Summing in hierachical query

Question

Hello
I'm having real problems trying to put together a hierarchical query. I've not really done too much with them before other than really basic stuff, so any help would be greatly appreciated.
I'm trying to set up a structure that will let me represent time windows as part of a batch job controller. Each window can have a parent and can also have multiple dependencies...probably best if I use a diagram:
Main batch
|
|----Data Preparation
|    |
|    |
|    |-----60m
|
|----Thread 1 *Data preparation
|    |
|    |
|    |-----30m
|----Thread 2 *Data preparation
|    |
|    |
|    |-----90m
|----Thread 3 *Data preparation
|    |
|    |
|    |-----180m
|----Thread 4 *Data preparation
|    |
|    |
|    |-----10m
|
|----Reports *Thread 1,2,3,4
|    |
|    |
|    |-----130m
|
|-----11 hoursThis shows that the "Main Batch" window is 11 hours in total. It contains 6 further windows each with their own durations. Each window can also be dependent on the completion of one or more other windows. So in this case, the "Reports" window is owned by the "Main Batch" window, but is dependent on the completion of Threads 1-4. The text in itallics denotes a dependency. Represented another way:
            Main batch(11hr)
                  |
           data preparation(1hr)
                  |
    |         |        |         |
thread 1  thread 2 thread 3  thread 4
  (30m)    (1hr30m)  (3hr)     (10m)
    |         |        |         |
                  |
             Report(2hr10m)To represent this I've got 2 tables and data like so:
CREATE TABLE dt_test_window
(   id                  NUMBER,
    label               VARCHAR2(30),
    duration            INTERVAL DAY TO SECOND,
    parent_id           NUMBER
CREATE TABLE dt_test_window_dependency
(   dependent_id        NUMBER,
    dependee_id         NUMBER
INSERT
INTO
    dt_test_window
VALUES
    (   1,
        'Main batch window',
        TO_DSINTERVAL('0 11:00:00'),
        NULL
INSERT
INTO
    dt_test_window
VALUES
    (   2,
        'Data preparation',
        TO_DSINTERVAL('0 01:00:00'),
        1
INSERT
INTO
    dt_test_window
VALUES
    (   3,
        'Thread 1',
        TO_DSINTERVAL('0 00:30:00'),
        1
INSERT
INTO
    dt_test_window
VALUES
    (   4,
        'Thread 2',
        TO_DSINTERVAL('0 01:30:00'),
        1
INSERT
INTO
    dt_test_window
VALUES
    (   5,
        'Thread 3',
        TO_DSINTERVAL('0 03:00:00'),
        1
INSERT
INTO
    dt_test_window
VALUES
    (   6,
        'Thread 4',
        TO_DSINTERVAL('0 00:10:00'),
        1
INSERT
INTO
    dt_test_window
VALUES
    (   7,
        'Thread 0 Reports',
        TO_DSINTERVAL('0 02:10:00'),
        1
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   3,
        2
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   4,
        2
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   5,
        2
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   6,
        2
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   7,
        3
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   7,
        4
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   7,
        5
INSERT
INTO
    dt_test_window_dependency
VALUES
    (   7,
        6
/What I'd like to do is run a query that will show the duration of each window, and also sum up the intervals for each window that it depends on, and where there are multiple dependencies at the same level, pick the longest duration. This would allow me to put in a start time, and show the expected end time for each stage. For these data I would expect to see
start_time  := 12:00
label               duration    max_end_time
Main batch          11:00:00    23:00
Data preparation    01:00:00    13:00
Thread 1            00:30:00    13:30
Thread 2            01:30:00    14:30
Thread 3            03:00:00    16:00
Thread 4            00:10:00    13:10
Reports             02:10:00    18:10The exected time for the reports to finish would be 18:10 because they depend on threads 1-4. Thread 3 is the longest running so it cannot start until 16:00.
