The old kernel names are handed out by netname.sh which is called by
79-netname.rules. rename-network-interfaces.service does the rename by
stopping networkd, taking down the interfaces, re-adding the interfaces,
and then restarting networkd.
Note: for now we don't have any units to save/restore ipset data. We may
need to modify the current iptables units so looking into ipsets should
be included as part of that work.
Some VM platforms suspend machines by simply stopping them instead of
gracefully suspending them like real hardware would. This means that
when the system is resumed the kernel's time will be completely wrong
and it doesn't have a way to fix it. Additionally ntp will abort if the
clock offset is greater than 1000 seconds (conveniently without logging
any error messages). We can tune that in ntp.conf but ntpd has so many
knobs related to how it skews the clock and other update strategies that
the easiest option is to just restart.
The coreos-setup-environment script can now only be provided by OEMs,
the default script that used to run when the OEM didn't provide one was
not smart enough to wait for networking to be configured before writing
out /etc/environment *but* we cannot simply add such a wait because that
would make it impossible for users to use cloud config in configdrive or
similar to configure static networking.
Longer term coreos-cloudinit will replace at least some of the current
OEM coreos-setup-environment scripts and perhaps eventually integrate
with networkd to cover the default case reasonably well. Sorting this
out is still ongoing.
(This ebuild bump was supposed to land a week ago but we missed it.)
Along with the other services systemd is enabling in /etc the built in
SNTP client has been added to. Unlike the others we don't re-enable it
in /usr since we are currently using ntpd instead.
This is just a safety feature for real Gentoo systems on the off chance
that they crash moments after a package is installed. There isn't any
such concern for the CoreOS SDK so just disable it. A normal build_image
is slightly faster with this but I suspect it may help the build host
which is usually running many emerge jobs in parallel.
