Update the top row of each group - Oracle - oracle

Question

The need is to update only the top row of each group of a table from the data of other table.
I need to update table A with details from table B
Table A
---------
ID Name Date PCNO
1 abc 1/1/12 123
2 def 1/1/12 234
3 fgh 1/2/12 222
4 asd 1/2/12 234
TABLE B
-----------
ID Name Date PCNO
1 adsf 1/1/12 4343
2 sdf 1/2/12 9347
For each top record of table A grouped by "Date" and ordered by PCNO desc, I would like to update the values from table B.
Do i use rank for this purpose.?

Answer 1

You can use rank or dense-rank (or even row-number) to get identify the 'top' row, though you may need t consider what to do if ties are possible in your real data:
select a.id, a.name, a.date_col, a.pcno,
dense_rank() over (partition by date_col order by pcno desc) as rnk
from table_a a;
ID NAME DATE_COL PCNO RNK
---------- ---- ---------- ---------- ----------
2 def 2012-01-01 234 1
1 abc 2012-01-01 123 2
4 asd 2012-01-02 234 1
3 fgh 2012-01-02 222 2
And you can join to table B to get the new values for the top-ranked:
select a.id, a.name, a.date_col, a.pcno,
dense_rank() over (partition by a.date_col order by a.pcno desc) as rnk,
case when dense_rank() over (partition by a.date_col order by a.pcno desc) = 1
then b.name else a.name end as new_name,
case when dense_rank() over (partition by a.date_col order by a.pcno desc) = 1
then b.pcno else a.pcno end as new_pcno
from table_a a
join table_b b on b.date_col = a.date_col;
ID NAME DATE_COL PCNO RNK NEW_ NEW_PCNO
---------- ---- ---------- ---------- ---------- ---- ----------
2 def 2012-01-01 234 1 adsf 4343
1 abc 2012-01-01 123 2 abc 123
4 asd 2012-01-02 234 1 sdf 9347
3 fgh 2012-01-02 222 2 fgh 222
and you can then use that in a merge statement:
merge into table_a target
using (
select a.id, a.name, a.date_col, a.pcno,
dense_rank() over (partition by a.date_col order by a.pcno desc) as rnk,
case when dense_rank() over (partition by a.date_col order by a.pcno desc) = 1
then b.name else a.name end as new_name,
case when dense_rank() over (partition by a.date_col order by a.pcno desc) = 1
then b.pcno else a.pcno end as new_pcno
from table_a a
join table_b b on b.date_col = a.date_col
) source
on (source.id = target.id)
when matched then update
set target.name = source.new_name, target.pcno = source.new_pcno
where source.rnk = 1;
or maybe
merge into table_a target
using (
select a.id, a.name, a.date_col, a.pcno,
case when dense_rank() over (partition by a.date_col order by a.pcno desc) = 1
then b.name else a.name end as new_name,
case when dense_rank() over (partition by a.date_col order by a.pcno desc) = 1
then b.pcno else a.pcno end as new_pcno
from table_a a
join table_b b on b.date_col = a.date_col
) source
on (source.id = target.id)
when matched then update
set target.name = source.new_name, target.pcno = source.new_pcno
where target.name != source.new_name or target.pcno != source.new_pcno;
either of which reports 2 rows merged, and then:
select * from table_a;
ID NAME DATE_COL PCNO
---------- ---- ---------- ----------
1 abc 2012-01-01 123
2 adsf 2012-01-01 4343
3 fgh 2012-01-02 222
4 sdf 2012-01-02 9347
You may need to adjust it if there isn't always going to be a match for a date, though the inner join ought to take care of that.
db<>fiddle demo