Get number of rows per partition

How can you get the number of rows in each partition of your partitioned tables?
Here is a simple query to retrieve that data, including the boundary value for each partition:

     p.partition_id                             as PartitionID
    ,t.name                                     as [Table]
    ,ps.name                                    as PartitionScheme
    ,pf.name                                    as PartitionFunction
    ,p.partition_number                         as [Partition]
    ,p.rows                                     as [Rows]
    ,prv.value                                  as Boundary
    --,pf.type_desc                               as BoundaryType
    ,case when pf.boundary_value_on_right = 1 
        then 'Right'
        else 'Left'
        end                                     as BoundarySide

    sys.tables t
    inner join sys.partitions p 
        on t.object_id = p.object_id 
        and p.index_id = 1
    inner join sys.indexes i on p.object_id = i.object_id and p.index_id = i.index_id
    inner join sys.data_spaces ds on i.data_space_id = ds.data_space_id
    inner join sys.partition_schemes ps on ds.data_space_id = ps.data_space_id
    inner join sys.partition_functions pf on ps.function_id = pf.function_id
    left outer join sys.partition_range_values prv 
        on pf.function_id = prv.function_id
        and p.partition_number = prv.boundary_id
    is_ms_shipped = 0   
    and exists (select 1 from sys.partitions p2
                where t.object_id = p2.object_id
                and partition_number > 1)
order by 
    ,p.partition_number desc

This will show you

  • Tables that have at least two partitions. If you also want to see the number of rows for all the other tables, remove the … and exits (…) from the where-clause.
  • Only one entry per partition. If you want to see an entry per partitioned index, remove and p.index_id = 1 from the join-condition.
  • Bit columns

    I am not a big friend of the bit data type. It’s one of these things that seem useful and helpful, but in the end they cause trouble, or at least annoyance. I’m not saying that they shouldn’t be used, but I think one should be aware what that entails.
    Like integer, bit is a very simple data type, so there is no problem with the processing time. It only takes up one byte, which is the least any data type can use, so I’m also okay with that (kind of. Why does something that is actually just a bit, even is called a bit, need a byte?). But that’s about it. Everything else about them (or their usage) annoys me.

    Nullable bit columns

    What’s the sense of that? A bit column makes sense as a flag, defining if a state is true or false. If a third state is needed, like ‘undefined’ or ‘who cares’ use another data type (see further down).

    Bit columns without a default

    I admit, sometimes the state of something isn’t known or doesn’t matter at insert time. To my understanding this is a third state (see above), but let’s for now assume it is not and we don’t want to include the column in every insert where it is not necessary and we actually don’t know, if the state is true. Then please define a default of 0 = false and make the column NOT NULL.
    Just because you’re too lazy to write the column into every insert statement, doesn’t mean it has to be NULL.

    Not 0 is 1

    Everything you insert into a bit column, that is not 0 (zero) is automatically 1 (true). So even if you insert -42, it will be 1. That’s fine, and part of the data type, but it’s contrary to some other programming languages.
    Let’s try this:

    declare @a bit = -100
    select @a as bitVariable
    select case when @a = -100 then 'True' else 'False' end bitVarCheck

    So it sets -100 as TRUE, but an equality check on -100 as FALSE. That’s inconsistent. I wouldn’t expect the second query to return TRUE, but I would also expect everything except TRUE being FALSE when inserting data, and not the other way round. Then again it is called bit and not boolean, so the behaviour is actually correct (0 = no value, 1 = (any) value). But still.

    And on a related topic:

    select * from tablename where bitColumn <> 0;

    There are only two possible values. So if it is not 0 it is 1. It’s the same as:

    select * from tablename where bitColumn = 1;

    Equality operators are always more exact and need less processor time than inequality operators. And they help the query optimizer to decide on its plan faster. I’ll admit that the difference might be minuscule in such a case, but still: do it right. For your karma.

