"btrfs device delete missing" - why does it write on healthy device?

"btrfs device delete missing" - why does it write on healthy device?

[Tomasz Chmielewski]

So I've removed a missing device, which took some time:

# time btrfs device delete missing /home

real    1512m33.763s
user    0m0.000s
sys     121m37.740s

OK, it needs time, fine.


And shifted quite large amounts of data:

Device:            tps    MB_read/s    MB_wrtn/s    MB_read    MB_wrtn
sda             347.00         0.37        16.24      34050    1512249
sdb             910.89        13.46        18.77    1252768    1747442


New device was sda - so it's expected it wrote quite a lot of data
there.

But why did it write so much on sdb (even more than on sda)? With
software RAID-1, when replacing a failed device, the amount of reads
from one drive is equal to the writes on the other drive; there is no
writing on the healthy drive.


# btrfs filesystem show
Label: 'test-btrfs'  uuid: 60f1759c-45f6-4484-9f60-66a4e9bbf2b6
        Total devices 2 FS bytes used 1.19TB
        devid    3 size 2.56TB used 1.19TB path /dev/sdb4
        devid    4 size 2.56TB used 1.19TB path /dev/sda4

Btrfs v0.20-rc1-324-g650e656



-- 
Tomasz Chmielewski
http://wpkg.org

Re: "btrfs device delete missing" - why does it write on healthy device?

[Chris Murphy]

On Aug 29, 2013, at 1:01 PM, Tomasz Chmielewski <tch@virtall.com> wrote:

> So I've removed a missing device, which took some time:
> 
> # time btrfs device delete missing /home
> 
> real    1512m33.763s
> user    0m0.000s
> sys     121m37.740s
> 
> OK, it needs time, fine.

25 hours seems like a long time … 

> And shifted quite large amounts of data:
> 
> Device:            tps    MB_read/s    MB_wrtn/s    MB_read    MB_wrtn
> sda             347.00         0.37        16.24      34050    1512249


… for ~1.5TB.


> 
> But why did it write so much on sdb (even more than on sda)? With
> software RAID-1, when replacing a failed device, the amount of reads
> from one drive is equal to the writes on the other drive; there is no
> writing on the healthy drive.

During the balance, chunks are read and re-written normally. Normal for raid1 means it will write two copies of each read chunk, therefore it writes to both devices. I don't know if it's possible, or planned, to optimize out the (seemingly) redundant write to the device that already has a legitimate copy.


Chris Murphy