SQL Server - Merging/Gruppierung zusammenhängende Zeitstempel

Question

Ich habe eine Tabelle (RECORDS) mit Datensätzen von wann bis wann Elemente zu einem übergeordneten Element verknüpft wurden:

CHILD STARTDATE     ENDDATE      PARENT 
207  2013-12-18 12:45:59.017 2014-01-09 18:16:01.227  NULL 
207  2014-01-09 18:16:01.227 2016-03-03 09:54:28.757  NULL 
207  2016-03-03 09:54:28.757 2100-01-01 00:00:00.000  NULL 
10558 2016-03-03 18:10:34.877 2016-04-05 10:25:22.860  10557 
10558 2016-04-05 10:25:22.860 2016-04-05 11:33:10.493  10557 
10558 2016-04-05 11:33:10.493 2100-01-01 00:00:00.000  10557 
10596 2016-05-15 16:55:15.970 2016-05-16 10:00:00.000  10592 
10596 2016-05-16 10:00:00.000 2016-05-17 10:00:00.000  10593 
10596 2016-05-17 10:00:00.000 2100-01-01 00:00:00.000  10592 
10600 2012-12-18 12:45:59.017 2100-01-01 00:00:00.000  10599 

ich über den 4 wichtigsten Fälle angezeigt:
Für das gleiche Kind, neue Datensätze mit null-Wert (207)
für das gleiche Kind, neue Datensätze mit demselben Stammwert (10558)
für das gleiche Kind, neue Datensätze mit einem übergeordneten Wechsel (10596)
Aus demselben Kind, nur ein Datensatz (10600).

Als Kind Artikel zurück zu einer älteren Eltern ändern können, mit:

SELECT 
    CHILD 
, MIN(STARTDATE) as STARTDATE 
, MAX(ENDDATE) as ENDDATE 
, PARENT 
FROM RECORDS 
GROUP BY CHILD, PARENT 

für Artikel im folgenden unerwünschten Ergebnis führen würde, 10596:

207  2013-12-18 12:45:59.017  2100-01-01 00:00:00.000  NULL 
10558 2016-03-03 18:10:34.877  2100-01-01 00:00:00.000  10557 
10596 2016-05-15 16:55:15.970  2100-01-01 00:00:00.000  10592 
10596 2016-05-16 10:00:00.000  2016-05-17 10:00:00.000  10593 
10600 2012-12-18 12:45:59.017  2100-01-01 00:00:00.000  10599 

Das gewünschte Ergebnis wäre:

207  2013-12-18 12:45:59.017  2100-01-01 00:00:00.000  NULL 
10558 2016-03-03 18:10:34.877  2100-01-01 00:00:00.000  10557 
10596 2016-05-15 16:55:15.970  2016-05-16 10:00:00.000  10592 
10596 2016-05-16 10:00:00.000  2016-05-17 10:00:00.000  10593 
10596 2016-05-17 10:00:00.000  2100-01-01 00:00:00.000  10592 
10600 2012-12-18 12:45:59.017  2100-01-01 00:00:00.000  10599 

Irgendeine Idee, wie man das erreicht?

Answer 1

Im Folgenden wird angenommen, dass es Kontinuität ist (keine Überschneidungen) in Ihre Einträge:

WITH cte 
AS 
(
    SELECT 
     * 
     , ROW_NUMBER() OVER (PARTITION BY CHILD ORDER BY STARTDATE) r 
    FROM RECORDS 
) 

, 
cte2 
AS 
(
    SELECT 
     t1.CHILD CHILD1 
     , t1.STARTDATE STARTDATE1 
     , t1.ENDDATE ENDDATE1 
     , t1.PARENT PARENT1 
     , t1.r r1 
     , t2.CHILD CHILD2 
     , t2.STARTDATE STARTDATE2 
     , t2.ENDDATE ENDDATE2 
     , t2.PARENT PARENT2 
     , t2.r r2 
    FROM 
     cte t1 
     LEFT JOIN cte t2 ON 
      t1.CHILD = t2.CHILD 
      AND t1.r = t2.r - 1 
    WHERE 
     t1.PARENT != t2.PARENT 
     OR (t1.PARENT IS NULL AND t2.PARENT IS NOT NULL) 
     OR (t2.PARENT IS NULL AND t1.PARENT IS NOT NULL) 
     OR t2.r IS NULL 
) 

SELECT 
    CHILD1 CHILD 
    , 
     ISNULL 
      (
       (
        SELECT MIN(STARTDATE) 
        FROM RECORDS 
        WHERE 
         CHILD = CHILD1 
         AND (PARENT = PARENT1 OR (PARENT IS NULL AND PARENT1 IS NULL)) 
         AND STARTDATE >= 
          (
           SELECT ENDDATE1 
           FROM cte2 cte2INNER 
           WHERE 
            cte2INNER.CHILD1 = cte2OUTER.CHILD1 
            AND cte2INNER.r2 = 
             (
              SELECT MAX(cteINNER2.r2) 
              FROM cte2 cteINNER2 
              WHERE 
               cteINNER2.CHILD1 = cte2OUTER.CHILD1 
               AND cteINNER2.r2 <= cte2OUTER.r1 
             ) 
          ) 
       ) 
       , 
        (
         SELECT MIN(STARTDATE) 
         FROM RECORDS 
         WHERE CHILD = CHILD1 
        ) 
      ) 
      STARTDATE 
    , ENDDATE1 ENDDATE 
    , PARENT1 PARENT 
FROM cte2 cte2OUTER 

