On Tue, May 5, 2020 at 2:39 PM Andrew Pam <andrew@xxxxxxxxxxxxxx> wrote:
>
> On 5/5/20 7:51 pm, Graham Cobb wrote:
> > Is there actually a scrub in progress?
>
> The scrub has been going for a couple of days now, and has scrubbed
> considerably more data than exists on the disks. Will it ever finish?
A raid1 volume has twice as many bytes to scrub as data reported by
df. Can you tell us what kernel version? And also what you get for:
$ sudo btrfs fi us /mp/
$ df -h
I'm using progs 5.6 and kernel 5.6.8 for this:
$ sudo btrfs scrub status /mnt/third
UUID:
Scrub resumed: Tue May 5 08:45:41 2020
Status: finished
Duration: 2:41:12
Total to scrub: 759.43GiB
Rate: 79.57MiB/s
Error summary: no errors found
$ sudo btrfs fi us /mnt/third
Overall:
Device size: 931.49GiB
Device allocated: 762.02GiB
Device unallocated: 169.48GiB
Device missing: 0.00B
Used: 759.43GiB
Free (estimated): 85.17GiB (min: 85.17GiB)
Data ratio: 2.00
Metadata ratio: 2.00
Global reserve: 512.00MiB (used: 0.00B)
Data,RAID1: Size:379.00GiB, Used:378.56GiB (99.89%)
/dev/mapper/sdd 379.00GiB
/dev/mapper/sdc 379.00GiB
Metadata,RAID1: Size:2.00GiB, Used:1.15GiB (57.57%)
/dev/mapper/sdd 2.00GiB
/dev/mapper/sdc 2.00GiB
System,RAID1: Size:8.00MiB, Used:80.00KiB (0.98%)
/dev/mapper/sdd 8.00MiB
/dev/mapper/sdc 8.00MiB
Unallocated:
/dev/mapper/sdd 84.74GiB
/dev/mapper/sdc 84.74GiB
$ df -h
...
/dev/mapper/sdd 466G 381G 86G 82% /mnt/third
I think what you're seeing is a bug. Most of the size reporting in
btrfs commands is in btrfs-progs; whereas the scrub is initiated by
user space, most of the work is done by the kernel. But I don't know
where the tracking code is.
Some of the sizes you have to infer perspective. There is no one
correct perspective. So it's normal to get a bit confused about the
convention that applies. On mdadm/lvm raid1, the mirror isn't included
in any of the space reporting. It seems like it's reporting 1/2 the
storage. Meanwhile Btrfs reports all of the storage, and variably
shows data taking up twice as much space (as literally behind the
scenes each block group of extents has a mirror)
--
Chris Murphy