    Indexes on bit columns

    Again, bit columns only hold two possible values, so an index on a bit column gives you a diversification of data of 2. Most likely the break down will be something like 60-40, 70-30. This index will never be chosen because it returns too many rows. Unless you have a break down of 95-5 and always query on the 5% value, an index on a bit column makes no sense.
    A bit column in an index can make sense as

  • A part of a multiple column index with columns with larger diversification before it
  • A where –clause in a filtered index as an additional criteria to keep the index small.
  • The query optimizer sometimes even suggests an index on a bit column. But that’s just because it doesn’t know the actual statistics on the column. So don’t create indexes on bit columns, unless you only query the 5%.

    Data conversion

    This is my biggest issue with bit columns: Almost every query results in a data conversion.

    select * from tablename where bitColumn = 1;
    insert into	
    	bitColumn = 1
    	intColumn > 10
    	case when bitColumn = 1 then 'True' else 'False' end

    This is how we would write those queries, right? In stored procedures or in views, wherever. Each of these queries results in forcing SQL Server to do data conversions. As in most (all?) other programming languages the default for undefined whole numbers is integer. So all the 1s are interpreted as integers. Only when the actual plan is compiled SQL Server realises they are actually meant to be bit and has to convert them.
    In the above cases this is one conversion per statement and converting integer to bit is not the most costly conversion. But still.

    select * from tablename where bitColumn = cast(1 as bit);
    	case when bitColumn = cast(1 as bit) then 'True' else 'False' end

    Database Bit in other programming languages

    Every programming language interprets a database bit differently. Some map it to true and false, some return it as 1 and 0. Some return it as byte, some as boolean.
    I know, once you mapped it, you know and can use it accordingly. But still.

    Alternative to bit

    What I like to use instead of bit columns is tinyint with a check constraint that only allows values 0 and 1.
    The advantages are obvious:

  • Integers are just as simple as bits
  • They also need a byte for storage but can actually store eight bits in it, not just one
  • If for some reason at some point in my development I need a third state or even more, I don’t need to change the data type of the column or any of my queries. I simply change the check constraint.
  • The queries I write won’t cause data conversion, because intColumn = 1 is really meant to be an integer.
  • The check constraint will give an error if I try to insert 100 or -42. Which is more consistent.
  • varchar(max), varchar(1), varchar(n), char(100) ,nchar, nvarchar

    I recently had a discussion with developers, DBAs and Data Designers on the usage of the character data types in SQL Server. A topic, I thought, that has been discussed so many times that everything about it has to be common knowledge. I was wrong. Many things seem to be left to assumptions still. So let’s take a closer look at those types.

    All of them store character data. All of them can be given a size (1 to 8000 or max). But what does that actually mean.


    Char stores one byte for each character resp. one byte for its size predicate. Char(50) fi stores 50 bytes.

    if object_id('dbo.t1') is not null
      drop table dbo.t1;
    set ansi_padding on
    create table dbo.t1
      (col1 char(50) null
      ,col2 char(50) not null)
    insert dbo.t1 values ('1234567890', '1234567890')
    insert dbo.t1 values (null, '1234567890')
    insert dbo.t1 values ('12345678901', '12345678901')
      ,datalength(col1) as datalength
      ,len(col1)      as stringlength
      ,datalength(col2) as datalength
      ,len(col2)      as stringlength

    Here is what we get

    No surprises there. Char columns have a fixed data length independent of their content. When the content is less than the fixed length, the rest will be padded with zeros.
    Let’s do the same again but let’s set ansi_padding off:

    if object_id('dbo.t1') is not null
      drop table dbo.t1;
    set ansi_padding off
    create table dbo.t1
      (col1 char(50) null
      ,col2 char(50) not null)
    insert dbo.t1 values ('1234567890', '1234567890')
    insert dbo.t1 values (null, '1234567890')
    insert dbo.t1 values ('12345678901', '12345678901')
      ,datalength(col1) as datalength
      ,len(col1)      as stringlength
      ,datalength(col2) as datalength
      ,len(col2)      as stringlength

    And what we get is this:

    On a nullable column the datalength is the same as the content. No padding occurs. On the mandatory column col2 the content is stilled filled with zeros up to 50.
    SET ANSI_PADDING ON/OFF must be set at table creation time to have an effect.


