Introduction to Statistics

TiDB uses statistics as input to the optimizer to estimate the number of rows processed in each plan step for a SQL statement. The optimizer estimates the cost of each available plan choice, including index accesses and the sequence of table joins, and produces a cost for each available plan. The optimizer then picks the execution plan with the lowest overall cost.

Collect statistics

Automatic update

For the INSERT, DELETE, or UPDATE statements, TiDB automatically updates the number of rows and modified rows in statistics.

TiDB persists the update information regularly and the update cycle is 20 * stats-lease. The default value of stats-lease is 3s. If you specify the value as 0, TiDB stops updating statistics automatically.

TiDB persists the update information every 60 seconds.

Based upon the number of changes to a table, TiDB will automatically schedule ANALYZE to collect statistics on those tables. This is controlled by the tidb_enable_auto_anlyze system variable and the following tidb_auto_analyze% variables.

When the ratio of the number of modified rows to the total number of rows of tbl in a table is greater than tidb_auto_analyze_ratio, and the current time is between tidb_auto_analyze_start_time and tidb_auto_analyze_end_time, TiDB executes the ANALYZE TABLE tbl statement in the background to automatically update the statistics on this table.

To avoid the situation that modifying data on a small table frequently triggers the automatic update, when a table has less than 1000 rows, modifications do not trigger the automatic update in TiDB. You can use the SHOW STATS_META statement to view the number of rows in a table.

Manual collection

Currently, TiDB collects statistical information as a full collection. You can execute the ANALYZE TABLE statement to collect statistics.

You can perform full collection using the following syntax.

• To collect statistics of all the tables in TableNameList:

  ANALYZE TABLE TableNameList [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];
  

• WITH NUM BUCKETS specifies the maximum number of buckets in the generated histogram.

• WITH NUM TOPN specifies the maximum number of the generated TOPNs.

• WITH NUM CMSKETCH DEPTH specifies the depth of the CM Sketch.

• WITH NUM CMSKETCH WIDTH specifies the width of the CM Sketch.

• WITH NUM SAMPLES specifies the number of samples.

• WITH FLOAT_NUM SAMPLERATE specifies the sampling rate.

WITH NUM SAMPLES and WITH FLOAT_NUM SAMPLERATE correspond to two different algorithms of collecting samples.

See Histograms, Top-N and CMSketch (Count-Min Sketch) for detailed explanations. For SAMPLES/SAMPLERATE, see Improve collection performance.

For information on persisting the options for easier reuse, see Persist ANALYZE configurations.

Types of statistics

Histogram

Histogram statistics are used by the optimizer to estimate selectivity of an interval or range predicate, and might also be used to determine the number of distinct values within a column for estimation of equal/IN predicates in Version 2 of statistics (refer to Versions of Statistics).

A histogram is an approximate representation of the distribution of data. It divides the entire range of values into a series of buckets, and uses simple data to describe each bucket, such as the number of values ​​falling in the bucket. In TiDB, an equal-depth histogram is created for the specific columns of each table. The equal-depth histogram can be used to estimate the interval query.

Here "equal-depth" means that the number of values ​​falling into each bucket is as equal as possible. For example, for a given set {1.6, 1.9, 1.9, 2.0, 2.4, 2.6, 2.7, 2.7, 2.8, 2.9, 3.4, 3.5}, you want to generate 4 buckets. The equal-depth histogram is as follows. It contains four buckets [1.6, 1.9], [2.0, 2.6], [2.7, 2.8], [2.9, 3.5]. The bucket depth is 3.

For details about the parameter that determines the upper limit to the number of histogram buckets, refer to Manual Collection. When the number of buckets is larger, the accuracy of the histogram is higher; however, higher accuracy is at the cost of the usage of memory resources. You can adjust this number appropriately according to the actual scenario.

Count-Min Sketch

Count-Min Sketch is a hash structure. When an equivalence query contains a = 1 or IN query (for example, a IN (1, 2, 3)), TiDB uses this data structure for estimation.

A hash collision might occur since Count-Min Sketch is a hash structure. In the EXPLAIN statement, if the estimate of the equivalent query deviates greatly from the actual value, it can be considered that a larger value and a smaller value have been hashed together. In this case, you can take one of the following ways to avoid the hash collision:

• Modify the WITH NUM TOPN parameter. TiDB stores the high-frequency (top x) data separately, with the other data stored in Count-Min Sketch. Therefore, to prevent a larger value and a smaller value from being hashed together, you can increase the value of WITH NUM TOPN. In TiDB, its default value is 20. The maximum value is 1024. For more information about this parameter, see Manual collection.
• Modify two parameters WITH NUM CMSKETCH DEPTH and WITH NUM CMSKETCH WIDTH. Both affect the number of hash buckets and the collision probability. You can increase the values of the two parameters appropriately according to the actual scenario to reduce the probability of hash collision, but at the cost of higher memory usage of statistics. In TiDB, the default value of WITH NUM CMSKETCH DEPTH is 5, and the default value of WITH NUM CMSKETCH WIDTH is 2048. For more information about the two parameters, see Manual collection.

