如何在具有Group By子句的SQL查询的where子句中使用算术表达式，而不会出现算术溢出？

Question

I have an Azure SQL DB and my query keeps throwing an arithmetic overflow error when it runs over too many rows. This is a sanitized version of my query (SET_END_TIME and SET_START_TIME are FLOAT type):

select t1.value1
from TABLE1 t1
    join TABLE2 t2
    on t1.ID = t2.ID
where
    (t1.SET_END_TIME - t1.SET_START_TIME) > 2
    and
    t2.CREATED_AT > '2017-06-11T00:00:00'
group BY
    t1.value1;

If I run this with another and in the where clause that restricts the date to two days, it runs fine. Eg if I added

and t2.CREATED_AT < '2017-06-13T00:00:00'

So I know the issue is somehow related to the number of rows it is running over, but I don't understand why since the actual arithmetic operation should only be done row by row. Does anyone know what could be causing the overflow error and how I can fix my query to be able to run over many rows?

EDIT: I know I have invalid values. I can see SET_START_TIME values of 1.79769313486232E+308 and -8.98846567431158E+307. Is there anyway I can tell the query to ignore invalid values (eg any Set start time > 100 and < 0) so it won't try to run the arithmetic on them?

Answer 1

I would be suspicious of the values in SET_END_TIME and SET_START_TIME and that calculation causing you the error. Can you output each column and see if perhaps SET_START_TIME is negative at some point and that produces the error you get?

Answer 2

Try removing the group by.

If you want distinct values for t1.value add the distinct keyword in the select.

(just thinking out loud)

Answer 3

Turns out the root cause was invalid values in the columns I was using for the arithmetic in my where clause, and the solution was to use a case expression inside a subquery to first sanitize the values I want to perform the arithmetic on, then filter on them in the outer query. Example:

select
    t1.value1,
    count(*) as total_values
from
    (
        select
            t1.ID as innerID,
            (case when t1.SET_START_TIME < 10000 and t1.SET_START_TIME > 0 then t1.SET_START_TIME else 0 end) as set_start,
            (case when t1.SET_END_TIME < 10000 and t1.SET_END_TIME > 0 then t1.SET_END_TIME else 0 end) as set_end
        from TABLE1 t1
    ) innerTable
    join TABLE2 t2
    on innerID = t2.ID
where
    (set_end - set_start) > 2
    and
    t2.CREATED_AT > '2017-06-11T00:00:00'
group by
    t1.value1;