docker has been known to crash from time to time in odd situations. Auto
restart docker 1 second after an unexpected exit so that people can go
about their business.
Thanks to dsal & fkautz in #coreos
Biggest diff here: coreos-init has a Makefile that supports the usual
'install' and 'test' targets so no file copying is required now.
coreos-c10n has moved to init from etcd and has its own service now.
This version of init also includes support for automounting virtfs
filesystems under qemu for use with an updated version of c10n but for
now c10n remains unchanged. Optionally unit tests are available too!
This adds the following patch: (sent upstream, waiting on response)
"9p: send uevent after adding/removing mount_tag attribute"
Also enable PCI hotplug to take advantage of more qemu fun! Now
adding/removing virtio devices (which are represented as PCI devices)
at runtime via the qemu monitor console works.
We don't have a valid kernel (or use-case to have one) for "cros_host"
(the SDK) so just fake it. Also remove some unused flags.
This change prevents the latest kmod ebuild from pulling in
coreos-kernel, bootengine, and friends into the SDK.
argparse is bundled into python-2.7 so dev-python/argparse cannot be
installed. Instead the virtual package should be used to work with any
python version combination.
They have some patches to fix cross-compiling in this build environment.
Start off simply by adding 2.7 as a new parallel install target, will
switch to 2.7 as the primary version in a later step.
This flag enables 'at_console' policy support using logind. I don't
think we actually have a use for that and having it disabled hasn't
caused anything weird that I know of so far so leave it disabled.
Enabling this flag causes a circular dependency between systemd and dbus
which is resolved in catalyst bootstrapped builds like the SDK but for
target builds this is a problem.
No need for me to maintain a similar profile in two entirely different
ways. This is also one tiny step towards cleaning up our profiles in
general. Original here: https://github.com/marineam/systemd-only-overlay
As part of this change the baselayout dependency on openrc is now
handled via a use flag instead of package.provided. We didn't previously
include a virtual/init package but Gentoo has one and I needed it for my
generic systemd-only overlay so might as well include it here if it is
needed in the future.
No source directory exists so change the value of S, otherwise the
implied cd $S prior to src_install fails. This isn't an issue in oem-ami
which I was using as reference because it declares EAPI=2 which doesn't
make errors fatal by default.
ACPI power buttons are input devices! Without this it isn't possible to
trigger a graceful shutdown via qemu's command 'system_powerdown' or
whatever libvirt and similar APIs that are layered on top of it.
Probably applicable to other things too that we just don't know about.
Create a meta-ebuild for the SDK based on the packages currently listed
explicitly in bootstrap_sdk.sh and a buildhost ebuild that expands on
that, adding packages that are required in containers used by build
slave instances.
Creating a new category for this, coreos-base is overused and dev-python
didn't seem right for custom infrastructure tools. Going forward I'd
like to put SDK and build host specific stuff in this category when
previously coreos-base would be used. Things that actually land in
images would stick with coreos-base.