Top-N values

Top-N values are values with the top N occurrences in a column or index. Top-N statistics are often referred to as frequency statistics or data skew.

TiDB records the values and occurrences of Top-N values. The default value is 20, meaning the top 20 most frequent values are collected. The maximum value is 1024. For details about the parameter that determines the number of values collected, see Manual collection.

Selective statistics collection

Collect statistics on indexes

To collect statistics on all indexes in IndexNameList in TableName, use the following syntax:

ANALYZE TABLE TableName INDEX [IndexNameList] [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];

When IndexNameList is empty, this syntax collects statistics on all indexes in TableName.

Collect statistics on some columns

In most cases, the optimizer only uses statistics on columns in the WHERE, JOIN, ORDER BY, and GROUP BY statements. These columns can be referred to as PREDICATE COLUMNS.

If a table has many columns, collecting statistics on all the columns can cause a large overhead. To reduce the overhead, you can collect statistics on only specific columns (that you choose) or PREDICATE COLUMNS to be used by the optimizer. To persist the column list of any subset of columns for reuse in future, see Persist column configurations.

• To collect statistics on specific columns, use the following syntax:

  ANALYZE TABLE TableName COLUMNS ColumnNameList [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];
  

  In the syntax, ColumnNameList specifies the name list of the target columns. If you need to specify more than one column, use comma , to separate the column names. For example, ANALYZE table t columns a, b. Besides collecting statistics on the specific columns in a specific table, this syntax collects statistics on the indexed columns and all indexes in that table at the same time.

• To collect statistics on PREDICATE COLUMNS, do the following:

  Warning

  Currently, collecting statistics on PREDICATE COLUMNS is an experimental feature. It is not recommended that you use it in production environments.

  1. Set the value of the tidb_enable_column_tracking system variable to ON to enable TiDB to collect PREDICATE COLUMNS.

     After the setting, TiDB writes the PREDICATE COLUMNS information to the mysql.column_stats_usage system table every 100 * stats-lease.

     After the setting, TiDB writes the PREDICATE COLUMNS information to the mysql.column_stats_usage system table every 300 seconds.

  2. After the query pattern of your business is relatively stable, collect statistics on PREDICATE COLUMNS by using the following syntax:

     ANALYZE TABLE TableName PREDICATE COLUMNS [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];
     

     Besides collecting statistics on PREDICATE COLUMNS in a specific table, this syntax collects statistics on indexed columns and all indexes in that table at the same time.

     Note

     • If the mysql.column_stats_usage system table does not contain any PREDICATE COLUMNS recorded for that table, the preceding syntax collects statistics on all columns and all indexes in that table.
     • Any columns excluded from collection (either by manually listing columns or using PREDICATE COLUMNS) will not have their statistics overwritten. When executing a new type of SQL query, the optimizer will use the old statistics for such columns if it exists or pseudo column statistics if columns never had statistics collected. The next ANALYZE using PREDICATE COLUMNS will collect the statistics on those columns.

• To collect statistics on all columns and indexes, use the following syntax:

  ANALYZE TABLE TableName ALL COLUMNS [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];
  

Collect statistics on partitions

• To collect statistics on all partitions in PartitionNameList in TableName, use the following syntax:

  ANALYZE TABLE TableName PARTITION PartitionNameList [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];
  

• To collect index statistics on all partitions in PartitionNameList in TableName, use the following syntax:

  ANALYZE TABLE TableName PARTITION PartitionNameList INDEX [IndexNameList] [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];
  

• If you only need to collect statistics on some columns of some partitions in a table, use the following syntax:

  Warning

  Currently, collecting statistics on PREDICATE COLUMNS is an experimental feature. It is not recommended that you use it in production environments.

  ANALYZE TABLE TableName PARTITION PartitionNameList [COLUMNS ColumnNameList|PREDICATE COLUMNS|ALL COLUMNS] [WITH NUM BUCKETS|TOPN|CMSKETCH DEPTH|CMSKETCH WIDTH]|[WITH NUM SAMPLES|WITH FLOATNUM SAMPLERATE];
  

Collect statistics of partitioned tables in dynamic pruning mode

When accessing partitioned tables in dynamic pruning mode (which is the default since v6.3.0), TiDB collects table-level statistics, which is called GlobalStats. Currently, GlobalStats is aggregated from statistics of all partitions. In dynamic pruning mode, a statistics update of any partitioned table can trigger the GlobalStats to be updated.

If partitions are empty, or columns for some partitions are missing, then the collection behavior is controlled by the tidb_skip_missing_partition_stats variable:

• When GlobalStats update is triggered and tidb_skip_missing_partition_stats is OFF:

  • If some partitions have no statistics (such as a new partition that has never been analyzed), GlobalStats generation is interrupted and a warning message is displayed saying that no statistics are available on partitions.

  • If statistics of some columns are absent in specific partitions (different columns are specified for analyzing in these partitions), GlobalStats generation is interrupted when statistics of these columns are aggregated, and a warning message is displayed saying that statistics of some columns are absent in specific partitions.

