Raid5 scrub really slow.

[mittkonto]

Hi, trying out raid5 with the "new" raid1c3/4 profiles. It's slow by it's nature, but..

Noticed scrub crawls 12MiB/s when I usually get 40MiB/s, also I'd rather scrub one disc at the time. It seems to be a thing.

I added "BTRFS_SCRUB_MOUNTPOINTS="/dev/sda1:/dev/sdb1:/dev/sdc1:/dev/sdd1" in /etc/default/btrfsmaintenance

But it does not seem to catch on when I: sudo systemctl start btrfs-scrub:

btrfs-scrub.service - Scrub btrfs filesystem, verify block checksums
     Loaded: loaded (/usr/lib/systemd/system/btrfs-scrub.service; static)
     Active: inactive (dead) since Tue 2020-09-15 17:57:46 CEST; 2s ago
TriggeredBy: ● btrfs-scrub.timer
       Docs: man:fstrim
    Process: 63436 ExecStart=/usr/share/btrfsmaintenance/btrfs-scrub.sh (code=exited, status=0/SUCCESS)
   Main PID: 63436 (code=exited, status=0/SUCCESS)

Sep 15 17:57:46 host btrfs-scrub[63442]: Path /dev/sda1 is not btrfs, skipping
Sep 15 17:57:46 host btrfs-scrub[63442]: Running scrub on /dev/sdb1
Sep 15 17:57:46 host btrfs-scrub[63442]: Path /dev/sdb1 is not btrfs, skipping
Sep 15 17:57:46 host btrfs-scrub[63442]: Running scrub on /dev/sdc1
Sep 15 17:57:46 host btrfs-scrub[63442]: Path /dev/sdc1 is not btrfs, skipping
Sep 15 17:57:46 host btrfs-scrub[63442]: Running scrub on /dev/sdd1
Sep 15 17:57:46 host btrfs-scrub[63442]: Path /dev/sdd1 is not btrfs, skipping
Sep 15 17:57:46 host systemd[1]: btrfs-scrub.service: Succeeded.

[likeman9]

Only need RAID1 for metadata on RAID5 if 2 disks fail you lose everything anyway but not really a bad thing having more metadata redundancy (it's RAID6 you use RAID1c3 so it can still function with 2 disks missing)

But yes it be nice if the btrfs scrub command was smarter on scrubbing on raid56 so it's not acing 2-4x read impact on each disk witch is very bad for hdds

Should script a batch run for scrub 1 at a time per disk (just have it wait until it has finished)

[bdelwood]

You get <device> is not btrfs because the helper function that checks for a btrfs filesystem, is_btrfs, uses stat, which returns tmpfs on block devices.

A fix for this would be to change the helper function in btrfsmaintenance-functions so that it uses blkid, which can handle block devices.

So is_btrfs should look like this:

is_btrfs() {
    local FS=$(blkid -o value -s TYPE "$1")
    [ "$FS" = "btrfs" ] && return 0
    return 1
}

I just replaced stat -f --format=%T "$1" with blkid -o value -s TYPE "$1".

Then running a scrub with the block devices in BTRFS_SCRUB_MOUNTPOINTS will scrub each disk separately in the order given.