1:27 p.m.
On 02/03/2014 10:50 AM, David Kranz wrote:
There has been a lot of interest in running tempest against real RDO
clusters. Having an rpm would make that a lot easier. There are several
issues peculiar to tempest that need to be resolved.
1. The way tempest configures itself and does test discovery depends
heavily on the tests being run from a directory containing the tests.
2. Unlike OpenStack python client libraries, tempest changes from
release to release in incompatible ways, so you need "havana tempest" to
test a havana cluster and an "icehouse tempest" to test icehouse.
The tempest group has little interest in changing either of these
behaviors. Additionally, it would be desirable if a tempest user could
install tempest rpms to test different RDO versions on the same machine.
Here is a proposal for how this could work and what the user experience
would be.
Dealing with these tempest issues suggests that the the tempest code
should be copied to /var/lib/tempest/{4.0,5.0, etc.} and the user should
configure a separate directory for each OpenStack cluster to be tested.
Each directory needs to contain:
.testr.conf
an etc directory containing the tempest.conf and logging.conf files
a symlink to the tempest test modules for the appropriate version
a copy of the test run scripts that are in the tools directory of tempest
To help the user create such directories, there should be a global
executable "configure-tempest-directory" that takes an optional version.
If multiple versions are present in /var/lib/tempest and no version is
specified then the user will be asked which version to configure.
Would it make sense to make the package name itself to have it?
This way, you could say install tempest for grizzly and update tempest
for havana - e.g.:
yum install tempest-grizzly
yum update -y tempest-havana
... and the RPMs would not obsolete each other. If we use RPM
versioning, "yum update" will blow away testing environments of
"older"
tempest versions.
We could perhaps do it using sub-rpms - which would allow us to add and
remove tempest suites as we move along:
* ship no files except perhaps license in 'tempest' itself
and stand-up environment bits
* use subpackages for tempest-grizzly/tempest-havana/tempest-icehouse/...
* get a special exception to *not* make tempest-* child
RPMs require a fully-versioned tempest base package all
the time just in case you wanted to have older tempest
for a given release
e.g. tempest-2.0.0
tempest-grizzly-1.0.0
tempest-havana-2.0.0
... could all be installed, and if done right, you could
'yum update -y tempest-grizzly' to 3.0.0 without breaking
the tempest-havana-2.0.0 package.
Just a thought.
User experience:
1. Install tempest rpm: yum install tempest-4.0
The above would mean:
yum install -y tempest-havana
yum install -y tempest-grizzly
etc.
2. Run configure-tempest-directory
3. Make changes to tempest.conf to match the cluster being tested (and
possibly logging.conf and .testr.conf as well)
4. Run tempest with desired test selection using
tools/pretty_tox_serial.sh or tools/ pretty_tox_serial
Does any one have any comments/suggestions about this?
-- Lon