Selecting GETDATE() function twice in a select lis

问题:

I have a SELECT statement that uses GETDATE() for two different column values. I'm wondering if by the nature of things those two separate function calls in the same SELECT will return identical values every time?

回答1:

No they aren't guaranteed to return identical values every time. Each individual reference to GetDate() is a runtime constant and will keep its value throughout the query...

SELECT GETDATE()
FROM large_table

will return the same value in all rows regardless of how long the query takes to run.

But there is no guarantee that different references will have the same value.

You can see this as below

SET NOCOUNT ON;

DECLARE @T TABLE 
(
rownum INT IDENTITY(1,1) PRIMARY KEY,
d1 DATETIME,
d2 DATETIME
)

WHILE (5 > (SELECT COUNT(*) FROM @T WHERE d1 <> d2))
    BEGIN
    DELETE FROM @T WHERE d1 = d2
    INSERT INTO @T 
    SELECT GETDATE(),GETDATE()
    END

SELECT * FROM @T

Example Results

rownum      d1                      d2
----------- ----------------------- -----------------------
22381       2011-05-18 12:24:14.433 2011-05-18 12:24:14.437
30912       2011-05-18 12:24:15.420 2011-05-18 12:24:15.423
43234       2011-05-18 12:24:16.717 2011-05-18 12:24:16.720
113360      2011-05-18 12:24:24.210 2011-05-18 12:24:24.213
147855      2011-05-18 12:24:27.817 2011-05-18 12:24:27.820

回答2:

Sorry to say it, but I just came up with a test to show that it will not always return the same value. It actually does get evaluated twice, and if the system clock happens to turn over during the time between those two evaluations, you could get slightly different times out of the two calls.

However, the others are right in saying that it won't be evaluated once per row: only once per column.

See Will GETUTCDATE() return the same value if used twice in the same statement?

回答3:

Most system functions are evaluated per query, not per row, except those that are by definition statistically unique such as NEWID(). This includes things like RAND() (which is not unique, only psuedo random) unless seeded randomly row by row with NEWD().

Determinism is not related to this evaluation because it means "the output is the same for any given input independent of, say, language or DMY/MDY)

This "once per query" evaluation makes sense of course, especially for GETDATE().

If I do a 10k row update, I want every row to have the same GETDATE() value. The entire update could easily take > 3.33 milliseconds and I don't want different values over my 10k rows.

回答4:

Yes, they will return the same date and time, to the millisecond.

SELECT GETDATE(), GETDATE()

Returns

2010-10-29 15:34:06.353 2010-10-29 15:34:06.353

I have tested this with 4000, GETDATE() and they all return the same.