Query stuck in a loop between two tables in evaluate_join_record

The query

PREPARE stmt FROM 'SELECT 5 AS Y, 6 AS M

FROM orders AS o

INNER JOIN (

    SELECT orderID

    FROM orderItems

    GROUP BY orderID

    ) AS oi ON oi.orderID = o.id

INNER JOIN addresses AS a ON a.id = o.shippingAddressID

WHERE o.isInStats = 1 AND a.countryID = ? AND YEAR(o.orderTime) = ?

ORDER BY 1 DESC, 2 DESC;';

EXECUTE stmt USING '1','2025';

EXECUTE stmt USING '50','2025';

EXECUTE stmt USING '65','2025';

EXECUTE stmt USING '151','2025';

is stuck in a loop sorting keys. This is a regression from 10.6 to 11.4.

There is a chat thread here about it: https://mariadb.zulipchat.com/#narrow/channel/118759-general/topic/Debug.20long.20running.20query/with/523721319

I am not yet able to produce a data set that I can attach here, but the current sql-file is 306M and if I decrease the number of inserts with around 1000 inserts, the query starts to behave.

Workarounds:
1) SET optimizer_switch='extended_keys=off' No longer works, I assume that dataset grew a bit.
2)

SET SESSION tmp_memory_table_size = 16384; SET SESSION max_heap_table_size = 16384;

3)

SET SESSION max_heap_table_size = 19198290;

Our default value of max_heap_table_size is 256M, it seems that going from the default 16M to 32M breaks the query.

Both work separately. 1) does not work when I replicate via docker, only on CloudLinux.

Changes that manage to make the query not ending up in a loop:
1) If I run the queries without prepare, they work.
2) They have to be in that order, if not, it just works.
3) By removing the

GROUP BY orderID

it works.