Oracle - SQL 查询根据列值将数据从一个表分装到另一个表？

Question

What SQL query to use to generate Table2 from Table1 (tables shown below)?

Table2 needs to bin the timestamps from Table1 into start and end times according to the Table1.Status field. As a result Table2.Status should toggle between 0 and 1 in every row.

Table2.StartTime = 1st value of Table1.TimeStamp whenever Table1.Status changes from 1 to 0 or viceversa.

Table2.EndTime = (1st value of Table1.TimeStamp of Table1.Status next change) - 1millisecond

The first record in Table1 can start from any Status - 1 or 0.

Table1

TimeStamp                               Status
11-NOV-11 12.45.45.445000000 AM           1
11-NOV-11 12.46.45.445000000 AM           1
11-NOV-11 12.47.45.445000000 AM           1
11-NOV-11 12.48.45.445000000 AM           0
11-NOV-11 12.49.45.445000000 AM           0
11-NOV-11 12.50.45.445000000 AM           1
11-NOV-11 12.51.45.445000000 AM           0
11-NOV-11 12.52.45.445000000 AM           0
11-NOV-11 12.53.45.445000000 AM           1
...

Table2

StartTime                                        EndTime                     Status
11-NOV-11 12.45.45.445000000 AM      11-NOV-11 12.48.45.444000000 AM            1
11-NOV-11 12.48.45.445000000 AM      11-NOV-11 12.50.45.444000000 AM            0
11-NOV-11 12.50.45.445000000 AM      11-NOV-11 12.51.45.444000000 AM            1
11-NOV-11 12.51.45.445000000 AM      11-NOV-11 12.53.45.444000000 AM            0
...

Any help would be appreciated.

Answer 1

select min(t2.TimeStamp) Start_time
  , lead(min(t2.TimeStamp), 1)over(order by min(t2.TimeStamp))  - INTERVAL '0.001' SECOND End_time
  , t2.status
from (
  select t1.TimeStamp TimeStamp, STATUS, group_tag, first_value(group_tag ignore nulls)over(order by t1.TimeStamp ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) group_tag_gap_filling
  from (
      select t.TimeStamp, Status, case when lead(status, 1, 2)over(order by t.TimeStamp) != status then row_number()over(order by t.TimeStamp)end group_tag
      from Table1 t
    )t1
  )t2
group by t2.group_tag_gap_filling, t2.status
;

First, in inline view "t1", I create temporary groups Then Secondly, I fill in the gap for all temporary groups Then I extract what is need using min aggregation function, and lead analytic function, and I extract 0.001 second from end_time

Answer 2

This is a gaps and islands problem, with a slight twist. The twist being that you don't want to report the most recent timestamp record for each island, but rather you want to report the record immediately following after that most recent record. So, we can use the difference in row numbers method, and also compute the lead of the timestamp column.

WITH cte AS (
    SELECT t.*, ROW_NUMBER() OVER (ORDER BY ts) rn1,
                ROW_NUMBER() OVER (PARTITION BY Status ORDER BY ts) rn2,
                LEAD(ts) OVER (ORDER BY ts) AS ts_lead
    FROM Table1 t
)

SELECT MIN(ts) AS StartTime, MAX(ts_lead) AS EndTime
FROM cte
GROUP BY Status, rn1 - rn2
HAVING MAX(ts_lead) IS NOT NULL
ORDER BY MIN(ts);

Demo

Answer 3

You can achieve it using the analytical functions and GROUP BY as follows:

SELECT MIN(TIMESTAMP) AS STARTTIME, MAX(TIMESTAMP) AS ENDTIME, STATUS 
  FROM
       (SELECT T.TIMESTAMP, T.STATUS, SUM(CHANGE_POINT) OVER (ORDER BY TIMESTAMP) AS SM 
          FROM
               (SELECT T.TIMESTAMP, T.STATUS, 
                       CASE WHEN LAG(STATUS) OVER (ORDER BY TIMESTAMP) <> STATUS THEN 1 END AS CHANGE_POINT
                  FROM YOUR_TABLE T
                ) T
        )
GROUP BY SM, STATUS

You can also use the MATCH_RECOGNIZE clause as follows:

SELECT * FROM YOUR_TABLE
MATCH_RECOGNIZE
(
 ORDER BY TIMESTAMP
  MEASURES
    FIRST( TIMESTAMP ) AS STARTDATE,
    LAST( TIMESTAMP ) AS ENDDATE,
    FIRST( STATUS ) AS STATUS
  ONE ROW PER MATCH
  PATTERN( same_field+ )
  DEFINE same_field AS FIRST(STATUS) = STATUS 
);

Answer 4

There are many ways to solve gaps-and-islands problems. In this case, the simplest method is two window functions and filtering:

select status, ts as start_time, lead(ts) over (order by ts) as end_time
from (select t1.*, lag(status) over (order by ts) as prev_status
      from table1 t1
     ) t1
where prev_status is null or prev_status <> status
order by start_time;

The logic is simple:

• Keep each row where the status changes, filtering out all other rows.
• Use lead() to get the ending timestamp.

Here is a db<>fiddle.