SQL：对记录进行分组以进行路径

Question

My SQL is a little rusty, thus the question, but it also means I am a little unsure how to frame the question exactly. I hope the example below explains it better.

I have a list of records, by user ID:

ID | Channel  | FiscalPeriod | Amount
======================================
A  | Online   | 201710       | 20.00
A  | Online   | 201709       | 20.00
A  | Voucher  | 201708       | 20.00
A  | Voucher  | 201707       | 20.00
A  | Voucher  | 201706       | 20.00
A  | Online   | 201705       | 20.00
A  | Online   | 201704       | 20.00

I need to group the "channel" field, but not as a complete group, as the ordering by fiscal period is important, so the result set would look like the following:

ID | Channel  | MAX(FiscalPeriod) | Amount
==========================================
A  | Online   | 201710            | 40.00
A  | Voucher  | 201708            | 60.00
A  | Online   | 201705            | 40.00

For pathing the output ordering above is really important, so I can see that online << voucher << online was the user path

Answer 1

You can use a difference of row numbers approach to classify consecutive same channels to a group (run the inner query to check this) and then use it to get sum and max.

select distinct id,channel
,max(fiscalperiod) over(partition by id,channel,grp) as max_fiscalperiod
,sum(amount) over(partition by id,channel,grp) as amount
from (select t.*,
      row_number() over(partition by id order by fiscalperiod)
      -row_number() over(partition by id,channel order by fiscalperiod) as grp
      from tbl t
     ) t

Answer 2

I would approach this using the difference of row numbers approach with aggregation:

select id, channel
       max(fiscalperiod) as max_fiscalperiod
       sum(amount) as amount
from (select t.*,
             row_number() over (partition by id order by fiscalperiod) as seqnum_i,
             row_number() over (partition by id, channel order by fiscalperiod) seqnum_ic
      from t
     ) t
group by id, channel, (seqnum_i - seqnum_ic);

To understand why the difference of row numbers works is a little cumbersome. However, if you stare at the results from the subquery, you'll see why the difference defines the groups that you want.