Skip to contentSkip to conversation

btrfs-balance convert from raid1 to single on specific device?

DISCUSSION Robert Krig 1 participant 2 messages Nov 10, 2015 Btrfs file system

lore
Share & tools
lore
#0 lore raw

Hi,

I'm running Kernel 4.3 and Btrfs-tools 4.3 on Debian Jessie. I compiled the tools and kernel myself.

Recently I added a new disk to my btrfs volume and wanted to proceed to convert from single to raid1.
Unfortunately the new disk seems to be faulty and started throwing a lot of errors.

The balance convert operation didn't manage to finish, and now I'm trying to get my btrfs volume back to single mode so that I can mount it read/write.

If I try to mount it degraded, then it complains that there are too few devices to mount writeable. So it will only allow me to mount read-only.

So far, so bad.

I have managed to boot and mount the volume once, with the faulty disk and cancel the balance operation.
Now I would like to convert it all to single again, so that I can finally remove that faulty drive.

However, if I simply say
btrfs fi balance start -dconvert=single -mconvert=single /btrfs-volume

Wouldn't that theoretically move some blocks to the faulty drive? Can I somehow specify to rebalance only the working drive?

I was thinking by specifying the device id, which seems possible according to the manpage.
I just wanted to check with you guys first, if that should work as expected, or will I run into further trouble?

Also, I understand -dconvert is for data, and -mconvert is for metadata, but is -sconvert? Is that data and metadata combined, or are system chunks something else entirely?

Thanks for your help.

#1 lore raw

Hi again,
I've been browsing through the mailing list and I came across a patch that might solve my issue a lot easier.

Specifically "[PATCH 02/15] btrfs: Do per-chunk check for mount time check" by Qu Wenruo

I'm a bit at a loss as to how to apply that patch. Since I'm not even sure for which kernel tree or git repo this patch refers to.

I've cloned the btrfs-next git repo from the btrfs wiki page. I've also searched through the gitweb site for that repo, but I couldn't find that patch included anywhere.
I'm guessing that it will probably take a while for patches to the mailing list to be merged into one of the repos.

What would be the easiest way for me to compile that specific patch? Which kernel version sources do I ideally need?
Is there a git repo that already has that patch included?

Thank you for your help.

Quoted 40 lines · Robert Krig

Hi,

I'm running Kernel 4.3 and Btrfs-tools 4.3 on Debian Jessie. I compiled the tools and kernel myself.

Recently I added a new disk to my btrfs volume and wanted to proceed to convert from single to raid1.
Unfortunately the new disk seems to be faulty and started throwing a lot of errors.

The balance convert operation didn't manage to finish, and now I'm trying to get my btrfs volume back to single mode so that I can mount it read/write.

If I try to mount it degraded, then it complains that there are too few devices to mount writeable. So it will only allow me to mount read-only.

So far, so bad.

I have managed to boot and mount the volume once, with the faulty disk and cancel the balance operation.
Now I would like to convert it all to single again, so that I can finally remove that faulty drive.

However, if I simply say
btrfs fi balance start -dconvert=single -mconvert=single /btrfs-volume

Wouldn't that theoretically move some blocks to the faulty drive? Can I somehow specify to rebalance only the working drive?

I was thinking by specifying the device id, which seems possible according to the manpage.
I just wanted to check with you guys first, if that should work as expected, or will I run into further trouble?

Also, I understand -dconvert is for data, and -mconvert is for metadata, but is -sconvert? Is that data and metadata combined, or are system chunks something else entirely?

Thanks for your help.