|
Replies:
21
-
Last Post:
Nov 20, 2008 5:11 AM
by: rarok
|
|
|
Posts:
3
From:
Sherwood, OR USA
Registered:
11/18/05
|
|
|
|
expanding raid-z possible?
Posted:
Nov 18, 2005 10:29 AM
|
|
Congratulations on the release of ZFS.
Is it possible to expand an existing raid-z in ZFS?
From what I read in the docs, it appears that I can create a pool with a raid-z and add another raid-z later to expand the size, but I don't see anything related to expanding the original raid-z.
For example, say I have a system with six 400GB drives in a raid-z and want to add two more 400GB drives to the pool at some later date (and keep redundancy of some kind). I can add the two new drives as a mirrored pair or a new 2 disk raid-z (which both waste a whole disk worth of space). I don't see any options to grow an existing raid-z. I guess I'm looking for something like "zpool expand raidz".
Did I miss this somewhere in the docs?
If this is not currently supported, are there plans to support this in the future?
Thanks for any help,
Jason
|
|
|
Posts:
25
From:
SE
Registered:
6/15/05
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 18, 2005 11:07 AM
in response to: jupton
|
|
I've not tested this but the %man zpool says..
---
zpool attach [-f] pool device new_device
Attach new_device to the mirror containing device. If
device is not currently part of any mirror or raidz,
then device automatically transforms into a two-way mir-
ror of device and new_device. In either case, new_device
begins to resilver immediately. The "zpool status" com-
mand reports the progress of the resilver.
---
/Per
|
|
|
|
|
Posts:
3
From:
Sherwood, OR USA
Registered:
11/18/05
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 18, 2005 11:26 AM
in response to: peobe
|
|
Thanks a bunch.
Guess I missed that - I was mainly looking at the "ZFS Administration Guide" (zfsadmin_1016.pdf) and don't see it covered in there. I had searched that doc for "raid" and "expand", but not the man pages. Sorry for not RTFMs fully...
I'm burning CDs now and I'll test this out once I get my system running.
Jason
|
|
|
|
|
Posts:
799
From:
Menlo Park, CA
Registered:
3/9/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 18, 2005 11:40 AM
in response to: peobe
|
|
Hmm, sounds like a manpage bug. The underlying kernel implementation
allows you to 'mirrorize' individual devices in a RAID-Z vdev (this is
how the 'replacing' vdev is implemented). But you cannot attach a
device to a RAID-Z vdev via the CLI.
- Eric
On Fri, Nov 18, 2005 at 11:07:34AM -0800, Per ?berg wrote:
> I've not tested this but the %man zpool says..
> ---
> zpool attach [-f] pool device new_device
>
> Attach new_device to the mirror containing device. If
> device is not currently part of any mirror or raidz,
> then device automatically transforms into a two-way mir-
> ror of device and new_device. In either case, new_device
> begins to resilver immediately. The "zpool status" com-
> mand reports the progress of the resilver.
> ---
>
> /Per
> This message posted from opensolaris.org
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at opensolaris dot org
> http://opensolaris.org/mailman/listinfo/zfs-discuss
--
Eric Schrock, Solaris Kernel Development http://blogs.sun.com/eschrock
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
3
From:
Registered:
11/18/05
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 18, 2005 11:40 AM
in response to: peobe
|
|
"Attach new_device to the mirror"
That quote specifically mentions that attach is used with a mirror.
The way I read it it will only work for mirrors, not raid-z devices. I hope I'm wrong, but that's the way I intarpret it.
/Magnus
|
|
|
|
|
Posts:
799
From:
Menlo Park, CA
Registered:
3/9/05
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 18, 2005 11:38 AM
in response to: jupton
|
|
No, you cannot expand a RAID-Z stripe. It's fundamentally a fixed width
stripe. For more information on RAID-Z, check out Jeff's blog:
http://blogs.sun.com/roller/page/bonwick?entry=raid_z
- Eric
On Fri, Nov 18, 2005 at 10:29:30AM -0800, Jason Upton wrote:
> Congratulations on the release of ZFS.
