| Age | Commit message (Collapse) | Author |
|---|
|
`sysctl -n hw.disknames` returns a trailing `\n`. We need to clean
this up. In addition, the criteria matching system is a source of
problem because:
- we don't have a way to look up the label of the partition
- we've got situation where an ISO image can be exposed through a virtio
block device.
So we just totally ignore the value of `criteria`. We end-up with a
slightly longer loop of mount-retry. But this way we're sure we don't
miss a configuration disk.
Tested on Kubvirt with the help of Brady Pratt @jbpratt.
|
|
Thanks to [1], the hostname is set prior to network bring-up.
The Azure data source has been bouncing the hostname during
setup(), occurring after the hostname has already been
properly configured.
Note that this doesn't prevent leaking the image's hostname
during Azure's _get_data() when it brings up ephemeral DHCP.
However, as are not guaranteed to have the hostname metadata
available from a truly "local" source, this behavior is to
be expected unless we disable `send host-name` from dhclient
config.
[1]: https://github.com/canonical/cloud-init/commit/133ad2cb327ad17b7b81319fac8f9f14577c04df
Signed-off-by: Chris Patterson <cpatterson@microsoft.com>
|
|
When the metadata is an ISO image and is exposed through a disk,
the device is called `/dev/sd?a` internally. For instance `/dev/sd1a`.
It can then be mounted with `mount_cd9660 /dev/sd1a /mnt`.
Metadata in the FAT32 format are exposed as `/dev/sd?i`.
With this change, we try to mount `/dev/sd?a` in addition to `/dev/sd?i`.
Closes: https://github.com/ContainerCraft/kmi/issues/12
|
|
Add a test helper to get top level directory
Many tests need to get the location of files & dirs within the cloud-init
project directory.
Tests implement this in various different ways, and often those ways
depend on the current working directory of the pytest invocation.
Create helper functions (and tests) that gets the path of the top
directory or any sub directory under the top directory. This function
does not depend on the environment.
|
|
Parametrized pytest tests get named by on their parameters. If a name
has random characters, it can break the test collection of when
using pytest-xdist. Replace random name with deterministic name.
|
|
|
|
|
|
Improve schema validation.
This adds strict validation of config module definitions at testing
time, with plumbing included for future runtime validation. This
eliminates a class of bugs resulting from schemas that have definitions
that are incorrect, but get interpreted by jsonschema as
"additionalProperties" that are therefore ignored.
- Add strict meta-schema for jsonschema unit test validation
- Separate schema from module metadata structure
- Improve type annotations for various functions and data types
Cleanup:
- Remove unused jsonschema "required" elements
- Eliminate manual memoization in schema.py:get_schema(),
reference module.__doc__ directly
|
|
If we set a dhcp server side like this:
$ cat /var/tmp/cloud-init/cloud-init-dhcp-f0rie5tm/dhcp.leases
lease {
...
option classless-static-routes 31.169.254.169.254 0.0.0.0,31.169.254.169.254
10.112.143.127,22.10.112.140 0.0.0.0,0 10.112.140.1;
...
}
cloud-init fails to configure the routes via 'ip route add' because to there are
two different routes for 169.254.169.254:
$ ip -4 route add 192.168.1.1/32 via 0.0.0.0 dev eth0
$ ip -4 route add 192.168.1.1/32 via 10.112.140.248 dev eth0
But NetworkManager can handle such scenario successfully as it uses "ip route append".
So change cloud-init to also use "ip route append" to fix the issue:
$ ip -4 route append 192.168.1.1/32 via 0.0.0.0 dev eth0
$ ip -4 route append 192.168.1.1/32 via 10.112.140.248 dev eth0
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
RHBZ: #2003231
|
|
This attempts to standardize unit test file location under test/unittests/
such that any source file located at cloudinit/path/to/file.py may have a
corresponding unit test file at test/unittests/path/to/test_file.py.
Noteworthy Comments:
====================
Four different duplicate test files existed:
test_{gpg,util,cc_mounts,cc_resolv_conf}.py
Each of these duplicate file pairs has been merged together. This is a
break in git history for these files.
