On 10/17/13 21:34, Scott McEachern wrote:
> On 10/17/13 20:57, Nick Holland wrote:
>> with the exception of the fact there's no code to rebuild a failed
>> disk, works great. that's a pretty big exception for most people. :)
>
> Hmm. That would present a problem.
>
> Let me make sure I'm absolutely clear here:
>
> A RAID5 array with four disks. I notice via bioctl(8) that one has
> failed. I pull the failed disk and replace it with my cold spare. I
> cannot use bioctl -R to incorporate the new disk into the array.
> Correct? (BTW, I know for a fact it works properly with RAID1.)
yep. works great for RAID1, not for RAID5.
If you think about it, it's a pretty different process between the two.
The RAID5 process is not yet written.
> So basically, if a drive fails, the RAID5 array is permanently borked
> until completely recreated with a new array, meanwhile I'll be trying to
> "back up" my data somewhere else.
yep. basically, your RAID5 turns into RAID0 (with a performance hit),
and you will need to dump/reload your data before it could be redundant
again.
Not that this might be without use. At my day job, I've got a machine
set up that people use for analyzing large chunks of data. They copy it
in, crunch it, then dump it. We are using RAID0 at the moment, but
softraid RAID5 could have been used to give a little more cushion when
the disk fails, so at least the current analysis could be continued.
>> you should be trying this stuff yourself. Doesn't matter if *I* can
>> rebuild your array, it matters if YOU can. You don't have to practice
>> on 3TB disks, you can practice on 10GB disks...though understanding
>> the time required for 3TB disks would be wise.
>
> As you've often advised in the past, test it yourself.
hey, I'm redundant! :D
> I plan on it,
> just to see what happens. BTW, I tried it once with a 3TB RAID1, and I
> believe it took two days, but I could be wrong and is
> hardware-dependent. Either way, it takes a /long/-ass time.
exactly. you don't want to promise "replace the drive tonight and
tomorrow all will be great" if it takes two days. RAID5 should have a
much longer rebuild time (RAID1: read once, write once. N disk RAID5:
read N-1, write once).
Also keep in mind that disk system performance during rebuild may be
unacceptable for some applications.
>> other than a 3TB disk is closer to 2.75TB than 3TB, yeah the math
>> works the same with softraid as it does with hw raid.
>
> Yes, sorry for not making clear I realize that. I didn't want to do the
> usable space, GB/GiB, TB/TiB dance. "You know what I mean..", hopefully.
yes, I knew what you meant. I'm just pissed at disk vendors. it's been
what, over 15 years they've been doing this? I'm still pissed. :)
> Either way, while I have the opportunity, I'm going to create some RAID5
> arrays of varying sizes, pull ("fail") disks, etc., and just see what
> happens for myself. Experimenting is fun!
>
> Unfortunately, if RAID5 cannot rebuild anything, it's about as useful as
> a leaky screwdriver and RAID1 remains the only viable option on
> OpenBSD. Damn.
>
> Thanks Nick, as always you're a gem of a resource.
:)
Nick.