• When GlobalStats update is triggered and tidb_skip_missing_partition_stats is ON:

  • If statistics of all or some columns are missing for some partitions, TiDB skips these missing partition statistics when generating GlobalStats so the generation of GlobalStats is not affected.

In dynamic pruning mode, the Analyze configurations of partitions and tables should be the same. Therefore, if you specify the COLUMNS configuration following the ANALYZE TABLE TableName PARTITION PartitionNameList statement or the OPTIONS configuration following WITH, TiDB will ignore them and return a warning.

Improve collection performance

TiDB provides two options to improve the performance of statistics collection:

• Collecting statistics on a subset of the columns. See Collecting statistics on some columns.
• Sampling.

Statistics sampling

Sampling is available via two separate options of the ANALYZE statement - with each corresponding to a different collection algorithm:

• WITH NUM SAMPLES specifies the size of the sampling set, which is implemented in the reservoir sampling method in TiDB. When a table is large, it is not recommended to use this method to collect statistics. Because the intermediate result set of the reservoir sampling contains redundant results, it causes additional pressure on resources such as memory.
• WITH FLOAT_NUM SAMPLERATE is a sampling method introduced in v5.3.0. With the value range (0, 1], this parameter specifies the sampling rate. It is implemented in the way of Bernoulli sampling in TiDB, which is more suitable for sampling larger tables and performs better in collection efficiency and resource usage.

Before v5.3.0, TiDB uses the reservoir sampling method to collect statistics. Since v5.3.0, the TiDB Version 2 statistics uses the Bernoulli sampling method to collect statistics by default. To re-use the reservoir sampling method, you can use the WITH NUM SAMPLES statement.

The current sampling rate is calculated based on an adaptive algorithm. When you can observe the number of rows in a table using SHOW STATS_META, you can use this number of rows to calculate the sampling rate corresponding to 100,000 rows. If you cannot observe this number, you can use the TABLE_KEYS column in the TABLE_STORAGE_STATS table as another reference to calculate the sampling rate.

The memory quota for collecting statistics

Since TiDB v6.1.0, you can use the system variable tidb_mem_quota_analyze to control the memory quota for collecting statistics in TiDB.

To set a proper value of tidb_mem_quota_analyze, consider the data size of the cluster. When the default sampling rate is used, the main considerations are the number of columns, the size of column values, and the memory configuration of TiDB. Consider the following suggestions when you configure the maximum and minimum values:

Persist ANALYZE configurations

Since v5.4.0, TiDB supports persisting some ANALYZE configurations. With this feature, the existing configurations can be easily reused for future statistics collection.

The following are the ANALYZE configurations that support persistence:

Enable ANALYZE configuration persistence

The ANALYZE configuration persistence feature is enabled by default (the system variable tidb_analyze_version is 2 and tidb_persist_analyze_options is ON by default).

The ANALYZE configuration persistence feature is disabled by default. To enable the feature, ensure that the system variable tidb_persist_analyze_options is ON and set the system variable tidb_analyze_version to 2.

You can use this feature to record the persistence configurations specified in the ANALYZE statement when executing the statement manually. Once recorded, the next time TiDB automatically updates statistics or you manually collect statistics without specifying these configuration, TiDB will collect statistics according to the recorded configurations.

To query the configuration persisted on a specific table used for auto analyze operations, you can use the following SQL statement:

SELECT sample_num, sample_rate, buckets, topn, column_choice, column_ids FROM mysql.analyze_options opt JOIN information_schema.tables tbl ON opt.table_id = tbl.tidb_table_id WHERE tbl.table_schema = '{db_name}' AND tbl.table_name = '{table_name}';

TiDB will overwrite the previously recorded persistent configuration using the new configurations specified by the latest ANALYZE statement. For example, if you run ANALYZE TABLE t WITH 200 TOPN;, it will set the top 200 values in the ANALYZE statement. Subsequently, executing ANALYZE TABLE t WITH 0.1 SAMPLERATE; will set both the top 200 values and a sampling rate of 0.1 for auto ANALYZE statements, similar to ANALYZE TABLE t WITH 200 TOPN, 0.1 SAMPLERATE;.

Disable ANALYZE configuration persistence

To disable the ANALYZE configuration persistence feature, set the tidb_persist_analyze_options system variable to OFF. Because the ANALYZE configuration persistence feature is not applicable to tidb_analyze_version = 1, setting tidb_analyze_version = 1 can also disable the feature.

After disabling the ANALYZE configuration persistence feature, TiDB does not clear the persisted configuration records. Therefore, if you enable this feature again, TiDB continues to collect statistics using the previously recorded persistent configurations.