The test suite appears to have a dependency on test order. Changing test
order causes some tests to fail. This should be rectified, but for now
some tests have been modified in
tests/unittests/config/test_set_passwords.py.
A helper class name starts with "Test" which causes pytest to try
executing it as a test case, which then throws warnings "due to Class
having __init__()". Silence by changing the name of the class.
# helpers.py is imported in many test files, import paths change
cloudinit/tests/helpers.py -> tests/unittests/helpers.py
# Move directories:
cloudinit/distros/tests -> tests/unittests/distros
cloudinit/cmd/devel/tests -> tests/unittests/cmd/devel
cloudinit/cmd/tests -> tests/unittests/cmd/
cloudinit/sources/helpers/tests -> tests/unittests/sources/helpers
cloudinit/sources/tests -> tests/unittests/sources
cloudinit/net/tests -> tests/unittests/net
cloudinit/config/tests -> tests/unittests/config
cloudinit/analyze/tests/ -> tests/unittests/analyze/
# Standardize tests already in tests/unittests/
test_datasource -> sources
test_distros -> distros
test_vmware -> sources/vmware
test_handler -> config # this contains cloudconfig module tests
test_runs -> runs
|
|
Given that there are additional network management tools that we haven't
yet supported with activators, we should log a warning and continue
without network activation here, especially since this was a no-op for
years.
LP: #1948681
|
|
(#1123)
Allow #cloud-config and cloud-init query to use underscore-delimited
"jinja-safe" key aliases for any instance-data.json keys
containing jinja operator characters.
This provides a means to use Jinja's dot-notation instead of square brackets
and quoting to reference "unsafe" obtain attribute names.
Support for these aliased keys is available to both #cloud-config user-data and
`cloud-init query`.
For example #cloud-config alias access can look like:
{{ ds.config.user_network_config }}
- instead of -
{{ ds.config["user.network-config"] }}
|
|
|
|
GCE currently fetches metadata after network has come up. There's no
reason we can't fetch at init-local time, so update GCE to fetch at
init-local time to be more performant and consistent with other
datasources.
|
|
When cloud-init is configured to show SSH user key fingerprints during
boot two of the same message appears for each user. This appears to be as
the util.multi_log call defaults to send to both console directly and to
stderr (which also goes to console).
This change sends them only to console directly.
|
|
Some references were missed in the removal of the agent command
in PR #799. This simply removes the remaining references.
Signed-off-by: Chris Patterson <cpatterson@microsoft.com>
|
|
testing: monkeypatch system_info call in unit tests
system_info can make calls that read or write from the filesystem, which
should require special mocking. It is also decorated with 'lru_cache',
which means test authors often don't realize they need to be mocking.
Also, we don't actually want the results from the user's local
machine, so monkeypatching it across all tests should be reasonable.
Additionally, moved some of 'system_info` into a helper function to
reduce the surface area of the monkeypatch, added tests for the new
function (and fixed a bug as a result), and removed related mocks that
should be no longer needed.
|
|
Also simplify a path and fix a spelling error while in the file
|
|
Add growpart integration test and associated unit tests
Additionally, a small runcmd check for a commented line.
|
|
Move more tests into test_combined.py and remove the CI mark from module
tests that aren't updated often or don't represent core functionality.
|
|
- Added to list of expected warnings on Oracle when opc user has
no ssh key
- Added retries to tests that read from syslog as that can sometimes
take time to reflect in the log
- Updated test_apt.py to remove proxy info into its own test as that
can cause failures in updating, which will immediately traceback
out of the module and prevent us from running further class tests
- Updated test_apt.py to use a more updated ppa in the test_keyserver
- Added basic rsyslog test to test_combined.py
- Added basic puppet test as test_puppet.py
|
|
On Bionic and Xenial, pycloudlib sets user.vendor-data config in lxd
to ensure that lxd-agent is setup on those images.
Adapt the lxd_discovery integration test to assert the appropriate
user.vendor-data config key exists if we are on xenial or bionic.