>
> Is it possible to expand an existing raid-z in ZFS?
>
> >From what I read in the docs, it appears that I can create a pool with a raid-z and add another raid-z later to expand the size, but I don't see anything related to expanding the original raid-z.
>
> For example, say I have a system with six 400GB drives in a raid-z and want to add two more 400GB drives to the pool at some later date (and keep redundancy of some kind). I can add the two new drives as a mirrored pair or a new 2 disk raid-z (which both waste a whole disk worth of space). I don't see any options to grow an existing raid-z. I guess I'm looking for something like "zpool expand raidz".
>
> Did I miss this somewhere in the docs?
>
> If this is not currently supported, are there plans to support this in the future?
>
> Thanks for any help,
> Jason
> This message posted from opensolaris.org
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at opensolaris dot org
> http://opensolaris.org/mailman/listinfo/zfs-discuss
--
Eric Schrock, Solaris Kernel Development http://blogs.sun.com/eschrock
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
3
From:
Registered:
11/18/05
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 18, 2005 11:52 AM
in response to: eschrock
|
|
Should I take this to mean that there will never be a way to make a bigger raid-z out of an existing raid-z without having somewhere else to store all the data on the raid-z?
I really hope that's not the case. If I want to add a disk to an array it's because I need more space. I'm rather unlikely to have enough space available to store the data while recreating the array with more disks.
Adding another raid-z to the pool seems a quite poor alternative to me, as the amount of redundant data is doubled in the pool(assuming identically sized discs) without fault tolerance being improved at all.
Please tell me that there's hope for a future way of expanding a raid-z in place :) Not neccessarily while in use of course.
edit:spelling
|
|
|
|
|
Posts:
799
From:
Menlo Park, CA
Registered:
3/9/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 18, 2005 12:17 PM
in response to: smoked
|
|
On Fri, Nov 18, 2005 at 11:52:59AM -0800, Magnus Lidbom wrote:
> Should I take this to mean that there will never be a way to make a
> bigger raid-z out of an existing raid-z without having somewhere else
> to store all the data on the raid-z?
Yes, that is a relatively safe bet. Among other reasons, all the DVAs
(device virtual addresses) are calculated from the fixed size of the
RAID-Z device. That means if you were to add a new device to the raid-z
stripe, all your existing addresses would be wrong.
Also, in order to do a complete replacement, we would need to support
mirrors of raid-Z vdevs. This is planned for a future release.
> I really hope that's not the case. If I want to add a disk to an array
> it's because I need more space. I'm rather unlikely to have enough
> space available to store the data while recreating the array with more
> disks.
Why not just use dynamic striping? This may not make sense if you want
to add a single disk, but makes much more sense when adding a group of
disks. Two 4-way RAID-Z devices is a much better configuration than a
single 8-way in many ways.
> Adding another raid-z to the pool seems a quite poor alternative to
> me, as the amount of redundent data is doubled in the pool(assuming
> identically sized discs) without fault tolerance being improved at
> all.
Actually, no data is made redundant. Adding another raid-z to the pool
will cause data to be dynamically striped between the two vdevs, not
mirrored. And you will get better performance as well.
Adding a device to a raid-z stripe will actually _decrease_ fault
tolerance, not increase it. For example, imagine two scenarios:
a) 8-disk RAID-Z
b) 2 x 4-disk RAID-Z
In the first case, you can handle one bad device per 8 in your system.
In the latter case, you can handle one bad device per group of 4 in your
system. The latter has better fault tolerance (while taking only a
small space penalty for the cost of parity information).
> Please tell me that there's hope for a future way of expanding a
> raid-z in place :) Not neccessarily while in use of course.
Yes, this is theoretically possible to do while the pool is offline.
But besides the incredible complexity in migrating the data in the first
place, all ZFS operations have been designed to operate while the pool
is online. It would be a shame to start a trend in the opposite
direction.
That being said, if you think you can develop a tool to do this, let us
know. The source code is out there for a reason!
