Hi Piotr,

Unfortunately, the current implementation does not allow to customize the
driver via the agent.properties.
Instead, it evaluates Libvirt and Qemu versions in order to assess if it is
supported.

I considered the libvirt & qemu documentation where address support for
io_uring  and release notes.
According to release notes, io_urin is supported in:
    Qemu >= 5.0 [1], and
    Libvirt >= 6.3.0 [2].
(for reference) The kernel feature (io_uring itself) is described in this
lwn.net article [3].

However, I am not confident (anymore) that RedHat is supporting it [4,5].
It also might be that there has been a regression in the newer Qemu release.

For the record, I've just re-checked a KVM host running 4.16.0.0 and I have
indeed VMs running with io_uring configured in the "domain.xml".
Host details: Ubuntu 20.04, Qemu 5.0.0, Libvirt 6.3.0.

As a conclusion, considering that it is not as stable and widely supported
as assumed before, I will propose a PR that would allow customizing the
disk IO_Driver in the "agent.properties".
Additionally, I will double-check documentation, release announcements, and
bugs raised in qemu, libvirt, redhat, & ubuntu scopes.

[1] https://wiki.qemu.org/ChangeLog/5.0
[2] https://libvirt.org/news.html#v6-3-0-2020-05-05
[3] https://lwn.net/Articles/776703/
[4] https://bugzilla.redhat.com/show_bug.cgi?id=1769597
[5] https://bugzilla.redhat.com/show_bug.cgi?id=1769598

Any feedback and tests with different OS and Libvurt/Qemu releases are very
appreciated.
Best Regards,
Gabriel.

Em qua., 24 de nov. de 2021 às 09:37, Piotr Pisz <pi...@piszki.pl> escreveu:

> Hi!
>
>
>
> Gabriel, could you suggest how to disable io_uring with agent.properties?
>
> In the second lab, I turned off advanced virtualization repo and downgrade
> qemu to native for CentOS 8 and everything works fine.
>
> Generally, I use a lot of nested virtualization and I need the latest
> version of qemu, hence the advanced virtualization repo.
>
>
>
> Regards,
>
> Piotr
>
>
>
>
>
> *From:* Gabriel Bräscher <gabrasc...@gmail.com>
> *Sent:* Wednesday, November 24, 2021 8:54 AM
> *To:* users <users@cloudstack.apache.org>; pi...@piszki.pl
> *Cc:* Дикевич Евгений Александрович <evgeniy.dikev...@becloud.by>
> *Subject:* Re: [!!Mass Mail]io uring is not supported by this QEMU binary
>
>
>
> It worked when I tested in Ubuntu with the following qemu/libvirt:
>
>  ~# qemu-system-x86_64 --version
>
>  QEMU emulator version 5.0.0 (Debian 1:5.0-5ubuntu6~ppa0)
>
>   ~# libvirtd --version
>
>  libvirtd (libvirt) 6.6.0
>
>
>
> I will check if there has been any regressions in qemu.
>
>
>
> Maybe the best would be to extract that IO_URING configuration into an
> agent.properties variable. Just in case it is needed to be disabled, or
> changed to another disk IO Driver.
>
>
>
> Regards,
>
> Gabriel.
>
>
>
> On Mon, Nov 22, 2021, 13:59 Piotr Pisz <pi...@piszki.pl> wrote:
>
> Hi,
>
> I seem to meet these requirements:
>
> In one LAB I have a custom kernel: 5.4.129-1.el8.elrepo.x86_64 and
> libvirtd (libvirt) 7.6.0 with qemu 6.0.0
>
> But in the second, it comes with the system: 4.18.0-240.22.1.el8_3.x86_64
> and libvirtd (libvirt) 7.0.0 with qemu-kvm-5.2.0-16.el8
>
> And I have the same error in both ☹
>
> Colleagues who tested it can boast in what environments it was done?
>
> Regards,
> Piotr
>
>
>
>
> -----Original Message-----
> From: Дикевич Евгений Александрович <evgeniy.dikev...@becloud.by>
> Sent: Monday, November 22, 2021 1:37 PM
> To: users@cloudstack.apache.org; pi...@piszki.pl
> Subject: RE: [!!Mass Mail]io uring is not supported by this QEMU binary
>
> Hi
> Mb this is answer for you)
>
> https://github.com/apache/cloudstack/pull/5012
>
>
>
> -----Original Message-----
> From: Piotr Pisz [mailto:pi...@piszki.pl]
> Sent: Monday, November 22, 2021 3:23 PM
> To: users@cloudstack.apache.org
> Subject: [!!Mass Mail]io uring is not supported by this QEMU binary
>
> Hi,
>
>
>
> I found the reason why the VM did not start in 4.16 after the upgrade.
>
> The following message appears during the start-up:
>
>
>
> libvirt.LibvirtException: unsupported configuration: io uring is not
> supported by this QEMU binary
>
>
>
> KVM Is CentOS 8.5 with advanced virtualization repo, now qemu has version:
> qemu-kvm-6.0.0-33.el8.
>
> This configuration works without any problem with CS 4.15, the question
> is, if anything has changed in the requirements or handling of kvm in 4.16?
>
>
>
> Regards,
>
> Piotr
>
>
>
>
>
> Внимание!
> Это электронное письмо и все прикрепленные к нему файлы являются
> конфиденциальными и предназначены исключительно для использования лицом
> (лицами), которому (которым) оно предназначено. Если Вы не являетесь лицом
> (лицами), которому (которым) предназначено это письмо, не копируйте и не
> разглашайте его содержимое и удалите это сообщение и все вложения из Вашей
> почтовой системы. Любое несанкционированное использование, распространение,
> раскрытие, печать или копирование этого электронного письма и прикрепленных
> к нему файлов, кроме как лицом (лицами) которому (которым) они
> предназначены, является незаконным и запрещено. Принимая во внимание, что
> передача данных посредством Интернет не является безопасной, мы не несем
> никакой ответственности за любой потенциальный ущерб, причиненный в
> результате ошибок при передаче данных или этим сообщением и прикрепленными
> к нему файлами.
>
> Attention!
> This email and all attachments to it are confidential and are intended
> solely for use by the person (or persons) referred to (mentioned) as the
> intended recipient (recipients). If you are not the intended recipient of
> this email, do not copy or disclose its contents and delete the message and
> any attachments to it from your e-mail system. Any unauthorized use,
> dissemination, disclosure, printing or copying of this e-mail and files
> attached to it, except by the intended recipient, is illegal and is
> prohibited. Taking into account that data transmission via Internet is not
> secure, we assume no responsibility for any potential damage caused by data
> transmission errors or this message and the files attached to it.
>
>

Reply via email to