    Now let’s look at varchar columns. Varchar doesn’t preemtively store whatever size it has in it’s predicate, but only stores as much as it needs, up to its size. To do that, varchar needs to additionally store the size of its content, which takes up additional two bytes.
    So varchar(50) always stores 2 bytes + (n Characters * 1 byte) up to 52 bytes.

    if object_id('dbo.t1') is not null
      drop table dbo.t1;
    set ansi_padding on
    create table dbo.t1
      (col1 varchar(50) null
      ,col2 varchar(50) not null)
    insert dbo.t1 values ('1234567890', '1234567890')
    insert dbo.t1 values (null, '1234567890')
    insert dbo.t1 values ('12345678901', '12345678901')
      ,datalength(col1) as datalength
      ,len(col1)      as stringlength
      ,datalength(col2) as datalength
      ,len(col2)      as stringlength

    And what we get is this:

    Independent of the ansi_padding setting and the null/not null constraint, the datalength is always the same as the content length.
    Looks like char can be just as variable as varchar with the appropriate settings, doesn’t it? No, it can’t. Don’t mistake datalength for storage length. Char still stores a byte for each character. Char(50) reserves 50 bytes for every row that is created. Varchar only stores the data that is actually there. Plus 2 bytes.
    The ansi_padding setting may have some benefits in application development, but it does not change the storage needs of character columns.

    Varchar(1) vs. Char(100)

    So when to use which data type? Well, the answer seems pretty obvious:

  • 1 to 3 characters => use char
  • 4 to 7 characters => depends. If most of your data fills the 7 characters, use char. Storage wise the maintenance of varchar has a slight overhead, so char could be the wiser decision.
  • More than 7 characters => use varchar. Unless, of course, your data has a fixed length that never varies.
  • You often hear to make the cut at 15, 20 or even 50 characters. Fact is, that even at as little as 7 or 8 characters varchar is the more efficient choice.


    A common practice I have seen many times is to use varchar(255) or even varchar(max) just for everything. Hey, it only stores what is there, so who cares?
    First, varchar(255). What is this artificial 255 threshold? I know, I am also old enough to remember database systems that only allowed 255 characters. And of course, 11111111 is 255, bla bla. But come on, 64bit systems, 32 GB of RAM or more, TBs of data storage, so please get rid of the 255. Or do your users really enter exactly 255 characters?


    Then varchar(max). Varchar(max) is not really a character data type. It is treated as large data field, just like varbinary, text, ntext and image.
    Well, it is a hybrid. If the data fits on the data page it is treated as varchar, if it does not fit, it is treated as a large object. This means, that instead of the data itself a 16 byte pointer is stored that points to a pointer tree containing pointers to the actual data. This means at least one additional read operation up to several, how many it takes to get all the pointers to all the data together.
    Note, this does not dependent on how much data is actually stored in the field, but solely on if the data still fits in the data row on the page.

    We can force SQL Server to treat varchar(max) as In-Row data by switching off the large value types out of row option (default is on, which makes more sense).

    sp_tableoption 'tablename', 'large value types out of row', 'OFF'

    With this set to off, SQL Server will store the data in-row, which basically means that less rows per data page can be stored. Use this option only when the large data is the main read and write target of the table. (For more on this topic see here)
    Oh, and even with this option off the maximum value that is stored in-row is 8000 bytes, so why not use varchar(8000) instead of varchar(max)?

    External applications like SSIS treat varchar(max) as LOB, which means for SSIS that no matter what data is in there, data transfer will always be done by swapping the data to disk and then reading again from disk. Which makes data transfer a hell lot slower.

    And of course you cannot create indexes on varchar(max) columns. But you can still use it in the include-clause of an index. If you like really big indexes.
    If after all of this you still feel the need to create varchar(max) columns, consider limiting the length with a constraint. That doesn’t really help the issues described above, but it could limit the size of your database. Do you really want users to load 2GB of data in every data row? Why should you allow them to if all you wanna store is some small binary files?

    create table dbo.t1
      (col1 varchar(50) null
      ,col2 varchar(50) not null
      ,col3 varchar(max) null)
    alter table dbo.t1 add constraint CHK_col3_MaxLength
      CHECK (len(col3) <= 32000)

    This leads to the foremost reason why you just shouldn’t set every varchar() column to 255 or 8000 or max besides all the technical details:

    The key to a well designed, well performing database is to maintain data integrity.
    Don’t store data you don’t need.
    Define every data you store as well as possible.