- Eric
--
Eric Schrock, Solaris Kernel Development http://blogs.sun.com/eschrock
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
3
From:
Registered:
11/18/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 18, 2005 1:06 PM
in response to: eschrock
|
|
> Lidbom wrote:
> > Should I take this to mean that there will never be
> a way to make a
> > bigger raid-z out of an existing raid-z without
> having somewhere else
> > to store all the data on the raid-z?
>
> Yes, that is a relatively safe bet. Among other
> reasons, all the DVAs
> (device virtual addresses) are calculated from the
> fixed size of the
> RAID-Z device. That means if you were to add a new
> device to the raid-z
> stripe, all your existing addresses would be wrong.
> Also, in order to do a complete replacement, we would
> need to support
> mirrors of raid-Z vdevs. This is planned for a
> future release.
There's plenty in your comment that went straight above my head. What I got out of it was basically: "It's quite difficult to implement". I hope that won't stop you, or someone else, from implementing it in the end.
> Why not just use dynamic striping? This may not make
> sense if you want
> to add a single disk, but makes much more sense when
> adding a group of
> disks. Two 4-way RAID-Z devices is a much better
> configuration than a
> single 8-way in many ways.
Exactly. For major upgrades a new raid-z makes sense. For an upgrade on a budget where you add just one disk it's not an option. For adding two disks it's possible, but half the space is "wasted" by parity data.
> Actually, no data is made redundant. Adding another
> raid-z to the pool
> will cause data to be dynamically striped between the
> two vdevs, not
> mirrored.
I thought dynamic striping was just splitting data across vdevs, with no parity information or fault tolerance unless the vdevs provide it. Doesn't the new raid-z have it's own parity information? Making for double the amount of parity data that you would have with just one raid-z (assuming all physical discs are the same size)?
>And you will get better performance as
> well.
I'm probably just flaunting my ignorance here, but I don't get it. Why would multible smaller raid-z devices provide better performance than one bigger raid-z using the same disks? Could you recommend some source of documentation that explains performance considerations for different raid-z configurations? I've read all I could find in the Administration guide and the zpool man page about raid-z, but saw no mention of this.
> > Please tell me that there's hope for a future way
> of expanding a
> > raid-z in place :) Not neccessarily while in use of
> course.
>
> Yes, this is theoretically possible to do while the
> pool is offline.
> But besides the incredible complexity in migrating
> the data in the first
> place, all ZFS operations have been designed to
> operate while the pool
> is online. It would be a shame to start a trend in
> the opposite
> direction.
>
> That being said, if you think you can develop a tool
> to do this, let us
> know. The source code is out there for a reason!
I get a minor hubris infection every now and then, not severe enough to make me believe I'm capable of implementing such o tool though :)
Best regards /Magnus
|
|
|
|
|
Posts:
55
From:
Registered:
11/29/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 29, 2005 9:18 AM
in response to: eschrock
|
|
> Two 4-way RAID-Z devices is a much better configuration than a single 8-way in many ways.
I thought I read somewhere you (well, the zfs team) were working on a version of raid-z with increased redundancy (resist to 2 disk failures for instance). In this case, it would make sense to put all the disks together.
|
|
|
|
|
Robert Prus
Robert.Prus@Sun.COM
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 29, 2005 5:44 AM
in response to: jupton
|
|
Hi,
I will cite original email instead of citing last email from this thread
;) .
There are some options to expand RAID-Z group on the fly, i.e. with zero
or minimum downtime.
As it was mentioned earlier it is not possible to expand RAID-Z group
from raidz(disk1, disk2, disk3, disk4, disk5, disk6) to raidz(disk1,
disk2, disk3, disk4, disk5, disk6, disk7) due to architectural reasons,
but ...
Here I will describe few approaches:
I) Using dynamic LUN extension capability of ZFS and zpool replace.
