Need to maintain "Flag" column

Question

Hello,
I am trying to maintain a "Flag" that identifies an address as the primary. The latest updated address for each member_id should be the primary with its primary_flag=Y and all the others for the member_id set to N. I have added a small sample of the data but my table has 2.3 million rows so performance is key. The records are updated and inserted using a MERGE query into my members table from an external table and this works very well.
I am on 11gR1
After my MERGE this is what I might see:
member_id    address_id  primary_flag   last_updated
     12         232          N             1/2/2011
     12         324          Y             2/14/2011
     12         376          N             2/29/2011
     25         256          N             1/13/2011
     28         278          N             1/20/2011
     28         291          Y             1/25/2011
     30         389          Y             3/1/2011What I would like to see:
member_id    address_id  primary_flag   last_updated
     12          232          N             1/2/2011
     12          324          N             2/14/2011
     12          376          Y             2/29/2011
     25          256          Y             1/13/2011
     28          278          N             1/20/2011
     28          291          Y             1/25/2011
     30          389          Y             3/1/2011I have come up with a few 2-Pass (SQL Updates) but I would like to do it in one (It would be real nice if I could do it in the MERGE statement).
Thanks for any help!

Answer 1

Hi,
Thanks for posting the sample data. The results you said you wanted included only 4 columns; don't you want to set the other columns, in particular last_update_date, as well? I included all columns in the INSERT nad UPDATE; you can easily remove them if you don't want them.
This does what I think you want:
merge into
    MEMBER_ADRESS MA
    using (
     WITH     new_data     AS
        select
         NULL                    AS address_id,
            ML.MEMBER_ID,
            SUBSTRB (ML.ADDRESS_1, 1, 30)     AS address_1,
            SUBSTRB (ML.ADDRESS_2, 1, 30)     AS address_2,
            SUBSTRB (ML.CITY,         1, 25)     AS city,
            ML.STATE,
            ML.ZIP_CODE          AS zipcode,
         SYSDATE          AS creation_date,
         SYSDATE          AS last_update_date,
         NULL          AS primary_flag,
            ML.ADDITIONAL_COMPANY_INFO,
            to_number ( regexp_replace ( ML.ADDITIONAL_COMPANY_INFO     ||
                          SUBSTRB (ML.ADDRESS_1, 1, 30)     ||
                          SUBSTRB (ML.ADDRESS_2, 1, 30)     ||
                          SUBSTRB (ML.CITY,     1, 25)     ||
                          ML.STATE                ||
                          ML.ZIP_CODE
                          , '[^[:digit:]]'
                )      as MATCH_VALUE,
         NULL          AS match_cnt
        from MEMBER_LOAD_V2 ML
     ,     primary_flag_y     AS
          SELECT     o.*
          ,     COUNT ( CASE
                        WHEN  o.match_value = n.match_value
                        THEN  1
                    END
                     )  OVER ( PARTITION BY  o.member_id
                          )          AS match_cnt
          FROM     member_adress     o
          JOIN     new_data     n  ON       o.member_id     = n.member_id
          WHERE     o.primary_flag     = 'Y'
     ,     combined_data     AS
          SELECT  *
          FROM     primary_flag_y
          WHERE     match_cnt     = 0
         UNION
          SELECT     *
          FROM     new_data
     SELECT     address_id, member_id, address_1, address_2, city, state
     ,     zipcode, creation_date, last_update_date
     ,     CASE
               WHEN  ROW_NUMBER () OVER ( PARTITION BY  member_id
                                           ORDER BY        last_update_date  DESC
                               ) = 1
               THEN  'Y'
               ELSE  'N'
          END               AS primary_flag
     ,     additional_company_info, match_value
     FROM     combined_data
     ) L
on (MA.MEMBER_ID = L.MEMBER_ID and MA.MATCH_VALUE = L.MATCH_VALUE)
when matched then
update set
    MA.ADDRESS_ID          = L.ADDRESS_ID,
    MA.ADDRESS_1           = L.ADDRESS_1,
    MA.ADDRESS_2           = L.ADDRESS_2,
    MA.CITY                = L.CITY,
    MA.STATE                = L.STATE,
    MA.ZIPCODE                = L.ZIPCODE,
    MA.CREATION_DATE          = L.CREATION_DATE,
    MA.LAST_UPDATE_DATE          = L.LAST_UPDATE_DATE,
    MA.PRIMARY_FLAG           = L.PRIMARY_FLAG,
    MA.ADDITIONAL_COMPANY_INFO      = L.ADDITIONAL_COMPANY_INFO
when not matched then
insert (MA.ADDRESS_ID, MA.MEMBER_ID, MA.ADDRESS_1, MA.ADDRESS_2, MA.CITY, MA.STATE, MA.ZIPCODE,
        MA.CREATION_DATE, MA.LAST_UPDATE_DATE, MA.PRIMARY_FLAG, MA.ADDITIONAL_COMPANY_INFO)
values (
        MEMBER_CNTCT_SEQ.NEXTVAL,
        L.MEMBER_ID,
        L.ADDRESS_1,
        L.ADDRESS_2,
        L.CITY,
        L.STATE,
        L.ZIPCODE,
     L.CREATION_DATE,
     L.LAST_UPDATE_DATE,
        L.PRIMARY_FLAG,
        L.ADDITIONAL_COMPANY_INFO
        );The cruical part is that, for each member_id in the source table, it finds the row with primary_flag='Y' that's already in the table (if a row with with the same member_id and match_value isn't already in the source table), and does a UNION to include those rows, so that an existing primary_flag may get changed from 'Y' to 'N'.
It looks like the address_1, address_2 and city columns are the same length in both tables, so I don't see why you're using SUBSTR. I left it in, in case you have some reason.
You might consider a user-defined function to caluculate match_value, and use that function bioth in the definition of the virtuyal column and in the MERGE. That way, you can be sure it is being computed the same in both places, and if, in the future, you change how match_value is computed, you only need to change it in one place. Match_value looks like just the kind of thing you might refine in the future, For example, you might decide that, on the rare occassions when city or state contains digits, it suits your needs better to ignore those digits.