RAID VFS Module SoC Project - Requirements Doc

[Ming Wong]

Hello Peter,

On 01/06/06, Peter Stamfest <peter at stamfest.at> wrote:
> >> [snipped]
> >> So why bother setting up a SW RAID (more work), when all you need to do
>
> SW RAID most of the time is real easy to do (If you know how). Eg. RAID 1:
>
> 1. Buy a new disk and hook it up
> 2. Create a degraded RAID 1 device from it
> 3. Copy the data from the old disk to the new (degraded RAID)
> 4. Unmount the old disk, remount the RAID-based FS
> 5. Hot-add the old disk to the RAID and have it resync.
>
> This (minus the actual resyncing/copying) takes no more than 15 Minutes.
>

Yep, I know it is easy. But something even easier is achieved with the
vfs module here, for example, you can even make mirrors using
directories. It's flexible. :)
(Of course, more issues will need be taken care of as you mentioned below. )

> OTOH, a RAID VFS always has to be prepared to find differences between the
> RAID dirs, because some local user messes around with one of the copies.
>

Yes, that too needs be taken care of; however, it's not a major
concern at the moment.

> >> is, create a share on samba whilst having a redundant mechanism in place.
> >> :-)
>
> While laziness is one of the virtues of a *real* sysadmin, it should not
> be encouraged. ;-)
>

heh, that's right. I personally was an extremely lazy sysadmin. :-D

> And to really make this fly, you would always have to assure some
> properties of the RAID directories:
>
>   * Free space should always be equal on both RAID Filesystems (and not
>     just when samba starts or the RAID VFS gets set up). This basically
>     means that it only makes sense for Filesystems used exclusively for
>     that particular samba share.

I think this was mentioned in the requirement I drafted.

>   * Disallow access to the RAID directories behind the back of samba
>   * Recheck/Rebuild the RAID after every hard crash of the machine - how do
>     you find out about such a crash from within samba? Do you implement
>     some sort of RAID superblock? Where do you put it? In a file everybody
>     (or at least the sysadmin) can mess around with?
>

The *most* we could do so far is, check if files can be read from the
disks; if files reading failed after some certain trials, samba would
try to access the particular files from the other disks/paths.

>
> And: What do you do if a "RAID" disk *really* fails? The operating systems
> I know do not always deal gracefully with *that*. A process (samba) might
> then just wait for ages or just crash. Or the entire system crashes or at
> least requires a reboot to make sure it works again. This is just not what
> an OS gets prepared for. Real RAID is designed for this. So I think this
> might give a false sense of security: While the data may be safe due to
> the second copy, the system likely won't be any more reliable/stable.
> Basically it means: You get the data security properties of RAID-1 with
> the failure tolerance of a RAID-0. Might be called "RAID-0 minus 1"
> (RAID0-1)  ;-))

That's true. HW RAID can certainly deal with fails fabulously.
Speaking of reliability, I'd admit that if raid vfs is applied on the
primary drive(the one boots up the O/S and similar), it'd not be
reliable. And so does SW RAID.

Failure can possibly be handled iff the OS is not relying on the
disks, which are enabled for raid vfs, for its regular processes.

>
> Now I just doubt that this makes the setup/maintenance any easier. And if
> you tell a sysadmin to use the RAID VFS module you have to teach him/her
> ahout these properties. Better save the time and teach the sysadmin real
> RAID properties.
>

Well. it's really easy to use. I doubt it's going to take longer than
a few minutes to pick it up. ;-)  (Of course, the sysadmin must know
how to configure samba for sure. )

> > One thing I'd like to add is, I should admit that HW RAID is portable whilst
> > SW is extremely non-portable.
>
> I disagree. As long as you stay with one software platform, SW RAID is
> more portable than HW RAID (eg. in linux you can even take out IDE disks,
> put them into USB-IDE cases and attach them to a linux system that does no
> longer have enough IDE controllers. Not nice, but it *works*). And have
> you ever been in a situation where a HW RAID controller fails, just to
> find out that there is no replacement anywhere in the market?
>

Yep, thats right. SW RAID is portable with certain platform.

Heh, and luckily, I still haven't come across HW RAID controller
failure whilst having to desperately look for a replacement for the
same model of controller.

> But I disgress.
>
> > But with the implementation happening here, it has no concept of O/S
> > specific partitioning issues. So this RAID vfs module is going to be
> > portable.
>
> How often have you moved a disk from one operating system to another?
> While this happens, it certainly is not common.
>
Hmm, maybe with a major system breakdown?
Just making a joke. lol
I know it is not common to move a disk from one system to another,
however, it is a useable feature. :)

cheers,
Ming