    This means, limit the size of every column (not only character columns) to the size and data type you actually need. Not, what is easy to implement or you might need in 100 years time. Varchar() is the most flexible data type of them all and devours almost every data you throw into it. The least we can do is limit the amount.

    Nchar, nvarchar

    Nchar and nvarchar are the unicode-equivalents to char and varchar. They need two bytes per character for storage, but can store any unicode character. Apart from that they behave exactly like char and varchar.
    Well, actually it is the other way around. Char and varchar are actually subsets of nchar and nvarchar. So every varchar column can implicitly be converted to nvarchar, but not the other way around!
    Beware of things like this

    where colVarchar = N'Unicode Text'

    This forces SQL Server to explicitly convert colVarchar to nvarchar, and convert, even if it is executed in the background by SQL Server, is a function. This means that the rows of the query have to be evaluated row-by-row (hidden r-bar) and no indexes on colVarchar can be used.

    where colNVarchar = 'Non-Unicode Text'

    This can be converted implicitly and indexes can be used, since varchar is a subset of nvarchar.


  • Choose varchar over char also for small values.
  • Use varchar(max) only when you really, actually, without doubt need it.
  • Better still, use nvarchar.
  • Table Variables vs. Temporary Tables

    Many articles have been written already about this topic, many online discussions carried out. I won’t recap all of the arguments here, but merely concentrate on the maybe lesser known facts (at least they were new to me not so long ago) about temp tables and table variables.


    Temporary tables are created like every other table, only the name is prefixed by a hash sign.

    create table #tabX
      (ID int not null identity(1,1) primary key clustered
      ,Col1 bit not null
      ,Col2 datetime2 not null
      ,Col3 varchar(20) not null
      ,Col4 tinyint not null);

    Temp tables are not only created like other tables, you also can do (almost) everything with them as you would with a normal table, like creating constraints, indexes asf.

    Table variables are not created but declared.

    declare @tabX as table (ID int not null identity(1,1) primary key clustered
          ,Col1 bit not null default 0
          ,Col2 datetime2 not null default sysdatetime()
          ,Col3 varchar(20) not null unique
          ,Col4 tinyint not null check (Col4 >=1 and Col4 <=5));

    As you can see, you can declare any constraint on a table variable, but you cannot create any indexes.

    Both tables can have constraints, only temporary tables can have indexes.


    A common myth is that temporary tables are created in tempdb and table variables are created in memory. That’s nonsense. Both are created in tempdb.


    Table variables only exist in the same scope as variables, i.e. within the same block of code, defined by begin… end or ended with go.
    Temporary tables exist in the scope of the session. So if you create a temporary table it will exist as long as you drop it or disconnect the session, but only for this session.
    So you can create a temporary table, call a stored procedure and use that same table within the stored procedure. This can be handy sometimes, but it also can lead to very nasty behaviour, if you are forgetting to drop tables or where exactly now you are getting your data from.
    You can even create global temporary tables by prefixing the name with two hash signs:

    create table ##tabX
      (ID int ...

    This table is not exclusive to the session, but is available to all sessions until it is dropped or SQL Sevrer stopped.

    Table variables only exist in the code block scope, temporary tables in the session scope and global temporary tables in the server scope.


    Another common misbelieve is, that you need to generate unique names for temporary tables, since they are created in the tempdb and no two objects may have the same name.
    Yes, objects must have different names, but in case of temp tables, SQL Server takes care of the uniqueness.
    To proof this, we open two sessions and in each session we create a temp table #tabX. Then we run

    select name, object_id from tempdb.sys.objects where name like '#tabX%'

    and have this result:

    SQL Server adds a unique number for each session to the name of the temporary table, thus taking care of the uniqueness of names. Our session can still use it as #tabX, not having to care about the suffix. Isn’t that awesome?
    Oh, and if we look at global temp tables

    No suffix here, so for global temp tables we really have to take care of uniqueness. Makes sense.