Here is simple example:
[root@sun] mkfile 80m /test/1
[root@sun] mkfile 80m /test/2
[root@sun] mkfile 80m /test/3
[root@sun] mkfile 80m /test/4
[root@sun] mkfile 160m /test/5
[root@sun] mkfile 160m /test/6
[root@sun] mkfile 160m /test/7
[root@sun] mkfile 160m /test/8
[root@sun] zpool create -f pool raidz /test/1 /test/2 /test/3 /test/4
[root@sun] zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
pool 302M 54.0K 302M 0% ONLINE -
[root@sun] zpool status
pool: pool
state: ONLINE
scrub: none requested
config:
NAME STATE READ WRITE CKSUM
pool ONLINE 0 0 0
raidz ONLINE 0 0 0
/test/1 ONLINE 0 0 0
/test/2 ONLINE 0 0 0
/test/3 ONLINE 0 0 0
/test/4 ONLINE 0 0 0
[root@sun] zpool replace pool /test/1 /test/5
[root@sun] zpool replace pool /test/2 /test/6
[root@sun] zpool replace pool /test/3 /test/7
[root@sun] zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
pool 302M 211K 302M 0% ONLINE -
[root@sun] zpool status
pool: pool
state: ONLINE
scrub: resilver completed with 0 errors on Tue Nov 29 13:13:31 2005
config:
NAME STATE READ WRITE CKSUM
pool ONLINE 0 0 0
raidz ONLINE 0 0 0
/test/5 ONLINE 0 0 0
/test/6 ONLINE 0 0 0
/test/7 ONLINE 0 0 0 14.5K resilvered
/test/4 ONLINE 0 0 0
[root@sun] zpool replace pool /test/4 /test/8
[root@sun] zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
pool 622M 262K 622M 0% ONLINE -
[root@sun] zpool status
pool: pool
state: ONLINE
scrub: resilver completed with 0 errors on Tue Nov 29 13:14:11 2005
config:
NAME STATE READ WRITE CKSUM
pool ONLINE 0 0 0
raidz ONLINE 0 0 0
/test/5 ONLINE 0 0 0
/test/6 ONLINE 0 0 0
/test/7 ONLINE 0 0 0
/test/8 ONLINE 0 0 0 11.0K resilvered
What we get at the end is expanded RAID-Z group with zero downtime!
You can use this method to replace disks in RAID-Z group one by one with
larger capacity disks.
II) If your RAID-Z group consists of disks/LUNs from storage array with
RAID controller, you can expand LUNs size at RAID controller level. ZFS
will see LUN size expansion immediately. I cannot test it now, but this
is what I know about ZFS behaviour.
III) To move data from raidz(disk1, disk2, disk3, disk4, disk5, disk6)
to raidz(disk6, disk7, disk8, disk9, disk10, disk11, disk12) group you
can use zfs backup/restore for every dataset. To minimize downtime you
can use full and incremental backups. In this case short application
downtime will be required. This method can be considered as RAID-Z
expansion.
IV) In future one can imagine RAID-Z expansion from raidz(disk1, disk2,
disk3, disk4, disk5, disk6) to raidz(concat(disk1,disk7),
concat(disk2,disk8), concat(disk3,disk9), concat(disk4,disk10),
concat(disk5,disk11), concat(disk6,disk12)) group, where size of disk1
is not necessary equal to size of disk7. But ZFS developers should
answer for themselves if this ZFS enhancement is worth of their effort.
In my opinion - yes.
V) In future basing on approach (IV) adding one disk to raidz(disk1,
disk2, disk3, disk4, disk5, disk6) could be considered as adding of
slices in following manner: raidz(concat(disk1,disk7slice1),
concat(disk2,disk7slice2), concat(disk3,disk7slice3),
concat(disk4,disk7slice4), concat(disk5,disk7slice5),
concat(disk6,disk7slice6)). But of course this approach has serious
performance implications.
Greetings,
Robert Prus
Jason Upton wrote:
>Congratulations on the release of ZFS.
>
>Is it possible to expand an existing raid-z in ZFS?
>
>>From what I read in the docs, it appears that I can create a pool with a raid-z and add another raid-z later to expand the size, but I don't see anything related to expanding the original raid-z.