Persist column configurations

If you want to persist the column configuration in the ANALYZE statement (including COLUMNS ColumnNameList, PREDICATE COLUMNS, and ALL COLUMNS), set the value of the tidb_persist_analyze_options system variable to ON to enable the ANALYZE configuration persistence feature. After enabling the ANALYZE configuration persistence feature:

• When TiDB collects statistics automatically or when you manually collect statistics by executing the ANALYZE statement without specifying the column configuration, TiDB continues using the previously persisted configuration for statistics collection.
• When you manually execute the ANALYZE statement multiple times with column configuration specified, TiDB overwrites the previously recorded persistent configuration using the new configuration specified by the latest ANALYZE statement.

To locate PREDICATE COLUMNS and columns on which statistics have been collected, use the following syntax:

SHOW COLUMN_STATS_USAGE [ShowLikeOrWhere];

The SHOW COLUMN_STATS_USAGE statement returns the following 6 columns:

In the following example, after executing ANALYZE TABLE t PREDICATE COLUMNS;, TiDB collects statistics on columns b, c, and d, where column b is a PREDICATE COLUMN and columns c and d are index columns.

SET GLOBAL tidb_enable_column_tracking = ON;
Query OK, 0 rows affected (0.00 sec)

CREATE TABLE t (a INT, b INT, c INT, d INT, INDEX idx_c_d(c, d));
Query OK, 0 rows affected (0.00 sec)

-- The optimizer uses the statistics on column b in this query.
SELECT * FROM t WHERE b > 1;
Empty set (0.00 sec)

-- After waiting for a period of time (100 * stats-lease), TiDB writes the collected `PREDICATE COLUMNS` to mysql.column_stats_usage.
-- Specify `last_used_at IS NOT NULL` to show the `PREDICATE COLUMNS` collected by TiDB.
SHOW COLUMN_STATS_USAGE
WHERE db_name = 'test' AND table_name = 't' AND last_used_at IS NOT NULL;
+---------+------------+----------------+-------------+---------------------+------------------+
| Db_name | Table_name | Partition_name | Column_name | Last_used_at        | Last_analyzed_at |
+---------+------------+----------------+-------------+---------------------+------------------+
| test    | t          |                | b           | 2022-01-05 17:21:33 | NULL             |
+---------+------------+----------------+-------------+---------------------+------------------+
1 row in set (0.00 sec)

ANALYZE TABLE t PREDICATE COLUMNS;
Query OK, 0 rows affected, 1 warning (0.03 sec)

-- Specify `last_analyzed_at IS NOT NULL` to show the columns for which statistics have been collected.
SHOW COLUMN_STATS_USAGE
WHERE db_name = 'test' AND table_name = 't' AND last_analyzed_at IS NOT NULL;
+---------+------------+----------------+-------------+---------------------+---------------------+
| Db_name | Table_name | Partition_name | Column_name | Last_used_at        | Last_analyzed_at    |
+---------+------------+----------------+-------------+---------------------+---------------------+
| test    | t          |                | b           | 2022-01-05 17:21:33 | 2022-01-05 17:23:06 |
| test    | t          |                | c           | NULL                | 2022-01-05 17:23:06 |
| test    | t          |                | d           | NULL                | 2022-01-05 17:23:06 |
+---------+------------+----------------+-------------+---------------------+---------------------+
3 rows in set (0.00 sec)

Versions of statistics

The tidb_analyze_version variable controls the statistics collected by TiDB. Currently, two versions of statistics are supported: tidb_analyze_version = 1 and tidb_analyze_version = 2.

• For TiDB Self-Hosted, the default value of this variable changes from 1 to 2 starting from v5.3.0.
• For TiDB Cloud, the default value of this variable changes from 1 to 2 starting from v6.5.0.
• If your cluster is upgraded from an earlier version, the default value of tidb_analyze_version does not change after the upgrade.

Version 2 is preferred, and will continue to be enhanced to ultimately replace Version 1 completely. Compared to Version 1, Version 2 improves the accuracy of many of the statistics collected for larger data volumes. Version 2 also improves collection performance by removing the need to collect Count-Min sketch statistics for predicate selectivity estimation, and also supporting automated collection only on selected columns (see Collecting statistics on some columns).

The following table lists the information collected by each version for usage in the optimizer estimates:

Switch between statistics versions

It is recommended to ensure that all tables/indexes (and partitions) utilize statistics collection from the same version. Version 2 is recommended, however, it is not recommended to switch from one version to another without a justifiable reason such as an issue experienced with the version in use. A switch between versions might take a period of time when no statistics are available until all tables have been analyzed with the new version, which might negatively affect the optimizer plan choices if statistics are not available.

Examples of justifications to switch might include - with Version 1, there could be inaccuracies in equal/IN predicate estimation due to hash collisions when collecting Count-Min sketch statistics. Solutions are listed in the Count-Min Sketch section. Alternatively, setting tidb_analyze_version = 2 and rerunning ANALYZE on all objects is also a solution. In the early release of Version 2, there was a risk of memory overflow after ANALYZE. This issue is resolved, but initially, one solution was to set tidb_analyze_version = 1 and rerun ANALYZE on all objects.

To prepare ANALYZE for switching between versions:

• If the ANALYZE statement is executed manually, manually analyze every table to be analyzed.

