Open Stack

On 1/6/23 9:12 PM, Clark Boylan wrote:
> On Fri, Jan 6, 2023, at 3:56 PM, Ghanshyam Mann wrote:
>> ---- On Fri, 06 Jan 2023 10:41:43 -0800  Ghanshyam Mann  wrote ---
>>   >  ---- On Fri, 06 Jan 2023 05:42:45 -0800  Brian Rosmaita  wrote ---
[snip]
>> Pinning in run-both.yaml playbook did not fix the python-cinderclient
>> issue and pinning tox<4 in
>> tox.ini is the way forward for this case.
> 
> I don't think this is a proper fix. This goes back to the concern I already mentioned on this thread. The correct way to fix this is to ensure we aren't installing tox multiple times with the final install being the version we want. We should ensure we install it once with the correct version.
> 
> The reason the python-cinderclient change failed is that devstack is blindly installing tox here: https://opendev.org/openstack/devstack/src/branch/master/lib/neutron_plugins/ovn_agent#L369-L370 which is installing latest tox per this log: https://zuul.opendev.org/t/openstack/build/961c429cd9fc4d649e8714aba67f052d/log/job-output.txt#9211-9279.
> 
> The problem with adding requires = tox<4 in tox.ini is that this will cause tox to install a new tox in a new venv unnecessarily simply to run the target under an older tox. If we fix devstack instead then we can install tox once and everything should work.

I think we have two separate issues here.  The cinderclient functional 
test job just wants devstack to be up and running so that tox-based 
cinderclient tests can be run against devstack.  I don't see that it's 
necessary that cinderclient have to use the same tox version to conduct 
its tests that devstack has installed for whatever reason devstack is 
installing tox.  There may be good reasons for using different versions.

In other words, it's not obvious to me that making devstack istelf 
tox-consistent implies that other projects running tox-based jobs 
against devstack have to use that same tox version.

> 
>>
>> -gmann
>