On Thu, Mar 11, 2021 at 11:58:54PM +0100, Ilya Maximets wrote: > On 3/11/21 11:29 PM, Ben Pfaff wrote: > >>>>>> Few bugfixes was accepted recently to OVS for ovsdb-cs code > >>>>>> that are required for OVN build. > >>> I actually thought that the submodule was a recommended OVS version as > >>> well as what OVN builds against. If that's not so, then yes we should > >>> clarify somehow. > >> > >> It's a recommended OVS version to build with. I'd not recommend using > >> it to actually build ovs-vswitchd or ovsdb-server, because it will > >> mean that you're running non-released version of OVS that is always > >> behind even the master branch. > > > > Hmm, OK. > > > > I would have guessed that for release branches we would point this to a > > release tag of OVS and that for the master branche we would point this > > to a commit in the master branch of OVS. > > This might be a good approach, but there is one major issue: > OVN has 2x as many releases as OVS. This means that every > second OVN release will have to point to some commit on OVS > master anyway if we want some new features there. > > I'd say that eventually OVN should become less feature hungry > or less dependent from OVS changes. Then we should consider > to use only stable released OVS as a build dependency and > actually wait for the next major OVS release if some new features > required from it. But right now we can't afford that.
We had almost the same problem at Nicira when we were separately developing OVS and our controller NVP. We sort-of solved it as the OVS team by trying to stay a couple of releases ahead of NVP. It mostly worked because the OVS team was more nimble than the NVP team and we were usually able to point out why they didn't really need a new feature from us anyway. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
