Re: system hangs when running btrfs balance

[Johannes Rohr <jorohr@gmail.com>]

Dear Qu, dear all,

hat the fix also been backported to the 5.6 kernel series? If so, from
which version on?

Cheers,

Johannes


Am 10.08.20 um 22:23 schrieb Johannes Rohr:
> Thanks so much, Qu, for your advice and the background
>
> We had some issues with btrfs, but I definitely want to continue using
> it. So the incredible responsiveness of btrfs devs like you is
> definitely on the plus side..
>
> Apparently, Ubuntu is preparing a kernel update to v 5.4.54
> https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1889669 so our
> problem should be solved soon!
>
> Cheers,
>
> Johannes
>
>
> Am 10.08.20 um 11:59 schrieb Qu Wenruo:
>> On 2020/8/10 下午5:22, Johannes Rohr wrote:
>>> Dear devs,
>>>
>>> since I upgraded our system from Ubuntu 18.04 LTS to 20.04 LTS, the file
>>> system completely freezes when I run a btrfs balance on it. The only way
>>> to get a usable system for the time being is with the mount option
>>> "skip_balance".
>>>
>>> The server has a raid1 with 4 SSDs with 500 GB each. 
>>> [Sun Aug  9 12:21:35 2020] CPU: 1 PID: 4537 Comm: btrfs-balance Tainted: G           O      5.4.47 #1
>> A quick git log glance shows that, some reloc tree related fixes haven't
>> landed in v5.4.47.
>>
>> E.g. (commits are upstream commits, not stable tree commits)\
>>
>> 1dae7e0e58b484eaa43d530f211098fdeeb0f404 btrfs: reloc: clear
>> DEAD_RELOC_TREE bit for orphan roots to prevent runaway balance
>> 51415b6c1b117e223bc083e30af675cb5c5498f3 btrfs: reloc: fix reloc root
>> leak and NULL pointer dereference.
>>
>> And above fixes only landed in v5.4.54, so I guess you have to update
>> your kernel anyway.
>>
>>> There has been a related bug report at kernel.org for a year,
>>> https://bugzilla.kernel.org/show_bug.cgi?id=203405 and I have found
>>> similar reports here and there, some pertaining to quite old kernel
>>> versions, but we have only been hit with kernel 5.4. After this first
>>> occurred, I had no better luck though, with older kernels (4 something
>>> from Debian buster, also 4 something from Ubuntu 18.04).
>> Nope, the mentioned one is another bug, we had some clue on this, but
>> need some time to solve it.
>> (It's mostly related to some special timing in canceling, leading to
>> parted dropped trees).
>>
>>> Apart from fixing the underlying issue, would there be any wordaround
>>> for it?
>> Update your kernel to at least v5.4.54, then mount with skip_balance and
>> finally  "btrfs balance cancel <mnt>".
>> After that, doing whatever you like should be fine.
>>
>> I prefer to do a btrfs check on the unmounted or at least ro mounted fs
>> to ensure your fs is sane in the first place.
>>
>> Thanks,
>> Qu
>>
>>> Currently the balance for the fs is in suspended status. Since
>>> there is quite a few people who depend on this server, I can't just play
>>> around with it at random. That's why I am asking for advice here...
>>>
>>> Thanks so much for any suggestions you might have!
>>>
>>> Johannes
>>>
>