This is the query I have and the result I get, can anyone tell me where I'm going wrong?
tylerd@DEV2> WITH batch AS
  2  (   SELECT
  3          TO_DATE('23-JAN-2007 12:00:00','DD-MON-YYYY HH24:MI:SS') start_date
  4      FROM
  5          dual
  6  )
  7  SELECT
  8      id,
  9      label,
10      parent_id,
11      batch.start_date + max_duration
12  FROM
13      (   SELECT
14              win.id,
15              win.label,
16              win.parent_id,
17              MAX(    (   SELECT
18                          SUM_INTERVAL(win_parent.duration)
19                      FROM
20                          dt_test_window win_parent
21                      WHERE
22                          win_parent.id = win_dep.dependee_id
23                      CONNECT BY
24                          PRIOR dependee_id = dependent_id
25                      ) + win.duration
26                 ) max_duration
27          FROM
28              dt_test_window win,
29              dt_test_window_dependency win_dep
30          WHERE
31              win.id = win_dep.dependent_id(+)
32          GROUP BY
33              win.id,
34              win.label,
35              win.parent_id
36      ),
37      batch
38  ORDER BY
39      id
40  /
        ID LABEL                           PARENT_ID BATCH.START_DATE+MA
         1 Main batch window                         23/01/2007 23:00:00
         2 Data preparation                        1 23/01/2007 13:00:00
         3 Thread 1                                1 23/01/2007 13:30:00
         4 Thread 2                                1 23/01/2007 14:30:00
         5 Thread 3                                1 23/01/2007 16:00:00
         6 Thread 4                                1 23/01/2007 13:10:00
7 Thread 0 Reports 1 23/01/2007 17:10:00<-should be 18:10
7 rows selected.Thank you
David
p.s.
Here's the code for sum_interval:
CREATE OR REPLACE TYPE SumInterval AS OBJECT
(     runningSum INTERVAL DAY(9) TO SECOND(9),
          Responsible for initialising the aggregation context
     STATIC FUNCTION ODCIAggregateInitialize
     (      actx IN OUT SumInterval
     ) RETURN NUMBER,
          Adds values to the aggregation context - NULLS are ignored
     MEMBER FUNCTION ODCIAggregateIterate
     (      self  IN OUT SumInterval,
          val   IN       DSINTERVAL_UNCONSTRAINED
     ) RETURN NUMBER,
          Routine invoked by Oracle to combine two aggregation contexts.  This happens when an
          aggregate is invoked in parallel
     MEMBER FUNCTION ODCIAggregateMerge
     (     self  IN OUT SumInterval,
          ctx2 IN      SumInterval
     ) RETURN NUMBER,
          Terminate the aggregation context and return the result
     MEMBER FUNCTION ODCIAggregateTerminate
     (     self             IN   SumInterval,
          returnValue  OUT DSINTERVAL_UNCONSTRAINED,
          flags           IN   NUMBER
     ) RETURN NUMBER
CREATE OR REPLACE TYPE BODY SumInterval AS
     STATIC FUNCTION ODCIAggregateInitialize
     (      actx IN OUT SumInterval
     )      RETURN NUMBER
     IS
     BEGIN
          IF actx IS NULL THEN
               actx := SumInterval (INTERVAL '0 0:0:0.0' DAY TO SECOND);
          ELSE
               actx.runningSum      := INTERVAL '0 0:0:0.0' DAY TO SECOND;
          END IF;
          RETURN ODCIConst.Success;
     END;
     MEMBER FUNCTION ODCIAggregateIterate
     (      self  IN OUT SumInterval,
          val   IN     DSINTERVAL_UNCONSTRAINED
     ) RETURN NUMBER
     IS
     BEGIN
          self.runningSum      := self.runningSum + val;
          RETURN ODCIConst.Success;
     END;
     MEMBER FUNCTION ODCIAggregateTerminate
     (      self        IN  SumInterval,
          ReturnValue OUT DSINTERVAL_UNCONSTRAINED,
          flags       IN  NUMBER
     ) RETURN NUMBER
     IS
     BEGIN
          returnValue := self.runningSum;
          RETURN ODCIConst.Success;
     END;
     MEMBER FUNCTION ODCIAggregateMerge
     (     self IN OUT SumInterval,
          ctx2 IN     SumInterval
     ) RETURN NUMBER
     IS
     BEGIN
          self.runningSum      := self.runningSum + ctx2.runningSum;
          RETURN ODCIConst.Success;
     END;
END;
CREATE OR REPLACE FUNCTION sum_interval
(      x DSINTERVAL_UNCONSTRAINED
) RETURN DSINTERVAL_UNCONSTRAINED
PARALLEL_ENABLE
AGGREGATE USING SumInterval;
/     

