[nmglug] Production server question

[Brendan]

Gary Sandine wrote:
> I very much agree with all of this.  However, I find hot-swapping
> out failed drives to be much easier with certain hardware RAID cards
> (I have experience with 3ware cards, and it has been easy for the
> most part).  Unless something has changed, doing that safely with
> software RAID requires mdadm commands and echoing carefully crafted
> remove-single-device and add-single-device commands to
> /proc/scsi/scsi followed by more mdadm commands.  It's
> straightforward but not for everyone.


For SCSI, that's true, but SATA (generally) has native hot swap 
capability without notifying the disk subsystem.  It's still good form 
to tell md that you're yanking out the failed device, though (as you 
suggest).  Hopefully you're always running with a hot spare (or second 
mirror) so there's no in-the-head-of-the-moment, drive swapping going 
on.  The only thing worse than having no RAID is, after RAID saves your 
ass, accidentally crashing the system anyhow and being unable to restart 
your RAIDs because a drive is missing ;-)

-Brendan