Probably because of a change made in MCOL-1403.
Using a padded string (i.e., "abc ") in a where clause should compare favorably with a column that has no padding. It no longer works that way.
regression: dictionary de-duplication cache bleeding between columns
Different trailing behaviour between columnstore and innodb
multibyte space (zenkaku) breaks strings comparisation
Fix remaining errors in working_tpch1
Fix for MCOL-1559 introduced performance lost
Replace glibc with google's re2 for regex processing