My table looks something like this:
datetime | field_a | field_b | field_c | field_d | field_e | field_f | updated_at
Actually, the number of fields is larger than that, more about 20, the a-f numbering is just for brevity.
This table is updated on a regular basis and the same rows can appear more than once but with more recent values of updated_at
.
What I want to achieve is to select rows with the most recent updated_at
so as to avoid duplicates (rows A and are duplicates if the only difference is the value of updated_at
).
My initial attempt is something like this:
WITH temp AS (
SELECT *,
ROW_NUMBER() OVER (PARTITION BY datetime, field_a, field_b, ... field_f ORDER BY updated_at DESC) rnk
FROM some_table)
)
SELECT * FROM temp WHERE rnk = 1
At first, I had thought that using datetime
in the PARTITION BY
clause might be enough, but it seems that I have to include all the fields so that the desired deduplication can happen.
Does this approach make sense? Am I correct in that all fields should be included in the window function? Is there a more elegant way to achieve what I want?
Sample input:
datetime | field_a | field_b | field_c | field_d | field_e | field_f | updated_at
2022-04-05 | a | b | c | d | e | f | 2022-04-05T20:11:42.864086
2022-04-05 | a | b | c | d | e | f | 2022-04-05T20:22:42.864086
2022-04-04 | a | b | c | d | e | f | 2022-04-05T19:11:42.864086
2022-04-04 | a | b | c | d | e | f | 2022-04-05T19:22:42.864086
The query should return:
2022-04-05 | a | b | c | d | e | f | 2022-04-05T20:22:42.864086
2022-04-04 | a | b | c | d | e | f | 2022-04-05T19:22:42.864086
That is, rows where all fields are the same (except for updated_at
), and updated_at
is the largest. In other words, the most recent row for each unique combination of (datetime, field_a, field_b, field_c, field_d, field_e, field_f)
.
Consider below approach
select * from your_table t
qualify 1 = row_number() over win
window win as (partition by to_json_string((select as struct * except(updated_at) from unnest([t]))) order by updated_at desc)
if applied to sample data in your question - output is