| Age | Commit message (Collapse) | Author |
|---|
|
Adds a udev script which will invoke a hotplug hook script on all net
add events. The script will write some udev arguments to a systemd FIFO
socket (to ensure we have only instance of cloud-init running at a
time), which is then read by a new service that calls a new 'cloud-init
devel hotplug-hook' command to handle the new event.
This hotplug-hook command will:
- Fetch the pickled datsource
- Verify that the hotplug event is supported/enabled
- Update the metadata for the datasource
- Ensure the hotplugged device exists within the datasource
- Apply the config change on the datasource metadata
- Bring up the new interface (or apply global network configuration)
- Save the updated metadata back to the pickle cache
Also scattered in some unrelated typing where helpful
|
|
Also added a new (currently experimental) systemd-networkd renderer,
and includes a small refactor to cc_resolv_conf.py to support the
resolved.conf used by systemd-resolved.
|
|
Rocky Linux is a RHEL-compatible distribution so all changes that have
been made should be trivial.
|
|
AlmaLinux OS is RHEL-compatible so all the changes needed are trivial.
|
|
This ensures that KVP data can be correctly reported to Azure.
|
|
(#546)
Lintian spotted the following issue:
The specified systemd .service file contains both DefaultDependencies=no
and Conflicts=shutdown.target directives without Before=shutdown.target.
This can lead to problems during shutdown because the service may linger
until the very end of shutdown sequence as nothing requests to stop it
before (due to DefaultDependencies=no).
There is race condition between stopping units and systemd getting a
request to exit the main loop, so it may proceed with shutdown before
all pending stop jobs have been processed.
Please add Before=shutdown.target.
[1] https://lintian.debian.org/tags/systemd-service-file-shutdown-problems.html
Co-authored-by: Rick Harding <rharding@mitechie.com>
|
|
We use "rhel" consistently everywhere else.
|
|
The name `redhat' is not used but rather `rhel' to identify the distro.
Signed-off-by: Eduardo Otubo <otubo@redhat.com>
|
|
In Centos/RHEL 8, NetworkManager is installed as the networking service.
cloud-init.service needs to run After this service to allow the OS to
bring networking up first.
LP: #1843334
|
|
Change the startup order. On SUSE distros cloud-init.service
should start after wicked.service. This reflects the same
start up order other distributions utilize.
|
|
Update the template to use libexec prefix path to ds-identify on redhat
systems.
LP: #1833264
|
|
The systemd generator used had a hard coded path for the location target
file to create. This path does not apply to all distributions.
Make the generator and template to have the path set during build time.
|
|
With db50bc0d9 the sysconfig renderer was enabled for
openSUSE and SUSE Linux Enterprise. This requires that
cloud-init.service starts before wicked.
LP: #1799709
|
|
This makes cloud-config.service (and as a result cloud-final.service)
run After snap.seeded.service. This is required to ensure that
pre-seeded snaps can be used by cloud-init or user-data input.
The snap.seeded.service was added to snapd at:
https://github.com/snapcore/snapd/pull/5124
Note that the following would be a workaround:
snap:
commands:
00: snap wait system seed.loaded
LP: #1767131
|
|
On openSUSE and SLES systemd is allowed to pull the default dependencies,
thus setting "Before=basic.target" is incorrect and creates a cycle leading
to the deletion of the cloud-init-local.service
|
|
Depending on distribution the default number of tasks (threads) maybe
unexpectedly low or it may be the default systemd setting (512).
Setting TasksMax to "infinity" in cloud-init-final.service removes the
restriction on tasks created.
LP: #1717969
|
|
Ordering on apt service should only be set up on Debian based
distributions. This changes is really a net-zero in runtime result.
But, mentioning apt on a rpm based distro could be confusing.
|
|
Things done here:
- identify 'suse' as a variant in util.system_info and
also tools/render-cloudcfg.
- update systemd and cloud.cfg templates for suse specific changes.
LP: #1718640
|
|
Revert "centos: do not package systemd-fsck drop-in."
Revert "systemd: make systemd-fsck run after cloud-init.service"
The systemd-fsck drop-in caused regressions by introducing ordering
The change reverts the original commit that added systemd-fsck drop-in
and another commit that had removed that from the centos packaging:
1f5489c258a26f4e26261c40786537951d67df1e
8a5296c41db45be3a172862f324ad44e732a2250
The result is to no longer provide the systemd-fsck drop-in.
