[tempest] what is a proper way to install a package into a vm instance spawned by tempest?

Roman Safronov rsafrono at redhat.com
Mon Mar 23 17:51:04 UTC 2020

Actually we need such test because we are testing Openstack as a whole
product (including Neutron+openvswitch+OVN+Nova+libvirt+Octavia etc.)
that's why I think neutron functional test would be not enough. We are
creating tests covering scenarios that our customers tried to use and
encountered issues.
For example this is a bug reported downstream on an issue happened in this
scenario: https://bugzilla.redhat.com/show_bug.cgi?id=1707241
There were more reported issues on similar use case and we would like to
catch such issues before the product is released.

Anyway, as I said this specific test runs stable in downstream CI on
virtual multi node environments with nested virtualization. It usually runs
with RHEL8 image but I also tried it with standard Ubuntu-18.04 guest image
used by upstream CI gates. The only issue is that keepalive package
installation by 'apt install' for some reason does not work when running on
upstream gates causing the test to be skipped. I just would like to
understand if running 'apt install/yum install' inside VMs spawned by
upstream tempest tests is not acceptable at all or I am missing something
(maybe proxy settings?).

On Mon, Mar 23, 2020 at 5:36 PM Clark Boylan <cboylan at sapwetik.org> wrote:

> On Sun, Mar 22, 2020, at 9:10 AM, Roman Safronov wrote:
> > Hi,
> >
> > I wrote a tempest test
> > <
> https://review.opendev.org/#/c/710975/https://review.opendev.org/#/c/710975/)>
> which requires keepalived to be running inside vm instance.
> > The test tries to install keepalived package inside vm by running "apt
> > install/yum install".
> > However as I see in upstream gates logs this does not work while the
> > same code works perfectly in our downstream CI using the same image.
> >
> > Do vm instances spawned on upstream gates during tests have a route to
> > the internet?
> > Is there an alternative way that I can use in order to install a
> > package?
> By default the tempest jobs use a cirros image. This is a very small,
> quick to boot linux without a package manager. If you need services to be
> running in the image typically you'll need to boot a more typical linux
> installation. Keep in mind that nested virt isn't on by default as it isn't
> available everywhere and has been flaky in the past. This makes these types
> of installations very small which may make reliable VRRP testing difficult.
> Thinking out loud here, I'm not sure how much benefit there is to testing
> VRRP failures in this manner. Do we think that OpenStack sitting on top of
> libvirt and OVS will somehow produce different results with VRRP than using
> those tools as is? To put this another way: are we testing OpenStack or are
> we testing OVS and libvirt?
> One option here may be to construct this as a Neutron functional test and
> run VRRP on Neutron networks without virtualization mixed in.
> >
> > Thanks in advance
> >
> > --
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20200323/862fe267/attachment.html>

More information about the openstack-discuss mailing list