btrfs balance resume + raid5/6

btrfs balance resume + raid5/6

[Tim Castle]

Greetings! 

I'm testing raid6, and recently added two drives. 
I haven't been able to properly resume a balance operation: the number of
total chunks is always too low.  

It seems that the balance starts and pauses properly, but always resumes
with ~7 chunks. 

Here's an example: 

vendikar tim # uname -r
3.10.0-031000rc4-generic
vendikar tim # btrfs fi sho
Label: 'bank'  uuid: 682bd1f5-1bd3-479b-8882-867dfe32375f
	Total devices 7 FS bytes used 4.70TB
	devid    4 size 1.82TB used 1.44TB path /dev/dm-9
	devid    1 size 2.73TB used 1.44TB path /dev/dm-8
	devid    5 size 2.73TB used 1.44TB path /dev/dm-7
	devid    6 size 931.51GB used 207.00GB path /dev/dm-6
	devid    2 size 2.73TB used 1.44TB path /dev/dm-5
	devid    3 size 2.73TB used 1.44TB path /dev/dm-4
	devid    7 size 931.51GB used 207.00GB path /dev/dm-3

Btrfs v0.20-rc1-324-g650e656
vendikar tim # btrfs fi df bank/
Data, RAID6: total=4.71TB, used=4.69TB
System, RAID1: total=32.00MB, used=376.00KB
System: total=4.00MB, used=0.00
Metadata, RAID1: total=8.00GB, used=6.54GB

vendikar tim # btrfs balance start bank/ 
vendikar tim # btrfs balance status bank/
Balance on 'bank/' is running
3 out of about 1479 chunks balanced (4 considered), 100% left

vendikar tim # btrfs balance pause bank/
vendikar tim # btrfs balance status bank/
Balance on 'bank/' is paused
4 out of about 1479 chunks balanced (4 considered), 100% left

vendikar tim # btrfs balance resume bank
vendikar tim # btrfs balance status bank/
Balance on 'bank/' is running
1 out of about 4 chunks balanced (5 considered),  75% left

All the best, 
Tim 

Re: btrfs balance resume + raid5/6

[Ilya Dryomov]

On Sun, Jun 16, 2013 at 06:00:10AM +0000, Tim Castle wrote:
> Greetings! 
> 
> I'm testing raid6, and recently added two drives. 
> I haven't been able to properly resume a balance operation: the number of
> total chunks is always too low.  
> 
> It seems that the balance starts and pauses properly, but always resumes
> with ~7 chunks. 

Hi Tim,

This is the expected behaviour.  Balance does not actually remember
which chunks it had processed before it was paused, instead it uses a
heuristic which is based on the amount of space used out of each chunk.
As you have an almost full filesystem, all of your chunks, except for
those ~7, are fully (or almost fully) used, and balance resume simply
skips them.  If you really want to balance everything (why?) you would
have to run balance to completion, w/o pausing it.

Thanks,

		Ilya