Get CI back running
kmuehlbauer opened this issue · 3 comments
tl;dr Pinning versions to those available in August 2022 and using older github-actions fixed the docker image and book build. This is not a longterm solution. This needs some discussion how to proceed here.
- #100 Trying to get CI running, almost wasted effort, since changing the wrong workflow 🙄
- #101 Fixing changes from #100, now using
bionic
as base image which corresponds tolrose-core
18.04 version - #103 fixes GLIBC issue with newer actions
- #104 Checking and Pinning dependencies to get book build running again, forcing repos2docker version to 2022.02.0
Fixed:
- container/image builds again on main and was pushed to https://github.com/openradar/erad2022/pkgs/container/erad2022
- closed all dependabot PR's until root cause of actions/checkout#1590 is identified and mitigated
- pinned Py-Art to 1.12.7 and jupyter-book to 0.12.3 make the linkchecker work again
- fixed links in wradlib-notebooks, linkchecker works again!
- fixed pinnings and versions, get book building again
Issues:
- actions/checkout#1590, need to use checkout v3
- sphinx 4.5.0 is introduced by pip installing (pyart and/or sphinx-pythia-theme) which in turn break book-build, fixed by pinning jupyter-book
- force juypter-repo2docker version to 2022.02.0, this fixes remaining build issues
- The above will only work until github restricts actions to use node20, see https://github.blog/changelog/2023-09-22-github-actions-transitioning-from-node-16-to-node-20/, that means we are bound to upgrade our base_image to include newer GLIBC anyway.
@mgrover1 #100, #101, #103 and #104 iteratively fixed the CI.
This just forced the CI, docker image etc back to versions around August 2022. That seemed to work, at least the rendered book contains the nicely rendered notebooks. See above this is no longterm solution.
@mike-dixon @leavesntwigs @DanielMichelson
If there is any movement forward on how to add lrose and baltrad to conda-forge this would really help for long-term maintenance of such repo here.
I'll leave the issue open for a while just that everyone who gets directed to this repo immediately sees what's going on.
Thanks for all the work here @kmuehlbauer - I think this is a reasonable solution.