Information about how values are distributed in a column, which the warehouse optimiser uses to estimate plan costs. Fabric builds and updates it for you; manual CREATE, UPDATE and DROP STATISTICS are limited to single-column histograms.
Also called query optimisation statistics.
Read more: Microsoft Learn
In the Ultra Transcenders books
Each book explains Statistics in context, with comparison tables and the common traps.
Terms in this definition
- VALUES
Returns in DAX the distinct column values, or table rows, still visible after filters are applied, sometimes with an extra blank entry. CALCULATE often takes the result as a table filter.
- Warehouse
Fabric item offering complete T-SQL support (DML, DDL, multi-table transactions) over Delta tables held in OneLake; lakehouse SQL analytics endpoints, by contrast, are read-only.
- CRUD
Shorthand for create, read, update and delete, the four basic things you do with data. Data-plane roles in Azure Cosmos DB, for instance, authorise those operations on items.
- DROP
A DDL (Data Definition Language) statement that deletes a database object like a table. Once a table is dropped, its rows are gone unless a backup exists.
Related terms
- ANALYZE TABLE
The
ANALYZE TABLE ... COMPUTE STATISTICSstatement gathers statistics on a table and its columns so the cost-based optimiser can plan queries better. On Unity Catalog managed tables, predictive optimization runsANALYZEfor you. - Automatic statistics
When queries run against the Fabric Data Warehouse or the SQL analytics endpoint, the engine creates statistics (histograms, cardinality and average column length) on columns used for filtering, joining, grouping, sorting or DISTINCT. The histogram objects are named with a
_WA_Sys_prefix. - Cold start
The extra time a Fabric warehouse query can take on its first run, because data has to be pulled from OneLake into the cache, statistics built or nodes woken up. In query insights, a
data_scanned_remote_storage_mbvalue above zero is the tell-tale sign. - Data profiling
One part of data quality monitoring. Once a table has a profile attached, statistics and drift are calculated for it over time, with results saved to tables and shown on a dashboard; anomaly detection is the equivalent that covers an entire schema.
- Data profiling tools
Three Power Query views: column quality, giving the share of valid, empty and error values; column distribution, giving distinct and unique counts; and column profile, giving statistics and a value breakdown. Unless you switch them to the full data set, only the first 1,000 rows are examined.
- Data skipping
Avoiding reads of files that cannot hold matching rows, using column statistics such as min, max and null counts gathered per file at write time. Liquid clustering and Z-ordering improve how much gets skipped.
- Filtered index
A nonclustered rowstore index restricted by a WHERE clause to a clearly defined group of rows, which keeps it small and its statistics precise. Only simple comparisons are allowed in the filter, so LIKE can't be used.
- ORC
Optimized Row Columnar (ORC), an Apache column-oriented format that stores indexes and statistics inside the file. You can use it with Auto Loader and external tables but not with managed tables.