NUMBER DEFAULT NULL, 
 flags NUMBER DEFAULT NULL,
 statown VARCHAR2 DEFAULT NULL);

tabname : Name of the table.

partname : Name of the table partition in which to store the statistics. If the table is partitioned and partname is NULL, then the statistics are stored at the global table level.

stattab : User stat table identifier describing where to store the statistics. If stattab is NULL, then the statistics are stored directly in the dictionary.

statid : Identifier (optional) to associate with these statistics within stattab (Only pertinent if stattab is not NULL).

numrows : Number of rows in the table (partition).

numblks : Number of blocks the table (partition) occupies.

avgrlen : Average row length for the table (partition).

flags : For internal Oracle use (should be left as NULL).

statown : Schema containing stattab (if different than ownname).

Get Table Stats

DBMS_STATS.get_table_stats (
 ownname VARCHAR2, 
 tabname VARCHAR2, 
 partname VARCHAR2 DEFAULT NULL,
 stattab VARCHAR2 DEFAULT NULL, 
 statid VARCHAR2 DEFAULT NULL,
 numrows OUT NUMBER, 
 numblks OUT NUMBER,
 avgrlen OUT NUMBER,
 statown VARCHAR2 DEFAULT NULL);

tabname : Name of the table to which this column belongs.

partname : Name of the table partition from which to get the statistics. If the table is partitioned and if partname is NULL, then the statistics are retrieved from the global table level.

stattab : User stat table identifier describing from where to retrieve the statistics. If stattab is NULL, then the statistics are retrieved directly from the dictionary.

statid : Identifier (optional) to associate with these statistics within stattab (Only pertinent if stattab is not NULL).

numrows : Number of rows in the table (partition).

numblks : Number of blocks the table (partition) occupies.

avgrlen : Average row length for the table (partition).

statown : Schema containing stattab (if different than ownname).

Get Index Stats

DBMS_STATS.GET_INDEX_STATS (
 ownname VARCHAR2, 
 indname VARCHAR2,
 partname VARCHAR2 DEFAULT NULL,
 stattab VARCHAR2 DEFAULT NULL, 
 statid VARCHAR2 DEFAULT NULL,
 numrows OUT NUMBER, 
 numlblks OUT NUMBER,
 numdist OUT NUMBER, 
 avglblk OUT NUMBER,
 avgdblk OUT NUMBER, 
 clstfct OUT NUMBER,
 indlevel OUT NUMBER,
 statown VARCHAR2 DEFAULT NULL);

indname : Name of the index.

partname : Name of the index partition for which to get the statistics. If the index is partitioned and if partname is NULL, then the statistics are retrieved for the global index level.

statid : Identifier (optional) to associate with these statistics within stattab (Only pertinent if stattab is not NULL).

numrows : Number of rows in the index (partition).

numlblks : Number of leaf blocks in the index (partition).

numdist : Number of distinct keys in the index (partition).

avglblk : Average integral number of leaf blocks in which each distinct key appears for this index (partition).

avgdblk : Average integral number of data blocks in the table pointed to by a distinct key for this index (partition).

clstfct : Clustering factor for the index (partition).

indlevel : Height of the index (partition).

statown : Schema containing stattab (if different than ownname).

Automated table monitoring and stale statistics gathering example

在oracle10g中 statistics_level 初始化参数作为一个全局设置影响对表的监控操作，本文下面涉及的alter_schema_tab_monitoring已经不再被使用，但是到我们调用这些过程时，不会报错，只是没有任何事情发生。

You can automatically gather statistics or create lists of tables that have stale or no statistics.

To automatically gather statistics, run the DBMS_STATS.GATHER_SCHEMA_STATS and DBMS_STATS.GATHER_DATABASE_STATS procedures with the OPTIONS and objlist parameters. Use the following values for the options parameter:

GATHER : Gathers statistics on all tables. (default)

GATHER EMPTY : Gathers statistics only on tables without statistics.

LIST STALE : Creates a list of tables with stale statistics.（通过*_tab_modifications视图）

LIST EMPTY : Creates a list of tables that do not have statistics.

The objlist parameter identifies an output parameter for the LIST STALE and LIST EMPTY options. The objlist parameter is of type DBMS_STATS.OBJECTTAB.