Also add assertions that /var/lib/cloud/nocloud-net/meta-data still
exists in the images because we want NoCloud to be a viable fallback
datasource if LXD config security.lxddev = false or LXD datasource
discovery encountered an unexpected error.
|
|
Integration test runs get unique log directories at
/tmp/cloud_init_test_logs/$DATE_TIME. Make
/tmp/cloud_init_test_logs/last always point to the most recent
integration test directory.
|
|
Currently any attempt to run an apt command while another process holds
an apt lock will fail. We should instead wait to acquire the apt lock.
LP: #1944611
|
|
Chef tests attempt to reach out to test URLs, which will get blocked by
our on our openstack installs.
|
|
Without UDF support, DS Azure cannot mount the provisioning ISO,
which contains platform metadata necessary to support
pre-provisioning. The required metadata is made available in IMDS
starting with api version 2021-08-01. This change will leverage IMDS
to obtain the required metadata to support pre-preprovisioning if
provisioning ISO was not available.
|
|
In our integration tests, a few tests were modifying the environment and
then calling 'install_new_cloud_init'. This is problematic because it
updates the environment for all future tests.
Other instances of 'install_new_cloud_init' aren't problematic because
they aren't modifying the underlying environment.
|
|
Add DataSourceLXD which knows how to talk to the dev-lxd socket to
obtain all instance metadata API:
https://linuxcontainers.org/lxd/docs/master/dev-lxd.
This first branch is to deliver feature parity with the existing
NoCloud datasource which is currently used to intialize LXC instances
on first boot.
Introduce a SocketConnectionPool and LXDSocketAdapter to support
performing HTTP GETs on the following routes which are surfaced by the
LXD host to all containers:
http://unix.socket/1.0/meta-data
http://unix.socket/1.0/config/user.user-data
http://unix.socket/1.0/config/user.network-config
http://unix.socket/1.0/config/user.vendor-data
These 4 routes minimally replace the static content provided in the
following nocloud-net seed files:
/var/lib/cloud/nocloud-net/{meta-data,vendor-data,user-data,network-config}
The intent of this commit is to set a foundation for LXD socket
communication that will allow us to build network hot-plug features
by eventually consuming LXD's websocket upgrade route 1.0/events to
react to network, meta-data and user-data config changes over time.
In the event that no custom network-config is provided, default to the
same network-config definition provided by LXD to the NoCloud
network-config seed file.
Supplemental features above NoCloud datasource:
surface all custom instance data config keys via cloud-init query ds
which aids in discoverability of features/tags/labels as well as
conditional #cloud-config jinja templates operations based on custom
config options.
TBD: better cloud-init query support for dot-delimited keys
|
|
When we added the install hotplug module, we forgot to update the
redhet/cloud-init.spec.in file and allow for execution on /usr/libexec.
This PR adds that functionality.
|
|
|
|
Also, add the "signed by" option to source definitions. This enables
users to limit the scope of trust for individual keys.
LP: #1836336
|
|
This commit removes automatically installing udev rules for hotplug
and adds a module to install them instead.
Automatically including the udev rules and checking if hotplug was
enabled consumed too many resources in certain circumstances. Moving the
rules to a module ensures we don't spend extra extra cycles on hotplug
if hotplug functionality isn't desired.
LP: #1946003
|
|
The main idea is to introduce a second module that takes care of
writing files, but in the 'final' stage.
While the introduction of a second module would allow for choosing
the appropriate place withing the order of modules (and stages),
there is no addition top-level directive being added to the cloud
configuration schema. Instead, 'write-files' schema is being extended
to include a 'defer' attribute used only by the 'write-deffered-files'
modules.
The new module 'write-deferred-files' reuses as much as
possible of the 'write-files' functionality.
|
|
In jsonschema 4, hostname validation was changed to have an optional
dependency on the fqdn package. Since we don't have this dependency
in cloud-init, attempting this validation will no longer fail for
a string that isn't a valid hostname.
|
|
Various modules restart services and they all have logic to try and
detect if they are running on a system that needs 'systemctl' or
'service', and then have code to decide which order the arguments
need to be etc. On top of that, not all modules do this in the same way.
The duplication and different approaches are not ideal but this also
makes it hard to add support for a new distribution that does not use
either 'systemctl' or 'service'.
