Are we talking about this on top of the 1/10 sampling CI on master/4.x already? I don't think it'll help much.
On Wed, Nov 18, 2026 at 3:31 PM Hyukjin Kwon via dev <[email protected]> wrote: > > Your idea is right in principle, but on master it would save almost > nothing ... > > ⚰️☠️🪦 > > On Thu, 19 Nov 2026 at 08:22, Hyukjin Kwon <[email protected]> wrote: > >> Yes yes. Let me share the result later today >> >> On Thu, Nov 19, 2026 at 8:20 AM Nicholas Chammas < >> [email protected]> wrote: >> >>> I think I misunderstood your proposal. I think you are talking about a >>> more limited case where PR CI passes against the latest master commit, >>> and is then merged before master moves. Correct? I think that’s safe to >>> do. >>> >>> Curious what your research on the potential savings shows. >>> >>> >>> On Nov 18, 2026 at 6:16:09 PM, Nicholas Chammas < >>> [email protected]> wrote: >>> >>>> The main objection to this idea is that master may progress after PR >>>> CI passes, so the build can fail silently post merge if we do not either a) >>>> test every commit on master, or b) require that a PR pass CI *on the >>>> latest* master before merging. >>>> >>>> b) is not practical without a merge queue, because PRs will constantly >>>> be chasing a moving target. >>>> >>>> I am in favor of building our own merge queues. I think it’s the right >>>> long-term approach. Happy to invest in this if there is support for the >>>> idea. >>>> >>>> >>>> On Nov 18, 2026 at 6:10:00 PM, Hyukjin Kwon via dev < >>>> [email protected]> wrote: >>>> >>>>> Let me actually do a quick research and how much we can save by doing >>>>> so >>>>> >>>>> On Thu, 19 Nov 2026 at 08:09, Hyukjin Kwon <[email protected]> >>>>> wrote: >>>>> >>>>>> Hi all, >>>>>> >>>>>> To reduce the CI usage, I am thinking about skipping the CI build >>>>>> IFF the CI passed against the latest commit in the master branch. WDYT? >>>>>> >>>>>
