Re: [pve-devel] [PATCH kernel] Backport two io-wq fixes relevant for io_uring

[Fabian Ebner <f.ebner@proxmox.com>]

Am 08.03.22 um 17:19 schrieb Mark Schouten:
> Hi,
> 
> So should I try and find someone who is able to reproduce this with a test-machine and is able to give you remote access to debug? Would that help?
>

It would certainly increase the likelihood of finding the issue. Since
it only happens on 7.x, it's likely a regression. Ideally, there needs
to be a snapshot of a problematic VM before the reboot, so that it can
be quickly tested against with e.g. different builds of QEMU/kernel.
Providing such a VM with snapshot state would of course be an
alternative to remote access.

> — 
> Mark Schouten, CTO
> Tuxis B.V.
> mark@tuxis.nl
> 
> 
> 
>> On 8 Mar 2022, at 10:12, Fabian Ebner <f.ebner@proxmox.com> wrote:
>>
>> Am 07.03.22 um 15:51 schrieb Mark Schouten:
>>> Hi,
>>>
>>> Sorry for getting back on this thread after a few months, but is the Windows-case mentioned here the case that is discussed in this forum-thread:
>>> https://forum.proxmox.com/threads/windows-vms-stuck-on-boot-after-proxmox-upgrade-to-7-0.100744/page-3 <https://forum.proxmox.com/threads/windows-vms-stuck-on-boot-after-proxmox-upgrade-to-7-0.100744/page-3>
>>>
>>> ?
>>
>> Hi,
>> the symptoms there sound rather different. The issue addressed by this
>> patch was about a QEMU process getting completely stuck on I/O while the
>> VM was live already. "completely" meant that e.g. connecting for the
>> display also would fail and there would be messages like
>>
>> VM 182 qmp command failed - VM 182 qmp command 'query-proxmox-support'
>> failed - unable to connect to VM 182 qmp socket - timeout after 31 retries
>>
>> in the syslog. The issue described in the forum thread reads like it
>> happens only upon reboot from inside the guest and nobody mentioned
>> messages like the above.
>>
>>>
>>> If so, should this be investigated further or are there other issues? I have personally not had the issue mentioned in the forum, but quite a few people seem to be suffering from issues with Windows VMs, which is currently holding us back from upgrading from 6.x to 7.x on a whole bunch of customer clusters.
>>
>> I also haven't seen the issue myself yet and haven't heard from any
>> colleagues either. Without a reproducer, it's very difficult to debug.
>>
>>>
>>> Thanks,
>>>
>>> — 
>>> Mark Schouten, CTO
>>> Tuxis B.V.
>>> mark@tuxis.nl
>>
> 
> 
>