LP: #1717477
|
|
cloud-init.service may write filesystems (fs_setup) or re-partition
(disk_setup) disks.
If systemd-fsck is running on a device while that is occuring
then the partitioning or mkfs might fail due to the device being busy.
Alternatively, the fsck might fail and cause subsequent mount to fail.
LP: #1691489
|
|
Under el7, cloud-init systemd files need some unit tweaks to ensure
they run at the right time. Pull in current el7 downstream systemd unit
changes.
|
|
This changes all cloud-init systemd units to run 'Before' the apt processes
that run daily and may cause a lock on the apt database.
apt-daily-upgrade.service contains 'After=apt-daily.service'.
Thus following order is enforced, so we can just be 'Before' the first.
apt-daily.service
apt-daily-upgrade.service
Note that this means only that apt-daily* will not run until
cloud-init has entirely finished. Any other processes running apt-get
operations are still affected by the global lock.
LP: #1693361
|
|
Simply fix a typo in a comment. No functional change.
|
|
Network configuration version 2 format is implemented in a package
called netplan (nplan)[1] which allows consolidated network config
for multiple network controllers.
- Add a new netplan renderer
- Update default policy, placing eni and sysconfig first
This requires explicit policy to enable netplan over eni
on systems which have both (Yakkety, Zesty, UC16)
- Allow any network state (parsed from any format cloud-init supports) to
render to v2 if system supports netplan.
- Move eni's _subnet_is_ipv6 to common code for use by other renderers
- Make sysconfig renderer always emit /etc/syconfig/network configuration
- Update cloud-init.service systemd unit to also wait on
systemd-networkd-wait-online.service
1. https://lists.ubuntu.com/archives/ubuntu-devel/2016-July/039464.html
|
|
ds-identify is run here from the generator. If ds-identify does
not see any datasources, it can completely disable cloud-init.
The big value in this is that if there is no datasource, no python
will ever be loaded, and cloud-init will be disabled.o
The default policy being added here is:
search,found=all,maybe=all,notfound=disabled
That means:
- enable (in 'datasource_list') all sources that are found.
- if none are found, enable all 'maybe'.
- if no maybe are found, then disable cloud-init.
On platforms without DMI (everything except for aarch64 and x86),
the default 'notfound' setting is 'enabled'. This is because many of
the detection mechanisms rely on dmi data, which is present only on
x86 and aarch64.
|
|
While cloud-init writes its data to /var/lib/cloud, we
previously lazily added RequiresMountsFor=/var/lib . It is more correct
to list the more complete path.
LP: #1642062
|
|
cloud-init.service cannot currently run
After=systemd-networkd-wait-online.service
as systemd-networkd.service is After dbus.service, but
cloud-init.service is:
Before=sysinit.target
and sysinit.target is before dbus.service.
The result is that we are temporarily preferring for cases
where there is systemd-networkd in place to have cloud-init.service run
without networking properly configured.
See bug 1636912 for more information.
|
|
sysinit.target happens earlier in boot than basic.target.
cloud-init-local.service had:
Before=basic.target
but cloud-init.service had:
After=cloud-init-local.service
Before=sysinit.target
The result was that cloud-init-local.service was indirectly
Before sysinit.target anyway, so this change is just to more
specifically state that.
|
|
Drop mention of local-fs.target from both cloud-init-local.service
and cloud-init.service. This could actually could cause a loop in
ordering due to cc_mounts writing mount points with:
x-systemd.requires=cloud-init.service
Rather, we specifically list /var/lib/ and systemd-remount-fs.service
in cloud-init-local.service. cloud-init.service will run after
cloud-init-local and thus will have these satisfied.
|
|
Here we drop the Requires=networking.service but keep the
After=networking.service. The change there is that we no longer
force networking.service to run, as the system may not have it or
be configured to use it.
Second, we add After=systemd-networkd-wait-online.service.
That causes us to run after systemd-networkd has brought networking
up.
The end result is that we now run after both ifupdown or
systemd-networkd (if they were going to run) but do not force either
to run.
One other change is that by dropping Requires=networking.service
cloud-init will now run even if networking failed to come up.
LP: #1636912
|
|
Per discussion on bug 1636912 and subsequent feedback on bug 1629797,
using:
Before=sysinit.target
is a simpler solution than what we had implemented:
Before=basic.target
Before=dbus.target
LP: #1629797
|
|
The change reported to fix bug 1629797 was incorrect. It added:
Before=dbus.target
and its commit message referenced dbus.service.
dbus.target does not exist, and dbus.service would not solve the
problem. The change did not fix the problem nor actually make any
difference.
