How to get columns from multiple rows in a single row in SQL - sql

Question

I want to get 2 columns col_a and col_b's values for min and max of some other column. For example:
id
last_updated
col_a
col_b
1
2021-01-01
abc
xyz
1
2021-01-02
abc_0
xyz_0
1
2021-01-03
abc_1
xyz_1
1
2021-01-04
abc_2
xyz_2
2
2021-01-01
abc
xyz
2
2021-01-01
abc
xyz
...
I want to get the result:
|1|abc|abc_2|xyz|xyz_2|
That is the result of grouping by id, and getting the values of these columns while putting the condition of min and max on some other column(last_updated).
I came up with the following query:
select id, max(last_updated), min(last_updated)
from my_table
group by id
This gives me the id and min and max dates but not the other 2 columns. I'm not sure how to get the values for the other 2 columns for both dates in same query.

Answer 1

You can use MIN and MAX analytical function as follows:
select id,
max(case when mindt = last_updated then col_a end) as min_col_a,
max(case when maxdt = last_updated then col_a end) as max_col_a,
max(case when mindt = last_updated then col_b end) as min_col_b,
max(case when maxdt = last_updated then col_b end) as max_col_b
from
(select t.*,
min(last_updated) over (partition by id) as mindt,
max(last_updated) over (partition by id) as maxdt
from your_table t) t
group by id

Answer 2

We can use ROW_NUMBER, twice, to find the first and last rows, as ordered by last_updated, for each id group of records. Then, aggregate by id and pivot out columns for the various col_a and col_b values.
WITH cte AS (
SELECT *, ROW_NUMBER() OVER (PARTITION BY id ORDER BY last_updated) rn_min,
ROW_NUMBER() OVER (PARTITION BY id ORDER BY last_updated DESC) rn_max
FROM yourTable
)
SELECT
id,
MAX(CASE WHEN rn_min = 1 THEN col_a END) AS col_a_min,
MAX(CASE WHEN rn_max = 1 THEN col_a END) AS col_a_max,
MAX(CASE WHEN rn_min = 1 THEN col_b END) AS col_b_min,
MAX(CASE WHEN rn_max = 1 THEN col_b END) AS col_b_max
FROM cte
GROUP BY id;

Answer 3

Not the neatest solution but demonstrates another way to obtain the data you want. We join the table on itself as we normally want data from 2 rows, then we use cross apply to restrict it to first and last.
select T1.id, T2.col_a, T1.col_a, T2.col_b, T1.col_b
from #my_table T1
inner join #my_table T2 on T1.id = T2.id
cross apply (
select id, max(last_updated) MaxLastUpdated, min(last_updated) MinLastUpdated
from #my_table
group by id
) X
where T1.last_updated = X.MaxLastUpdated and T2.last_updated = X.MinLastUpdated;
With the sample data provided this appear to perform worse than the row_number() solution. The fastest solution is the analytical functions.

Answer 4

select id,max(last_updated) last_updatedMax, min(last_updated) last_updatedMin,max(col_aMax) col_aMax, max(col_aMin) col_aMin,max(col_bMax) col_bMax, max(col_bMin) col_bMin
from
(
select
*
, first_value(col_a) OVER (PARTITION BY id ORDER BY last_updated desc) as col_aMax
, first_value(col_a) OVER (PARTITION BY id ORDER BY last_updated asc) as col_aMin
, first_value(col_b) OVER (PARTITION BY id ORDER BY last_updated desc) as col_bMax
, first_value(col_b) OVER (PARTITION BY id ORDER BY last_updated asc) as col_bMin
from my_table
) t
group by id