    Temporary tables don’t need to generate a unique name. SQL Server does that internally.

    Transaction scope

    A huge difference is the transaction scope of temp tables and table variables. Only for temporary tables transaction logs are recorded, which is essential to rollback transactions.

    begin transaction
      insert #tabX (Col3, Col4) values ('New Value', 1);
      insert @tabX (Col3, Col4) values ('New Value', 1);

    So if we run something like this, #tabX will have rolled back and not have the new line, @tabX will still have the new line.

    Only temp table write transaction logs and writes can be rolled back.

    Compilation scope

    A stored procedure that creates and/or drops a temporary table can never have a precompiled execution plan. Since a temporary table behaves like a normal table, only with a different scope, it is also treated as a normal object.
    If you were to create a real table and/or drop it, after using it for storage or data aggregation in a stored procedure, the execution plan will also have to be recompiled every time.
    If you use a table variable, the execution plan can be precompiled, since not an object is created, but the mere image of an object.

    Only table variables can let a stored procedure have precompiled, reusable execution plans.


    There are many, many articles with benchmarks and performance test scripts. As far as I gather, the conclusion is there is no real difference. A common opinion is that temporary tables perform better with large data sets. I don’t really know if that is true, but for now I believe it.
    So for everything over 50,000 rows I rather tend towards a temp table, below that to a table variable.
    But the main decision for or against one or the other shouldn’t be based on performance but on other facts like scope, transaction handling, execution plans.

    No real performance difference. On larger volumes of data temp tables believed to be better.

    Unique advantages

    Temporary tables

    Temporary tables can be used in select… into… statements. Just as real tables can be created with select… into… , temporary tables can also be created with it.

    select top(100)
      ID > 100

    This can be very handy at times, but it has the same problems as with a real table (see my post on select… into…).

    Table variables

    Since SQL Server 2008 you can pass table variables as parameters to stored procedures. This is a great functionality. It has some restrictions (http://www.sqlteam.com/article/sql-server-2008-table-valued-parameters) but it’s a really cool and powerful feature for complex application development.


    As always with development, there is no real answer which one has the most advantages. It all depends on what you actually need. Base your decision on the advantages and disadvantages of both options, the result you want to accomplish, the scope and lifetime of the objects, the transactional scope.
    I tend to rather use table variables, because I am a developer who does a lot with stored procedures, so I like that the execution plan can still be precompiled with variables and that I don’t need to care to clean up, since at the end of the execution the variable is gone anyway.

    select… into…

    A pattern I see frequently used, especially in conjunction with temporary tables is this

      ID > 100

    This syntax automatically creates a table for the result set and inserts the records of the result set into it. It is fast and efficient. And it can be really handy, especially in complex stored procedures or scripts, storing data from complex queries for other complex queries.
    What makes it so popular is the possibility to create not only manifest tables, but also temporary tables:

      ID > 100

    You need not define or create the table before in fact it must not be defined before.
    This is all very neat, and sometimes I use it myself, but only sometimes and never in stored procedures. More likely in some ad-hoc scripts.

    And here is why:

    Schema Locks

    select… into… always creates a Schema Lock on the table it is selecting from. Even with (nolock) it needs to create a schema lock to get the schema for the table it creates. A schema lock may not sound too bad and is actually not that big a deal.
    But the problem is not so much the lock itself or what it does, but what it takes to get it. Imagine an OLTP system with a high data load, a table with approx. 1000 records inserted per second. And you have a select.. into… from this table, possibly joint with two or three other tables, and your statement, even with the nolock hint, needs to be able to place a schema lock on all four tables at once.
    This means the query has to wait for a window, when it can actually place these locks. Sometimes nanoseconds, sometimes a few milliseconds, in worst case, although unlikely, even seconds.
    While this may not seem much it can amount to quite a lot in the long run. And what you get in any case are highly different, unreliable execution times.

