Description of problem:
With the move towards Python3 we need a regression job (nightly initially) that exercises the tests on a platform that is Python3 by default.
Current regression on CentOS7 uses Python2 and to ensure we are working on both versions of Python we need a new job for Python3, and later as we move up CentOS versions or the default platform moves to Python3 we need to retain a Python2 regression job against an older platform till we support the same (this is just a note for later).
The tests can be filtered out to just run those that utilize Python code in the code base, but to start with let us get a full regression and once the pared down list is complete, we can reduce the scope of the job.
So, any version of python3 would do the trick, or do we have one in mind ?
survey of the platforms I build gluster for have:
fedora 30 (rawhide): 3.7
fedora 29: 3.7
fedora 28: 3.6.6
debian 10 (buster/sid): 3.6.7rc1
debian 9: 3.5.3
debian 8¹: 3.4.2
ubuntu 16.04 (xenial): 3.5.2
ubuntu 18.04 (bionic): 3.6.4+
ubuntu 18.10 (cosmic): 3.6.6
centos7 (w/ EPEL)²: 3.4.9 or 3.6.6
¹ We will not be building gluster-5 packages on debian 8.
² Information only. The packages in the CentOS Storage SIG will not use EPEL and will be python2.)
I'm not a python expert but I'd be inclined to be aggressive and use 3.7; 3.6.6 would be the lowest version I'd be interested in.
I have to wonder if rhel8 will update to 3.7 before it ships.
Going to target this for Sprint 5.