>
>For example, say I have a system with six 400GB drives in a raid-z and want to add two more 400GB drives to the pool at some later date (and keep redundancy of some kind). I can add the two new drives as a mirrored pair or a new 2 disk raid-z (which both waste a whole disk worth of space). I don't see any options to grow an existing raid-z. I guess I'm looking for something like "zpool expand raidz".
>
>Did I miss this somewhere in the docs?
>
>If this is not currently supported, are there plans to support this in the future?
>
>Thanks for any help,
>Jason
>This message posted from opensolaris.org
>_______________________________________________
>zfs-discuss mailing list
>zfs-discuss at opensolaris dot org
>http://opensolaris.org/mailman/listinfo/zfs-discuss
>
>
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
29
From:
Registered:
11/18/05
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 29, 2005 11:26 AM
in response to: Robert Prus
|
|
All I can say is WOW!
I tried this 2 ways:
First like the post above to simulate adding extra drives to the system, then "moving" the pool to the new drives.
Then I tried an in-place replace (make the backing file bigger, then zpool replace pool /test/{1-4} ) to simulate "upgrading" a drive at a time. This also worked great, but you have to wait for each disk to resilver before moving on (or you will lose the array).
Note: both of these were tested with an iso bigger than one disk (to ensure it was on multiple disks) and md5sums were ok after each test.
So if you limit your raidz's to 3 or 4 disks, you not only get added redundancy, but you have a reasonable upgrade path (a little riskier if you do it in-place) by replacing 3 or 4 disks at a time.
I didn't test mirroring, but I assume that it would work.
Now I just can't wait for the SATA framework ...
|
|
|
|
|
Posts:
91
From:
Menlo Park, CA
Registered:
3/9/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 29, 2005 6:00 PM
in response to: jebc4
|
|
On Tue, Nov 29, 2005 at 11:26:59AM -0800, Jeb Campbell wrote:
> All I can say is WOW!
>
> I tried this 2 ways:
>
> First like the post above to simulate adding extra drives to the
> system, then "moving" the pool to the new drives.
>
> Then I tried an in-place replace (make the backing file bigger, then
> zpool replace pool /test/{1-4} ) to simulate "upgrading" a drive at a
> time. This also worked great, but you have to wait for each disk to
> resilver before moving on (or you will lose the array).
Did you actually try multiple replaces at the same time, or are you
assuming it won't work? It actually should work just fine. If not,
please file a bug.
--Bill
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
29
From:
Registered:
6/14/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 29, 2005 7:11 PM
in response to: billm
|
|
Bill Moore wrote:
> On Tue, Nov 29, 2005 at 11:26:59AM -0800, Jeb Campbell wrote:
>
>>All I can say is WOW!
>>
>>I tried this 2 ways:
>>
>>First like the post above to simulate adding extra drives to the
>>system, then "moving" the pool to the new drives.
>>
>>Then I tried an in-place replace (make the backing file bigger, then
>>zpool replace pool /test/{1-4} ) to simulate "upgrading" a drive at a
>>time. This also worked great, but you have to wait for each disk to
>>resilver before moving on (or you will lose the array).
>
>
> Did you actually try multiple replaces at the same time, or are you
> assuming it won't work? It actually should work just fine. If not,
> please file a bug.
I think he's simulating yanking a drive and replacing it with a higher
capacity drive. The simulation should really have included DD'ing zeros
all over the newly enlarged backing file. :-) So, just like a RAID
running in degraded mode, any failure on one of the other drives during
the reconstruction to the new drive will bork your data. After the
resilver completes, repeat process until all the drives in your system
are replaced. For those of us who can't afford multiple JBOD
enclosures, this seems like a useful if slightly risky technique.
BTW, there doesn't seem to be a way to designate a hot spare for a zpool
or set of zpools. Is this a planned ZFS feature?
-Jason
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
29
From:
Registered:
11/18/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 29, 2005 8:52 PM
in response to: snorbert
|
|
Sorry, I was being quick.
For the in-place replace, I actually offline'd the file/drive, then moved a new file in -- and yes I should have zero'd it, but I was confident in ZFS ;)
Anyway, I will test again on true disks/arrays whenever the SATA framework is merged.