This change adds a new manage_service() method to the distro class
and updates several modules to use it.
|
|
When self.failed_desired_api_version was added to DataSourceAzure, the
attribute was never added to the _unpickle method using the upgrade
framework. This commit adds the attribute.
LP: #1946644
|
|
There is no reason for the ISO missing this functionality.
As discussed in https://github.com/canonical/cloud-init/pull/947/files#r707338489
|
|
|
|
PyYAML upgraded from 5.4.1 to 6.0.0. 6.0.0 always requires a `Loader`
arg to `yaml.load()`
|
|
Also added supporting distro/datasource classes and updated tests
that have a `get_cloud` call.
|
|
Due to multiarch, the libdeployPkgPlugin.so is deployed into dir
/usr/lib/<multiarch name>/open-vm-tools, we need to add this path
into search_paths.
LP: #1944946
|
|
Allow comments in runcmd and report failed commands correctly
A `runcmd` script may fail to parse properly, but does not mark
`runcmd` as failed when that occurs. Additionally `shellify()` fails
to correctly parse scripts that contain a comment line.
Rectify both issues and add unit tests to verify correct behavior.
LP: #1853146
|
|
In #919 (81299de), we refactored some of the code used to bring up
networks across distros. Previously, the call to bring up network
interfaces during 'init' stage unintentionally resulted in a no-op
such that network interfaces were NEVER brought up by cloud-init, even
if new network interfaces were found after crawling the metadata.
The code was altered to bring up these discovered network interfaces.
On ubuntu, this results in a 'netplan apply' call during 'init' stage
for any ubuntu-based distro on a datasource that has a NETWORK
dependency. On GCE, this additional 'netplan apply' conflicts with the
google-guest-agent service, resulting in an instance that can no
be connected to.
This commit adds a 'disable_network_activation' option that can be
enabled in /etc/cloud.cfg to disable the activation of network
interfaces in 'init' stage.
LP: #1938299
|
|
OpenNebula 6.1.80 (current dev. version) is introducing new IPv6 gateway
contextualization variable ETHx_IP6_GATEWAY, which mimics existing
variable ETHx_GATEWAY6. The ETHx_GATEWAY6 used until now will
be depracated in future relase (ET spring 2022).
See:
- new variable - https://github.com/OpenNebula/one/commit/e4d2cc11b9f3c6d01b53774b831f48d9d089c1cc
- deprecation tracking issue - https://github.com/OpenNebula/one/issues/5536
Also, added support for SET_HOSTNAME context variable, which is
currently widely used variable to configure guest VM hostname. See
https://docs.opennebula.io/6.0/management_and_operations/references/template.html#context-section
|
|
Cloud tests have been replaced with integration tests
|
|
|
|
Offload Vultr's vendordata assembly to the backend, correct vendordata
storage and parsing, allow passing critical data via the useragent,
better networking configuration for additional interfaces.
|
|
tox: bump the pinned flake8 and pylint version
* pylint: fix W1406 (redundant-u-string-prefix)
The u prefix for strings is no longer necessary in Python >=3.0.
* pylint: disable W1514 (unspecified-encoding)
From https://www.python.org/dev/peps/pep-0597/ (Python 3.10):
The new warning stems form https://www.python.org/dev/peps/pep-0597,
which says:
Developers using macOS or Linux may forget that the default encoding
is not always UTF-8. [...] Even Python experts may assume that the
default encoding is UTF-8. This creates bugs that only happen on Windows.
The warning could be fixed by always specifying encoding='utf-8',
however we should be careful to not break environments which are not
utf-8 (or explicitly state that only utf-8 is supported). Let's silence
the warning for now.
* _quick_read_instance_id: cover the case where load_yaml() returns None
Spotted by pylint:
- E1135 (unsupported-membership-test)
- E1136 (unsubscriptable-object)
LP: #1944414
|
|
In #1006, we set Azure to apply networking config every
BOOT_NEW_INSTANCE because the BOOT_LEGACY option was causing problems
applying networking the second time per boot. However,
BOOT_NEW_INSTANCE is also wrong as Azure needs to apply networking
once per boot, during init-local phase.
|
|
|
