AFAIK, the RBD suite already does this where possible (apparently w/ one missed directory [1]). Let me know if I am incorrect in my understanding of how to properly link to the random distro. [1] https://github.com/ceph/ceph/tree/master/qa/suites/rbd/librbd On Fri, Apr 12, 2019 at 4:31 PM Gregory Farnum <gfarnum@xxxxxxxxxx> wrote: > > Hey folks, > A while ago we changed the lab from being a mix of long-lived OS > installations to being freshly imaged on every test. It appears we > never updated the non-RADOS suites for that world. > > What does that mean? > When the machine installations were long-lived, we had a pretty even > mixture of the OSes we tested in the lab — at that time, CentOS and > Ubuntu LTS releases. But now, the test needs to select an OS to > install or it gets the default, which is apparently Ubuntu 16.04. > A small number of tasks always selected a fixed OS for one reason or > another (things like known valgrind issues, or packages that weren't > built, etc). But most of them don't. > > How do I fix it? > There's a ceph.git/qa/distros/ folder that lets you select (by > symlinking them into your suite, same as "clusters"; see the rados > suites for examples) from a "supported" matrix (rhel, centos, and > ubuntu latest), a "supported-all-distros" matrix (centos and ubuntu > latest, plus rhel7.5 and ubuntu 16.04), or a > "supported-random-distro$" that will select a random distro on every > scheduled run from the supported-all-distros list. > > It'd be great if we can get the rbd, rgw, and cephfs suites updated > for a better distribution cross-section. > Thanks! > -Greg -- Jason
