Bug#563602: I had a similar issue

[Виталий]

Hi!

I was running MySQL InnoDB databases on mdadm RAID5 partition with XFS  
file system when I discovered that if /etc/cron.d/mdadm (checkarray) of  
this partition concurs with some database updates, database file is  
corrupted and needs some kind of recovery, which is not always simple.

This bug does reproduce on my server with, for example, on FeedOnFeeds  
database (server-side rss aggregator) when it's cronjob is updating rss  
feeds.

RAID array, though, does not degrade, and it seems that MyISAM  
(non-transactional) tables are OK even when checkarray concurs with  
database updates. So I suppose the data loss is caused by some  
sync/barrier issues... Is that so? And if so, what I'm supposed to do to  
get rid of this problem?

-- 
With best regards,
   Vitaliy Filippov