On 2023-02-07 23:37, Chris Johns wrote:
On 7/2/2023 9:31 pm, Christian MAUDERER wrote:
On 2023-02-07 07:03, Chris Johns wrote:
On 30/1/2023 10:12 pm, Christian MAUDERER wrote:
Hello,
recently the following tickets were added (beneath a few more related ones):
https://devel.rtems.org/ticket/4790 - Setup Gitlab instance
https://devel.rtems.org/ticket/4791 - Update BuildBot
It's great that a patch review system and a CI/CD that builds every patch for
RTEMS starts to get within reach. Thanks a lot to all involved in that for the
efforts.
It is exciting to see this happening. It will benefit the project and all who
give their time.
I reviewed earlier discussions related to CI/CD. From my point of view there are
mainly two points that are missing in the tickets:
First: From my point of view, we should make it simple for new users to
register. Adding authentication using well-known services can help with that.
GitLab supports (for example):
- GitHub: https://docs.gitlab.com/ee/integration/github.html
- GitLab.com: https://docs.gitlab.com/ee/integration/gitlab.html
- Google: https://docs.gitlab.com/ee/integration/google.html
- ...
I think it would be good to select the most common ones (at least the three
mentioned above) and add them as a goal to the ticket or a new one. What do you
think?
I suggest a ticket to handle authentication. If you create a ticket please
indicate it is unfunded. If this is handled else where and in another ticket
this one can be closed with a suitable reason.
https://devel.rtems.org/ticket/4840
Thanks.
The addition of these authentication methods can be done when someone funds the
work. If a person or organisation thinks it is important please reach out to
Amar directly.
Second: It's still a bit unclear for me how the CI/CD with BuildBot will work.
Will it be possible for anyone to help improve the CI/CD? An example to make it
clear what I want to know: Let's assume an unprivileged developer has a patch
set that allows building device tree files using the RTEMS build system, but the
patches require a new tool like dtc. Let's further assume that the idea has been
discussed and everyone agrees that it is a good idea (currently not yet the case
for dtc). Problem is: The patches trigger a CI error because the new tool is
missing and therefore can't be merged yet. How can the developer suggest a fix
so that the patches can be accepted faster without having to wait for one
specific maintainer to have enough time for adapting the CI config?
There are details that will need to be worked out. Deployment of tools for
building in a CI flow is important. How complex and automated this will be will
depends on the funding provided. At this point in time the push is to get a
basic framework up that allows us to handle simple merge requests. The approach
is more organic in nature so funding can be targeted at the most important
foundation pieces. Further funding can build on that base. Before we get to
Gitlab and CI we need to rebase the servers and software on them. This part of
the effort is funded and under way. Amar is updating the tickets with the
progress.
Maybe my question hasn't been clear enough. Of course part of it depends on what
is implemented. But every selected system also adds limitations. At the moment
BuildBot is the suggested system in the tickets. It is a well known service and
has a lot of usefull features for us that certainly make it a good choice.
But I never really worked with it, so I basically wanted to know what I have to
expect. Some systems like (I think it was) jenkins.io can be configured to
behave quite different depending on how you use it. You can either configure it
via a static configuration that is put somewhere where only admins can access
it. But you can also configure it in a way that build jobs are defined by yaml
files in the repository similar to popular services like GitHub, GitLab or
Travis.
Basically what I wanted to know is: What is possible / usual with BuildBot and
which of the possible solutions do we want to use?
A developer with a gitlab account will be encouraged to move their personal repo
to their gitlab account and doing that allows them to use the gitlab CI tools.
Gitlab will be running on FreeBSD and in a jail so any dependent services a
developer needs will need to be requested via a ticket. I hope you appreciate
there are limit here related to resources and the hardened nature of the
environment.
Of course there are limits. That's OK.
Do we configure BuildBot
statically and every change will be done only if someone pays for it?
The main CI build flows will be controlled and changed via a ticket. We need to
QA control that process to ensure the patches merged are suitable. I have no
idea yet how this will look and work but it will need to allow the project to
make updates. Wether this is widely available on the first pass or pass ten of
the work that needs to be done I cannot say.
Of course every process should be checked before it is executed on our
runners. My concern is that everyone should be able to suggest such
changes in the form of patches (or similar) to the build scripts.
The worst case (from my point of view) would be if I have to write a
ticket "please add new tool x to the toolchain" into a ticket and only
one person can add it. In this case the one person will be a bottleneck
and new features can only be added as soon as that person has time.
The best case would be "I added tool x to the toolchain. Here is a patch
that I tested on a local instance. Please review and merge it." We still
have a bottleneck because only few people can merge it. But at least
it's less work for them.
Will there
be a repo with config files where everyone can suggest and help but only one or
two admins can accept changes if they have time? Or will every repo contain a
config yaml (or similar format) and every maintainer can accept a change to
that? Or is the currently discussed solution something completely different?
Your personal repo will have the standard gitlab features. I am not sure what
they are. A merge request to a project repo that changes the gitlab CI config
will need to be approved. It should be treated like any other part of the
process. I also do not know how the project CI configs will operate in a
personal repo. That is another issue to work out.
I think that's nearly what I described as best case above: I can create
a merge request for the CI config even if I maybe can't test it before
the merge request.
That shouldn't be a pure decision by the one who pays for the work but one that
is (in the optimal case) discussed and specified in the tickets.
I did not know that was happening. I am sorry if you think it is and if I gave
you that impression.
There have been relatively new tickets that changed the status from
"needs-funding" to "funded" nearly without delay. That was a bit
surprising for me.
It seems that the tickets are still updated based on feedback so that is
fine and it's great that someone funded them. It just would have been
nice if there would have been some announcement for two or three days
like "Someone wants to fund the tickets 1, 2 and that part of 4 exactly
like they are written now. If there are big objections with the
direction, please speak up now."
That way
everyone in the community has at least a chance to have some influence on the
system that he will have to use later.
That is fair. Every attempt is being made to involve the community. The tickets
are detailed and open. Anyone can fund the work. I will point out the funding is
not via a foundation and so it can be directed. It is not ideal but this is the
best solution I can find so far.
Of course, it's OK that funding is directed to the tickets that someone
wants to have funded. It's like most RTEMS development works at the
moment and I wouldn't make that different. Like I said above: Would just
have been nice to announce it a bit more clearly which parts of which
tickets are funded and give a short period to object in case one of the
tickets has a direction that is not wanted by some community members.
At the moment I don't need all tickets from my point of view. But there
are no tickets that I would see as a problem. So again: I'm happy that
our system will get better.
Best regards
Christian
Chris
--
--------------------------------------------
embedded brains GmbH
Herr Christian MAUDERER
Dornierstr. 4
82178 Puchheim
Germany
email: christian.maude...@embedded-brains.de
phone: +49-89-18 94 741 - 18
mobile: +49-176-152 206 08
Registergericht: Amtsgericht München
Registernummer: HRB 157899
Vertretungsberechtigte Geschäftsführer: Peter Rasmussen, Thomas Dörfler
Unsere Datenschutzerklärung finden Sie hier:
https://embedded-brains.de/datenschutzerklaerung/
_______________________________________________
devel mailing list
devel@rtems.org
http://lists.rtems.org/mailman/listinfo/devel
