Re: [PATCH v1] generic/476: requires 27GB scratch size

[Chuck Lever III <chuck.lever@oracle.com>]

> On Jul 21, 2022, at 11:29 AM, Zorro Lang <zlang@redhat.com> wrote:
> 
> On Thu, Jul 21, 2022 at 10:29:59AM +0800, bxue@redhat.com wrote:
>> From: Boyang Xue <bxue@redhat.com>
>> 
>> The test requires larger scratch dev size when running on top of NFS other
>> than ext4 and xfs. It requires at least 27GB in my test. Without this
>> requirement, the test run never finishes on NFS, leaving 100% scratch disk
>> space use.
>> 
>> Signed-off-by: Boyang Xue <bxue@redhat.com>
>> ---
>> Hi,
>> 
>> I find generic/476 easily goes into an infinite run on top of NFS. When it
>> happens, the common pattern is 100% disk space use of SCRATCH_MNT, and
>> `nfsiostat` shows 50% write error on SCRATCH_MNT. When I run it with a large
>> enough SCRATCH_MNT, the problem disappears. So I post this patch to add the size
>> requirement.
>> 
>> Please help review this patch. Thanks!
>> 
>> -Boyang
>> 
>> tests/generic/476 | 1 +
>> 1 file changed, 1 insertion(+)
>> 
>> diff --git a/tests/generic/476 b/tests/generic/476
>> index 212373d1..dcc7c3da 100755
>> --- a/tests/generic/476
>> +++ b/tests/generic/476
>> @@ -24,6 +24,7 @@ _cleanup()
>> _supported_fs generic
>> 
>> _require_scratch
>> +_require_scratch_size $((27 * 1024 * 1024)) # 27GB
> 
> At first, most of other filesystems don't need this requirement, that will
> reduce test coverage of other fs suddently. Second, there's not a clear
> reason to prove NFS (or others) need a 27+GB device to run this test. Due to
> the generic/476 does nothing special, except random I/Os, even running with
> ENOSPC...
> 
> So one difference of running with small device or large enough device is the
> chance to run with ENOSPC. I think the device size isn't the root cause of
> nfs hang you hit, I doubt if it's a NFS bug with ENOSPC, or something else
> bug which is triggered by someone random I/O operation.
> 
> We'd better not to skip a known generic test (from upstream fstests directly)
> if without a clear reason. That might cause we miss bugs or test coverage,
> better to make sure if it's a real bug at first. Then think about if we need
> to improve the case or fix a bug.

+1

I can help troubleshoot the NFS-related aspects of this further, if needed.


> Thanks,
> Zorro
> 
>> _require_command "$KILLALL_PROG" "killall"
>> 
>> echo "Silence is golden."
>> -- 
>> 2.27.0
>> 
> 

--
Chuck Lever