RAID5 to RAID6 migration

RAID5 to RAID6 migration

[Michał Przyłuski]

Good day,

I've been a happy user of md RAID5 for over 8 months now. However,
a few days ago, one (I wonder which one too) of my (3) WD RE2 750s
decided to start clicking every few seconds during write. No disk
shows any abnormalities from others on SMART tests, and raid5
sync_action check, checks all right. Seems it's all safe for now, it's
raid5 after all, but that brought to my attention a need, and
possibility, to improve my redundancy, by migrating to RAID6. I was
planning to grow the array anyways. Therefore, I sit back to google,
and encountered a few issues I couldn't really resolve myself.

First of all, what is the current state of RAID5->6 conversion?
I've went through a lot of blogs, and it seem that this matter has
been considered log ago to be added to mdadm, however I was unable to
find any definitive information about that for now. I've read, well
skimmed, latest mdadm source and I cannot really say I see it there.
What's more, I recall reading about future possibility (or lack of it)
to change the raid5/6 chunk size, in a way of reshape. I've not found
any updates on that since then. It's mostly out of curiosity, as it
seems I'll have to totally recreate my raid anyways.

My second question, is a bit of theoretical one.
I've read "The mathematics of RAID-6" by H. Peter Anvin (raid6.pdf).
In section 4, he describes a situation of a single disk *corruption*.
That makes me wonder, if md implementation, during a sync_action
repair/check follows that procedure. I remember reading up on raid5,
which of course cannot properly detect & correct such things, that all
it can do (again, of course) is set a new parity block. That's why
I'm wonder if raid6 repair/check does the same, just setting new
parity blocks, or it tries to detect which one is erroneous and take
appropriate actions.


Now, back to the point. What should I do? :) I want to migrate my
1.1TiB of data, from 3 hdd RAID5 to, most likely, a 5 hdd RAID6.
I guess it's the very last moment, size-of-data-wise, to accommodate
all data during migration.

I see the following ways, supposing there's no reshape from raid5
to raid6.

1. Copy stuff to 2 new hdds (not raid'ed), create a slightly degraded
raid6 out of old 3 hdds (making raid6 looks like it should have had 4),
copy stuff over to raid6, and then add/reshape/etc.

2. Create a totally degraded raid6 with 2 new hdds, lacking 2 drives.
Copy from current, operational, raid5, and then add/reshape/etc.

3. (risky) Check current raid5, and if it's checks ok, degrade it!
Create a 4 hdd raid6 (4 out of 4 operational), copy stuff from
degraded raid5, add/reshape/etc.

I'll be thankful for any kind of input on that matters.
Kind regards,
MiKylie
// Michał Przyłuski

Re: RAID5 to RAID6 migration

[Conway S. Smith]

On Mon, 4 Aug 2008 14:44:41 +0200
"Michał Przyłuski" <mikylie@gmail.com> wrote:
> Good day,
> 
> I've been a happy user of md RAID5 for over 8 months now. However,
> a few days ago, one (I wonder which one too) of my (3) WD RE2 750s
> decided to start clicking every few seconds during write. No disk
> shows any abnormalities from others on SMART tests, and raid5
> sync_action check, checks all right. Seems it's all safe for now,
> it's raid5 after all, but that brought to my attention a need, and
> possibility, to improve my redundancy, by migrating to RAID6. I was
> planning to grow the array anyways. Therefore, I sit back to google,
> and encountered a few issues I couldn't really resolve myself.
> 
> First of all, what is the current state of RAID5->6 conversion?
> I've went through a lot of blogs, and it seem that this matter has
> been considered log ago to be added to mdadm, however I was unable
> to find any definitive information about that for now. I've read,
> well skimmed, latest mdadm source and I cannot really say I see it
> there. What's more, I recall reading about future possibility (or
> lack of it) to change the raid5/6 chunk size, in a way of reshape.
> I've not found any updates on that since then. It's mostly out of
> curiosity, as it seems I'll have to totally recreate my raid
> anyways.
> 

My understanding from when I asked a while ago is that RAID5->6
is still a planned feature, but shouldn't be expected or counted on
for the foreseeable future.  I don't know anything about changing
chunk size on a live array.

