Tools » teuthology

Currently, we have several ways forcing multiple OS/versions for scheduled runs.

1. suites (like rados) scheduled to run on the baremetal lab (smithi) and the baremetal lab has nodes with desired imaged (e.g. trusty, centos etc)

2. suites (like upgrades) have distros yaml configuartions (https://github.com/ceph/ceph-qa-suite/tree/master/distros/supported) and essentially run a full set of tests on every OS

3. use --distro arg (not often used in nightlies ATM)

Now we want to run large suites (rados) on vps and this presents an issue of running tests with sufficient coverage on supported distros. If we use option 2, we would practically multiply amount of jobs by a number of OS'es, which is impractical for thousands of tests in rados.

It'd be nice to be able to specify, for example in teuthology-suite cli, a mixture of supported distros (defined somewhere on the list) and have jobs scheduled run on proportional mix of OS'es

Open issue - how would we schedule jobs with multiple roles ?