  SELECT DISTINCT(CONCAT('ANALYZE TABLE ', table_schema, '.', table_name, ';'))
  FROM information_schema.tables JOIN mysql.stats_histograms
  ON table_id = tidb_table_id
  WHERE stats_ver = 2;
  

• If TiDB automatically executes the ANALYZE statement because the auto-analysis has been enabled, execute the following statement that generates the DROP STATS statement:

  SELECT DISTINCT(CONCAT('DROP STATS ', table_schema, '.', table_name, ';'))
  FROM information_schema.tables ON mysql.stats_histograms
  ON table_id = tidb_table_id
  WHERE stats_ver = 2;
  

• If the result of the preceding statement is too long to copy and paste, you can export the result to a temporary text file and then perform execution from the file like this:

  SELECT DISTINCT ... INTO OUTFILE '/tmp/sql.txt';
  mysql -h ${TiDB_IP} -u user -P ${TIDB_PORT} ... < '/tmp/sql.txt'
  

View statistics

You can view the ANALYZE status and statistics information using the following statements.

ANALYZE state

When executing the ANALYZE statement, you can view the current state of ANALYZE using the following SQL statement:

SHOW ANALYZE STATUS [ShowLikeOrWhere]

This statement returns the state of ANALYZE. You can use ShowLikeOrWhere to filter the information you need.

Currently, the SHOW ANALYZE STATUS statement returns the following 11 columns:

Starting from TiDB v6.1.0, the SHOW ANALYZE STATUS statement supports showing cluster-level tasks. Even after a TiDB restart, you can still view task records before the restart using this statement. Before TiDB v6.1.0, the SHOW ANALYZE STATUS statement can only show instance-level tasks, and task records are cleared after a TiDB restart.

SHOW ANALYZE STATUS shows the most recent task records only. Starting from TiDB v6.1.0, you can view the history tasks within the last 7 days through the system table mysql.analyze_jobs.

When tidb_mem_quota_analyze is set and an automatic ANALYZE task running in the TiDB background uses more memory than this threshold, the task will be retried. You can see failed and retried tasks in the output of the SHOW ANALYZE STATUS statement.

When tidb_max_auto_analyze_time is greater than 0 and an automatic ANALYZE task running in the TiDB background takes more time than this threshold, the task will be terminated.

mysql> SHOW ANALYZE STATUS [ShowLikeOrWhere];
+--------------+------------+----------------+-------------------------------------------------------------------------------------------+----------------+---------------------+---------------------+----------+-------------------------------------------------------------------------------|
| Table_schema | Table_name | Partition_name | Job_info                                                                                  | Processed_rows | Start_time          | End_time            | State    | Fail_reason                                                                   |
+--------------+------------+----------------+-------------------------------------------------------------------------------------------+----------------+---------------------+---------------------+----------+-------------------------------------------------------------------------------|
| test         | sbtest1    |                | retry auto analyze table all columns with 100 topn, 0.055 samplerate                      |        2000000 | 2022-05-07 16:41:09 | 2022-05-07 16:41:20 | finished | NULL                                                                          |
| test         | sbtest1    |                | auto analyze table all columns with 100 topn, 0.5 samplerate                              |              0 | 2022-05-07 16:40:50 | 2022-05-07 16:41:09 | failed   | analyze panic due to memory quota exceeds, please try with smaller samplerate |

Metadata of tables

You can use the SHOW STATS_META statement to view the total number of rows and the number of updated rows.

SHOW STATS_META [ShowLikeOrWhere];

The syntax of ShowLikeOrWhereOpt is as follows:

Currently, the SHOW STATS_META statement returns the following 6 columns:

Health state of tables

You can use the SHOW STATS_HEALTHY statement to check the health state of tables and roughly estimate the accuracy of the statistics. When modify_count >= row_count, the health state is 0; when modify_count < row_count, the health state is (1 - modify_count/row_count) * 100.

The syntax is as follows:

SHOW STATS_HEALTHY [ShowLikeOrWhere];

The synopsis of SHOW STATS_HEALTHY is:

Currently, the SHOW STATS_HEALTHY statement returns the following 4 columns:

Metadata of columns

You can use the SHOW STATS_HISTOGRAMS statement to view the number of different values and the number of NULL in all the columns.

Syntax as follows:

SHOW STATS_HISTOGRAMS [ShowLikeOrWhere]

This statement returns the number of different values and the number of NULL in all the columns. You can use ShowLikeOrWhere to filter the information you need.

Currently, the SHOW STATS_HISTOGRAMS statement returns the following 10 columns:

Buckets of histogram

You can use the SHOW STATS_BUCKETS statement to view each bucket of the histogram.

The syntax is as follows:

SHOW STATS_BUCKETS [ShowLikeOrWhere]

The diagram is as follows:

This statement returns information about all the buckets. You can use ShowLikeOrWhere to filter the information you need.

Currently, the SHOW STATS_BUCKETS statement returns the following 11 columns:

Top-N information

You can use the SHOW STATS_TOPN statement to view the Top-N information currently collected by TiDB.