Es läuft schnell mit einer temporären Tabelle:

WITH cte 
AS 
(
    SELECT 
     * 
     , ROW_NUMBER() OVER (PARTITION BY CHILD ORDER BY STARTDATE) r 
    FROM RECORDS 
) 


SELECT 
    t1.CHILD CHILD1 
    , t1.STARTDATE STARTDATE1 
    , t1.ENDDATE ENDDATE1 
    , t1.PARENT PARENT1 
    , t1.r r1 
    , t2.CHILD CHILD2 
    , t2.STARTDATE STARTDATE2 
    , t2.ENDDATE ENDDATE2 
    , t2.PARENT PARENT2 
    , t2.r r2 
INTO #RECORDSTEMP 
FROM 
    cte t1 
    LEFT JOIN cte t2 ON 
     t1.CHILD = t2.CHILD 
     AND t1.r = t2.r - 1 
WHERE 
    t1.PARENT != t2.PARENT 
    OR (t1.PARENT IS NULL AND t2.PARENT IS NOT NULL) 
    OR (t2.PARENT IS NULL AND t1.PARENT IS NOT NULL) 
    OR t2.r IS NULL 

SELECT 
    CHILD1 CHILD 
    , 
     ISNULL 
      (
       (
        SELECT MIN(STARTDATE) 
        FROM RECORDS 
        WHERE 
         CHILD = CHILD1 
         AND (PARENT = PARENT1 OR (PARENT IS NULL AND PARENT1 IS NULL)) 
         AND STARTDATE >= 
          (
           SELECT ENDDATE1 
           FROM #RECORDSTEMP rtmpINNER 
           WHERE 
            rtmpINNER.CHILD1 = rtmpOUTER.CHILD1 
            AND rtmpINNER.r2 = 
             (
              SELECT MAX(rtmpINNER2.r2) 
              FROM #RECORDSTEMP rtmpINNER2 
              WHERE 
               rtmpINNER2.CHILD1 = rtmpOUTER.CHILD1 
               AND rtmpINNER2.r2 <= rtmpOUTER.r1 
             ) 
          ) 
       ) 
       , 
        (
         SELECT MIN(STARTDATE) 
         FROM RECORDS 
         WHERE CHILD = CHILD1 
        ) 
      ) 
      STARTDATE 
    , ENDDATE1 ENDDATE 
    , PARENT1 PARENT 
FROM #RECORDSTEMP rtmpOUTER 

DROP TABLE #RECORDSTEMP 

Answer 2

Dank für Ihre Antwort danken, es funktioniert.

ich auch diese Lösung erhalten:

WITH DISTINCTRECORDS as (
       SELECT 
       CHILD 
       , 
       CASE 
       WHEN 
       (
        CHILD = lag(CHILD) over(order by CHILD,STARTDATE, ENDDATE, PARENT) 
        AND 
        STARTDATE = lag(ENDDATE) over(order by CHILD,STARTDATE, ENDDATE, PARENT) 
        AND 
        (PARENT = lag(PARENT) over(order by CHILD,STARTDATE, ENDDATE, PARENT) OR (PARENT is null AND lag(PARENT) over(order by CHILD,STARTDATE, ENDDATE, PARENT) is null )) 
       ) 
        OR 
       (
        CHILD = lead(CHILD) over(order by CHILD,STARTDATE, ENDDATE, PARENT) 
        AND 
        ENDDATE = lead(STARTDATE) over(order by CHILD,STARTDATE, ENDDATE, PARENT) 
        AND 
        (PARENT = lead(PARENT) over(order by CHILD,STARTDATE, ENDDATE, PARENT) OR (PARENT is null AND lead(PARENT) over(order by CHILD,STARTDATE, ENDDATE, PARENT) is null )) 
       ) 
       THEN 1 
       ELSE 0 
       END ToBeMerged 
       , STARTDATE 
       , ENDDATE 
       , PARENT 

       FROM RECORDS 
) 

SELECT CHILD, min(STARTDATE) STARTDATE, max(ENDDATE) ENDDATE, PARENT FROM DISTINCTRECORDS WHERE ToBeMerged = 1 GROUP BY CHILD, PARENT 
UNION ALL 
SELECT CHILD, STARTDATE, ENDDATE, PARENT FROM DISTINCTRECORDS WHERE ToBeMerged = 0 

, welche die beste Leistung haben würde?