Re: mkfs.xfs "concurrency" change concerns

On 6/15/26 9:19 AM, Christoph Hellwig wrote:
> On Thu, Oct 09, 2025 at 03:13:47PM -0500, Eric Sandeen wrote:
>> Hey all -
>>
>> this got long, so tl;dr:
>>
>> 1) concurrency geometry breaks some xfstests for me
>> 2) concurrency behavior is not consistent w/ loopback vs. imagefile
>> 3) concurrency defaults to the mkfs machine not the mount machine
>>
>> In detail:
>>
>> So, I realize I'm late to the game here and didn't review the patches
>> before they went in, but it looks like the "concurrency" mkfs.xfs
>> arguments and defaults are breaking several xfstests.
>>
>> 4738ff0 mkfs: allow sizing realtime allocation groups for concurrency
>> c02a1873 mkfs: allow sizing internal logs for concurrency
>> 9338bc8b mkfs: allow sizing allocation groups for concurrency
>>
>> Specifically, xfs/078, xfs/216, and xfs/217 are failing for us
>> on various machines with between 8 and 128 CPUS, due to the
>> fundamental change in geometry that results from the new
>> concurrency behavior, which makes any consistent golden
>> output that involves geometry details quite difficult.
> 
> Did anything ever come out of this?  We're hitting errors in exactly
> those tests in the zoned XFS CI (which also tests regular XFS).
> Shin'ichiro recently spent some time tracking these down, which
> finally made me remember this thread.

No not really. IIRC, I don't think Darrick felt that there was any
problem with the code as it is, I wasn't getting my points across,
and I decided not to pursue it further. (That's the sanitized version,
in reality I got pretty grumpy about it, sorry.)

If we have more evidence that this really is a problem, happy to re-
engage with discussion about a path forward to resolve it.

-Eric