The syntax is as follows:

SHOW STATS_TOPN [ShowLikeOrWhere];

Currently, the SHOW STATS_TOPN statement returns the following 7 columns:

Delete statistics

You can run the DROP STATS statement to delete statistics.

DROP STATS TableName

The preceding statement deletes all statistics of TableName. If a partitioned table is specified, this statement will delete statistics of all partitions in this table as well as GlobalStats generated in dynamic pruning mode.

DROP STATS TableName PARTITION PartitionNameList;

This preceding statement only deletes statistics of the specified partitions in PartitionNameList.

DROP STATS TableName GLOBAL;

The preceding statement only deletes GlobalStats generated in dynamic pruning mode of the specified table.

Load statistics

By default, depending on the size of column statistics, TiDB loads statistics differently as follows:

• For statistics that consume small amounts of memory (such as count, distinctCount, and nullCount), as long as the column data is updated, TiDB automatically loads the corresponding statistics into memory for use in the SQL optimization stage.
• For statistics that consume large amounts of memory (such as histograms, TopN, and Count-Min Sketch), to ensure the performance of SQL execution, TiDB loads the statistics asynchronously on demand. Take histograms as an example. TiDB loads histogram statistics on a column into memory only when the optimizer uses the histogram statistics on that column. On-demand asynchronous statistics loading does not affect the performance of SQL execution but might provide incomplete statistics for SQL optimization.

Since v5.4.0, TiDB introduces the synchronously loading statistics feature. This feature allows TiDB to synchronously load large-sized statistics (such as histograms, TopN, and Count-Min Sketch statistics) into memory when you execute SQL statements, which improves the completeness of statistics for SQL optimization.

To enable this feature, set the value of the tidb_stats_load_sync_wait system variable to a timeout (in milliseconds) that SQL optimization can wait for at most to synchronously load complete column statistics. The default value of this variable is 100, indicating that the feature is enabled.

After enabling the synchronously loading statistics feature, you can further configure the feature as follows:

• To control how TiDB behaves when the waiting time of SQL optimization reaches the timeout, modify the value of the tidb_stats_load_pseudo_timeout system variable. The default value of this variable is ON, indicating that after the timeout, the SQL optimization process does not use any histogram, TopN, or CMSketch statistics on any columns. If this variable is set to OFF, after the timeout, SQL execution fails.
• To specify the maximum number of columns that the synchronously loading statistics feature can process concurrently, modify the value of the stats-load-concurrency option in the TiDB configuration file. The default value is 5.
• To specify the maximum number of column requests that the synchronously loading statistics feature can cache, modify the value of the stats-load-queue-size option in the TiDB configuration file. The default value is 1000.

During TiDB startup, SQL statements executed before the initial statistics are fully loaded might have suboptimal execution plans, thus causing performance issues. To avoid such issues, TiDB v7.1.0 introduces the configuration parameter force-init-stats. With this option, you can control whether TiDB provides services only after statistics initialization has been finished during startup. Starting from v7.2.0, this parameter is enabled by default.

Starting from v7.1.0, TiDB introduces lite-init-stats for lightweight statistics initialization.

• When the value of lite-init-stats is true, statistics initialization does not load any histogram, TopN, or Count-Min Sketch of indexes or columns into memory.
• When the value of lite-init-stats is false, statistics initialization loads histograms, TopN, and Count-Min Sketch of indexes and primary keys into memory but does not load any histogram, TopN, or Count-Min Sketch of non-primary key columns into memory. When the optimizer needs the histogram, TopN, and Count-Min Sketch of a specific index or column, the necessary statistics are loaded into memory synchronously or asynchronously.

The default value of lite-init-stats is true, which means to enable lightweight statistics initialization. Setting lite-init-stats to true speeds up statistics initialization and reduces TiDB memory usage by avoiding unnecessary statistics loading.

After enabling the synchronously loading statistics feature, you can control how TiDB behaves when the waiting time of SQL optimization reaches the timeout by modifying the value of the tidb_stats_load_pseudo_timeout system variable. The default value of this variable is ON, indicating that after the timeout, the SQL optimization process does not use any histogram, TopN, or CMSketch statistics on any columns. If this variable is set to OFF, after the timeout, SQL execution fails.

Import and export statistics

Export statistics

The interface to export statistics is as follows:

• To obtain the JSON format statistics of the ${table_name} table in the ${db_name} database:

  http://${tidb-server-ip}:${tidb-server-status-port}/stats/dump/${db_name}/${table_name}
  

  For example:

  curl -s http://127.0.0.1:10080/stats/dump/test/t1 -o /tmp/t1.json
  

• To obtain the JSON format statistics of the ${table_name} table in the ${db_name} database at specific time:

  http://${tidb-server-ip}:${tidb-server-status-port}/stats/dump/${db_name}/${table_name}/${yyyyMMddHHmmss}
  

Import statistics

Generally, the imported statistics refer to the JSON file obtained using the export interface.

Syntax:

LOAD STATS 'file_name'

file_name is the file name of the statistics to be imported.

