Re: smaller cloud images, raw cloud images
On Fri, Aug 07, 2020 at 06:49:14PM +0200, Thomas Goirand wrote:
> On 8/7/20 1:55 PM, Thomas Lange wrote:
> > Can't we just run a script on petterson after each release (not on the
> > dailys) to untar the disk image, rename it (because it's always
> > disk.raw) and generate a sha512sums file? Yes, the raw will be of 2G
> > size, but this is much less than the 310 subdirectories in our daily
> > builds. We also can have the sha512sum of the raw in the json file and
> > check this.
>
> Well, can't we just STOP the non-sense of building images on Salsa all
> together?
You've voiced concerns about using gitlab-ci, but I haven't heard that
anyone else agrees with you. You've raised that we're not building on
casulana, but that's not correct. Are there other reasons for your
objection?
gitlab-ci provides us a very nice suite of features that are nontrivial
to find or implement:
- gitlab ci is a well-documented and well-supported. It's easy to
learn, even if you can only access salsa or gitlab.com.
- it provides built-in support for parallel jobs, dependency
relationships, conditional builds, test result reporting, etc.
- It makes the build process easy to observe and check, even for people
that have no access to salsa or any Debian system.
- Official build jobs execute on casulana, while development jobs run on
unprivileged shared runners using exactly the same API/interfaces.
> This was an unilateral move of Waldi without any discussion,
> even yet an approval, from the team. What we were suppose to do was
> building the images on Casulana.
This doesn't match with my memory. As I recall: using gitlab-ci was
raised at the 2017 sprint, but folks wanted to see an example
implementation. Between that sprint and the next, Bastian built the
first release of the gitlab-ci pipeline. After an initial thread on the
list, we discussed it at the 2018 sprint. You were concerned that the
images were being built on the shared runners. Bastian explained that
the builds were in fact running on casulana, salsa just orchestrates the
process. This is still true: every official job runs on casulana.
I don't recall any other objections. Most people seemed to think it
looked okay. Am I forgetting part of the discussion?
> At this point, we see that it is bad in many regards, the most blocking
> point being that it is impossible to save an artifact bigger than 256GB.
> This prevents me from building an Octavia image, for example...
I agree, this limit is an issue. It sounds like we might have some
progress on shrinking the images more. But we'll inevitably run into
this limit again in the future. The last time it came up, Bastian said
the salsa admins would not consider changing it.
> I also would like to discuss (here) if we could move the regular meeting
> time to business hours, because really, I hate the evening meetings
> because either I'm busy, or I just forget. If I remember well, the
> reason it was on the evening was because of Serpent, who's not part of
> the team anymore. Is there anyone else for who a late afternoon (to be
> US timezone inclusive...) meeting wouldn't be convenient?
>From the previous thread, it looks like 17:00UTC was the next best time.
Would that count as business hours? If so, I'll raise this during this
week's meeting.
Ross
Reply to: