openstack-qa-team team mailing list archive

[corvus@xxxxxxxxxxxx (James E. Blair)]

Jay Pipes <jaypipes@xxxxxxxxx> writes:

> On 09/07/2012 08:13 PM, Dan Smith wrote:
>> 
>> Running tempest tests in parallel is definitely a good goal to
>> have. However, I believe that all the tests run in a gate are done
>> sequentially right now, correct? Seems like if we kicked off tempest at
>> the same time as the regular gate stuff, much of their execution would
>> overlap. I'm not sure if the current CI stuff does any parallel jobs
>> like that, but I think it shouldn't be too hard to do and would help
>> with the latency-to-merge issue.

The gate jobs are run in parallel -- so gate-nova-python27 is started at
the same time as gate-tempest-devstack-vm, assuming there are nodes
available to run them.  Depending on whether it's running the smoke or
full tests, and depending on whether it gets a fast or slow compute
instance, the tests take about 15-40 minutes, which makes them either
slightly slower than the nova unit tests, or much slower.

> Part of the problem is that the CI system runs Tempest on the same CI
> node as the OpenStack environment that devstack installs. If we could
> run multiple Tempest workers that each fired a subset of the full
> tempest run against a CI node, that would solve the parallel problem.
> But unfortunately, that's just not how Jenkins works :(

You could, however, spawn them from the devstack-gate script, wait for
them, then OR their results.

But how much more RAM would parallel tests require?  If we need to
increase the node size (we're at 4G now), we can, but I don't want to do
so unnecessarily.

-Jim