Lock statistics

Starting from v6.5.0, TiDB supports locking statistics. After the statistics of a table or a partition are locked, the statistics of the table cannot be modified and the ANALYZE statement cannot be executed on the table. For example:

Create table t, and insert data into it. When the statistics of table t are not locked, the ANALYZE statement can be successfully executed.

mysql> CREATE TABLE t(a INT, b INT);
Query OK, 0 rows affected (0.03 sec)

mysql> INSERT INTO t VALUES (1,2), (3,4), (5,6), (7,8);
Query OK, 4 rows affected (0.00 sec)
Records: 4  Duplicates: 0  Warnings: 0

mysql> ANALYZE TABLE t;
Query OK, 0 rows affected, 1 warning (0.02 sec)

mysql> SHOW WARNINGS;
+-------+------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Level | Code | Message                                                                                                                                                                                                               |
+-------+------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Note  | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t, reason to use this rate is "Row count in stats_meta is much smaller compared with the row count got by PD, use min(1, 15000/4) as the sample-rate=1" |
+-------+------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.00 sec)

Lock the statistics of table t and execute ANALYZE. The warning message shows that the ANALYZE statement has skipped table t.

mysql> LOCK STATS t;
Query OK, 0 rows affected (0.00 sec)

mysql> SHOW STATS_LOCKED;
+---------+------------+----------------+--------+
| Db_name | Table_name | Partition_name | Status |
+---------+------------+----------------+--------+
| test    | t          |                | locked |
+---------+------------+----------------+--------+
1 row in set (0.01 sec)

mysql> ANALYZE TABLE t;
Query OK, 0 rows affected, 2 warnings (0.00 sec)

mysql> SHOW WARNINGS;
+---------+------+-----------------------------------------------------------------------------------------------------------------------------------------+
| Level   | Code | Message                                                                                                                                 |
+---------+------+-----------------------------------------------------------------------------------------------------------------------------------------+
| Note    | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t, reason to use this rate is "use min(1, 110000/8) as the sample-rate=1" |
| Warning | 1105 | skip analyze locked table: test.t                                                                                                       |
+---------+------+-----------------------------------------------------------------------------------------------------------------------------------------+
2 rows in set (0.00 sec)

Unlock the statistics of table t and ANALYZE can be successfully executed again.

mysql> UNLOCK STATS t;
Query OK, 0 rows affected (0.01 sec)

mysql> ANALYZE TABLE t;
Query OK, 0 rows affected, 1 warning (0.03 sec)

mysql> SHOW WARNINGS;
+-------+------+-----------------------------------------------------------------------------------------------------------------------------------------+
| Level | Code | Message                                                                                                                                 |
+-------+------+-----------------------------------------------------------------------------------------------------------------------------------------+
| Note  | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t, reason to use this rate is "use min(1, 110000/8) as the sample-rate=1" |
+-------+------+-----------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.00 sec)

In addition, you can also lock the statistics of a partition using LOCK STATS. For example:

Create a partition table t, and insert data into it. When the statistics of partition p1 are not locked, the ANALYZE statement can be successfully executed.

mysql> CREATE TABLE t(a INT, b INT) PARTITION BY RANGE (a) (PARTITION p0 VALUES LESS THAN (10), PARTITION p1 VALUES LESS THAN (20), PARTITION p2 VALUES LESS THAN (30));
Query OK, 0 rows affected (0.03 sec)

mysql> INSERT INTO t VALUES (1,2), (3,4), (5,6), (7,8);
Query OK, 4 rows affected (0.00 sec)
Records: 4  Duplicates: 0  Warnings: 0

mysql> ANALYZE TABLE t;
Query OK, 0 rows affected, 6 warning (0.02 sec)

mysql> SHOW WARNINGS;
+---------+------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Level   | Code | Message                                                                                                                                                                                                                              |
+---------+------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Warning | 1105 | disable dynamic pruning due to t has no global stats                                                                                                                                                                                 |
| Note    | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t's partition p0, reason to use this rate is "Row count in stats_meta is much smaller compared with the row count got by PD, use min(1, 15000/4) as the sample-rate=1" |
| Warning | 1105 | disable dynamic pruning due to t has no global stats                                                                                                                                                                                 |
| Note    | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t's partition p1, reason to use this rate is "TiDB assumes that the table is empty, use sample-rate=1"                                                                 |
| Warning | 1105 | disable dynamic pruning due to t has no global stats                                                                                                                                                                                 |
| Note    | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t's partition p2, reason to use this rate is "TiDB assumes that the table is empty, use sample-rate=1"                                                                 |
+---------+------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
6 rows in set (0.01 sec)

Lock the statistics of partition p1 and execute ANALYZE. The warning message shows that the ANALYZE statement has skipped partition p1.

