Re: [LSF/MM/BPF TOPIC] development/troubleshooting workflows for BPF

[Muhammad Usama Anjum <musamaanjum@gmail.com>]

On Wed, 2024-02-28 at 20:24 -0800, Manu Bretelle wrote:
> Hi all,
> 
> Building upon the learnings and knowledge of BPF CI [0], I would like to present
> ways that can help developpers make their workflow easier, and improve their
> development and troubleshooting velocity.
> 
> While there is probably thousand different ways to get to the same result, and
> people may have their personal preferences, I will be presenting an approach
> which is close to what is used by the CI.
> Feedback and suggestions would be greatly appreciated.
> 
> 
> Some topics I have in mind:
> 
> - Reproducing CI error in CI environemt. Until recently, it was challenging to
>   Re-using artifacts produced by the CI, how to reproduce
>   the issue within the CI environment, and poke at the system.
I've been looking into the BPF CI. Sometimes vmtest.sh isn't effective
in reproducing the failures from BPF CI. I've shared this info over
the mailing list. But I've not got any response over what should be
done when vmtest.sh doesn't reproduce the same results as the BPF CI.
So if we can discuss what are the possible ways to circumvent this
problem, it would be very beneficial.

> - Building and running tests for a foreign architecture. Method to cross-compile
>   and run tests for a foreign architecture.
> - local development workflow. While most people may already have their own setup,
>   here I am proposing a workflow which is simple and allow for fast iteration.
It would be beneficial as well. vmtest.sh solves some problems. But
there may be even faster way to test regressions?

> - bisecting issues: leveraging bpf selftests and vm testing, how to leverage
>   `git bisect run` to "quickly", or at least with less friction, identify a bad
>   commit.
> 
> Thanks,
> 
> Manu
> 
> [0]: https://github.com/kernel-patches/bpf