So, statistics. The information relating to statistics comes from [sys].[stats] and [sys].[stats_columns], so let's get straight in and have a look at those two tables:
object_id | This is the ID of the object on which the statistics were created. |
|---|---|
name | This is the name of the statistics. Like indexes, you cannot have more than one set of stats on an object with the same name, but sets of stats on different objects can share the same name. Auto-generated stats have names starting with '_WA_Sys'. |
stats_id | This is the unique ID of the statistics within the entry. |
auto_created | This is 1 if the statistics were auto-created by the query processor. These will only appear if AUTO_CREATE_STATISTICS has been set on for the database using ALTER DATABASE. |
user_created | This is 1 if the statistics were created manually |
no_recompute | This is 1 if the statistics were created with the NORECOMPUTE option, meaning that statistics are not recomputed automatically. This can be a good idea for static data, but seeing as statistics recompution is triggered by 20% of the rows of the underlying object changing, is mostly without merit. |
has_filter | Like indexes, from SQL Server 2008 onwards, statistics could be filtered using a WHERE clause style filter. Again, like indexes, this column will be 1 if such a filter exists. |
filter_definition | This column will contain the definition of the statistics filter if one is specified, or NULL if not. |
So, that's the basics of statistics covered - so what about the involved columns? For that information, we look to [sys].[stats_columns]:
object_id | This is the ID of the object on which the statistics were created. |
|---|---|
stats_id | This is the same as the stats_id column from [sys].[stats]. |
stats_column_id | This is the ID of the column within in the statistics (i.e. giving a sequential number to the columns involved). |
column_id | This is the ID of the column covered by the statistics, and can be matched with [sys].[columns] for any given object_id value. |
Job done - we now know everything we need to about stats and their metadata. Notice that there are a lot less columns than for indexes, because there isn't any scope for included columns or partitioning, and the order and ascendency of the columns sampled does not matter.
We'll take a break from the metadata series for a while, and posts will be on a variety of topics relating to SQL Server.