mysql> LOCK STATS t PARTITION p1;
Query OK, 0 rows affected (0.00 sec)

mysql> SHOW STATS_LOCKED;
+---------+------------+----------------+--------+
| Db_name | Table_name | Partition_name | Status |
+---------+------------+----------------+--------+
| test    | t          | p1             | locked |
+---------+------------+----------------+--------+
1 row in set (0.00 sec)

mysql> ANALYZE TABLE t PARTITION p1;
Query OK, 0 rows affected, 2 warnings (0.01 sec)

mysql> SHOW WARNINGS;
+---------+------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Level   | Code | Message                                                                                                                                                              |
+---------+------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Note    | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t's partition p1, reason to use this rate is "TiDB assumes that the table is empty, use sample-rate=1" |
| Warning | 1105 | skip analyze locked table: test.t partition (p1)                                                                                                                     |
+---------+------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2 rows in set (0.00 sec)

Unlock the statistics of partition p1 and ANALYZE can be successfully executed again.

mysql> UNLOCK STATS t PARTITION p1;
Query OK, 0 rows affected (0.00 sec)

mysql> ANALYZE TABLE t PARTITION p1;
Query OK, 0 rows affected, 1 warning (0.01 sec)

mysql> SHOW WARNINGS;
+-------+------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Level | Code | Message                                                                                                                                                              |
+-------+------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Note  | 1105 | Analyze use auto adjusted sample rate 1.000000 for table test.t's partition p1, reason to use this rate is "TiDB assumes that the table is empty, use sample-rate=1" |
+-------+------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.00 sec)

Behaviors of locking statistics

• If you lock the statistics on a partitioned table, the statistics of all partitions on the partitioned table are locked.
• If you truncate a table or partition, the statistics lock on the table or partition will be released.

The following table describes the behaviors of locking statistics:

Manage ANALYZE tasks and concurrency

Terminate background ANALYZE tasks

Since TiDB v6.0, TiDB supports using the KILL statement to terminate an ANALYZE task running in the background. If you find that an ANALYZE task running in the background consumes a lot of resources and affects your application, you can terminate the ANALYZE task by taking the following steps:

1. Execute the following SQL statement:

   SHOW ANALYZE STATUS
   

   By checking the instance column and the process_id column in the result, you can get the TiDB instance address and the task ID of the background ANALYZE task.

2. Terminate the ANALYZE task that is running in the background.

   • If enable-global-kill is true (true by default), you can execute the KILL TIDB ${id}; statement directly, where ${id} is the ID of the background ANALYZE task obtained from the previous step.
   • If enable-global-kill is false, you need to use a client to connect to the TiDB instance that is executing the backend ANALYZE task, and then execute the KILL TIDB ${id}; statement. If you use a client to connect to another TiDB instance, or if there is a proxy between the client and the TiDB cluster, the KILL statement cannot terminate the background ANALYZE task.

   To terminate the ANALYZE task, you can execute the KILL TIDB ${id}; statement, where ${id} is the ID of the background ANALYZE task obtained from the previous step.

For more information on the KILL statement, see KILL.

Control ANALYZE concurrency

When you run the ANALYZE statement, you can adjust the concurrency using system variables, to control its effect on the system.

The relationships of the relevant system variables are shown below:

tidb_build_stats_concurrency, tidb_build_sampling_stats_concurrency, and tidb_analyze_partition_concurrency are in an upstream-downstream relationship, as shown in the preceding diagram. The actual total concurrency is: tidb_build_stats_concurrency * (tidb_build_sampling_stats_concurrency + tidb_analyze_partition_concurrency). When modifying these variables, you need to consider their respective values at the same time. It is recommended to adjust them one by one in the order of tidb_analyze_partition_concurrency, tidb_build_sampling_stats_concurrency, tidb_build_stats_concurrency, and observe the impact on the system. The larger the values of these three variables, the greater the resource overhead on the system.

tidb_build_stats_concurrency

When you run the ANALYZE statement, the task is divided into multiple small tasks. Each task only works on statistics of one column or index. You can use the tidb_build_stats_concurrency variable to control the number of simultaneous small tasks. The default value is 2. The default value is 4 for v7.4.0 and earlier versions.

tidb_build_sampling_stats_concurrency

When analyzing ordinary columns, you can use tidb_build_sampling_stats_concurrency to control the concurrency of executing sampling tasks. The default value is 2.

tidb_analyze_partition_concurrency

When running the ANALYZE statement, you can use tidb_analyze_partition_concurrency to control the concurrency of reading and writing statistics for a partitioned table. The default value is 2. The default value is 1 for v7.4.0 and earlier versions.

tidb_distsql_scan_concurrency

When you analyze regular columns, you can use the tidb_distsql_scan_concurrency variable to control the number of Regions to be read at one time. The default value is 15. Note that changing the value will affect query performance. Adjust the value carefully.

tidb_index_serial_scan_concurrency

When you analyze index columns, you can use the tidb_index_serial_scan_concurrency variable to control the number of Regions to be read at one time. The default value is 1. Note that changing the value will affect query performance. Adjust the value carefully.

See also

• LOAD STATS
• DROP STATS
• LOCK STATS
• UNLOCK STATS
• SHOW STATS_LOCKED

• LOAD STATS
• LOCK STATS
• UNLOCK STATS
• SHOW STATS_LOCKED