The test innodb.instant_alter_crash is supposed to flush the redo log before each time it kills the server. For some reason, it does not seem to be doing that. I was only able to repeat
MDEV-18007 by adding a sleep before each kill:
This bug could be affecting earlier versions as well.
I used to blame this on
MDEV-15740, but it could also be something else. I think that this can be pretty serious. Basically, losing committed transactions is breaking the Durability in ACID.