SQL 查询，select 顶部结果或底部结果取决于行是否存在

Question

I need to select rows:

One for each distinct centerid and studentid
The column status is not 'D' and the oldest one (I use autoincremented id ).
If no such row exists, find the newest column with createdDate today and status='D'.
No row for that centerid, studentid, if no such row exists.

Below is the query I'm currently using:

SELECT *
FROM DailyJournals as DJ
WHERE IFNULL(
        (SELECT id FROM DailyJournals AS x WHERE DJ.studentid = x.studentid AND DJ.centerid = x.centerid
          AND NOT status = 'D' ORDER BY id LIMIT 1),
        (SELECT id FROM DailyJournals AS x WHERE DJ.studentid = x.studentid AND DJ.centerid = x.centerid
          AND createdDate = DATE(NOW()) ORDER BY id DESC LIMIT 1)
      ) = DJ.id;

Trying to find the best way to add indexes and structure the query and DB to keep this reasonably optimized while keeping it simple. Load is very small for now, and the table gets pruned, so performance isn't currently an issue.

MariaDB 10.4
centerid and studentid are 3 character alphanumeric strings. id is autoincremented indexed primary key. createdDate is a date string for when the row was created.

Some additional information:

(There are only a few possible values for column status and it's a 2 character string) which I've set up an indexed virtual boolean column for (done=0 or done=1)

alter table DailyJournals add column `done` bool as (status='D');
create index done on DailyJournals(done);

Answer 1

You can try using row_number() :

select dj.*
from (select dj.*,
             row_number() over (partition by centerid, studentid
                                order by (status <> 'D') desc,
                                         (case when status <> 'D' then id end) desc,
                                         id asc
                               ) as seqnum
      from dailyjournals dj
     ) dj
where seqnum = 1;

EDIT:

From a performance perspective, probably the best you can do is a computed column with an index.

alter table DailyJournals
    add column magic_d_id int as 
        (status = 'D' then id else - id end);

create index dailyjournals_magic_d_id on DailyJournals(centerid, dataid, magic_d_id);

You can try row_number() :

row_number() over (partition by centerid, studentid order by magic_d_id)

I'm not sure if that will use the index. You can also use:

select dj.*
from dailyjournals dj
where dj.magic_d_id = (select dj2.magic_d_id
                       from dailyjournals dj2
                       where dj2.centerid = dj.centerid and
                             dj2.studentid = dj.studentid and
                       order by dj2.magic_d_id
                       limit 1
                      );

This assumes that id is an integer and never negative.