Planning Council/Helios retrospective
Helios Planning Council retrospective
These notes were collected at the end of the Helios Release, at 7/7/2010 Planning Council call, specifically just to collect them. And not to solve issues, or even necessarily to suggest solutions, but just to capture issues (good and bad) while the release was still fresh on our minds. Where solutions are suggested below, it is intended to primarily better capture the issue discussed .. not to dictate a or pre-judge a solution. Action plans and solutions will be discussed in the Fall.
Helios is the fifth simultaneous release, following Callisto, Europa, Ganymede, Galileo. The Planning Council met regularly to firm come up with plan and requirements. For meeting minutes, see http://wiki.eclipse.org/Planning_Council.
Comments from PC during 7/7 meeting
items from John's note to planning council list
- First, I think the post-mortem is a bit early. Last year we did this in August, and we haven't had a chance to have all the necessary discussions within our own project to provide good input yet. I hope we will also have some time in the August call for further discussion on it.
- For planning purposes, it would be useful to know the subsequent release name earlier. We only selected the Indigo name a few weeks before the Helios release was out. I'm sure many projects start thinking about their next release before June and it would help to have a name associated with it. Perhaps we should even select the next two or three release names all at once - both for continuity/consistency of naming and avoid the annual churn of the name selection process (I vaguely remember this being done before).
- Every release train milestone and RC starting with M1 was delivered on time! This an amazing accomplishment and the first time this has been on time for the entire release cycle as far as I remember. Actually there was one exception: M2 was released a day early.
- Having the +0,+1,+2,+3 dates on contiguous days felt a bit tight. There was no room for build failures or last minute problems. I.e., a project team has only 24 hours from when their prerequisites are available until they must deliver their final bits for the milestone. If they discover a problem caused by their prerequisite they have very little time to diagnose and iterate with the upstream project. Having said that, we did manage to ship on schedule every time so maybe it is ok. One possible improvement would be to encourage projects to deliver "candidate" builds to the staging repository before their milestone release date, with the caveat that their bits could still change. That would allow downstream projects to build/test against the candidate, and report problems upstream before it is too late.
- If we are going to test and enforce legal files in bundles (about.html, etc), we need to start doing it earlier. It appears this was only tested *after* RC4 and it caused a last minute firestorm for several projects (bug 316720). I know this is the responsibility of each project, but if we have the tools available to run on the release train repository we should try doing it earlier (say M7).
- There were a few cases of unexpected contributions sneaking into the build unannounced after their due date. I think we should have the flexibility to accomodate late changes but they at least need to be communicated so anyone downstream can react. It would be nice if we could avoid it, but maybe we need to "lock" the build at the end of the +2 date and only rebuild if requested on the list for all to see. I can give examples where this happened but it's nicer to avoid naming names ;)
See also [[Planning_Council/Galileo_postmortem| last year's retrospective]