I filed the following bug today. Basically it looks like in certain conditions Innodb may be started twice in the parallel on the same data files which can hurt database badly. It is true it does not seems to be 100% repeatable but I’ve seen it multiple times in different environments. I’m not sure if this is some platform which has issues with locking, Some race condition which allows to copies to start or anything else.
In any case you’we been warned – be careful and ensure MySQL is not already started when you’re thinking to start one. Now when
MySQL starts to listen to the socket after Innodb Recovery is completed it is very easy to make a mistake and think MySQL is not running at all when recovery is in progress.