Question

EDIT :-

With PK on the Id column following is produced both for EXISTS and COUNT(*) > 0

Table 'TableWithHugeData'. Scan count 0, logical reads 3, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

---

IF OBJECT_ID('TableWithHugeData') IS NOT NULL
BEGIN
DROP TABLE TableWithHugeData
END

CREATE TABLE TableWithHugeData
(
ID UNIQUEIDENTIFIER,
Name varchar(max)
)


DECLARE @Counter INT
SELECT @Counter = 0
WHILE (@Counter < 50000)
BEGIN
INSERT INTO TableWithHugeData
VALUES (NewId(),'Ashish ' + Convert(varchar(5000),@Counter))
SELECT @Counter = @Counter +1
END

EXISTS

DBCC FreeProcCache
DBCC DROPCLEANBUFFERS
SET  STATISTICS IO ON
DECLARE @Id UNIQUEIDENTIFIER
SELECT @Id = '28BD1F4C-7D89-4731-9D2C-21ECB20500F8'
IF EXISTS (SELECT * FROM TableWithHugeData WHERE Id = @Id)
BEGIN
SELECT CONVERT(BIT, 1)
END
ELSE
BEGIN
SELECT CONVERT(BIT, 0)
END

IO statistics for EXISTS :-

Table ‘TableWithHugeData’. Scan count 1, logical reads 6, physical reads 1, read-ahead reads 270, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

COUNT(*) >0

DBCC FreeProcCache
DBCC DROPCLEANBUFFERS
DECLARE @Id UNIQUEIDENTIFIER
SELECT @Id = '28BD1F4C-7D89-4731-9D2C-21ECB20500F8'
IF (SELECT COUNT(*) FROM TableWithHugeData WHERE Id = @Id)>0
BEGIN
SELECT CONVERT(BIT, 1)
END
ELSE
BEGIN
SELECT CONVERT(BIT, 0)
END

*IO Statistics For Count() >0 ** :-

Table ‘TableWithHugeData’. Scan count 1, logical reads 6, physical reads 1, read-ahead reads 270, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

As you see, the IO statistics for EXISTS and COUNT(*)>0 are same. I thought EXISTS would be faster and would perform less reads. Am I missing something here? Would defining a Primary key would make any difference?

Answer 1

As you see, the IO statistics for EXISTS and COUNT(*)>0 are same. I thought EXISTS would be faster and would perform less reads. Am I missing something here?

SQL Server reads data block-wise.

If the record you are looking for happens to be in the last block (or not found at all) then the EXISTS would still have to scan all blocks.

Would defining a Primary key would make any difference?

Yes.

With the PRIMARY KEY, EXISTS and COUNT(*) would always be the same (single index seek).

Without one, EXISTS would stop on first match while COUNT(*) would continue to the end of the table.

Update:

Sorry, I was wrong earlier.

It seems that SQL Server optimizes COUNT(*) > 0 to EXISTS, so yes, the statistics would be the same.

DBCC FreeProcCache
CHECKPOINT
DBCC DROPCLEANBUFFERS
SET  STATISTICS IO ON
DECLARE @id UNIQUEIDENTIFIER
SET @id = '7C65EBB1-1242-4084-8BB4-3CC92CA2BE51'
IF      (
        SELECT  COUNT(*)
        FROM    tablewithhugedata
        WHERE   id = @id
        ) > 0
BEGIN
        SELECT CONVERT(BIT, 1)
END
ELSE
BEGIN
        SELECT CONVERT(BIT, 0)
END

DBCC execution completed. If DBCC printed error messages, contact your system administrator.
DBCC execution completed. If DBCC printed error messages, contact your system administrator.
Table 'TableWithHugeData'. Scan count 1, logical reads 1, physical reads 0, read-ahead reads 270, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

(1 row(s) affected)

And a little bit different query:

DBCC FreeProcCache
CHECKPOINT
DBCC DROPCLEANBUFFERS
SET  STATISTICS IO ON
DECLARE @id UNIQUEIDENTIFIER
SET @id = '7C65EBB1-1242-4084-8BB4-3CC92CA2BE51'
IF EXISTS
        (
        SELECT  1
        FROM    tablewithhugedata
        WHERE   id = @id
        )
BEGIN
        SELECT CONVERT(BIT, 1)
END
ELSE
BEGIN
        SELECT CONVERT(BIT, 0)
END

DBCC FreeProcCache
CHECKPOINT
DBCC DROPCLEANBUFFERS
SET  STATISTICS IO ON
DECLARE @id UNIQUEIDENTIFIER
SET @id = '7C65EBB1-1242-4084-8BB4-3CC92CA2BE51'
IF      (
        SELECT  COUNT(*)
        FROM    tablewithhugedata
        WHERE   id = @id
        ) BETWEEN 1 AND 2
BEGIN
        SELECT CONVERT(BIT, 1)
END
ELSE
BEGIN
        SELECT CONVERT(BIT, 0)
END

DBCC execution completed. If DBCC printed error messages, contact your system administrator.
DBCC execution completed. If DBCC printed error messages, contact your system administrator.
Table 'TableWithHugeData'. Scan count 2, logical reads 266, physical reads 6, read-ahead reads 270, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

(1 row(s) affected)

The second query needs to know exact value of COUNT(*) so it reads all blocks (266 logical reads vs. 1 for EXISTS / COUNT(*) > 0).