Getting SATA merged would be a great gift for the holidays . . . ;)
|
|
|
|
|
Posts:
799
From:
Menlo Park, CA
Registered:
3/9/05
|
|
|
|
Re: Re: expanding raid-z possible?
Posted:
Nov 30, 2005 2:56 PM
in response to: snorbert
|
|
On Wed, Nov 30, 2005 at 02:11:41PM +1100, Jason Ozolins wrote:
>
> BTW, there doesn't seem to be a way to designate a hot spare for a zpool
> or set of zpools. Is this a planned ZFS feature?
>
Yes. See the ZFS FAQ on opensolaris.org.
- Eric
--
Eric Schrock, Solaris Kernel Development http://blogs.sun.com/eschrock
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://mail.opensolaris.org/zfs-discuss
|
|
|
|
|
Robert Prus
Robert.Prus@Sun.COM
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 30, 2005 3:05 AM
in response to: Robert Prus
|
|
Hi,
Small correction to my previuos posting:
> III) To move data from raidz(disk1, disk2, disk3, disk4, disk5,
disk6) to raidz(disk6, disk7, disk8, disk9, disk10, disk11, disk12)
Sorry, I meant:
III) To move data from raidz(disk1, disk2, disk3, disk4, disk5, disk6)
to raidz(disk7, disk8, disk9, disk10, disk11, disk12, disk13)
-----------------
To continue some thoughts about dynamic expansion of ZFS storage pools.
VI) You can use mirrors (i.e. zpool attach/detach) to dynamically expand
ZFS in some cases.
[root@sun] mkfile 80m /test/1
[root@sun] mkfile 160m /test/2
[root@sun] zpool create pool /test/1
[root@sun] zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
pool 75.5M 32.5K 75.5M 0% ONLINE -
[root@sun] zpool status
pool: pool
state: ONLINE
scrub: none requested
config:
NAME STATE READ WRITE CKSUM
pool ONLINE 0 0 0
/test/1 ONLINE 0 0 0
[root@sun] zpool attach pool /test/1 /test/2
[root@sun] zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
pool 75.5M 51.0K 75.5M 0% ONLINE -
[root@sun] zpool status
pool: pool
state: ONLINE
scrub: resilver completed with 0 errors on Wed Nov 30 11:07:27 2005
config:
NAME STATE READ WRITE CKSUM
pool ONLINE 0 0 0
mirror ONLINE 0 0 0
/test/1 ONLINE 0 0 0
/test/2 ONLINE 0 0 0 32.5K resilvered
[root@sun] zpool detach pool /test/1
[root@sun] zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
pool 156M 89.5K 155M 0% ONLINE -
[root@sun] zpool status
pool: pool
state: ONLINE
scrub: resilver completed with 0 errors on Wed Nov 30 11:07:27 2005
config:
NAME STATE READ WRITE CKSUM
pool ONLINE 0 0 0
/test/2 ONLINE 0 0 0 32.5K resilvered
At the end you get expanded ZFS storage pool!
The same applies to LUN expansion via storage array RAID controller.
VII) In future one can imagine extension of this approach working in
following manner:
a) Creating RAID-Z group using standard syntax, e.g.:
zpool create -f pool raidz disk1 disk2 disk3 disk4 disk5 disk6
b) Extending storage pool to configuration like:
mirror( raidz(disk1, disk2, disk3, disk4, disk5, disk6), raidz(disk7,
disk8, disk9, disk10, disk11, disk12, disk13) )
This will cause resilvering.
Due to man zpool(1) this task is not allowed at this moment!
> Virtual devices cannot be nested, so a mirror or raidz vir-
> tual device can only contain files or disks. Mirrors of mir-
> rors (or other combinations) are not allowed.
c) Detaching first vdev, i.e. raidz(disk1, disk2, disk3, disk4, disk5,
disk6).
