On 9 January 2013 14:14, steve-ALUG@hst.me.uk wrote:
I spotted from the logs you're using ext4. Providing your kernel version is 2.6.30 or above it should be as robust as ext3.
As I understand it the biggest issue is RAID5 itself; a failure to write cached writes physically to disk will do horrible things to the RAID array.
I doubt that you could turn Hardware Write Caching off, and from the sounds of it, the disk remains powered even after the whoopsie, so any pending disk writes should get through once sent from disk controller.
It's a good point that hardware write caching should be ensured by the fact the drive retains power. (The caddy is on a UPS as well, which will help in other circumstances.)
Re: Software write caching.
sudo hdparm -i /dev/sdf
$ sudo hdparm -i /dev/sdf /dev/sdf: HDIO_GET_IDENTITY failed: Invalid argument
I assume that's due to it being on USB?
However: $ sudo hdparm -I /dev/sdf Commands/features: Enabled Supported: * SMART feature set Security Mode feature set * Power Management feature set Write cache
.. if that helps (I assume that's just confirming that hardware write caching is disabled).