    Crappy Schema

    The created table gets the data from the selected data. So f.i. if for Column X the longest selected string value has 4 characters, the field will be varchar(4). Might be ok, but when I later need space for longer values, I am in trouble.

       1      as ID
      ,'XXX'    as Col1
      ,getdate()  as Col2
      ,200.23   as Col3
      ,N'djjdh' as Col4

    This gives you this schema

    CREATE TABLE dbo.tabInto
      (ID int NOT NULL
      ,Col1 varchar(3) NOT NULL
      ,Col2 datetime NOT NULL
      ,Col3 numeric(5, 2) NOT NULL
      ,Col4 nvarchar(5) NOT NULL)

    A heap with no primary key, all the columns are exactly the maximum length the data had, and all columns are not null, just because there were not any nulls in the data.
    Of course you can create your keys and indexes afterwards, but why let the engine work twice to create a decent table. And we all know what increasing the size of not null-columns does to a table (basically tear it apart).
    If you just leave it as the heap it is, the Query Optimizer will have a hard time working with it.

    And speaking of execution plans:

    Non-Reuseable Execution Plans

    If you have a create table-command in your stored procedure, which you implicitly have with select… into…, the procedure has to be recompiled for every execution. It will never get a reusable execution plan. This increases the execution time of the procedure immensely. And it doesn’t matter if it is a real table or a temporary table, they are treated the same way.

    Whereas if you use a well-defined table variable, insert .. select… into it, you will get a reusable execution plan for your procedure, and, if you defined all the data types correctly and used constraints where possible, you will also get the best possible execution plan for your data.

    Generating random numbers in T-SQL

    Every once in a while the necessity arises to generate some random numbers in your T-SQL code, be it for some random selection of price winners, be it to randomize a sort order or whatever reason.
    SQL Server provides a built-in function rand() which returns a random number. Great, you might say, until you try it.

    select rand()

    This returns a float value between 0 and 1. To make that useful you have to tweak it. I chose to create a function that returns a random integer in between the boundaries I need.

    -- view that returns a random number
    begin try
        exec('create view dbo.vw_RandNumber as select 1');
    end try begin catch end catch
    alter view dbo.vw_RandNumber
        -- return a random number using the built-in rand() function
            rand() as RandNumber    -- returns a float value between 0 and 1
    -- function to return a random integer value
    begin try
        exec('create function dbo.fn_RandNumber() returns int');
    end try begin catch end catch
    alter function dbo.fn_RandNumber
        (@Min int   -- lower boundary of desired random value
        ,@Max int)  -- upper boundary of desired random value
        returns int
             @rand float    -- the built-in rand() function returns a float,
                            -- so we need one here
            ,@ret int       -- our return value
        while @ret is null  -- just to be sure, we really return sth valid
            -- rand() returns a float between 0 and 1, so we need to tweak it
            -- to get a random number between our boundaries
            -- f.i. @Min = 1, @Max = 100 =&gt; random number between 1 and 100
            select      @rand = @Min
                                + (select RandNumber
                                from dbo.vw_RandNumber)
                                * (@Max-@Min)
            -- now we convert the float into an integer
            -- simply by assigning it to an int variable, thus cutting of the decimals
            -- Note: this is still an implicit conversion, even if no convert() or
            --       cast() is used meaning it still needs the same resources
            select @ret = @rand   -- 94,56873936663979 =&gt; 94
        -- return the result
        return @ret;

    This way I can get a random number within any range I want (as long as it’s integer).

    select dbo.fn_RandNumber(1, 100)
    select dbo.fn_RandNumber(1, 500000)
    select dbo.fn_RandNumber(800, 900)
    select dbo.fn_RandNumber(-50, 0)

    You might ask why we need this construct with a view. Simple answer, the usage of rand() is not allowed within a user defined function. It will throw an error message:
    Invalid use of a side-effecting operator ‘rand’ within a function.

    Therefore we need to ‘fool’ SQL Server by creating a view that returns just the one random number and then select from the view within the function.


