Counting running total on time series where condition is X - sql

Question

I have a SQL table of date entries with three columns: date, item, and status. The table appears like this:
date
item
status
2023-01-01
A
on
2023-01-01
B
on
2023-01-01
C
off
2023-01-02
A
on
2023-01-02
B
off
2023-01-02
C
off
2023-01-02
D
on
2023-01-03
A
on
2023-01-03
B
off
2023-01-03
C
off
2023-01-03
D
off
Looking at the most recent entries, I need grouped by item, the latest date and status, and a count on the running total of entries where status has not changed. For example, the output I am looking for would be:
latest_date
item
current_status
number_of_days_on_current
2023-01-03
A
on
3
2023-01-03
B
off
2
2023-01-03
C
off
3
2023-01-03
D
off
1
How would I get the output I want in PostgreSQL 13.7?
This returns the latest date, item, and current status, but does not correctly count the number of days the item has been on the current status:
WITH CTE AS (
SELECT
item,
date,
status,
LAG(status) OVER (PARTITION BY item ORDER BY date) AS prev_status,
ROW_NUMBER() OVER (PARTITION BY item ORDER BY date DESC) AS rn
FROM
schema.table
)
SELECT
MAX(date) AS latest_date,
item,
status AS current_status,
SUM(CASE WHEN prev_status = status THEN 0 ELSE 1 END)
OVER (PARTITION BY item ORDER BY date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS number_of_days
FROM
CTE
WHERE
rn = 1
GROUP BY
item, status, prev_status, date
ORDER BY
item

Answer 1

Using a cte to build the runs of consecutive statuses:
with recursive cte(s_date, date, item, status, s_count, result) as (
select e.date, e.date, e.item, e.status, 1, '[]'::jsonb from entries e
left join entries e1 on e1.item = e.item and e.date - interval '1 day' = e1.date where e1.date is null
union all
select c.s_date, e.date, c.item, e.status,
case when e.status = c.status then c.s_count + 1 else 1 end,
case when e.status = c.status then c.result else c.result || jsonb_build_object('s', c.status, 'c', c.s_count) end
from cte c join entries e on e.item = c.item and c.date + interval '1 day' = e.date
)
select date(t1.md), t1.item, e.status, (select max(((v -> 'c')#>>'{}')::int)
from jsonb_array_elements(r::jsonb) v where (v -> 's')#>>'{}' = e.status)
from (select t.s_date, t.item, max(t.date) md, max(t.result::text) r
from (select c.s_date, c.date, c.item, c.result || jsonb_build_object('s', c.status, 'c', c.s_count) result from cte c) t
group by t.s_date, t.item) t1
join entries e on e.item = t1.item and date(e.date) = date(t1.md)
See fiddle.

Answer 2

According to your comment, you want to find the max count of consecutive statuses where status = last status value, this became a gaps and islands problem. This can be solved using a difference between two row_numbers and the last_value function as the following:
with last_status as
(
select *,
last_value(status) over (partition by item order by date_
range between unbounded preceding and unbounded following) current_status,
max(date_) over (partition by item) latest_date,
row_number() over (partition by item order by date_) -
row_number() over (partition by item, status order by date_) grp
from table_name
),
consecutive_status_counts as
(
select latest_date, item, current_status, status, count(*) cnt
from last_status
where current_status = status
group by latest_date, item, current_status, status, grp
)
select latest_date,
item,
current_status,
max(cnt) number_of_days_on_current
from consecutive_status_counts
group by latest_date, item, current_status
order by item
See demo