雪花条件 Window Function

Question

I am trying to write a window function to help me retrieve a single field from a second table that I need to join to my existing table. The issue is the only way to figure out which value out of many possible values is the correct one requires matching two IDs and then out of those options (options where IDs match) pulling the most recent one that is not before a date (which is different for each record and pulled from the initial table).

Right now I have written:

Select distinct primary_id,
       first_value(desired_column) over partition by id_1, id_2, order by date desc)
From base_table
Left join second_table 
on second_table.id_1 = base_table.id_1 and 
   second_table.date <= base_table.date

However, this is still returning incorrect values. The returned table should have the same row count as base_table, but with the desired_column added based on whichever record matches ids but also happens before the base_table date (each desired_column value should be one result, the most recent one before the base_table date that matches the ids). This has the same row count, but it's returning desired_column values that are completely incorrect (I suspect that it is because I don't break down the second date <= base in the window function directly, but that isn't possible? I'm not sure how to proceed.)

Thank you in advance.

Edit to add:

Sample Base Table

Primary Key	ID1	ID2	Date
1	123	321	01/22/2021
2	123	654	09/02/2022
3	234	432	02/02/2019

Sample Second Table

Desired_Column	ID1	ID2	Date
q	123	321	01/21/2021
r	123	654	09/03/2022
w	234	432	02/01/2019
s	234	432	03/20/2022
a	123	439	02/20/2022
w	999	999	09/10/2022
null	234	987	10/10/2020

Desired Output

Primary Key	ID1	ID2	Date	Desired_Column
1	123	321	01/22/2021	q
2	123	654	09/02/2022	null
3	234	432	02/02/2019	w

Answer 1

so making some CTE's for the data:

with base_table(primary_id, ID_1, ID_2, Date) as (
    select * from values
    (1, 123, 321, '01/22/2021'::date),
    (2, 123, 654, '09/02/2022'::date),
    (3, 234, 432, '02/02/2019'::date)
), second_table(Desired_Column, ID_1, ID_2, Date) as (
    select * from values
    ('q'    ,123, 321, '01/21/2021'::date),
    ('r'    ,123, 654, '09/03/2022'::date),
    ('w'    ,234, 432, '02/01/2019'::date),
    ('s'    ,234, 432, '03/20/2022'::date),
    ('a'    ,123, 439, '02/20/2022'::date),
    ('w'    ,999, 999, '09/10/2022'::date),
    (null   ,234, 987, '10/10/2020'::date)
)

and then correcting your SQL:

Select distinct b.primary_id,
       first_value(s.desired_column) over (partition by b.id_1, b.id_2 order by s.date desc)
From base_table as b
Left join second_table as s
on s.id_1 = b.id_1 and 
   s.date <= b.date

gives:

PRIMARY_ID	FIRST_VALUE(S.DESIRED_COLUMN) OVER (PARTITION BY B.ID_1, B.ID_2 ORDER BY S.DATE DESC)
1	q
2	a
3	w

but the distinct is the hint, this is not the method you are looking for...

dropping the FIRST_VALUE which get a result for very row, and using a QUALIFY and ROW_NUMBER to RANK the rows, just keep the best (aka 1)

Select b.primary_id,
       s.desired_column
From base_table as b
Left join second_table as s
on s.id_1 = b.id_1 and 
   s.date <= b.date
qualify row_number() over (partition by b.id_1, b.id_2 order by s.date desc) = 1

gives:

PRIMARY_ID	DESIRED_COLUMN
1	q
2	a
3	w

but also allow accessing all the other values from the two tables:

Select b.*,
       s.*   
From base_table as b
Left join second_table as s
on s.id_1 = b.id_1 and 
   s.date <= b.date
qualify row_number() over (partition by b.id_1, b.id_2 order by s.date desc) = 1

And given you want to match "both ID's" you should use this SQL:

Select b.primary_id,
    b.id_1,
    b.id_2,
    s.desired_column 
From base_table as b
Left join second_table as s
on s.id_1 = b.id_1 and 
   s.id_2 = b.id_2 and 
   s.date <= b.date
qualify row_number() over (partition by b.id_1, b.id_2 order by s.date desc) = 1