8.9.6 Optimizer Statistics

The column_statistics data dictionary table stores histogram statistics about column values, for use by the optimizer in constructing query execution plans. To perform histogram management, use the ANALYZE TABLE statement; see Section 13.7.3.1, “ANALYZE TABLE Syntax”.

The column_statistics table has these characteristics:

The column_statistics table is not directly accessible by users because it is part of the data dictionary. Histogram information is available using INFORMATION_SCHEMA.COLUMN_STATISTICS, which is implemented as a view on the data dictionary table. COLUMN_STATISTICS has these columns:

Column histograms contain buckets for parts of the range of values stored in the column. Histograms are JSON objects to permit flexibility in the representation of column statistics. Here is a sample histogram object:

{
  "buckets": [
    [
      1,
      0.3333333333333333
    ],
    [
      2,
      0.6666666666666666
    ],
    [
      3,
      1
    ]
  ],
  "null-values": 0,
  "last-updated": "2017-03-24 13:32:40.000000",
  "sampling-rate": 1,
  "histogram-type": "singleton",
  "number-of-buckets-specified": 128,
  "data-type": "int",
  "collation-id": 8
}

Histogram objects have these keys:

To extract particular values from the histogram objects, you can use JSON operations. For example:

The optimizer uses histogram statistics, if applicable, for columns of any data type for which statistics are collected. The optimizer applies histogram statistics to determine row estimates based on the selectivity (filtering effect) of column value comparisons against constant values. Predicates of these forms qualify for histogram use:

For example, these statements contain predicates that qualify for histogram use:

The requirement for comparison against a constant value includes functions that are constant, such as ABS() and FLOOR():

Histogram statistics are useful primarily for nonindexed columns. Adding an index to a column for which histogram statistics are applicable might also help the optimizer make row estimates. The tradeoffs are:

The optimizer prefers range optimizer row estimates to those obtained from histogram statistics. If the optimizer determines that the range optimizer applies, it does not use histogram statistics.

For columns that are indexed, row estimates can be obtained for equality comparisons using index dives (see Section 8.2.1.2, “Range Optimization”). In this case, histogram statistics are not necessarily useful because index dives can yield better estimates.

In some cases, use of histogram statistics may not improve query execution (for example, if the statistics are out of date). To check whether this is the case, use ANALYZE TABLE to regenerate the histogram statistics, then run the query again.

Alternatively, to disable histogram statistics, use ANALYZE TABLE to drop them. A different method of disabling histogram statistics is to turn off the condition_fanout_filter flag of the optimizer_switch system variable (although this may disable other optimizations as well):

If histogram statistics are used, the resulting effect is visible using EXPLAIN. Consider the following query, where no index is available for column col1:

If histogram statistics indicate that 57% of the rows in t1 satisfy the col1 < 24 predicate, filtering can occur even in the absence of an index, and EXPLAIN shows 57.00 in the filtered column.