    There are two major caveats to consider:

    Firstly, rand() returns a float value, we also need a float variable in the function. Using floats is never really a good idea. Normal CPUs are not optimized for floating point calculations, meaning that operations with floats need additional CPU cycles. Unless you have an additional floating point processor in your SQL Server machine (which is very unlikely), this will take up some CPU time.

    There are other ways to create random numbers, you’ll find them when you google. Here is a neat example by Jeff Moden.

         abs(checksum(newid())) % (@Max - @Min + 1) + 1 as randNum

    I prefer the construct with the view and the function, since I think it’s quite flexible (I can have functions for decimal, money, whatever), easy to read in stored procedure code and easy to maintain.

    Using Fully Qualified Object Names and DEFAULT_SCHEMA for performance gain

    As a rule of thumb you should always use fully qualified object names in your T-SQL code. I know that many find it inconvenient, many are not even using different schemas, but still, using fully qualified object names can give your database a performance boost.

    In SQL server you can define schemas and you can define on object creation, in which schema this object should be created. When accessing the object you can use the fully qualified name of the object, consisting of up to three parts


    We will ignore the first part, the database, for now. The database should be defined by the connection anyway. This article concentrates on the schema qualifier.

    select ID from myschema.mytable;
    exec dbo.uspMyProcedure;
    select * from dbo.myView;
    select ID, dbo.myFunction(myColumn) as calcVal from dbo.myView;

    But why should you when this

    select ID from mytable;

    works just as well?

    Well, that´s the point: it doesn’t work as well.


    Let´s take a look at what happens when we execute the statement without schema qualifier. We’ll assume we are executing the statement as userX. SQL Server first looks for the object in the user’s default schema, which pre-SQL Server 2005 was the user’s own. So SQL Server looks for userX.mytable. The object does not exist, so SQL Server looks in the dbo (the overall default schema), so it looks for dbo.mytable, finds it and returns the data.

    Note: If the table was in another schema than userX or dbo, the second statement wouldn’t work at all, even if the userX has access to the other schema. SQL Server stops its search. You will have to use a schema qualifier in any case.

    One step to solve this is to set the DEFAULT_SCHEMA for userX. Since SQL Server 2005 the default for the DEFAULT_SCHEMA on user creation is dbo. You can look up the default schema for all users in the column default_schema_name of the DMV sys.database_principals.

    select * from sys.database_principals

    You can also set the default schema with this


    This means Server directly looks in dbo for the object dbo.mytable.

    So all is good, isn’t it? Well, not quite.

    SQL Server still has to look up the user’s default schema. Yes, we saved one of two unnecessary reads, but why not save all? Just tell SQL Server where the object is. One additional read, mostly from the cache, may not sound like much, but just imagine a big database system with hundreds of transactions and queries per second. Every unnecessary operation you save can make a difference, and by schema-qualifying your objects you can save at least one, depending on your user-schema-default schema-architecture even two operations for every statement.

    Finding Objects

    In an architectare where you actually do have several schemas, fully qualifying the object names is an absolute must! Imagine userX and userY each have their own schemas and there is a dbo schema as well.

    Your application issues the statement

    truncate table customer;

    UserX executes the statement in his default schema, which contains the table userX.customer with his/her private customer table.

    truncate table userX.customer;

    UserY doesn’t have a table customer in his/her default schema, so dbo.customer is truncated.

    truncate table dbo.customer;

    UserX, finding the private table truncated drops the table in the schema userX. He/she executes the statement again, this time, since userX.customer doesn’t exist anymore, on dbo.customer….

    And so on.. I hope you see where I am going with this.

    Reusability of Execution Plans

    Last and not least, a big performance advantage of SQL Server is the reusability of execution plans. In a scenario like the one above SQL Server cannot rely on the object targeted being the same on every execution. It depends on the user executing the command. So in this case SQL Server will not reuse the execution plan but will recompile the statement every time.

    All this may not make a difference in every database and every project you do, but if you make a habit out of using schema.object as your way of writing T-SQL, you will never have to think about, if it could.

    %d bloggers like this: