Re: Critical error from Tree-checker

[Qu Wenruo <wqu@suse.com>]

在 2024/9/13 15:24, Archange 写道:
> Le 13/09/2024 à 09:29, Qu Wenruo a écrit :
>> 在 2024/9/13 14:55, Archange 写道:
>> [...]
>>>> And it's indeed a false alert.
>>>>
>>>> In that case, as long as you still have unallocated space, you can
>>>> just relocate the system chunks:
>>>>
>>>> # btrfs balacne start -s <mnt>
>>>>
>>>> Which should move the system chunks to new locations and will not
>>>> utilize the first 1MiB reserved space.
>>>
>>> # btrfs balance start -s /
>>> ERROR: Refusing to explicitly operate on system chunks.
>>> Pass --force if you really want to do that.
>>>
>>> According to https://btrfs.readthedocs.io/en/latest/btrfs-balance.html,
>>> -s requires -f, so I guess I should continue with that?
>>
>> Yes.
> 
> Hum, no success:
> 
> # btrfs balance start -s --force /
> ERROR: error during balancing '/': No space left on device
> There may be more info in syslog - try dmesg | tail
> 
> # dmesg
> [ 2919.917607] BTRFS info (device dm-0): balance: start -f -s
> [ 2919.918105] BTRFS info (device dm-0): 1 enospc errors during balance
> [ 2919.918108] BTRFS info (device dm-0): balance: ended with status: -28
> 
> Indeed,
> 
> # btrfs filesystem show /dev/mapper/root
> Label: 'root'  uuid: e6614f01-6f56-4776-8b0a-c260089c35e7
>      Total devices 1 FS bytes used 439.69GiB
>      devid    1 size 476.87GiB used 476.87GiB path /dev/mapper/root
> 
> There is unused space though, but not sure how to reclaim it.
> 
> $ btrfs filesystem df /
> Data, single: total=472.87GiB, used=438.21GiB
> System, single: total=4.00MiB, used=80.00KiB
> Metadata, single: total=4.00GiB, used=1.48GiB
> GlobalReserve, single: total=512.00MiB, used=0.00B

This means most of the "free" space is inside data chunks.

You will need to free up some continuous data space, or use different 
"usage=" filters for data and metadata, or both.

Thankfully your metadata has enough free space, so not an urgent situation.

Thanks,
Qu
> 
> As advised on the balance page, I’ve tried to run with `usage=0` as 
> filter (for both m, s and d), but the result is always:
> 
> Done, had to relocate 0 out of 480 chunks
> 
>> And I also recommend to convert your metadata and system chunks to DUP,
>> if there are enough unallocated space.
>> (If have more devices then RAID1).
>>
>> It looks like the old mkfs defaults to SINGLE for SSDs, but nowadays we
>> keep DUP no matter if it's SSD or not.
> 
> Alright, but I guess I need to solve -ENOSPC first…
> 
> Thanks,
> Archange
>