Why Should I Care About … Partitioned Views? Attendees : I will upload my sample scripts but I’m still cleaning up my DB creation code. Check back in a few days. Why Should I Care About … Partitioned Views? Let’s Partition Like It’s 1999
Frederick (Rick) Lowe rick@data-flowe.com DataFLowe
Overview of Partitioned Tables (Opposed to Partitioned Views) Appears to be a normal table Table is broken into multiple partitions based on the value of a single column Migrating a partition in or out is a metadata-only operation (i.e. basically instantaneous) Select statements can eliminate partitions Enterprise edition, MSSQL 2008 and higher
Working With Partitioned Tables Create a partition function and scheme Build clustered and nonclustered indexes on the partition scheme Views, CRUD etc reference a partitioned table pretty much like any other table SQL Server does all the heavy lifting
Partitioned Table Update/Insert/Select MyTable 2005? 2006? 2007? 2008? Partition Scheme Associated With Table P0 … P1 … P2 … P3 …
Defining a Partitioned Table CREATE PARTITION FUNCTION MyFunction(DATE) AS RANGE RIGHT FOR VALUES (…); CREATE PARTITON SCHEME MyPartitionScheme AS PARTITION MyFunction TO( … ); CREATE TABLE Sales.SalesOrderHeader( OrderDate DATETIME2(3), SalesOrderID INT IDENTITY NOT NULL, … CONSTRAINT PK_SalesOrderHeader PRIMARY KEY CLUSTERED(OrderDate, SalesOrderID) ON MyPartitionScheme( OrderDate ) );
Working With Partitioned Views Build individual tables Typically, create check constraints on tables View unions together all underlying tables *Create procedure determines which table to insert the data into *Update/Delete also need to either determine which table to work on or operate on all * indicates operations on view can save work
Partitioned View Update Insert Select UpdateMyTable CreateMyTable vwMyTable 2005? 2006? 2007? 2008? Data Boundaries Coded Into View / Procs MyTbl_05 … MyTbl_06 … MyTbl_07 … MyTbl_08 …
Defining a Partitioned View CREATE VIEW Sales.vwSalesOrderHeader AS SELECT OrderDate, SalesOrderID, … FROM Sales.SalesOrderHeader_2005 UNION ALL SELECT OrderDate, SalesOrderID, … FROM Sales.SalesOrderHeader_2006 SELECT OrderDate, SalesOrderID, … FROM Sales.SalesOrderHeader_2007 SELECT OrderDate, SalesOrderID, … FROM Sales.SalesOrderHeader_2008; CREATE PROCEDURE Sales.SalesOrderHeader_Create … AS INSERT INTO Sales.SalesOrderHeader_2008( OrderDate, SalesOrderID, ….) VALUES( …. )
Limitations of Partitioned Tables Enterprise feature Horizontal partitioning scheme only Prior to MSSQL 2014, Statistics tracked at table (not partition) level Indexing defined at the table level Partitioning can only occur on one column Prior to MSSQL 2014, online rebuild only supported at table level
Vertical Partitioning Horizontal partitioning separates groups of rows into partitions (table) or tables (view) Vertical partitioning on the other hand separates columns from the rest of the table Can be driven by normalization In practice, this term is more often associated with row splitting
Horizontal Partition
Vertical Partition
Before Row Splitting ID_Int ID_GUID Name … Big_xml LastDoc 1 1a23… Anna <… /> 2 2ff0… Bubba 9001 10ab… Goku
After Row Splitting ID_Int ID_GUID Name … 1 1a23… Anna 2 2ff0… Bubba 9001 10ab… Goku ID_Int Big_xml LastDoc 1 <… /> … 2 9001
Statistics Sampling
Common Answer
Possible Answer
Moving Data Around – Partitioned Table To extract an entire partition from a table Use ALTER TABLE SWITCH to instantly switch the partition out to a new table Archive / delete / manipulate the new table To add existing data to partitioned table Get data into table with same structure/indexing Add check constraints Switch the new table into the partitioned table
Moving Data Around – Partitioned View To separate a table from the view Alter the view so that it no longer references table Update stored procedures if necessary To add existing data to partitioned view Get data into table with same structure Ideally, add check constraints Update view and stored procedures to reference new table
Demo Startup Conditions in Execution Plan
Thank You Rick Lowe – rick@data-flowe.com DataFLowe