Here’s a fun little problem that a coworker and I ran into last week. We’re creating a database that isn’t quite a data vault, but will have some data-vault-like properties. Most of the data is coming from third-party services. Data that comes from third-parties may not be edited; it is what is. We also wanted to be able to track data that gets loaded somewhat together.

Most of our data tables have an InsertedDate column. We want everything that gets loaded together to have the same value in this column. Our question was this: If you wrap all of your operations in a transaction, will this set the timestamp at commit?

Here’s our simple test.

create table ThisIsJunk (
InsertedDate datetimeoffset(7)
);

begin transaction;

declare @counter int = 0;
while @counter < 1000
begin
insert into ThisIsJunk (InsertedDate) values (getutcdate());
set @counter = @counter + 1;
end

commit transaction;

select min(InsertedDate) as minval, max(InsertedDate) as maxval from ThisIsJunk;
drop table ThisIsJunk;


Here’s what we see for the output.

minval maxval
2017-08-21 13:06:27.8900000 +00:00 2017-08-21 13:06:28.2966667 +00:00

Obviously, these values are not the same. If you want all of your InsertedDate values to be the same, you must explicitly make a variable and save this variable.

create table ThisIsJunk (
InsertedDate datetimeoffset(7)
);

begin transaction;

declare @counter int = 0;
declare @insertedDate datetimeoffset(7) = getutcdate();
while @counter < 1000
begin
insert into ThisIsJunk (InsertedDate) values (@insertedDate);
set @counter = @counter + 1;
end

commit transaction;

select min(InsertedDate) as minval, max(InsertedDate) as maxval from ThisIsJunk;
drop table ThisIsJunk;


This time, we get an expected, deterministic result for our timestamps.

minval maxval
2017-08-21 13:09:54.0233333 +00:00 2017-08-21 13:09:54.0233333 +00:00

And now we know: we must explicitly set our InsertedDate if we want everything that is inserted together to share the same timestamp. The commit will not take care of this for us.