Online device removal pushed to the unstable tree

Online device removal pushed to the unstable tree

[Chris Mason]

Hello everyone,

The unstable tree is still format compatible with v0.14, and it now includes 
the ability to remove devices online.  In mirror configurations, IO errors to 
one of the mirrors are also handled, although it doesn't currently kick off a 
rebuild or other magic.

To try things out:

mkfs.btrfs -d raid1 -m raid1 /dev/sdb /dev/sdc
mount /dev/sdb /mnt
# put data on /mnt
# then add /dev/sdd to the FS
btrfs-vol -a /dev/sdd /mnt

At this point we've got a raid1 filesystem with data spread across sdb and 
sdc.  If we remove /dev/sdc, all the existing data will be remirrored onto 
sdd:

btrfs-vol -r /dev/sdc /mnt

There is also a mount -o degraded to handle situations where a device in the 
FS has been lost or removed:

mkfs.btrfs -d raid1 -m raid1 /dev/sd[bc]
mount /dev/sdb /mnt
# put data on /mnt
umount /mnt
# corrupt the drive
dd if=/dev/zero of=/dev/sdc bs=1M count=1000

# this mount will fail, dmesg will show that not enough devices were found
mount /dev/sdb /mnt

# this mount will succeed, but only one device will be used
mount -o degraded /dev/sdb /mnt

# add a spare
btrfs-vol -a /dev/sdd /mnt

# remove the corrupted drive
btrfs-vol -r missing /mnt

btrfs-vol -r missing is a special command that tells btrfs to reconstruct 
things without the first device it can find that is mentioned in the metadata 
but not present in the mounted FS.

I plan on hammering on this for a bit and then cutting v0.15

-chris

Re: Online device removal pushed to the unstable tree

[Miguel Sousa Filipe]

Hi all,

On Tue, May 13, 2008 at 9:08 PM, Chris Mason <chris.mason@oracle.com> wrote:
> Hello everyone,
>
>  The unstable tree is still format compatible with v0.14, and it now includes
>  the ability to remove devices online.  In mirror configurations, IO errors to
>  one of the mirrors are also handled, although it doesn't currently kick off a
>  rebuild or other magic.
>
>  To try things out:
>
>  mkfs.btrfs -d raid1 -m raid1 /dev/sdb /dev/sdc
>  mount /dev/sdb /mnt
>  # put data on /mnt
>  # then add /dev/sdd to the FS
>  btrfs-vol -a /dev/sdd /mnt
>
>  At this point we've got a raid1 filesystem with data spread across sdb and
>  sdc.  If we remove /dev/sdc, all the existing data will be remirrored onto
>  sdd:
>

Is there a way to check the progress of the remirror/rebuild ?
(somewhat similar to what /proc/mdstat provides when a mirror is being synced)



-- 
Miguel Sousa Filipe

Re: Online device removal pushed to the unstable tree

[Chris Mason]

On Wednesday 14 May 2008, Miguel Sousa Filipe wrote:
> Hi all,
>
> On Tue, May 13, 2008 at 9:08 PM, Chris Mason <chris.mason@oracle.com> wrote:
> > Hello everyone,
> >
> >  The unstable tree is still format compatible with v0.14, and it now
> > includes the ability to remove devices online.  In mirror configurations,
> > IO errors to one of the mirrors are also handled, although it doesn't
> > currently kick off a rebuild or other magic.
> >
> >  To try things out:
> >
> >  mkfs.btrfs -d raid1 -m raid1 /dev/sdb /dev/sdc
> >  mount /dev/sdb /mnt
> >  # put data on /mnt
> >  # then add /dev/sdd to the FS
> >  btrfs-vol -a /dev/sdd /mnt
> >
> >  At this point we've got a raid1 filesystem with data spread across sdb
> > and sdc.  If we remove /dev/sdc, all the existing data will be remirrored
> > onto sdd:
>
> Is there a way to check the progress of the remirror/rebuild ?
> (somewhat similar to what /proc/mdstat provides when a mirror is being
> synced)

Sorry, not yet.  Some information is printed to the console, but it is 
currently difficult to tell how far along the remirroring is.

The current code is definitely slower than it should be, I went for the most 
generic setup in hopes of stabilizing it quickly.

-chris