Practical uses of new T-SQL functionality in SQL Server 2012 & Windows Azure SQL Database.

Practical uses of new T-SQL functionality in SQL Server 2012 & Windows Azure SQL DatabaseTobias TernströmLead Program ManagerSQL Database Systems

Overview• Key generation• Exceptions & Messages• Calculations & Aggregations• Common tasks• Updates to Windows Azure SQL Database

Key generation• Use cases• Unique keys across multiple tables

Customers & Employees Contacts• Client-side/distributed-system key generation

Prep files with related items for import

• Options• GUIDs• Table to manage the keys• NEW: Sequences

Sequence Generators

• New Database Object, similar to the IDENTITY property

• Separates number-generation from table/column

• ANSI SQL standard compliant implementation

CREATE SEQUENCE [ schema_name . ] sequence_name

[ AS { <data_type> } ]

[ START WITH <constant> ]

[ INCREMENT BY <constant> ]

[ MINVALUE <constant> | NO MINVALUE ]

[ MAXVALUE <constant> | NO MAXVALUE ]

[ CYCLE | NO CYCLE ]

[ CACHE [ <constant> ] | NO CACHE ]

NEXT VALUE FOR

[ schema_name . ] sequence_name

[ OVER (<over_order_by_clause>) ]

EXEC sp_sequence_get_range …

• Guaranteed to be globally unique

• Can be made sequential• Caveats!

• Really hard to guess…(when not sequential)

• Use any data type for keys(for ex. strings)

• Can guarantee a continuous range of values (no wholes)• Including ranges

• Serializes number generation across transactions

• Supports all integer types, starting from TINYINT up to DECIMAL(38, 0)

• Supports range reservation

• Caching “knob” for performance tuning

• Large; requires 16 bytes storage

• Many systems don’t natively support GUIDs

• Causes fragmentation• Sequential generation

cannot be guaranteed even with NEWSEQUENTIALID()

• Access to the identifier is serialized across transactions• Performance issue

• Not native RDBMS functionality, more code

• Cannot guarantee a continuous range of values• E.g. can “lose”

values because of dirty shutdowns or rollbacks

GUIDs

Pros

Cons

Separate table Sequences

Exceptions & Messages• Use cases:• Send progress updates to the application• Log events• Raise & handle exceptions

• Options:• RAISERROR• xp_logevent• Logging table• NEW: THROW

New error handling with THROW• THROW <number>, <message>, <state>;• always aborts the batch• Batch aborts if not in SET XACT_ABORT ON where it

Transaction-aborts• Does not automatically use sys.messages• is even spelled correctly!

• re-THROW• BEGIN CATCH

…; THROW;END CATCH

• Flexible• Well known

• Logs events without sending a message to the client

• Multiple behaviors, easy to get it wrong

• Not supported on SQL Database

RAISERROR

Pros

Cons

xp_logevent

• Flexible, supports the attributes you specify

• Transactions need to be managed so log records are not rolled back

Logging table• Easy to use• Similar

behavior to other programming languages

• Re-throw capabilities

• Less flexible than RAISERROR

THROW

Calculations & Aggregations• Use cases:• Include aggregations in details

Cumulative sum, averages etc.• Reference other rows in a calculation

Finding gaps, trends, etc.• Options:• Imperative programming patterns• Correlated sub queries• NEW: Window Functions

Calculations across rows• Referencing same row is

simple• Transaction cost is $1.5/tran., what

was the total amount?• Amount + 1.5 AS TotalAmount

• What about other rows?• Current bank balance?• Current account balance?• Number of days since last tran.?• Amount of previous transaction?

TranId AccountId TranDate Amount

1 12011-12-19 120

2 22011-12-20 500

3 32011-12-21 430

4 12011-12-26 95

5 22011-12-27 50

6 32011-12-28 25

7 12012-01-02 250

8 22012-01-03 25

9 32012-01-04 5000

10 12012-01-09 75

3 trn. avg.