The fix here is just to use Before=dbus.socket.
LP: #1629797
|
|
This is the best work around we have for an issue seen when dns is used
during cloud-init.service on a system that is configured in nsswitch
to use systemd-resolved. The problem is that cloud-init.service was
blocking basic.target, and dbus would not be available until after
cloud-init.service finished.
LP: #1629797
|
|
|
|
The end result of all of these changes is to get mounts managed by
cloud-init to occur only after cloud-init.service is done. We need to do
that so that filesystems that are set up by cloud-init (in disk_setup)
do not get mounted by stale entries in /etc/fstab before
the setup occurs.
This can occur in 2 ways:
a.) new instance with old /etc/fstab
b.) same instance where disk needs adjusting (Azure resize will re-format
the ephemeral disk).
The list of changes here is:
- move mounts and disk_setup module to cloud-init.service rather
than config. cloud-init.service runs earlier in boot so it
can get those mount points done earlier.
- on systemd add 'x-systemd.requires=cloud-init.service' to fstab options
- cloud-init-local.service: add Before=basic.target
- cloud-init.service:
- extend After, Before, and Wants to multiple lines rather than one
long line.
- sort consistently with cloud-init-local.service
- add DefaultDependencies=no
- add Before=default.target
- add Conflicts=shutdown.target
LP: #1611074
|
|
When we moved cloud-final.service to run After multi-user.target it
caused a dependency loop (as cloud-init.target was still implied to be
Before multi-user.target).
journalctl would either show:
cloud-init.target: Breaking ordering cycle by deleting job
cloud-final.service/start
or
multi-user.target: Breaking ordering cycle by deleting job
cloud-init.target/start
The fix here is to clearly state that cloud-init.target is also
After multi-user.target
LP: #1623868
|
|
In systemd, package installation before the system is fully booted
(systemctl is-system-running == starting) may result in the package not
being started. Upgrade (package_upgrade: true) can also cause failure if
that is done during systemd boot.
The solution here is:
a.) move config modules that do or may do package installation to
'final_modules'. That list is:
- snappy
- package-update-upgrade-install
- fan
- landscape
- lxd
- puppet
- chef
- salt-minion
- mcollective
b.) move cloud-final.service to run as 'Type=idle'
LP: #1576692, #1621336
|
|
This adds some function to the generator to maintain the presense of a
flag file '/run/cloud-init/enabled' indicating that cloud-init is enabled.
Then, only run the dhclient hooks if on Azure and cloud-init is enabled.
The test for is_azure currently only checks to see that the board vendor
is Microsoft, not actually that we are on azure. Running should not be
harmful anywhere, other than slowing down dhclient.
The value of this additional code is that then dhclient having run
does not task the system with the load of cloud-init.
Additionally, some changes to config are done here.
* rename 'dhclient_leases' to 'dhclient_lease_file'
* move that to the datasource config (datasource/Azure/dhclient_lease_file)
Also, it removes the config in config/cloud.cfg that set agent_command
to __builtin__. This means that by default cloud-init still needs
the agent installed. The suggested follow-on improvement is to
use __builtin__ if there is no walinux-agent installed.
|
|
This didn't really work. See bug for more info.
LP: #1577844
|
|
syslog.target is obsolete in debian, this fixes a lintian warning.
LP: #1536964
|
|
|
|
|
|
this might work. And if it does means we could generally test this
as the file that the cloud-init-wait will wait for will actually get
created.
|
|
Note, still broken as cloud-init local is not going to ever touch the
CI_NET_READY file (/run/cloud-init/network-config-ready).
So as this is , it will actually just block for 60 seconds and go on.
|
|
|
|
|
|
This might need cleaning up in the future as I believe in some
containers /proc/cmdline is provided, and in that case it would
be preferred to pid 1's command line.
|
|
This should mean that cloud-init-local is running earlier now
(DefaultDependencies=no). And also blocking networking coming up
(Before=network-pre.target)
cloud-init.service should now actually block network-online.target
from being made (meaning it will run before anything that expects
that) but after networking.service, which is what actually does
the bringup on ifupdown / ubuntu.
|
|
|
|
no wants or after or before.
|
