nova-libvirt needs fixed libvirt on AArch64 servers with ThunderX cpu

[Marcin Juszkiewicz]

It was detected during Queens/Rocky cycle. We fixed it and once we moved to Debian 'buster' in Stein it happened again - just no one noticed. Now Linaro Developer Cloud is upgrading from Rocky to Stein to Train and they need it fixed again.

Problem exists in libvirt used in Debian 'buster' as fixes were merged after 5.0.0 was released.

Original bug report:

I am enabling a set of cavium thunderx hypervisors as compute nodes, however, libvirtd 3.10.0 fails with the following set of errors, one is related to the vf/pf network device mapping when the device node set is performed by libvirt.

None of the ethernet/vf devices are available for libvirt.

(nova-libvirt)[root@uk-dc-cavium-01 libvirt]# virsh nodedev-list | grep eth
(nova-libvirt)[root@uk-dc-cavium-01 libvirt]#

* At nova side, I see the following traceback:
https://pastebin.ubuntu.com/p/XSsFRz7H9J/
* At the libvirtd side, I see the following error messages:

2018-04-30 15:50:09.053+0000: 5069: info : libvirt version: 3.10.0,
package: 1~bpo9+1 (Marcin Juszkiewicz <email address hidden>
Thu, 14 Dec 2017 12:44:47 +0000)

2018-04-30 15:50:09.053+0000: 5069: info : hostname: uk-dc-cavium-01
2018-04-30 15:50:09.249+0000: 5069: error :
virNetDevGetPhysicalFunction:1391 : internal error: The PF device for
VF eth0 has no network device name
2018-04-30 15:50:09.251+0000: 5069: error :
virNetDevGetPhysicalFunction:1391 : internal error: The PF device for
VF eth1 has no network device name
2018-04-30 15:50:09.253+0000: 5069: error :
virNetDevGetPhysicalFunction:1391 : internal error: The PF device for
VF eth2 has no network device name
2018-04-30 15:50:09.255+0000: 5069: error :
virNetDevGetPhysicalFunction:1391 : internal error: The PF device for
VF eth3 has no network device name
2018-04-30 15:50:09.257+0000: 5069: error :
virNetDevGetPhysicalFunction:1391 : internal error: The PF device for
VF eth4 has no network device name

[Marcin Juszkiewicz (hrw)]

https://bugs.linaro.org/show_bug.cgi?id=3778 for all details.

[Radosław Piliszek (yoctozepto)]

addressed by https://review.opendev.org/712987

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/712987
Committed: https://git.openstack.org/cgit/openstack/kolla/commit/?id=245992d37c0d051ed50bd843e4ba4ba852373afe
Submitter: Zuul
Branch: master

commit 245992d37c0d051ed50bd843e4ba4ba852373afe
Author: Marcin Juszkiewicz <email address hidden>
Date: Fri Mar 13 16:30:08 2020 +0100

    kibana, nova-libvirt: handle it properly for non-x86

    Upstream kibana package contains NodeJS x86-64 binaries so can not be
    used directly on non-x86 architectures. I took upstream packages,
    removed NodeJS binary from it and added 'nodejs' dependency.

    Package is present in my Linaro OBS repository where I keep other
    packages needed for aarch64 kolla run. Via APT pinning I mark them as
    not wanted so they are not used on any architecture with two exceptions:

    1. libvirt - we need fixed version to be able to use ThunderX servers
    2. kibana - to be able to run it at all

    For x86-64 upstream kibana package is used.

    Closes-bug: #1867365

    Change-Id: I456402849022100bde1fffdfbf6292b35690f0f2

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/train
Review: https://review.opendev.org/713491

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/stein
Review: https://review.opendev.org/713492

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/713492
Committed: https://git.openstack.org/cgit/openstack/kolla/commit/?id=84003367bcd20c24f2e5884f5d86d2420506e816
Submitter: Zuul
Branch: stable/stein

commit 84003367bcd20c24f2e5884f5d86d2420506e816
Author: Marcin Juszkiewicz <email address hidden>
Date: Mon Mar 16 12:23:50 2020 +0100

    kibana, nova-libvirt: handle it properly for non-x86

    manual backport from master changes:

    - https://review.opendev.org/#/c/712987/
    - https://review.opendev.org/#/c/713403

    Upstream kibana package contains NodeJS x86-64 binaries so can not be
    used directly on non-x86 architectures. I took upstream packages,
    removed NodeJS binary from it and added 'nodejs' dependency.

    Package is present in my Linaro OBS repository where I keep other
    packages needed for aarch64 kolla run. Via APT pinning I mark them as
    not wanted so they are not used on any architecture with two exceptions:

    1. libvirt - we need fixed version to be able to use ThunderX servers
    2. kibana - to be able to run it at all

    For x86-64 upstream kibana package is used.

    Closes-bug: #1867365

    Change-Id: I955619d6992f8714ae753dc155e248d255340ecf

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/713491
Committed: https://git.openstack.org/cgit/openstack/kolla/commit/?id=1817a09150111da4486dbe27af052c3ba8aa801c
Submitter: Zuul
Branch: stable/train

commit 1817a09150111da4486dbe27af052c3ba8aa801c
Author: Marcin Juszkiewicz <email address hidden>
Date: Mon Mar 16 12:23:50 2020 +0100

    kibana, nova-libvirt: handle it properly for non-x86

    manual backport from master changes:

    - https://review.opendev.org/#/c/712987/
    - https://review.opendev.org/#/c/713403

    Upstream kibana package contains NodeJS x86-64 binaries so can not be
    used directly on non-x86 architectures. I took upstream packages,
    removed NodeJS binary from it and added 'nodejs' dependency.

    Package is present in my Linaro OBS repository where I keep other
    packages needed for aarch64 kolla run. Via APT pinning I mark them as
    not wanted so they are not used on any architecture with two exceptions:

    1. libvirt - we need fixed version to be able to use ThunderX servers
    2. kibana - to be able to run it at all

    For x86-64 upstream kibana package is used.

    Closes-bug: #1867365

    Change-Id: I955619d6992f8714ae753dc155e248d255340ecf