Re: Regression that causes data csum mismatch

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

On 2023/6/14 00:02, David Sterba wrote:
> On Tue, Jun 13, 2023 at 02:26:39PM +0800, Qu Wenruo wrote:
>> Hi,
>>
>> Recently I am testing scrub preparing for the incoming logical scrub.
>>
>> But I noticed some rare and random test cases failure from scrub and
>> replace groups.
>>
>> E.g. btrfs/072 has a chance of failure around 1/30.
>
> I'd like to get a list of tests that could potentially reproduce it.

For my last reproduce, it's pure btrfs/072 loop, no other extra test
involved.

>  I've
> started 072 in a loop but given the frequency and also a possible other
> factors this probably won't be enough.

Last night I ran it for around 4 hours, but surprisingly no reproduce
any more.

>
>> Initially I thought it's my scrub patches screwing things up, but with
>> more digging, it turns out that it's real data corruption.
>>
>> After scrubbing found errors, btrfs check --check-data-csum also reports
>> csum mismatch.
>
> It would be good to share the updates to fstests with the tests. I've
> added the data csum check to _check_btrfs_filesystem.
>
>> Furthermore this is profile independent, I have see all profiles hitting
>> such data corruption.
>
> How does the corruption look like?

Just some csum mismatch, which both scrub and btrfs check
--check-data-csum report.

One of my concern here is the temperature of my environment, with AC
running it no longer reproduces...

Hope it's really just a false alert.

Anyway thank you very much for the extra runs, I have to say my systems
are no longer reliable during north hemisphere summer...

Thanks,
Qu