[Cluster-devel] gfs2 hang in xfstests generic/361 (v3)

[Bob Peterson]

On 7/27/21 2:05 AM, Christoph Hellwig wrote:
> On Mon, Jul 26, 2021 at 01:50:11PM -0500, Bob Peterson wrote:
>> FYI: I just pushed a replacement patch to linux-gfs2/for-next.next4.
>> The patch I mentioned last time had problems, so this is a simplified
>> middle-ground between that patch and the one I suggested earlier.
>> Patch is a89a427e0afe.
> 
> With just that patch cherry picked the test still hangs.
> 
> The entire linux-gfs2/for-next.next4 completes a quick group xfstests
> run, but with a lot of failures:
> 
> Failures: generic/079 generic/082 generic/092 generic/094 generic/103 generic/219 generic/230 generic/235 generic/244 generic/294 generic/306 generic/347 generic/379 generic/380 generic/382 generic/383 generic/384 generic/385 generic/386 generic/400 generic/441 generic/452 generic/488 generic/545 generic/566 generic/587 generic/594 generic/600 generic/601 generic/603
> Failed 30 of 514 tests
> 

Hi Christoph,

Thanks for testing gfs2.
Andreas and I are fully aware of these failures. When I do a full run 
(which I do almost every day), I get these same 9 failures that you also 
reported:

generic/294
generic/306
generic/441
generic/452
generic/488
generic/545
generic/594
generic/619
generic/631

I've not done a detailed analysis of these 9 lately, but in the past 
I've looked at individual failures and they're usually because gfs2 does 
a "file system withdraw" due to IO errors writing to its journal (and 
such--as we want and expect) whereas most file systems don't do that. So 
these are kind of expected failures. Perhaps I should go back check them 
again.

The others you reported are are exempted from my testing (I've got them 
in my excludes file) because in most cases the problem is gfs2's lack of 
support for the usrquota / grpquota / prjquota mount options:

generic/082 - Lack of support for -o usrquota / grpquota / prjquota
generic/219
generic/230
generic/235
generic/244
generic/379
generic/380
generic/382
generic/383
generic/384
generic/385
generic/386
generic/400
generic/566
generic/587
generic/600
generic/601
generic/603

That leaves us:

generic/079 [expunged] - Incorrect permission check (doesn't match the 
one in notify_change)
generic/092 [expunged] - Lack of support for unwritten extents
generic/094 [expunged] - Lack of support for unwritten extents
generic/103       [expunged] - Command having problems: xfs_io -i -fc 
falloc 0 66574524k /mnt/scratch/spc
(I haven't taken the time to analyze this yet)
Inconsistent filesystem when running out of space on dm-thin
(I haven't taken the time to analyze this yet either)

So we know we've still got some work to do here, but we've got higher 
priorities at the moment. They're still on our to-do list.

Regards,

Bob Peterson