This is being added in no small part to better support running the
kubelet in more ways.
It adds up to a few hundred kbs of disk usage, and the benefit is some
tooling which desires to install the kubelet as a static binary on the
host can do so with fewer problems.
We could have picked either readline or ssl as a thing to remove in
package.use since the license issue is ssl XOR readline.
I arbitrarily picked ssl. The primary consumer of socat, the kubelet,
needs neither.
The changes in the commits shouldn't make a difference here, but it
is built from the same source as the gmerge package, so this makes
it look in sync.
The newer version of coreos-init removes the update-ssh-keys script in
preparation for the standalone update-ssh-keys package. Without this
bump, the update-ssh-keys package will step on coreos-init's toes.
coreos-metadata has been rewritten in rust
(github.com/coreos/coreos-metadata#62). This modifies the ebuild to
use the coreos-cargo eclass to build the rust project.
In some cases etcd is able to start before systemd-resolved has
populated the /etc/resolv.conf file. This commit addresses this by
adding a dependency on netowrk-online.target.
This will ensure that systemd-resolved finishes it's work before
etcd-member.service starts.
this adds the update-ssh-keys package, which is a utility for managing
groups of openssh authorized public keys. the previous incarnation of
this utility included in the operating system was a shell script
included in init. it is now a rust binary living at
github.com/coreos/update-ssh-keys.
The update-ssh-keys shell script is removed from init in
ed478168f0133cc140baa08d4d77e4333e8991d7, github.com/coreos/init#256.
This syncs the ebuild with Gentoo, while preserving cros_workon and
cross-compiler targets. It also bumps the commit to include a fix
for a use-after-free issue.
We've always installed it in the session bus directory, which is
useless, since update_engine only communicates on the system bus. This
prevented bus activation from working. We can't fix bus activation now
because that could cause update_engine to start in cases where the user
believes it to be disabled, so just drop the file.
Since we're picking up a kernel change that solves the same problem (see
previous commit), it's safe to not include this patch and wait on
upstream to pick it up.
This bumps the commit for 17.03 to remove that patch, and stops applying
it to 17.09.
1.12 never had it.
The script and udev rule were added to Gentoo due to
https://bugs.gentoo.org/556632 without much explanation. However,
iscsidev.sh has multiple problems:
- It doesn't run for all iSCSI devices, only those with a vendor string
of "IET". Thus it works for e.g. tgtd but not for enterprise iSCSI
targets.
- It was added in 2015, but udev's builtin path_id has supported iSCSI
targets -- and produced exactly the same /dev/disk/by-path links --
since systemd/systemd@a2fe9450bb in 2009.
- When the script decides not to do anything, it exits 1, producing a
series of systemd-udevd journal messages:
systemd-udevd[636]: Process '/usr/lib/udev/scripts/iscsidev.sh' failed with exit code 1.
The rule was added in 1b28d4e25452d0eacdc19074ccfbd445e88ccc11.
systemd 234 now ships a tmpfiles.d rule to create this file. Fixes
systemd-tmpfiles warning in journal:
[/usr/lib64/tmpfiles.d/var.conf:17] Duplicate line for path "/var/log/lastlog", ignoring.
By switching to the upstream rule, this changes the lastlog gid
from root to utmp and the mode from 0644 to 0664.
