Suddenly cannot change metadata profile any longer
Here's a new one (with kernel 3.18.7++):
$btrfs --version
Btrfs v3.18.2 (also tried 3.19-rc2)
$mount /dev/sdf1 /mnt/usb
$btrfs filesystem df /mnt/usb
Data, single: total=8.00MiB, used=256.00KiB
System, single: total=32.00MiB, used=16.00KiB
Metadata, single: total=1.00GiB, used=112.00KiB
GlobalReserve, single: total=16.00MiB, used=0.00B
Note: metadata=single - done right after mkfs for the initial balance.
$btrfs balance start -mconvert=dup /mnt/usb
Done, had to relocate 2 out of 3 chunks
root>btrfs filesystem df /mnt/usb
Data, single: total=8.00MiB, used=512.00KiB
System, single: total=32.00MiB, used=16.00KiB
Metadata, single: total=1.00GiB, used=192.00KiB
GlobalReserve, single: total=16.00MiB, used=0.00B
Wait, what? Still single?
This used to work, but somehow no longer does (verified on another filesystem). If it's dup, it stays dup; if it's single, it stays single.
What works is the very first balance after the mkfs - it successfully changes from the initial default dup to single, but subsequent balances don't work. They also don't really fail.
Any ideas what might have caused this? Must have been a recent patch.
Holger
Re: Suddenly cannot change metadata profile any longer (BISECTED)
Quoted 24 lines · Holger Hoffstätte
Here's a new one (with kernel 3.18.7++):
$btrfs --version
Btrfs v3.18.2 (also tried 3.19-rc2)$mount
/dev/sdf1/mnt/usb$btrfs filesystem df
/mnt/usb
Data, single: total=8.00MiB, used=256.00KiB
System, single: total=32.00MiB, used=16.00KiB
Metadata, single: total=1.00GiB, used=112.00KiB
GlobalReserve, single: total=16.00MiB, used=0.00BNote: metadata=single - done right after mkfs for the initial balance.
$btrfs balance start -mconvert=dup
/mnt/usb
Done, had to relocate 2 out of 3 chunksroot>btrfs filesystem df
/mnt/usb
Data, single: total=8.00MiB, used=512.00KiB
System, single: total=32.00MiB, used=16.00KiB
Metadata, single: total=1.00GiB, used=192.00KiB
GlobalReserve, single: total=16.00MiB, used=0.00B
Quoted 8 lines
Wait, what? Still single?
This used to work, but somehow no longer does (verified on another filesystem). If it's dup, it stays dup; if it's single, it stays single.
What works is the very first balance after the mkfs - it successfully changes from the initial default dup to single, but subsequent balances don't work. They also don't really fail.
Bisection found that commit 2f0810880f082fa8ba66ab2c33b02e4ff9770a5e [1] in the integration tree is the cause:
btrfs: delete chunk allocation attemp when setting block group ro
Reverting/omitting this patch lets me rebalance/convert at will.
thanks,
Holger
Re: Suddenly cannot change metadata profile any longer (BISECTED)
Quoted 10 lines · Holger Hoffstätte
Quoted 3 lines
What works is the very first balance after the mkfs - it successfully changes from the initial default dup to single, but subsequent balances don't work. They also don't really fail.
Bisection found that commit
2f0810880f082fa8ba66ab2c33b02e4ff9770a5e[1] in the integration tree is the cause:btrfs: delete chunk allocation attemp when setting block group ro
Reverting/omitting this patch lets me rebalance/convert at will.
Just for the list archives: Chris mailed me privately that they could reproduce the problem and will provide a fix post-3.20-rc1 (since that commit already went in with the merge window).
-h
Re: Suddenly cannot change metadata profile any longer (BISECTED)
Quoted 12 lines · Holger Hoffstätte
Quoted 10 lines · Holger Hoffstätte
Quoted 5 lines
What works is the very first balance after the mkfs - it
successfully
changes from the initial default dup to single, but subsequent
balances
don't work. They also don't really fail.Bisection found that commit
2f0810880f082fa8ba66ab2c33b02e4ff9770a5e[1]
in the integration tree is the cause:
Quoted 7 lines
Quoted 3 lines
btrfs: delete chunk allocation attemp when setting block group ro
Reverting/omitting this patch lets me rebalance/convert at will.
Just for the list archives: Chris mailed me privately that they could reproduce the problem and will provide a fix post-3.20-rc1 (since that commit already went in with the merge window).
Apparently reply-to-all is harder than it looks. Thanks for forwarding this along.
-chris