SQL help on group by query with inline query - sql

Question

I've these 2 tables. I want to count number of con_id which has remark '1' continuously for the last period(s).
ex: 2 for A1, 1 for A3, but 0 for A2 and B1 as they don't have '1' continuously for the latest result(s) for the following table.
t_conmast
con_id [pk]
off_code
con_id off_code
A1 1
A2 1
B1 2
A3 1
t_readbak
con_id [fk]
counter
remark
timestamp [not shown in the table; auto inserted by system]
con_id counter remark timestamp
A1 1 0
A1 3 1
A1 6 1
B1 1 1
B1 2 0
A2 1 0
A2 2 1
A2 3 0
A3 1 1
what I tried and failed (I added the off_code just to get result for a single office)
select con_id,
count(con_id)
from t_readbak
where remark=1 and timestamp > (select max(timestamp)
from t_readbak
where remark=0
group by con_id)
and con_id in (select con_id from t_conmast where off_code=1)
Expected output
con_id count(con_id)
A1 2
A2 0
A3 1
B1 0

Answer 1

This is the approach that I took to solving this. First, calculate a cumulative sum of remark going backwards for each con_id. Then, the first time that you hit a row where remark = 0, use the value on that row. You can find the first such row using row_number().
The complication is when you have no remarks with a value of 0. In that case, you just take the total number.
The following query combines this logic into SQL:
select rb.con_id,
(case when NumZeros = 0 then numRemarks else cumsum end) as count1
from (select rb.*,
SUM(remark) over (partition by con_id order by counter desc) as cumsum,
ROW_NUMBER() over (partition by con_id, remark order by counter desc) as remark_counter,
SUM(case when remark = 0 then 1 else 0 end) as NumZeros,
SUM(remark) over (partition by con_id) as numRemarks
from t_readbak rb
) rb
where (remark_counter = 1 and remark = 0) or
(NumZeros = 0 and remark_counter = 1)

Answer 2

A left self join might work. Something like this:
select con_id, count(*) records
from t_readback t1 left join t_readback t2 using (con_id, remark)
where remark = 1
and t1.counter < t2.counter
group by con_id

Answer 3

If you mean that you only want to include con_id counts if every remark in the period is 1, you can do something like this:
SELECT
con_id,
COUNT(CASE remark = 1 THEN 1 END) AS Remark1Count,
COUNT(CASE remark <> 1 THEN 1 END) AS RemarkNot1Count
FROM t_conmast
INNER JOIN t_readbak ON t_conmast.con_id = t_readbak.con_id
WHERE your-timestamp-condition
GROUP BY con_id
HAVING COUNT(CASE remark <> 1 THEN 1 END) = 0
The HAVING will filter out any con_id that has a remark <> 1.

Answer 4

get the maximum timestamp for each con_id where remark is 0.
thereafter, again for each con_id, count items with younger timestamps. remark is set to 1 in these records by construction:
select con_id
, count(*)
from t_readbak master
inner join t_conmast office on ( office.off_code = 1
and office.con_id = master.con_id )
inner join (
select con_id con_id
, max(timestamp) ts
from (
select con_id
, remark
, timestamp
from t_readbak
where remark = 0
) noremark
group by con_id
) cutoff
on ( master.con_id = cutoff.con_id )
where master.timestamp > cutoff.ts
group by master.con_id
;
replace timestamp ( max(timestamp) ) by counter ( min(counter)) and change the comparison operator if you can't trust your timestamp ordering.