How to get only those rows where the next row (by datestamp) has a different property in some field? - sql

Question

I have a table of data like this:
+--------------+-------------------------+----------+
| o_objguid | o_acttime | o_action |
+--------------+-------------------------+----------+
| 478n8937g990 | 2013-10-02 10:45:33.423 | 1012 |
| 478n8937g990 | 2013-10-02 11:21:57.207 | 1012 |
| 478n8937g990 | 2013-10-02 11:21:57.887 | 1012 |
| 478n8937g990 | 2013-11-15 13:42:11.983 | 1013 |
+--------------+-------------------------+----------+
I want a query to return only those rows where, for a given o_objguid, the next row in time sequence does not have an o_action of 1012.
I'm using the following query:
select d1.* from dbo.dms_audt d1
inner join
(select d2.o_objguid,d2.o_acttime,d2.o_action,
min(datediff(second,d1.o_acttime,d2.o_acttime)) as intervalToNext
from dbo.dms_audt d1
inner join
dbo.dms_audt d2
on
d1.o_objguid=d2.o_objguid
where
d2.o_acttime>d1.o_acttime
group by
d2.o_objguid,d2.o_acttime,d2.o_action) d2
on
d1.o_objguid=d2.o_objguid
where
datediff(second,d1.o_acttime,d2.o_acttime)=intervalToNext
and
d1.o_action=1012
and
d2.o_action<>1012
This query does not return the row with an o_acttime of 2013-10-02 10:45:33.423, because the next row has the same o_action. But because I'm using an argument of second in the datediff() function, the rows with these o_acttime:
2013-10-02 11:21:57.207
2013-10-02 11:21:57.887
are both treated as the same date value for calculation purposes, so both rows are returned, when really the only one that should be returned is the 2013-10-02 11:21:57.887 row.
I tried changing the datediff() argument to millisecond, but this resulted in an overflow error, probably because a date difference of several days or more will have too many milliseconds to fit in the return type.
I guess I can join the query's results against another query that will use max(o_acttime), so that only the highest o_acttime in a group of results that have the same intervalToNext will be returned. But I'm concerned about the size and performance of this query; there are a lot of nested Cartesian products here and it's already pretty slow against a set of 1000000+ rows. Is there a better way to get the result I want?

Answer 1

This should work as well without computing a CTE twice. Depending on the data and indexing strategy it may or may not be faster than Joachim's response.
SELECT
*
FROM dbo.dms_audt d1
CROSS APPLY ( -- get next action
SELECT TOP 1
*
FROM dbo.tbl_audt X
WHERE d1.o_objguid = X.o_objguid
AND d1.o_acttime < X.o_acttime
ORDER BY X.o_acttime
) D2
WHERE D1.o_action = 1012
AND D2.o_action != 1012

Answer 2

Sadly, SQL Server 2008 does not have LEAD() (a 2012 feature) which would have made the query trivial, but you can simulate it using ROW_NUMBER();
WITH cte AS (
SELECT *, ROW_NUMBER() OVER (ORDER BY o_acttime) rn
FROM table1
WHERE o_objguid = '478n8937g990'
)
SELECT a.*
FROM cte a
JOIN cte b
ON a.rn = b.rn - 1 AND b.o_action <> 1012;
An SQLfiddle to test with.