Window Functions - Different frame sizes

Calculate the average amount of the last 3 transactions

TranId AccountId TranDate Amount

1 1 2011-12-19 120

2 2 2011-12-20 500

3 3 2011-12-21 430

4 1 2011-12-26 95

5 2 2011-12-27 50

6 3 2011-12-28 25

7 1 2012-01-02 250

8 2 2012-01-03 25

9 3 2012-01-04500

0

10 1 2012-01-09 75

120.000

310.000

350.000

341.666

191.666

56.666

108.333

100.000

1758.3331700.000

The FRAMEWindow

PARTITIONThe FRAMEThe FRAMEThe FRAMEThe FRAMEThe FRAMEThe FRAMEWindow FRAME

Window Functions

Multiple partitions, e.g. for calculations for different accounts

BalanceTranId AccountId TranDate Amount

1 1 2011-12-19 120

4 1 2011-12-26 95

7 1 2012-01-02 250

10 1 2012-01-09 75

13 1 2012-01-16 125

16 1 2012-01-23 175

2 2 2011-12-20 500

5 2 2011-12-27 50

8 2 2012-01-03 25

11 2 2012-01-10 125

120

215

465

540

665

840

500

550

575

700

Window PARTITIONs

Window PARTITIONsThe FRAMEWindow FRAME

Window Functions• Aggregation• AVG, CHECKSUM_AGG, COUNT, COUNT_BIG, MIN, MAX, SUM, STDEV,

STDEVP, VAR, VARP• User-defined CLR Aggregates (excl. windowing)

• Analytical• ROW_NUMBER, NTILE, RANK, DENSE_RANK, CUME_DIST,

PERCENT_RANK

• Distribution• PERCENTILE_CONT, PERCENTILE_DISC

• Offset• LAG, LEAD, FIRST_VALUE, LAST_VALUE

• Approach that is commonly used in other programming languages

• Follows the SQL set based paradigm

• Improved optimization

• Easy to write• Following the SQL

set based paradigm• Not set based• Only individual

queries are optimized, not the operation as a whole

• Many lines of code Higher risk for bugs

• Bad performance due to plan with N2 complexity solution

• Non-trivial to write

• Not always the optimal solution

• Not all queries can be easily rewritten to leverage window functions

Imperative approach

Pros

Cons

Correlated sub queries Window Functions

Common tasks• Use cases:• Transform data to and from strings• Constructing date instances• Handle conversion errors• Paging

• Options:• Write yourself• NEW: Additional scalar functions

New Scalar Functions• New conversion functions for all types:• TRY_CONVERT(data_type[(length)], expression [,style])• TRY_CAST(expression AS data_type[(length)])

• New conversion functions to and from strings:• FORMAT(value, format [,culture])• PARSE(string_value AS data_type [USING culture])• TRY_PARSE(string_value AS data_type [USING culture])

• Other functions:• IIF(boolean_expr, true_value, false_value)• CHOOSE(index, val1, val2,... [,valN])• CONCAT(val1, val2,… [,val N])

New Scalar Functions• New date & time related functions:• EOMONTH(date [, months_to_add])• DATEFROMPARTS(year, month, day)• TIMEFROMPARTS(hour, minutes, seconds, fractions, scale)• DATETIME2FROMPARTS(year, month, day ,hour, minutes,

seconds, fractions, scale)• DATETIMEFROMPARTS(year, month, day, hour, minutes,

seconds, milliseconds)• SMALLDATETIMEFROMPARTS(year, month, day, hour, minutes)• DATETIMEOFFSETFROMPARTS (year, month, day, hour, minute,

seconds, fractions, hour_offset, minute_offset,

precision)

Query Constructs / Dynamic SQL enhancements

• OFFSET / FETCH• Support for paging result sets

• Enforce contract for Result Sets for EXEC

SELECT ...ORDER BY ... OFFSET <expr> ROWS FETCH NEXT <expr> ROWS ONLY

EXECUTE <proc|clr proc|remote proc|function> [WITH <execute_option>[,...n ]]{ RESULT SETS {UNDEFINED|NONE|(<result_sets_definition>)} }<result_sets_definition> ::= { <result_set_definition> | AS OBJECT [<object_location>.] {table_name | view_name | tvf} | AS TYPE [schema_name.]table_type_name | AS FOR XML | (…) [,...n ]}

Updates to Windows Azure SQL DatabaseThis year

• Recursive Triggers• Common Default Options

on CREATE TABLE & INDEX

• Support for SQL Database as a Linked Server• From SQL Server

• DBCC SHOW_STATISTICS• FW Rules @ DB Level

Coming• KILL <spid>• WITH XMLNAMESPACES• ORIGINAL_LOGIN,

HOST_NAME, HOST_ID• DDL Triggers:

EVENTDATA() • sys.column_type_usages,

sys.parameter_type_usages

Questions?