New f39 (official, not beta) image #312
Conversation
|
Re: Ugh, damn python dependency breakage. The tooling images are nearly all critical to CI. This might be a problem with missing/incompatible C libraries on the system, based on: I normally turn to google first, to see if anybody else reported this/similar (including on other OSes). Otherwise somebody probably needs to sit down with |
edsantiago
force-pushed
the
f39_released
branch
from
ac16318
to
ced2ecc
Compare
|
Pushed same fix as #311. I was just being (stupidly) optimistic by hoping that the bug would be fixed by now. |
edsantiago
force-pushed
the
f39_released
branch
2 times, most recently
from
29ee77e
to
3d35ebe
Compare
|
|
edsantiago
force-pushed
the
f39_released
branch
from
d52ce39
to
3d2ab5f
Compare
|
BTW, Thanks for taking this on @edsantiago I was going to ask this afternoon, then noticed you're already on top of it 👍 |
edsantiago
force-pushed
the
f39_released
branch
5 times, most recently
from
8d5121e
to
f3360b1
Compare
|
Cirrus CI build successful. Found built image names and IDs:
|
edsantiago
force-pushed
the
f39_released
branch
2 times, most recently
from
abbdf4a
to
595727b
Compare
|
Cirrus CI build successful. Found built image names and IDs:
|
|
@cevich I think this is ready for review. Podman test PR is containers/podman#20610 |
|
Nope, still broken, bad buildah
conmon
container-selinux
containers-common
criu
crun
golang
netavark
passt
podman
runc
skopeo
slirp4netns
systemd
|
|
Cirrus CI build successful. Found built image names and IDs:
|
|
Tighter format.
|
| # has python-3.12, which causes something to barf: | ||
| # aiohttp/_websocket.c:3744:45: error: ‘PyLongObject’ {aka ‘struct _longobject’} has no member named ‘ob_digit’ | ||
| # Possible cause: https://github.com/cython/cython/issues/5238 | ||
| ARG BASE_TAG=38 |
There was a problem hiding this comment.
Oof, another one. Maybe add the time-bomb to ci/validate.sh based on a search of all files in the repo?
Stepping-back, my main concern here is the adage "There's nothing more permanent than temporary". Maybe this is an annoyance that can be considered later and/or in a separate PR.
There was a problem hiding this comment.
Cannot add timebomb here... and am reluctant to add a complicated special-case to validation. It would need to find FIXMEs, then look for a YMD, then parse, then just yuk. Some day this will blow up again, and whoever fixes it can revert to latest (if that's really desirable) or f39 or do whatever is necessary then. This really can't be futureproofed.
There was a problem hiding this comment.
Okay, thanks for your considerations and explaining your thinking.
Wow, that would be really cool if ran automatically and the output followed the table of images/IDs. Should we collaborate in a separate PR to figure a way to make that happen? |
There was a problem hiding this comment.
LGTM - assuming it fixes the crun problem, I'm willing to live with this as-is. I'm sensitive to what a PITA it is to get a new Distro version built and passing tests.
Unless you feel my other comments should not be addressed and/or in separate PRs?
If not, please feel free to merge as appropriate. Oh, and assuming podman-CI is happy, would you mind tagging this version?
(That's the magic that makes Renovate deploy it across all other repos.)
edsantiago
force-pushed
the
f39_released
branch
from
b590e44
to
14dedc0
Compare
Some day, but not now. It's too kludgey and heuristic. A better fix will be to Since this is a complicated new push, I won't tag any images yet. |
First step: create new base images:
1minutetip$ make IMPORT_IMG_SFX
1minutetip$ make image_builder_debug ....
Second step:
home$ make IMG_SFX
Commit and push. Subsequent emergency management steps:
1) Change "-qq" to "-q" in debian apt-get, so we have some
hope of figuring out what is failing.
2) debian update of grub no longer works. Try a new way.
(We can no longer update grub-common, due to dependency
error. Old grub fails with a "version_find_latest" error.
So, new solution is to provide version_find_latest).
2a) New timebomb() function will ensure that temporary
workarounds like this one do not accumulate.
3) force-update crun on f38 so we get 1.11.2.
3a) use new timebomb(), see 2a above.
4) ccia is failing due to cython issue in newer Fedora.
Force using f38, which works. Cannot timebomb().
5) fedora-aws build kept timing out. Discover and add
AWS_SOMETHING envariables to .cirrus.yml
Signed-off-by: Ed Santiago <santiago@redhat.com>
edsantiago
force-pushed
the
f39_released
branch
from
14dedc0
to
d16ced3
Compare
|
Cirrus CI build successful. Found built image names and IDs:
|
| EOF | ||
| exit 1 | ||
| } | ||
|
|
There was a problem hiding this comment.
Nice! I haven't seen a stderr redirect done like that before, interesting.
Now you've got me thinking I should add this into the common automation library. There's probably lots of places we could use this.
Thanks!
|
I pushed a tag for this too BTW. Thanks again Ed. |
|
Signed-off-by: Ed Santiago santiago@redhat.com