OK, we all know that df lies with btrfs. But even in the cases where it _could_ tell the truth, it's still getting it wrong. I've finally managed to pin down a concrete example:
Label: 'amelia' uuid: 1da97c6f-5467-4591-ad79-5d283db800d4
Total devices 6 FS bytes used 7668.78GiB
devid 4 size 3722.02GiB used 3165.00GiB path /dev/sda2
devid 7 size 1393.02GiB used 1147.00GiB path /dev/sdd2
devid 9 size 1858.77GiB used 1614.03GiB path /dev/sdb2
devid 12 size 3722.02GiB used 3164.00GiB path /dev/sdj2
devid 13 size 3721.92GiB used 3627.03GiB path /dev/sdc2
devid 14 size 5589.03GiB used 2655.00GiB path /dev/sdi2
Data, RAID1: total=7674.00GiB, used=7657.94GiB
System, RAID1: total=0.03GiB, used=0.00GiB
Metadata, RAID1: total=12.00GiB, used=10.84GiB
GlobalReserve, single: total=0.50GiB, used=0.00GiB
Filesystem 1G-blocks Used Available Use% Mounted on
/dev/sda2 10004G 7670G 915G 90% /media/video
Now, by my calculations, there's unallocated space on the disks as
follows:
devid 4 unalloc 555 GiB
devid 7 unalloc 246 GiB
devid 9 unalloc 244 GiB
devid 12 unalloc 556 GiB
devid 13 unalloc 94 GiB
devid 14 unalloc 2937 GiB
Now, clearly the first five sum to less than the last one. With
RAID-1 on everything, this means that we'll fill devices 4-13 first,
and we can fit 1695 GiB more allocation on this FS. Added to the 16
GiB of space allocated but unused, the estimate should therefore be
1711 GiB of usable space, not the 915 GiB reported by df.
I don't know what it's doing in there, but whatever it is, it's
definitely *not* using the right algorithm for it (the one I wrote up
in detail a few months ago).
I think if the unallocated space on all the devices is similar, it
gets the answer nearly right. It seems to get worse as the imbalance
of unallocated space across the devices gets larger.
Hugo.
--
Hugo Mills | I know of three kinds: hot, cool, and
hugo@... carfax.org.uk | what-time-does-the-tune-start?
http://carfax.org.uk/ |
PGP: E2AB1DE4 | Jill Swinburne, The Beiderbecke Connection
signature.asc
Description: Digital signature
