On 06/19/2012 09:13 PM, Michael Butash wrote:
> So yeah, no raid is perfect...
>
> I've been using software raid1 (md) for a while now for my desktops and
> laptops work and home, and since my adventures in ati gpu land, I've
> twice now had video software/hardware cause my software raid to fail
> ugly, but both times survivable while I rebuilt the array manually. This
> was just a few days ago the last...
>
> Both times were using GL functions (this time toggling compositing
> on/off, last time i think minecraft) that caused the ati fglrx drivers
> to spew hardware errors seeming to glitch the card itself. Two separate
> cards as well now. Getting back into desktop went into visa with gpu
> unavailable. Then I saw my raid was degraded, again, same timestamp as
> the gpu glitch.
>
> First time prior one of the two disks in the md for boot went offline,
> simply added sdb1/2 back. This time one partition on each disk to the
> two md's (boot/else) to go offline alternatively (sda1/sdb2) - very odd.
> The second disk wouldn't respond to hdparm/fdisk query until a reboot
> that was done very hesitantly and not before I backed up anything I
> cared about to an nfs share. Data on both remained available which was
> really the odd part.
>
> To its testament, it rebooted, both disks reported healthy (hdparm,
> ubuntu disk utility), I re-added each partition, let it rebuild, and
> works again. Still worries me as my last set of ssd disks got unstable
> on one after less than 9 months of use and I'm probably about there with
> these that are known to get cranky. Smart reports them as ok, so I
> wonder how bad ati taints the kernel space that it causes disk
> controller/driver exceptions.
>
> Moral of story: know when/how to repair whatever raid, as software and
> hardware are seemingly still prone to exception from unlikely places.
> Last time a disk died with md, I just mounted the secondary in an
> enclosure, copied off data as pluggable, and copied to the new pair of
> raid disks. Hardware is never this easy, especially fakeraids.
>
> -mb
I use software raid strictly on servers, which are headless (of course).
I keep data on a server when practical, and run a daily rsync (offsite)
backup of data on my workstation. Interesting to note the problems with
video though. You might want to hop onto the raid list
(gmane.linux.raid) and see what they think about it.
I don't know why anyone would run SSDs in a raid. RAID IS NOT A BACKUP.
Of course if you're concerned about device failure and need to maintain
continuity, raid is entirely appropriate.
One of the nice things about sw raid-1 is that as long as you have one
good drive/partition, then you can recover. Just start the array in
degraded mode, and you're good to go. I know that grub can access a
single raid-1 partition w/out starting the array, which makes me wonder
if you can simply mount one of the raid-1 partitions straight away w/out
starting the array. I should try that so I know for sure if it's
possible or not.
BL, *never* use fakeraid, and avoid raid-5 if possible. Disk space is no
longer expensive enough to justify using raid-5.
--
-Eric 'shubes'
---------------------------------------------------
PLUG-discuss mailing list -
PLUG-discuss@lists.plug.phoenix.az.us
To subscribe, unsubscribe, or to change your mail settings:
http://lists.PLUG.phoenix.az.us/mailman/listinfo/plug-discuss
(text/plain)