VIII) In future one can imagine extension of RAID-Z group working as
follows:
a) Creating RAID-Z group using standard syntax, e.g.:
zpool create -f pool raidz disk1 disk2 disk3 disk4 disk5 disk6
b) Extending storage pool to configuration like:
zpool add pool raidz disk7 disk8 disk9 disk10 disk11 disk12 disk13
to create configuration like:
stripe( raidz(disk1, disk2, disk3, disk4, disk5, disk6), raidz(disk7,
disk8, disk9, disk10, disk11, disk12, disk13) )
c) Detaching first vdev, i.e. raidz(disk1, disk2, disk3, disk4, disk5,
disk6).
This will move data from raidz(disk1, disk2, disk3, disk4, disk5, disk6)
to raidz(disk7, disk8, disk9, disk10, disk11, disk12, disk13). Of
course, it could take many minutes/hours to perform this task, depending
on many factors, i.e. how fast are these disks, how many data is written
to them, what is current application workload, etc. .
What we need to perform approach (VII) and (VIII) is whole vdev removal
from ZFS control. In case of approach (VII) we need possibility of using
mirror( raidz(), raidz() ) configuration.
ZFS team comments are welcome ;) !!!
Greetings,
Robert Prus
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
10
From:
Registered:
7/8/08
|
|
|
|
Re: expanding raid-z possible?
Posted:
Jul 13, 2008 2:03 AM
in response to: jupton
To: Communities » zfs » discuss
|
|
A feature like that would be indeed great. It's the ONLY reason I'm not 100% sure about RAID-Z yet for my next system... 8-)
|
|
|
|
|
Posts:
75
From:
Registered:
6/23/08
|
|
|
|
Re: expanding raid-z possible?
Posted:
Jul 14, 2008 4:00 PM
in response to: evert
|
|
On Sun, Jul 13, 2008 at 2:03 AM, Evert Meulie <evert at meulie dot net> wrote:
> A feature like that would be indeed great. It's the ONLY reason I'm not 100% sure about RAID-Z yet for my next system... 8-)
The explanation that I've heard is that expanding raidz is of less
importance to enterprise users, who are the initial target for ZFS.
Most enterprise users would just attach a new drive tray and add that
as another raid-z to the zpool.
That being said, there is an RFE for expanding the width of a raidz:
http://bugs.opensolaris.org/view_bug.do?bug_id=6718209
-B
--
Brandon High bhigh at freaks dot com
"The good is the enemy of the best." - Nietzsche
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
7
From:
ES
Registered:
9/2/08
|
|
|
|
Re: expanding raid-z possible?
Posted:
Nov 20, 2008 4:30 AM
in response to: bhigh
To: Communities » zfs » discuss
|
|
There is any change about this? Any developer tried it?
|
|
|
|
|
Posts:
3,398
From:
Registered:
3/9/05
|
|
|
|
Re: [zfs-discuss] expanding raid-z possible?
Posted:
Nov 20, 2008 4:45 AM
in response to: rarok
|
|
>There is any change about this? Any developer tried it?
"It won't work unless someone implements it".
In the case of a RAIDZ, it means reading and rewriting all data.
The only thing you can do is add a new raidz to the pool.
So you start with
NAME STATE READ WRITE CKSUM
export ONLINE 0 0 0
raidz ONLINE 0 0 0
c0d0 ONLINE 0 0 0
c1d0 ONLINE 0 0 0
c2d0 ONLINE 0 0 0
And you add a second raidz
zpool add export raidz c3d0 c4d0 c5d0
Casper
_______________________________________________
zfs-discuss mailing list
zfs-discuss at opensolaris dot org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
|
|
|
|
|
Posts:
7
From:
ES
Registered:
9/2/08
|
|
|
|
Re: [zfs-discuss] expanding raid-z possible?
Posted:
Nov 20, 2008 5:11 AM
in response to: casper
To: Communities » zfs » discuss
|
|
Thanks for replying. I was asking because maybe someone implemente it could be as a stand-alone software no in the repository.
And I know that it could mean reading and rewriting all data, but I'm installing a personal NAS, so waiting even a week can be understable.
Thanks again for the reply.
|
|
|
|
|
