Reduce frequency of pushes to mozilla-central
Categories
(Testing :: General, enhancement, P3)
Tracking
(Not tracked)
People
(Reporter: marco, Unassigned)
References
(Blocks 1 open bug)
Details
The main drawback I see is that each push will contain more commits and so bisecting a failure would be a bit slower.
IIRC crash triage is done once per day, so it wouldn't be affected at all even if we went back to one Nightly per day.
|
||
Updated•5 years ago
|
|
Reporter | |
Comment 1•5 years ago
|
||
As discussed on Matrix, a first step could be to look at 2020 data to see how many patches land directly on m-c.
|
|
Reporter | |
Comment 2•5 years ago
|
||
Right now, we push to m-c every ~6 hours to have developers using central as their base use a recent commit (e.g. incorporating changes from coworkers touching the same files).
Do we know how many developers use m-c as their base vs autoland? In the review bot we saw a lot of revisions based on top of autoland (this was true at least a few months ago).
If we did one push every 8 hours (3 pushes per day) instead of one every 6 hours (4 pushes per day), we'd save ~1/4 of the m-c cost, which is not negligible.
|
|
Reporter | |
Comment 3•3 years ago
|
||
Do you see any drawback in doing this?
|
|
||
Comment 4•3 years ago
|
||
Every 6h/4 merges per day is the theoretical limit, the reality should be closer to 3 merges per weekday and 2 on the weekend. Release Management often uses Nightly bake time before they uplift changes to beta. Running nightlies only once per day would slow them down. Working with a mozilla-central base revision is the default (small check: 6/6 Try pushes based on mozilla-central). Developers want to avoid unrelated bustage/test failures.
|
||
Comment 5•3 years ago
|
||
keep in mind we have cronjobs on m-c to generate nightlies, run tier-3 jobs, etc. - these happen twice a day and I would imagine account for a good portion of our total m-c costs.
I think the savings are negligible, IIRC the remaining work when we did cost savings 2 years ago would be on retention policy of artifacts. Given the reality of a few pushes/day already I don't think a formal change in policy would result in different results. I counted 13 pushes to m-c in the last 7 days, to me that indicates we are barely getting 2x pushes/day.
|
|
Reporter | |
Updated•3 years ago
|
Description
•