Oracle SQL challenging subquery (aggregation, group by, count)

Question

Have a table given, holding the status history and creation of a user:

ID	USERID	STATUS	USER_CREATED	TIMESTAMP
1	A	active	01.02.2020	01.02.2020, 08:00:00
2	A	inactive	01.02.2020	02.02.2020, 11:00:00
3	A	active	01.02.2020	03.02.2020, 12:00:00
4	B	active	02.02.2020	02.02.2020, 13:00:00
5	B	inactive	02.02.2020	05.02.2020, 14:00:00
6	B	active	02.02.2020	07.02.2020, 16:00:00
7	C	active	02.02.2020	02.02.2020, 09:00:00
8	C	inactive	03.02.2020	03.02.2020, 11:00:00
9	D	active	03.02.2020	03.02.2020, 10:00:00
10	E	active	04.02.2020	04.02.2020, 10:00:00

I am working on a statistics overview for user creation, which works on a provided min/max date.

Desired result for create date: 01.02.2020 to 03.02.2020:

Create Date	Users created	Still active today	Total active users at create date
01.02.2020	1	1	1
02.02.2020	2	1	2
03.02.2020	1	1	3

The first two columns, Users created and Still active today are straight forward, I simply use a group by and select max(id) to get the latest status of every userid.

The challenge however comes with the last column, Total active users at create date .

SELECT USER_CREATED as "Create Date",
count(*) as "Users created",
sum(case when STATUS = 'active' then 1 else 0 end) as "Still active today",
??? as "Total active users at create date"
FROM myTable t1
WHERE t1.USER_CREATED >= to_date('01.02.2020', 'DD.MM.YYYY') and t1.USER_CREATED <= to_date('03.02.2020', 'DD.MM.YYYY')
AND and t1.ID = (select max(ID) from myTable where USERID = t1.USERID)
GROUP BY t1.USER_CREATED;

For the column Total active users at create date , I am a bit clueless. The point is I have to make sure that the timestamp is below the user_created field in my counting. Ie on create date 02.02.01, only timestamps below 02.02.01 must be considered. More precisely, the timestamp CLOSEST to the create date tells the actual status of a user.

Current approach (not working):

SELECT USER_CREATED as "Create Date",
count(*) as "Users created",
sum(case when STATUS = 'active' then 1 else 0 end) as "Still active today",
(select count(*) from (select USERID, max(STATUS) keep (dense_rank first order by USER_CREATED desc) as STATUS
      from myTable t2
      where t2.timestamp <= t1.timestamp
      group by USERID) where STATUS = 'active')
as "Total active users at create date"
FROM myTable t1
WHERE t1.USER_CREATED >= to_date('01.02.2020', 'DD.MM.YYYY') and t1.USER_CREATED <= to_date('03.02.2020', 'DD.MM.YYYY')
AND and t1.ID = (select max(ID) from myTable where USERID = t1.USERID)
GROUP BY t1.USER_CREATED;

I need to Any help is appreciated

Answer 1

I'd use CROSS APPLY rather than a correlated sub-query, it's just a bit easier to read.

Then, the sub-query would need to get every users' last event preceding the end of the day in question (for which I chose to use ROW_NUMBER() ) . Then count the number of users who's last event is an active event.

WITH
  dates AS
(
  SELECT
    User_Created,
    COUNT(*)                                             AS count_created,
    SUM(CASE WHEN STATUS = 'active' THEN 1 ELSE 0 END)   AS count_still_active
  FROM
  (
    SELECT
      myTable.*,
      ROW_NUMBER() OVER (PARTITION BY UserID ORDER BY Timestamp DESC)  AS event_id
    FROM
      myTable
  )
    sorted
  WHERE
        User_Created BETWEEN DATE'2020-02-01' AND DATE'2020-02-03'
    AND event_id = 1
  GROUP BY
    User_Created
)
SELECT
  *
FROM
  dates
OUTER APPLY
(
  SELECT
    COUNT(*) AS active_users_at_end_of_day
  FROM
  (
    SELECT
      myTable.*,
      ROW_NUMBER() OVER (PARTITION BY UserID ORDER BY Timestamp DESC)  AS event_id
    FROM
      myTable
    WHERE
      myTable.Timestamp < dates.User_Created+1
  )
    filtered_sorted
  WHERE
        event_id = 1
    AND status = 'active'
)
  daily_stats
ORDER BY
  User_Created

Demo...

