Re: [btrfs tools] ability to fail a device...

[Anand Jain <anand.jain@oracle.com>]

On 09/09/2015 03:34 AM, Hugo Mills wrote:
> On Tue, Sep 08, 2015 at 09:18:05PM +0200, Ian Kumlien wrote:
>> Hi,
>>
>> Currently i have a raid1 configuration on two disks where one of them
>> is failing.
>>
>> But since:
>> btrfs fi df /mnt/disk/
>> Data, RAID1: total=858.00GiB, used=638.16GiB
>> Data, single: total=1.00GiB, used=256.00KiB
>> System, RAID1: total=32.00MiB, used=132.00KiB
>> Metadata, RAID1: total=4.00GiB, used=1.21GiB
>> GlobalReserve, single: total=412.00MiB, used=0.00B
>>
>> There should be no problem in failing one disk... Or so i thought!
>>
>> btrfs dev delete /dev/sdb2 /mnt/disk/
>> ERROR: error removing the device '/dev/sdb2' - unable to go below two
>> devices on raid1
>
>     dev delete is more like a reshaping operation in mdadm: it tries to
> remove a device safely whilst retaining all of the redundancy
> guarantees. You can't go down to one device with RAID-1 and still keep
> the redundancy.
>
>     dev delete is really for managed device removal under non-failure
> conditions, not for error recovery.
>
>> And i can't issue rebalance either since it will tell me about errors
>> until the failing disk dies.
>>
>> Whats even more interesting is that i can't mount just the working
>> disk - ie if the other disk
>> *has* failed and is inaccessible... though, i haven't tried physically
>> removing it...
>
>     Physically removing it is the way to go (or disabling it using echo
> offline >/sys/block/sda/device/state). Once you've done that, you can
> mount the degraded FS with -odegraded, then either add a new device
> and balance to restore the RAID-1, or balance with
> -{d,m}convert=single to drop the redundancy to single.

  its like you _must_ add a disk in this context otherwise the volume 
will render unmountable in the next mount cycle. the below mentioned 
patch has more details.


>> mdam has fail and remove, I assume for this reason - perhaps it's
>> something that should be added?
>
>     I think there should be a btrfs dev drop, which is the fail-like
> operation: tell the FS that a device is useless, and should be dropped
> from the array, so the FS doesn't keep trying to write to it. That's
> not implemented yet, though.


  There is a patch set to handle this..
     'Btrfs: introduce function to handle device offline'

Thanks, Anand

>     Hugo.
>