Step 1 : Perform a quick analyze to load in base statistics

BEGIN
DBMS_STATS.GATHER_SCHEMA_STATS (
 ownname => 'scott',
 estimate_percent => null, -- Small table, lets compute
 block_sample => false,
 method_opt => 'FOR ALL COLUMNS',
 degree => null, -- No parallelism used in this example
 granularity => 'ALL',
 cascade => true, -- Make sure we include indexes
 options => 'GATHER' -- Gather mode
 );
END;
/

PL/SQL procedure successfully completed.

Step 2 : Examine the current statistics

SELECT table_name, num_rows, blocks, avg_row_len 
FROM user_tables
WHERE table_name='EMP';

TABLE_NAME NUM_ROWS BLOCKS AVG_ROW_LEN
------------------------------ ---------- ---------- -----------
EMP 1500 28 92

Step 3 : Turn on Automatic Monitoring

Now turn on automatic monitoring for the emp table. This can be done using the alter table method. Starting with Oracle 9i, you can also perform this at the "schema", and "entire database" level. I provide the syntax for all three methods below.

通过alter table 语句我们可以设置oracle数据库自动监控某张表的变化，从9i开始，我们还可以在schema或者数据库级别设置是否监控数据变化，监控结构会存储在*_tab_modifications视图中。

Monitor only the EMP table.

alter table emp monitoring;

Table altered.

BEGIN
 DBMS_STATS.alter_schema_tab_monitoring('scott', true);
END;
/

PL/SQL procedure successfully completed.

Note: Although the option to collect statistics for SYS tables is available via ALTER_DATABASE_TAB_MONITORING, Oracle continues to recommend against this practice until the next major release after 9i Release 2. Also note that the ALTER_DATABASE_TAB_MONITORING procedure in the DBMS_STATS package only monitors tables; there is an ALTER INDEX...MONITORING statement which can be used to monitor indexes. Thanks to Nabil Nawaz for providing this and pointing out an error I made in the previous version of this article.

BEGIN
 DBMS_STATS.alter_database_tab_monitoring (
 monitoring => true,
 sysobjs => false); -- Don't set to true, see note above.
END;
/

PL/SQL procedure successfully completed.

Step 4 : Verify that monitoring is turned on.

Note: The results of the following query are from running the alter table ... statement on the emp table only.

可以通过*_tables视图的monitoring字段来判断某张表是否开启了自动监控

SELECT table_name, monitoring
FROM user_tables
ORDER BY monitoring;

TABLE_NAME MONITORING
------------------------------ ----------
DEPT NO
EMP YES

Step 5 : Delete some rows from the database.

SQL> DELETE FROM emp WHERE rownum < 501;

500 rows deleted.

SQL> commit;

Commit complete.

Step 6 : Wait until the monitered data is flushed.

exec dbms_stats.flush_database_monitoring_info;

PL/SQL procedure successfully completed.

Step 7 : Check for what it has collected.

SELECT * FROM user_tab_modifications;

TABLE_NAME PARTITION_NAME SUBPARTITION_NAME INSERTS UPDATES DELETES TIMESTAMP TRUNCATED
---------- -------------- ----------------- ------- ------- ------- --------- ---------
EMP 0 0 500 18-SEP-02 NO

Step 8 : Execute DBMS_STATS to gather stats on all "stale" tables.

BEGIN
 DBMS_STATS.GATHER_SCHEMA_STATS(
 ownname => 'scott',
 estimate_percent => null,
 block_sample => false,
 method_opt => 'FOR ALL COLUMNS',
 degree => null,
 granularity => 'ALL',
 cascade => true,
 options => 'GATHER STALE');
END;
/

PL/SQL procedure successfully completed.

Step 9 : Verify that the table is no longer listed in USER_TAB_MODIFICATIONS.

SQL> SELECT * FROM user_tab_modifications;

no rows selected.

Step 10 : Examine some of new statistics collected.

SELECT table_name, num_rows, blocks, avg_row_len 
FROM user_tables where table_name='EMP';

TABLE_NAME NUM_ROWS BLOCKS AVG_ROW_LEN
------------------------------ ---------- ---------- -----------
EMP 1000 28 92

How to determine if dictionary statistics are RDBMS-generated or user-defined

This is crucial for development environments that are testing the performance of SQL statements with various sets of statistics. The DBA will need to know if the relying statistics are RDBMS-defined or user-defined.

RDBMS-generated statistics are generated by the following:（我们可以通过如下方式生成统计信息）

The column USER_STATS from DBA_TABLES, ALL_TABLES, USER_TABLES displays: