Calculate difference between one column over multiple rows using coalesce and join - mysql

Question

I have following table
+-----+--------+-----------+-----------+-------+
| id | job_id | source_id | target_id | value |
+-----+--------+-----------+-----------+-------+
| 204 | 5283 | 247 | 228 | 1201 |
| 349 | 4006 | 247 | 228 | 100 |
| 350 | 4007 | 247 | 228 | 500 |
| 351 | 4008 | 247 | 228 | 1000 |
| 352 | 4009 | 1 | 100 | 100 |
| 353 | 4010 | 1 | 100 | 500 |
| 354 | 4011 | 1 | 100 | 50 |
+-----+--------+-----------+-----------+-------+
I want to create a diff between the column value groupped by source_id and target_id. The older one (smaller id) should be compared with the newer one
I have searched a little bit and found coalesce. I have written a small query and it works in "general", but not as expexted:
SELECT
c.id, c.source_id, c.target_id, c.value, COALESCE(c1.value - c.value, -1) AS diff
FROM
changes c LEFT JOIN changes c1 ON (c1.source_id = c.source_id AND c1.target_id = c.target_id)
GROUP BY c.source_id, c.target_id, c.job_id
ORDER BY c.id
I got following result:
+-----+-----------+-----------+-------+------+
| id | source_id | target_id | value | diff |
+-----+-----------+-----------+-------+------+
| 204 | 247 | 228 | 1201 | 0 |
| 349 | 247 | 228 | 100 | 1101 |
| 350 | 247 | 228 | 500 | 701 |
| 351 | 247 | 228 | 1000 | 201 |
| 352 | 1 | 100 | 100 | 0 |
| 353 | 1 | 100 | 500 | -400 |
| 354 | 1 | 100 | 50 | 50 |
+-----+-----------+-----------+-------+------+
You can see the diff work for id 349 and 353, I want this for all rows like the following expected result:
+-----+-----------+-----------+-------+------+
| id | source_id | target_id | value | diff |
+-----+-----------+-----------+-------+------+
| 204 | 247 | 228 | 1201 | 1201 |
| 349 | 247 | 228 | 100 | 1101 |
| 350 | 247 | 228 | 500 | -400 |
| 351 | 247 | 228 | 1000 | -500 |
| 352 | 1 | 100 | 100 | 100 |
| 353 | 1 | 100 | 500 | -400 |
| 354 | 1 | 100 | 50 | 450 |
+-----+-----------+-----------+-------+------+
It would be no problem if the diff result is inverted.
What did I miss?
Thanks for any hints.

Answer 1

if you use user defined variables you don't need to join the table to itself. just do a row by row comparrision like so
SELECT
id,
job_id,
target_id,
if(#a = source_id, #b - value, value) as diff,
#b := value as value,
#a := source_id as source_id
FROM changes
CROSS JOIN (SELECT #a:=0, #b:=0)t
DEMO

Answer 2

I suspect that you're looking for something like this - although the COALESCE bit seems misleading to me...
SELECT a.*, COALESCE(b.value-a.value,a.value) diff
FROM
( SELECT x.* , COUNT(*) rank FROM changes x JOIN changes y ON y.id <= x.id GROUP BY x.id ) a
LEFT
JOIN
( SELECT x.* , COUNT(*) rank FROM changes x JOIN changes y ON y.id <= x.id GROUP BY x.id ) b
ON b.source_id = a.source_id
AND b.rank = a.rank - 1;

Answer 3

I think you want:
SELECT c.id,
c.source_id,
c.target_id,
c.value,
c.value - COALESCE(co.value, 0) delta
FROM changes c
LEFT JOIN (
SELECT ci.id, MAX(cio.id) prev_id
FROM changes ci
JOIN changes cio
ON cio.source_id = ci.source_id
AND cio.target_id = ci.target_id
AND cio.id < ci.id
GROUP BY ci.id
) link
ON link.id = c.id
LEFT JOIN changes co
ON co.id = link.prev_id
ORDER BY c.id
I have changed the logic slightly.
In your expected results, the first diff has gone from unknown (0?) to 1201 and is reported as a positive diff, but the second has gone from 1201 to 100 and is still reported as positive.
I have changed the name to delta, and given you the number required to move from the previous value to the new value. Obviously you can change this if you want to:
COALESCE(co.value-c.value, c.value) diff
which will get you the results you provided (with the diff 500 changed to -500, which I believe was a typo).