On 06/06/2013 07:10 AM, Prasanna Santhanam wrote:
On Wed, Jun 05, 2013 at 05:39:16PM +0000, Edison Su wrote:
I think we miss a VOTE from Jenkins, the vote from Jenkins should
be taken as highest priority in each release. This kind of
regression should be easily identified in Jenkins(If we have a
regression test for each environment).
+1 - need more people focussed on cloudstack-infra in general.
The 41 regression with local storage, that required 2 or more hosts to
duplicate, would be one example of an issue that would be detected by
automatic testing provided the testing is done on a sufficiently big
test fixture.
Q: How many hosts are used in daily testing now?
/Ove
-----Original Message-----
From: Marcus Sorensen [mailto:shadow...@gmail.com]
Sent: Wednesday, June 05, 2013 9:03 AM
To: dev@cloudstack.apache.org
Subject: KVM development, libvirt
It looks like a bug was probably introduced into 4.1, where stock Ubuntu
12.04 doesn't support part of the libvirt xml format for system vms. I feel bad
that it got in there, but I think it highlights an issue that needs to be
addressed within our development. Libvirt versioning is somewhat of a
moving target. Features are introduced rapidly, and versions vary quite a bit
between distributions and releases of those distributions. Despite this,
we've largely ignored what libvirt we are targeting, assuming "whatever is on
the distribution". There is the occasional discussion about this or that being
available in libvirt x.x.x during the development cycle, but when it comes to
qualifying the release we don't pay attention to it.
We should. Looking at the vote for 4.1.0, several people call out which
OS/distribution they use, but I'd like to see the libvirt version as well.
At the very least - we should test for the versions of OS we decide to
support. I use a base CentOS and install all packages from bare OS to
cloudstack management / agent start. But I haven't been able to
automate the kickstarts for Ubuntu. If *anyone* has got that working,
please enlighten the list, then I can include ubuntu tests for
packaging on the test infrastructure so we don't have to constantly
test these things manually.
Here are some initial thoughts, please add to these if you can:
1) When voting for a release, should we require a minimum number of votes
FOR EACH supported OS? Meaning that we require positive test results from
every OS listed as supported? In retrospect this seems like a no-brainer,
however it may change the bylaws.
I think we need to have that included for hypervisors too. I often
find that XCP becomes the bastard child in these situations. It would
be nice to have people include their version of hypervisor/OS but
since most are testing devcloud as recommended in our test procedure I
daresay we'll catch much. All this needs to be automated.
(Any cloudstack clouds out there want to provide some tenant accounts
where I can run some packaging tests for different OSes?)
2) Do we want to pull libvirt out as a standalone dependency? Meaning that
we code to a specific version and make that more visible. This could be a
"least common denominator" type thing where we pick the lowest version
from all supported OSes, or it could be independent of distribution,
whatever we decide, but we would make an effort to call out the version and
treat it independently of OS.
3) I can think of a few things we could do in packaging to help catch
versioning, but I'm not sure they would entirely address the issues.
Please bring them forward.
--
Ove Everlid
System Administrator / Architect / SDN & Linux hacker
Mobile: +46706662363
Office: +4618656913 (note EMEA Time Zone)