• https://dbfiddle.uk/?rdbms=oracle_18&fiddle=10a4d416c06afd37d8b882975f8b701f

Answer 2

You can use:

WITH expected_range (min_dt, max_dt) AS (
  SELECT DATE '2020-02-01',
         DATE '2020-02-04'
  FROM   DUAL
),
calendar (dt, max_dt) AS (
  SELECT MIN(LEAST(TRUNC(timestamp), min_dt)),
         MAX(GREATEST(TRUNC(timestamp), max_dt))
  FROM   myTable
         CROSS JOIN expected_range
UNION ALL
  SELECT dt + INTERVAL '1' DAY, max_dt
  FROM   calendar
  WHERE  dt + INTERVAL '1' DAY <= max_dt
),
activity (dt, userid, status, rn, day_rn, status_now, day_status) AS (
  SELECT c.dt,
         a.userid,
         status,
         ROW_NUMBER() OVER (PARTITION BY userid ORDER BY timestamp NULLS LAST),
         ROW_NUMBER() OVER (PARTITION BY userid, c.dt ORDER BY timestamp DESC),
         MAX(status)
           KEEP (DENSE_RANK LAST ORDER BY timestamp NULLS FIRST)
           OVER (PARTITION BY userid),
         COALESCE(
           LAST_VALUE(status)
             IGNORE NULLS
             OVER (PARTITION BY a.userid ORDER BY c.dt, a.timestamp),
           'inactive'
         )
  FROM   calendar c
         LEFT OUTER JOIN mytable a
         PARTITION BY (a.userid)
         ON (c.dt <= a.timestamp AND a.timestamp < c.dt + INTERVAL '1' DAY)
)
SELECT a.dt AS create_date,
       COUNT(CASE WHEN rn = 1 THEN 1 END) AS users_created,
       COUNT(CASE WHEN rn = 1 AND status_now = 'active' THEN 1 END)
         AS still_active,
       COUNT(CASE WHEN day_rn = 1 AND day_status = 'active' THEN 1 END)
         AS total_active_users_today
FROM   activity a
       INNER JOIN expected_range e
       ON (a.dt BETWEEN e.min_dt AND e.max_dt)
GROUP BY a.dt
ORDER BY a.dt

Which, for the sample data (with additional test cases):

CREATE TABLE myTable (ID, USERID, STATUS, TIMESTAMP) AS
SELECT  1, 'A', 'active',   TO_DATE('01.02.2020 08:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  2, 'A', 'inactive', TO_DATE('02.02.2020 11:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  3, 'A', 'active',   TO_DATE('03.02.2020 12:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  4, 'B', 'active',   TO_DATE('02.02.2020 13:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  5, 'B', 'inactive', TO_DATE('05.02.2020 14:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  6, 'B', 'active',   TO_DATE('07.02.2020 16:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  7, 'C', 'active',   TO_DATE('02.02.2020 09:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  8, 'C', 'inactive', TO_DATE('03.02.2020 11:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT  9, 'D', 'active',   TO_DATE('03.02.2020 10:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT 10, 'E', 'active',   TO_DATE('04.02.2020 10:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT 11, 'F', 'active',   TO_DATE('31.01.2020 00:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT 12, 'G', 'active',   TO_DATE('04.02.2020 00:00:00', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL UNION ALL
SELECT 13, 'G', 'inactive', TO_DATE('04.02.2020 23:59:59', 'DD.MM.YYYY HH24:MI:SS') FROM DUAL;

Outputs:

CREATE_DATE	USERS_CREATED	STILL_ACTIVE	TOTAL_ACTIVE_USERS_TODAY
2020-02-01 00:00:00	1	1	2
2020-02-02 00:00:00	2	1	3
2020-02-03 00:00:00	1	1	4
2020-02-04 00:00:00	2	1	5

db<>fiddle here