On Wed, Jan 22, 2020 at 02:12:11PM -0800, Clark Boylan wrote: > On Tue, Jan 7, 2020, at 1:45 AM, Roman Gorshunov wrote: > > Hello Clark, > > > > Thank you for your reply. Meeting time is OK for me. I have forwarded > > invitation to Pete Birley and Matt McEuen, they would hopefully join > > us. > > I wanted to make sure we got a summary of this meeting sent out. Notes were > kept at https://etherpad.openstack.org/p/Wqxwce1UDq > > Airship needs to test their All in One deployment tool. This tool deploys > their entire bootstrapping system into a single VM which is then used to > deploy other software which may be clustered. Because the production usage of > this tool is via a VM it is actually important to be able to test the > contents of that VM in CI and that is what creates the memory requirements > for Airship CI. > > We explained the benefits of being able to run Airship CI on less special > hardware. Airship gains redundancy as more than one provider can supply these > resources, reliability should be improved as nested virt has been known to be > flaky, and better familiarity within the community with global resources > means that debugging and working together is easier. > > However, we recognize that Airship has specific constraints today that > require more specialized testing. The proposed plan is to move forward with > adding a new cloud, but have it provide specialized and generic resources. > The intent is to address Airship's needs of today with the expectation that > they will work towards running on the generic resources. Having generic > resources ensures that the infra team has exposure to this new cloud outside > the context of Airship. This improves familiarity and debuggability of the > system. It is also more equitable as other donations are globally used. Also, > Nodepool doesn't actually allow us to prevent consumption of resources > exposed to the system; however, we would ask that specialized resources only > be used when necessary to test specific cases as with Airship. This is > similar to our existing high memory, multi numa node, nested virt enabled > test flavors. > > For next steps we'll work to add the new cloud with the two sets of flavors, > and Airship will begin investigating what a modified test setup looks like to > run on our generic resources. We'll see where that takes us. > > Let me know if this summary needs editing or updating. > > Finally, we'll be meeting again Wednesday January 29, 2020 at 1600UTC to > followup on any questions now that things should be moving. I recently used > jitsi meet and it worked really well so want to give that a try for this. > Lets meet at https://meet.jit.si/AirshipCICloudFun. Fungi says you can click > the circled "i" icon at that url to get dial in info if necessary. > > If for some reason jitsi doesn't work we'll fall back to the method used last > time: https://wiki.openstack.org/wiki/Infrastructure/Conferencing room 6001. > Regarding the dedicated cloud, it might be an interesting discussion point to talk with some of the TripleO folks from when tripleo-test-cloud-rh1 cloud was still a thing. As most infra people know, this was a cloud dedicated to running tripleo specific jobs.
There was an effort to make there jobs more generic, to run on any cloud infrastructure, which resulted IMO, to a large increase of testing (as there was much more capacity). While it took a bit of effort, I believe overall it was a better improvement for CI. Paul _______________________________________________ OpenStack-Infra mailing list OpenStack-Infra@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-infra