Answer 1

Hello
After some more experimentation I've found that I need to decode the dependent/dependee ids and group on that to get the correct result. Using id 7(Reports) as the starting point
tylerd@DEV2> SELECT
  2      win_lookup.parent_id,
  3      win_lookup.id,
  4      win_dep_lookup.dependent_id,
  5      win_dep_lookup.dependee_id,
  6      win_lookup.duration,
  7      CONNECT_BY_ISLEAF
  8  FROM
  9      dt_test_window_dependency win_dep_lookup,
10      dt_test_window win_lookup
11  WHERE
12      win_lookup.id = win_dep_lookup.dependee_id
13  START WITH
14      win_dep_lookup.dependent_id=7
15  CONNECT BY
16      PRIOR win_dep_lookup.dependee_id = win_dep_lookup.dependent_id
17  /
PARENT_ID         ID DEPENDENT_ID DEPENDEE_ID DURATION                       CONNECT_BY_ISLEAF
         1          3            7           3 +00 00:30:00.000000                            0
         1          2            3           2 +00 01:00:00.000000                            1
         1          4            7           4 +00 01:30:00.000000                            0
         1          2            4           2 +00 01:00:00.000000                            1
         1          5            7           5 +00 03:00:00.000000                            0
         1          2            5           2 +00 01:00:00.000000                            1
         1          6            7           6 +00 00:10:00.000000                            0
         1          2            6           2 +00 01:00:00.000000                            1
8 rows selected.I can see that if I group using decode and CONNECT_BY_ISLEAF like so:
tylerd@DEV2> SELECT
  2      DECODE(CONNECT_BY_ISLEAF, 0, win_dep_lookup.dependee_id,win_dep_lookup.dependent_id) grouping_id,
  3      SUM_INTERVAL(win_lookup.duration)
  4  FROM
  5      dt_test_window_dependency win_dep_lookup,
  6      dt_test_window win_lookup
  7  WHERE
  8      win_lookup.id = win_dep_lookup.dependee_id
  9  START WITH
10      win_dep_lookup.dependent_id=7
11  CONNECT BY
12      PRIOR win_dep_lookup.dependee_id = win_dep_lookup.dependent_id
13  GROUP BY
14      DECODE(CONNECT_BY_ISLEAF, 0, win_dep_lookup.dependee_id,win_dep_lookup.dependent_id)
15  /
GROUPING_ID SUM_INTERVAL(WIN_LOOKUP.DURATION)
          3 +000000000 01:30:00.000000000
          4 +000000000 02:30:00.000000000
          5 +000000000 04:00:00.000000000
          6 +000000000 01:10:00.000000000I get the correct result for each thread. The problem is, when I try to plug this into the main query to select the MAX of these dependencies, I get
tylerd@DEV2> WITH batch AS
  2  (   SELECT
  3          TO_DATE('23-JAN-2007 12:00:00','DD-MON-YYYY HH24:MI:SS') start_date
  4      FROM
  5          dual
  6  )
  7  SELECT
  8      id,
  9      label,
10      parent_id,
11      batch.start_date + max_duration
12  FROM
13      (   SELECT
14              win.id,
15              win.label,
16              win.parent_id,
17              MAX(    (   SELECT
18                              SUM_INTERVAL(win_lookup.duration)
19                          FROM
20                              dt_test_window_dependency win_dep_lookup,
21                              dt_test_window win_lookup
22                          WHERE
23                              win_lookup.id = win_dep_lookup.dependee_id
24                          START WITH
25                              win_dep_lookup.dependent_id=win.id
26                          CONNECT BY
27                              PRIOR win_dep_lookup.dependee_id = win_dep_lookup.dependent_id
28                          GROUP BY
29                              DECODE(CONNECT_BY_ISLEAF, 0, win_dep_lookup.dependee_id,win_dep_lookup.dependent_id)
30                      ) + win.duration
31                 ) max_duration
32          FROM
33              dt_test_window win,
34              dt_test_window_dependency win_dep
35          WHERE
36              win.id = win_dep.dependent_id(+)
37          GROUP BY
38              win.id,
39              win.label,
40              win.parent_id
41      ),
42      batch
43  ORDER BY
44      id
45  /
            MAX(    (   SELECT
ERROR at line 17:
ORA-01427: single-row subquery returns more than one rowWhich I do understand, I just don't know how to get round it.
Any ideas?
David