Rolling sum on sequentially linked data - sql

Question

I am working on large a legacy dataset with sequentially related data of which I lack the words to explain so I made a beautiful paint image. This isn't of course of the real dataset but it is close. In the example there are three sequences.
Each record has an ID and a value. It also has a pointer to the next related ID. The sequence length is random and stops when the next related ID hits a 0 value. All records are only used once in one sequence, meaning they cant merge or split. A sequence can consist of only one record.
What I need to accomplish is to get the rolling sum on each record of a sequence using a SQL query(SQL server 2014). I know how to do this if there is a common identifier in the sequence, but in this case there is not.
I have been able to accomplish it in Excel (for what it's worth) by finding the previous sum (if it exists) and adding the current value. But I'm unable to translate it to SQL. Does anyone know where to start to get to the end goal of the 'rolling sum result' column in SQL?
[previous sum] formule: =IFNA(INDEX([rolling sum formula],MATCH([#id],[next_pointer],0),0),0)
[rolling sum result] formula: =[#[previous sum]]+[#value]
*The data sequences aren't sorted like in the Excel example. This just makes it easier to read in the example.

Answer 1

You need something like a RECURSIVE query.
You can do this with CTE. This is a test with your data (the column you seek is "cumul", the others are there to help understand what's going on):
WITH sequenza AS (
SELECT
id,
value,
nextid,
id AS lastid,
value as cumul
FROM
items
WHERE nextid = 0
UNION ALL
SELECT
curr.id,
curr.value,
curr.nextid,
prev.lastid,
prev.cumul + curr.value AS cumul
FROM
items AS curr
INNER JOIN sequenza AS prev
ON prev.id = curr.nextid
)
SELECT * FROM sequenza
WHERE id = 31;
To do this in reverse order... there is probably more than one way. Off the top of my head I'd get, for each chain (identified by its lastid), the minimum and maximum cumul value, then I'd apply the ladder algorithm - in this case the descending rolling sum is VALMIN+VALMAX-ROLLING.
So, something like
WITH sequenza AS (
SELECT
id,
value,
nextid,
id AS lastid,
value as cumul
FROM
items
WHERE nextid = 0
UNION ALL
SELECT
curr.id,
curr.value,
curr.nextid,
prev.lastid,
prev.cumul + curr.value AS cumul
FROM
items AS curr
INNER JOIN sequenza AS prev
ON prev.id = curr.nextid
),
sequenza2 AS (
SELECT
id,
value,
nextid,
id AS lastid,
value as cumul
FROM
items
WHERE nextid = 0
UNION ALL
SELECT
curr.id,
curr.value,
curr.nextid,
prev.lastid,
prev.cumul + curr.value AS cumul
FROM
items AS curr
INNER JOIN sequenza2 AS prev
ON prev.id = curr.nextid
)
SELECT sequenza.*, m1+m2-cumul AS cumulasc FROM sequenza
JOIN (
SELECT lastid, MIN(cumul) AS m1, MAX(cumul) AS m2
FROM sequenza2
GROUP BY lastid
) AS cirpo ON (sequenza.lastid = cirpo.lastid)
ORDER BY sequenza.lastid, cumul DESC