DATEDIFF overflow - azure

Question

I am using the following code in azure sql datawarehouse
SELECT cast(DATEDIFF(ms,cast(Start as datetime2),cast(EndTime as datetime2)
) as float) AS [total]--difference to be calculated in millisecond
FROM systable
but coming across an error as
"The datediff function resulted in an overflow. The number of dateparts separating two date/time instances is too large. Try to use datediff with a less precise datepart.
"
My requirement is to have the difference in milliseconds and if thats changed then it will affect other results.
request you to please provide some help

Answer 1

This happens because the DATEDIFF() function returns an integer. An integer only allows values up to 2,147,483,647. In this case, you have more than ~2B values causing the data type overflow. You would ideally use the DATEDIFF_BIG() function which returns a bigint that allows for values up to 9,223,372,036,854,775,807 or ~9 Septillion. DATEDIFF_BIG() isn't supported in SQL Data Warehouse / Azure Synapse Analytics (as of Jan 2020).
You can vote for the feature here: (https://feedback.azure.com/forums/307516/suggestions/14781627)
Testing DATEDIFF(), you can see that you can get ~25 days and 20 hours of difference between dates before you run out of integers. Some sample code is below.
DECLARE #startdate DATETIME2 = '01/01/2020 00:00:00.0000';
DECLARE #enddate DATETIME2 = '01/01/2020 00:00:02.0000';
-- Support:
-- MILLISECOND: ~25 days 20 Hours
-- MICROSECOND: ~35 minutes
-- NANOSECOND: ~ 2 seconds
SELECT
DATEDIFF(DAY, #startdate, #enddate) [day]
, DATEDIFF(HOUR, #startdate, #enddate) [hour]
, DATEDIFF(MINUTE, #startdate, #enddate) [minute]
, DATEDIFF(SECOND, #startdate, #enddate) [second]
, DATEDIFF(MILLISECOND, #startdate, #enddate) [millisecond]
, DATEDIFF(MICROSECOND, #startdate, #enddate) [microsecond]
, DATEDIFF(NANOSECOND, #startdate, #enddate) [nanosecond]
In the interim, you could calculate the ticks since the beginning of the time for each value and then subtract the difference. For a DATETIME2, you can calculate ticks like this:
CREATE FUNCTION dbo.DATEDIFF_TICKS(#date DATETIME2)
RETURNS BIGINT
AS
BEGIN
RETURN
(DATEDIFF(DAY, '01/01/0001', CAST(#date AS DATE)) * 864000000000.0)
+ (DATEDIFF(SECOND, '00:00', CAST(#date AS TIME(7))) * 10000000.0)
+ (DATEPART(NANOSECOND, #date) / 100.0);
END
GO
You can then just run the function and determine the ticks and the difference between ticks.
DECLARE #startdate DATETIME2 = '01/01/2020 00:00:00.0000';
DECLARE #enddate DATETIME2 = '01/30/2020 00:00:00.0000';
SELECT
dbo.DATEDIFF_TICKS(#startdate) [start_ticks],
dbo.DATEDIFF_TICKS(#startdate) [end_ticks],
dbo.DATEDIFF_TICKS(#enddate) - dbo.DATEDIFF_TICKS(#startdate) [diff];
Here is a sample running 500 years of differences:
DECLARE #startdate DATETIME2 = '01/01/2000 00:00:00.0000';
DECLARE #enddate DATETIME2 = '01/01/2500 00:00:00.0000';
SELECT
dbo.DATEDIFF_TICKS(#startdate) [start_ticks],
dbo.DATEDIFF_TICKS(#startdate) [end_ticks],
dbo.DATEDIFF_TICKS(#enddate) - dbo.DATEDIFF_TICKS(#startdate) [diff];
The results:
start_ticks end_ticks diff
-------------------- -------------------- --------------------
630822816000000000 630822816000000000 157785408000000000