Why are you thinking you'll have to totally recreate your raid anyway?
(I'm assuming you meant even if RAID5->6 reshape was possible, you'd
still be recreating it instead of reshaping).

> <snip>
> 
> Now, back to the point. What should I do? :) I want to migrate my
> 1.1TiB of data, from 3 hdd RAID5 to, most likely, a 5 hdd RAID6.
> I guess it's the very last moment, size-of-data-wise, to accommodate
> all data during migration.
> 
> I see the following ways, supposing there's no reshape from raid5
> to raid6.
> 
> 1. Copy stuff to 2 new hdds (not raid'ed), create a slightly
> degraded raid6 out of old 3 hdds (making raid6 looks like it should
> have had 4), copy stuff over to raid6, and then add/reshape/etc.
> 
> 2. Create a totally degraded raid6 with 2 new hdds, lacking 2
> drives. Copy from current, operational, raid5, and then
> add/reshape/etc.
> 
> 3. (risky) Check current raid5, and if it's checks ok, degrade it!
> Create a 4 hdd raid6 (4 out of 4 operational), copy stuff from
> degraded raid5, add/reshape/etc.
> 

I'd actually lean towards a fourth option: keeping the existing
RAID5, possibly reshaped to add more space on a new disk, and then 
have another new disk as a hot spare.  Not as good redundancy as
moving to RAID6, but much simpler w/ the current md feature set.
Except above you said you thought you'd have to recreate the array
anyway, in which case you may as well go to RAID6 now.  Of the three
methods you show, I think #2 looks best, as it only involves one copy
step.  The totally degraded RAID6 will be slow, but I doubt as slow
as another copy step.


Good luck,
Conway S. Smith
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: RAID5 to RAID6 migration

[Michal Przyluski]

Thank you for your agile response, 

On Monday 04 August 2008 15:57, Conway S. Smith wrote:
> Why are you thinking you'll have to totally recreate your raid anyway?
> (I'm assuming you meant even if RAID5->6 reshape was possible, you'd
> still be recreating it instead of reshaping).

No actually, I'd really prefer to  reshape it, but I've been somehow 
suspecting the bad news incoming.

> > I see the following ways, supposing there's no reshape from raid5
> > to raid6.
> >
> > 1. Copy stuff to 2 new hdds (not raid'ed), create a slightly
> > degraded raid6 out of old 3 hdds (making raid6 looks like it should
> > have had 4), copy stuff over to raid6, and then add/reshape/etc.
> >
> > 2. Create a totally degraded raid6 with 2 new hdds, lacking 2
> > drives. Copy from current, operational, raid5, and then
> > add/reshape/etc.
> >
> > 3. (risky) Check current raid5, and if it's checks ok, degrade it!
> > Create a 4 hdd raid6 (4 out of 4 operational), copy stuff from
> > degraded raid5, add/reshape/etc.
>
> I'd actually lean towards a fourth option: keeping the existing
> RAID5, possibly reshaped to add more space on a new disk, and then
> have another new disk as a hot spare.  Not as good redundancy as
> moving to RAID6, but much simpler w/ the current md feature set.
> Except above you said you thought you'd have to recreate the array
> anyway, in which case you may as well go to RAID6 now.  Of the three
> methods you show, I think #2 looks best, as it only involves one copy
> step.  The totally degraded RAID6 will be slow, but I doubt as slow
> as another copy step.

Keeping a RAID5 would be a great idea, and with a still rather small array, 
I think it could resync with a hot spare in about 4 hours. I'm on the safe 
side, I guess. 
Problem is, that I'd really like to move onto a RAID6 at some 
point, and, if I have to copy all the data to a temporary location, it's 
easier to do now, than few months later, with, say, 2 times more data.

Crikey, I've realised that all my "ideas" are bad actually, they wouldn't 
survive any drive failure during the procedure. I think I'm gonna spend extra 
100E on 3rd drive, and set up a slightly less degraded raid6. Knowing 
tricks, things tend to fail in worst possible moment. 

Best regards, 
MK

PS: Sorry for some garbage at the end of my previous message, I have no idea 
what made gmail do that.