(Pardon me for asking a very basic question. I have come back to SQL
Server after five years of Oracle, and my memory is a tad rusty).

I have a bit of SHOW_PLAN output that I believe that I understand, but
I would appreciate your comments if I am wrong.

I have a requirement to produce a list of the primay key values in a
table along with the total count of rows:

Given this table and contents:

create table taCountDemo (ID char(01) primary key, someData
varchar(50));

insert into taCountDemo (ID,somedata)
select 'a', 'aaaa'
union
select 'b', 'bbbb';

the results should be:

ID counter
---- -----------
a 2
b 2

(don't wonder why - there's a SAS application on top which means that
ordinary rules of logic don't apply)

This is the query that I'm using:

select ID
, cnt.counter
from taCountDemo
cross join
(select counter
from ( select count(*) as counter
from taCountDemo
) as i
) as cnt;

The SHOW_PLAN output is like this:

|--Nested Loops(Inner Join)
|--Compute Scalar(DEFINE[Expr1006]=CONVERT_IMPLICIT(int,
[Expr1009],0)))
| |--Stream Aggregate(DEFINE[Expr1009]=Count(*)))
| |--Clustered Index Scan(OBJECT[master].[dbo].
[taCountDemo]. )
|--Clustered Index Scan(OBJECT[master].[dbo].
[taCountDemo].)

My understanding of this is that the count(*) is only executed ONCE,
and that the nested loop then combines the result (EXPR1006) with all
keys obtained by scanning the primary key index. Or in other words
that the single-"row" result of the count(*) is chosen as the "table"
that drives the loop.

Is that correctly understood?

Thanks for your input

Bo Brunsgaard

bbcworldtour@hotmail.com wrote:
> (Pardon me for asking a very basic question. I have come back to SQL
> Server after five years of Oracle, and my memory is a tad rusty).
>
> I have a bit of SHOW_PLAN output that I believe that I understand, but
> I would appreciate your comments if I am wrong.
>
> I have a requirement to produce a list of the primay key values in a
> table along with the total count of rows:
>
> Given this table and contents:
> (..)
> (don't wonder why - there's a SAS application on top which means that
> ordinary rules of logic don't apply)

> This is the query that I'm using:
>
> select ID
> , cnt.counter
> from taCountDemo
> cross join
> (select counter
> from ( select count(*) as counter
> from taCountDemo
> ) as i
> ) as cnt;


What about:

SELECT ID, (SELECT Count(*) FROM taCountDemo) as counter
FROM taCountDemo

?

Execution plan will be exactly the same but IMHO it looks much simpler.


> The SHOW_PLAN output is like this:
>
> |--Nested Loops(Inner Join)
> |--Compute Scalar(DEFINE[Expr1006]=CONVERT_IMPLICIT(int,
> [Expr1009],0)))
> | |--Stream Aggregate(DEFINE[Expr1009]=Count(*)))
> | |--Clustered Index Scan(OBJECT[master].[dbo].
> [taCountDemo]. )
> |--Clustered Index Scan(OBJECT[master].[dbo].
> [taCountDemo].)
>
> My understanding of this is that the count(*) is only executed ONCE,
> and that the nested loop then combines the result (EXPR1006) with all
> keys obtained by scanning the primary key index. Or in other words
> that the single-"row" result of the count(*) is chosen as the "table"
> that drives the loop.
>
> Is that correctly understood?


Yes, you're completely right.


--
Best regards,
Marcin Guzowski
http://guzowski.info

> My understanding of this is that the count(*) is only executed ONCE,
> and that the nested loop then combines the result (EXPR1006) with all
> keys obtained by scanning the primary key index. Or in other words
> that the single-"row" result of the count(*) is chosen as the "table"
> that drives the loop.

Maybe I am wrong but I think it is other way around. It puts the sql
statement in a nested loop and count(*) is executed many times.