SQL Server sample Update of Statistics misses highest RANGE_HI_KEY on ascending key column

Should SQL Server not identify the highest value for that key and use that as the maximum RANGE_HI_KEY? Or is this just one of the limits of update without using FULLSCAN?

It's a limitation of the current implementation of sampled statistics. As it stands, sampled statistics collection uses TABLESAMPLE SYSTEM, which uses an allocation-order scan and chooses pages from the scan to sample. Only chosen pages contribute to the histogram.

Since the scan is allocation-ordered (rather than index-ordered), there is no way to give preference to the first and last pages in key order.

For more information see this related question:

How does sampling work when updating statistics?

and my article, Allocation Order Scans

For workarounds, see Statistics on Ascending Columns by Fabiano Amorim


Quick guess: enable trace flag 4139.

You are already running SP2-CU1 so this should work

Note the guidelines about TF 2389 ad 2390 too and also see http://sql-sasquatch.blogspot.com.mt/2013/06/mssql-plan-guides-to-address-ascending.html

(Note, I've not really hit this problem and have not dived in that deep)

Tags:

Sql Server

Statistics

Sql Server 2012

Related

How to improve estimate of 1 row in a View constrained by DateAdd() against an index How do I detect execute permission granted to a role (when no ON clause was used) Is total_elapsed_time in DMV sys.dm_exec_requests completely inaccurate? Convert row into concatenated string with column names included What's the difference between WITH CTE & WITH CTE (<column_names>)? How to insert values into a table with default values from a select query in PostgreSQL? Is it worth making a separate DATE table or not? Migration from Cassandra 2.x to 3.x: schema_keyspaces table is gone. How do you query system keyspaces? Transaction in a stored procedure How can SQL Delete using a sub query Performance of a=0 and b=0 and ... z=0 vs a+b+c+d=0 Do varchar/text/longtext columns in MySQL occupy full size in filesystem even for partially filled cells?

Recent Posts