Assigning sequential value that resets to a group of records

Question

Hello all, and happy FRIDAY!!!
I'm doing some data conversion for a new system and I'm having trouble coming up with a query.
Running Oracle 10.2
Sample Data:
create table test (GRP number, PART varchar2(20), SEQ number);
insert into test values(9000, 'lskdjf', null);
insert into test values(9000, 'alsdk', null);
insert into test values(9000, '492kjsfjsldk', null);
insert into test values(9000, 'lkjasdf0982j', null);
insert into test values(9001, 'likfjajsd', null);
insert into test values(9001, '234-092838', null);
insert into test values(8934, '000-192893aj', null);
insert into test values(8934, 'anotherpart', null);
insert into test values(8934, 'jjjj0-aa-2001', null);
insert into test values(8934, 'encifudy', null);
insert into test values(8934, 'asfdjslkjdfklsj', null);
insert into test values(8934, 'lksjdflj', null);
insert into test values(4736, 'l;ask---jdflasj', null);
commit;
Select * from test;Doing the select * at this point will give you the null data elements for the third column.
Problem:
I'm trying to run a query that spits out a sequence for each "group" of records. If there were 5 records in a group, I would want the sequence to start at 1 and go to 5... maybe ordered by part. (not quite sure the ordering, but need to do babysteps here) :P
Here is an example of the data output I'd want from the above example:
GRP     PART               SEQ
9000     lskdjf               1
9000     alsdk               2
9000     492kjsfjsldk          3
9000     lkjasdf0982j          4
9001     likfjajsd          1
9001     234-092838          2
8934     000-192893aj          1
8934     anotherpart          2
8934     jjjj0-aa-2001          3
8934     encifudy          4
8934     asfdjslkjdfklsj          5
8934     lksjdflj          6
4736     l;ask---jdflasj          1In that result, SEQ is not applied to the record in part order, but ultimately that might be something that I'd like to do. So far just working toward wanting to get those numbers in there. I have read using rankover(), but that is still a bit confusing to me on how that works. I'll go research it more as I suspect that is probably what I'm going to have to use.
Ultimately this is part of a cursor that is being fed through a bulk collect. I might be able to populate that SEQ in the PL/SQL, but I figured it'd be easier to just get them in place at the query level so they're already part of the collection, instead of having to build the logic to create that numbering system prior to insertion to a staging table.
Anyhoo... any help would be greatly appreciated! If I left anything out or am un-clear in anyway, please let me know! Thank you!
Edited by: dvsoukup on Jul 27, 2012 4:25 PM

Answer 1

Hi,
dvsoukup wrote:
Running Oracle 10.2
Sample Data:
create table test (GRP number, PART varchar2(20), SEQ number);
insert into test values(9000, 'lskdjf', null); ...
Thanks for posting your version, and the CREATE TABLE and INSERT statemen ts; that's very helpful!
>
Doing the select * at this point will give you the null data elements for the third column.
Problem:
I'm trying to run a query that spits out a sequence for each "group" of records. If there were 5 records in a group, I would want the sequence to start at 1 and go to 5... maybe ordered by part. (not quite sure the ordering, but need to do babysteps here) :P
I have read using rankover(), but that is still a bit confusing to me on how that works. You've got that right! Analytic functions are very strange looking and confusing at first. After a while, they don't seem so strange, and then they even get less confusing.
One thing to remember: "PARTITION BY x" (this clause is always optional) means that each value of x is a world unto itself. It's as if a separate query is being done for each value of x, and when they're all finished, the results are UNIONed together. In this case, you want to PARTITION BY grp.
I'll go research it more as I suspect that is probably what I'm going to have to use.You're very close; only, as the first reply said, it's not RANK that you want, but its close relative ROW_NUMBER. The difference between the two is how they handle ties. If you have duplicate data, RANK assigns duplicate numbers. I assume you want unique numbers, even if you have identical parts in the same grp.
Ultimately this is part of a cursor that is being fed through a bulk collect. I might be able to populate that SEQ in the PL/SQL, but I figured it'd be easier to just get them in place at the query level so they're already part of the collection, instead of having to build the logic to create that numbering system prior to insertion to a staging table.It's easy enough to assign the numbers when you build the table.
If the situation is what you posted, that is, your table already exists, but the seq number isn't populated yet, then you can do this:
MERGE INTO     test     dst
USING   (
         SELECT  grp
         ,         part
         ,         ROW_NUMBER () OVER ( PARTITION BY  grp
                                                ORDER BY      part
                           )       AS seq
         FROM    test
     )          src
ON     (     src.grp          = dst.grp
     AND     src.part     = dst.part
WHEN MATCHED THEN UPDATE
SET     dst.seq   = src.seq
;The MERGE statement above assumes that the combination (grp, part) is unique.
After you run it, this query:
SELECT       *
FROM       test
ORDER BY  grp
,            part
;will produce these results:
`      GRP PART                        SEQ
      4736 l;ask---jdflasj               1
      8934 000-192893aj                  1
      8934 anotherpart                   2
      8934 asfdjslkjdfklsj               3
      8934 encifudy                      4
      8934 jjjj0-aa-2001                 5
      8934 lksjdflj                      6
      9000 492kjsfjsldk                  1
      9000 alsdk                         2
      9000 lkjasdf0982j                  3
      9000 lskdjf                        4
      9001 234-092838                    1
      9001 likfjajsd                     2The USING clause above is almost the same as the query posted in the last message, but, in the analytic clause, instead of "ORDER BY grp, part" it only says "ORDER BY part". It never makes any sense to PARTITION BY and ORDER BY the same column in the same function. Why? Discuss.