Use modularity in repoclosure #8231
Conversation
|
It looks like it still reads the modules.d from |
ekohl
force-pushed
the
rpm/develop-use-el8-modularity
branch
from
46645f1 to
01c73ab
Compare
|
This doesn't work because we use repoclosure on EL7 nodes, which don't understand modularity. |
repoclosure/repoclosure.sh
Outdated
|
|
||
| # repo names must be unique, or yum will get confused between different OSes and URLs | ||
| reponame=undertest-$(basename $yumorig .conf)-$(echo $url | cksum | sed 's/ /-/g') | ||
|
|
||
| yumconf=$TEMPDIR/yum.conf | ||
| cat $yumorig > $yumconf | ||
| cat >> $yumconf << EOF | ||
| modulesdir=$MODULESDIR |
There was a problem hiding this comment.
I saw this in https://github.com/rpm-software-management/libdnf/blob/eff7e688e2f1ad5d10a5ed05948c08141083f316/data/tests/modules/etc/dnf/dnf.conf.in#L6 but looking at https://github.com/rpm-software-management/libdnf/blob/dnf-4-master/libdnf/module/ModulePackageContainer.cpp it is hardcoded to /etc/dnf/modules.d. I think we have to switch to containers for proper modularity repoclosure.
There was a problem hiding this comment.
Looking at https://github.com/rpm-software-management/libdnf/blob/eff7e688e2f1ad5d10a5ed05948c08141083f316/libdnf/module/ModulePackageContainer.cpp#L297-L298, you should be able to pass --installroot to the DNF call and point it at a different "root" that has an /etc/dnf/modules.d
There was a problem hiding this comment.
Yep, that works. In that case you must also pass --releasever. However, that info is also quite useful to have anyway.
Another complication is that obal doesn't use this script anyway and has duplicated the logic.
ekohl
force-pushed
the
rpm/develop-use-el8-modularity
branch
from
01c73ab to
dab6edf
Compare
|
I verified this works: $ ./repoclosure.sh yum.conf 8 --check el8-foreman-nightly --repo el8-baseos
Modular dependency problem:
Problem: conflicting requests
- nothing provides module(postgresql:12) needed by module foreman:el8:999920220718215004:6b5f9578.x86_64
- nothing provides module(ruby:2.7) needed by module foreman:el8:999920220718215004:6b5f9578.x86_64
$ ./repoclosure.sh yum.conf 8 --check el8-foreman-nightly --repo el8-baseos --repo el8-appstream --repo el8-powertools --repo el8-puppet-6 |
| stream=el$releasever | ||
| profiles= | ||
| state=enabled | ||
| EOF |
There was a problem hiding this comment.
I take it from this that if the foreman module was the only one in the config, it would not cause its dependencies to get enabled and thus we must hard-code the ruby and postgresql module configs in?
There was a problem hiding this comment.
Correct - this is also what dnf module enable foreman:el8 does. If you don't, you get conflicts. I've tried, but sadly we'll have this duplication.
I'm not sure yet how we should deal with Katello, which we don't want to enable unless it should.
There was a problem hiding this comment.
It may be possible to run dnf --installroot $TEMPDIR --releasever $releasever module enable foreman instead.
There was a problem hiding this comment.
It isn't possible:
$ dnf --installroot /tmp/repoclosure-RSRkCQ --releasever 8 module enable -y foreman
Error: This command has to be run with superuser privileges (under the root user on most systems).We always use DNF for repoclosure and that doesn't implement failovermethod. It only causes warnings in the logs.
This is unused and makes the script a lot easier.
DNF generates a correct exit code
Bash should always use tabs. Heredoc indenting doesn't understand spaces.
This removes the module_hotfixes and excludepkgs hack. To use a different modules.d directory, the --installroot parameter must be passed. This in turn requires the releasever to be set. This is actually also useful in determining which module(s) need to be enabled. Passing an installroot also uses a correct cache directory and further isolates dnf from the host system. Additionally a template is passed to mktemp for easier debugging.
ekohl
force-pushed
the
rpm/develop-use-el8-modularity
branch
from
dab6edf to
d61794a
Compare
|
Closing in favor of #10514 |
This is a test. It's incomplete, but the principle should be testable.