Teradata - calculate difference between two dates from different row number - teradata

Question

I want to move my excel calculation to Teradata but not sure how to do it. In excel is rather easy and I use simple if to give me DIFF =IF(A2=A3, (C2-B3) * 24, "")
NO T_DATE L_DATE DIFF
AAA 10/08/2019 17:02:00 10/08/2019 20:35:00 5.83
AAA 10/08/2019 14:45:00 10/08/2019 15:10:00 11.78
AAA 10/08/2019 03:23:00 10/08/2019 10:25:00 17.32
AAA 09/08/2019 17:06:00 10/08/2019 01:11:00 25.70
AAA 08/08/2019 23:29:00 09/08/2019 10:27:00
BBB 08/08/2019 09:34:00 08/08/2019 21:19:00 22.23
BBB 07/08/2019 23:05:00 08/08/2019 06:09:00 18.03
BBB 07/08/2019 12:07:00 07/08/2019 20:25:00 22.32
BBB 06/08/2019 22:06:00 07/08/2019 08:53:00 22.77
BBB 06/08/2019 10:07:00 06/08/2019 19:44:00
Is there a way of doing it in Teradata? I want to have again the difference in hours between L_DATE and T_DATE for each NO.

Answer 1

You can use window functions to achieve this. It's important to note that when you subtract two dates or timestamps (in this case) you will be returned in INTERVAL type so you will need to specify what type of INTERVAL you want as well as it's size (SECOND, MINUTE, HOUR, DAY, etc)..
CREATE MULTISET VOLATILE TABLE yourtable(
ID VARCHAR(3)
,T_DATE TIMESTAMP(0)
,L_DATE TIMESTAMP(0)
,DIFF NUMERIC(6,2)
) PRIMARY INDEX (ID) ON COMMIT PRESERVE ROWS;
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('AAA','2019-10-08 17:02:00','2019-10-08 20:35:00',5.83);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('AAA','2019-10-08 14:45:00','2019-10-08 15:10:00',11.78);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('AAA','2019-10-08 03:23:00','2019-10-08 10:25:00',17.32);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('AAA','2019-09-08 17:06:00','2019-10-08 01:11:00',25.70);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('AAA','2019-08-08 23:29:00','2019-09-08 10:27:00',NULL);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('BBB','2019-08-08 09:34:00','2019-08-08 21:19:00',22.23);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('BBB','2019-07-08 23:05:00','2019-08-08 06:09:00',18.03);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('BBB','2019-07-08 12:07:00','2019-07-08 20:25:00',22.32);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('BBB','2019-06-08 22:06:00','2019-07-08 08:53:00',22.77);
INSERT INTO yourtable(ID,T_DATE,L_DATE,DIFF) VALUES ('BBB','2019-06-08 10:07:00','2019-06-08 19:44:00',NULL);
SELECT yourtable.*,
CAST(((LEAD(T_DATE) OVER (PARTITION BY ID ORDER BY T_DATE) - L_DATE) HOUR(4)) AS INTEGER)
FROM yourtable;
+-----+---------------------+---------------------+--------+-------------------------------------------+
| ID | T_DATE | L_DATE | DIFF | (LEAD (<value expression>) - L_DATE) HOUR |
+-----+---------------------+---------------------+--------+-------------------------------------------+
| AAA | 2019-08-08 23:29:00 | 2019-09-08 10:27:00 | <null> | 7 |
| AAA | 2019-09-08 17:06:00 | 2019-10-08 01:11:00 | 25.70 | 2 |
| AAA | 2019-10-08 03:23:00 | 2019-10-08 10:25:00 | 17.32 | 4 |
| AAA | 2019-10-08 14:45:00 | 2019-10-08 15:10:00 | 11.78 | 2 |
| AAA | 2019-10-08 17:02:00 | 2019-10-08 20:35:00 | 5.83 | <null> |
| BBB | 2019-06-08 10:07:00 | 2019-06-08 19:44:00 | <null> | 3 |
| BBB | 2019-06-08 22:06:00 | 2019-07-08 08:53:00 | 22.77 | 4 |
| BBB | 2019-07-08 12:07:00 | 2019-07-08 20:25:00 | 22.32 | 3 |
| BBB | 2019-07-08 23:05:00 | 2019-08-08 06:09:00 | 18.03 | 3 |
| BBB | 2019-08-08 09:34:00 | 2019-08-08 21:19:00 | 22.23 | <null> |
+-----+---------------------+---------------------+--------+-------------------------------------------+
The reason this looks so ugly is because you are trying to compare (subtract) values in two different records. In a database there is no relationship between one record and another. There is no ordering. They live independently of one another. This is radically different than excel where rows (records) have order (a row number).
We use the Window Function LEAD() to establish a group of records as being in a group (a partition) using the PARTITION BY clause, and we give that partition an ordering with the ORDER BY clause. Then we use that LEAD() to say "The very next T_DATE in this ordered partition for this record".
Then we do our date math and subtract the two timestamps. We specify that we want an INTERVAL of type HOUR(4) back. This will hold up to 9999 hours and it will error if it goes over 9999 hours.
Lastly we cast that thing to integer so you can do math on it. You do not, however, have to do the casting if you don't want. I added it because often times we want to add hours together and whatnot.
If you are working on an older version of Teradata that doesn't have the LEAD() function (it's a newer addition) you can use MAX() or MIN() and some extra syntax in your windowing definition to explicitely say "Just the next record's T_DATE" like:
MAX(T_DATE) OVER (PARTITION BY ID ORDER BY T_DATE ROWS BETWEEN 1 FOLLOWING AND 1 FOLLOWING)