Add birds-eye view of open PRs to release checklist #7198
Conversation
There was a problem hiding this comment.
I don't think this makes sense in the pre-release checklist. The PR list changes constantly, this is not something you can decide is "done" in a meaningful way. This objection goes away if there is an explicit timeline for this.
Furthermore, the release checklist should have explicit instructions for what needs to be done here. Is this a blocking concern? It should not be IMO.
Presumably the step to be taken for high-quality contributions is "ask the WG and PR author about whether this should make the release and what its landing timeline is/can be". I do not want releases blocked on PRs that we think are good but have authors who have not committed to a timeline. When we plan a release we do a lot of work to ensure things are on track.
And also, depending on what we consider to be "next steps" for this, timeline matters; I do not really consider two weeks leading up to a release to be sufficient for landing large things, for example. But it might be enough to at least trigger a conversation and see where the WG stands on prioritization wrt other release blockers.
I think this needs discussion.
|
How's this? |
| * [ ] Verify that the milestone and checklist are complete | ||
| * [ ] Verify with component owners that they're ready for release | ||
| * [ ] Verify that the semver breakages (listed by the build-test job) are acceptable | ||
| * [ ] A week before the planned release, go through each open PR and consult with the author to decide the timeline for landing it and whether it should make the release. PRs in draft status or with a "waiting-on-author" label can be skipped. |
There was a problem hiding this comment.
"new" open PR, for some definition of new. Probably "opened in the last two months" is sufficient.
We have 40 open PRs, that is a lot.
There was a problem hiding this comment.
I actually think waiting-on-author is fair game for consideration but we can be conservative here.
Zibi's PR was waiting on author last week I think, though it wasn't tagged as such (we don't use that label consistently for team members)
There was a problem hiding this comment.
I didn't say "new" open PR because I think even if a PR is several months old, it should either be merged, closed, converted to draft, or labeled with waiting-on-author. If those statuses aren't enough, we should add more.
There was a problem hiding this comment.
Anything that is not a new PR should have been part of the TC milestone planning
There was a problem hiding this comment.
The waiting-on-author label is something we stick on PRs that have been waiting for more than a ~week for the author to reply to reviewer feedback.
There was a problem hiding this comment.
We did actually check PRs before the release, we included #7186 because it was very obvious. I think Manish and I both did not consider host_info blocking given that it was not part of the milestone.
There was a problem hiding this comment.
Do you have a suggestion on how to encode mid-cycle prerequisites, things like triaging issues and PRs and compiling a changelog as we go?
There was a problem hiding this comment.
Yeah, I will note that this policy would not have helped with the hostinfo situation. Zibi had made it somewhat clear earlier that he didn't have a strict timeline for it, and we had explicitly discussed that it may or may not make the actual release (which is fine since it's a util!). As release driver I was aware of this PR, but I did not know of any reason to try and get it out for the release, based on explicit discussions we had already had. The desire to advertise it as a part of the release was only brought to my attention later.
I generally think a "check the open PRs" step is useful as a part of release management, but it wouldn't have helped here. The failure was earlier in the process.
There was a problem hiding this comment.
I acknowledge that this wouldn't have much impact on host info.
There was a problem hiding this comment.
@sffc I don't know! It's an "unfunded mandate", we could do something like having a checklist issue filed each month (each two months?), but then who would it be assigned to? (It being an unfunded mandate is precisely why I'm wary of making this bottom out to being the release drivers' problem)
Right now we do it haphazardly: we sometimes do this during meetings, and sometimes the two of us do triage. I also occasionally do stuff myself but usually my focus is the upcoming release milestone (regardless of whether I am release driver).
For the changelog, specifically, I have filed #7154 and there are ideas there. Making it a per-PR responsibility ought to help.
Explicitly stating another thing that I considered part of the release duty but wasn't explicit.