Use correct ssh executable and options in all cases on connection plugin
* Also nicer naming/comments
Co-authored-by: Sviatoslav Sydorenko <wk.cvs.github@sydorenko.org.ua>
Ability to add local variables into AnsibleJ2Vars was added in
18a9eff11f to fix#6653. Local variables
are added using ``AnsibleJ2Vars.add_locals()`` method when creating a
new context - typically when including/importing a template with
context. For that use case local template variables created using
``set`` should override variables from higher contexts - either from the
play or any parent template, or both; Jinja behaves the same way.
Also removes AnsibleJ2Vars.extras instance variable which is not used.
Also adds missing test for #6653.
Fixes#72262Fixes#72615
ci_complete
The `command` module does not return stdout & sterr when calling
a non existing executable or an unknown exception arises. This fix
lets the module return empty byte strings in those cases.
Change:
- Add Ubuntu 20.04 to CI now that venv is default instead of virtualenv in ansible-test.
Test Plan:
- CI
Tickets:
- Fixes#69203
Signed-off-by: Rick Elrod <rick@elrod.me>
The init script for the test container writes additional lines to
the .bashrc of the user. This was done via a `cat` multiline
instruction, which is implemented internally by writing a
temporary file to TMPDIR (/tmp in this case) first. Docker fails
to provide /tmp just after creation, which results in a race
condition that rarely makes the init fail. Changed the `cat`
statement to multiple `echo`s.
Co-authored-by: Matt Clay <matt@mystile.com>
* Fix YAML error message when error is at the end of the file
If a YAML file fails to load due to a syntax error in a file, or there is an error in the last line of a
file, PyYAML reports the last line number of the file as the index where the error occurred.
When reading the file lines, we use that index to the get the relevant line. If the index value is out
of range, the relevant line is lost for error reporting.
Subtract one from the index value to avoid the IndexError in this specific scenario. It is possible
to still get an IndexError, which will be handled as it is currently.
* Update existing tests and add new tests
* Allow restricting config values to enumerated list
* dont document internal entries
* also ignore private defs for ansible-config
* remove invalid value from tests
* added porting entry
When the pause module is run in the background and seconds parameter is provided,
do not warn.
* Add tests
* Fix existing tests
The test wasn't failing when it should have.
Change:
- On CentOS Stream, make distribution_release be "Stream"
- On CentOS Core, it continues to be "Core"
- Implement custom distribution file parser for CentOS, so we can look
for "CentOS Linux" and "CentOS Stream"
- Two new fixtures introduced (CentOS Linux 8.1 and CentOS Stream 8)
- Removed two dicts from `Distribution` class that were seemingly not
used anywhere.
Test Plan:
- ci_complete
- New test fixtures
Tickets:
- Fixes#73027
Signed-off-by: Rick Elrod <rick@elrod.me>
The test currently only expects the literal `ansible` followed
by a semver in the first output line of `ansible --version`.
When running from within a git checkout, additional information
like the currently checkout branch are attached, which lets
the test fail. This commit allows arbitrary information to
follow the semver.
`test_prepare_multipart` fails in non Debian environments since
Debian installations map the file ending `.key` to the MIME type
`application/pgp-keys`, which is not IANA conformant. This commit
explicitly sets the corresponding file type to
`application/octet-stream` and adjusts the expected serialized
result.
* Remove unused code in ansible-test.
* Remove obsolete endpoint logic from ansible-test.
* Remove obsolete region selection in ansible-test.
* Remove obsolete port logic in ansible-test.
* Clean up ansible-test remote providers.
Do the right thing on Linux when password lock and a password hash are provided by writing
out the password hash prepended by the appropriate lock string rather than using -U and -L.
This is the correct way to set and lock the account in one command.
On BSD, run separate commands as appropriate since locking and setting the password cannot
be done in a single action.
FreeBSD requires running several commands to get the account in the desired state. As a result,
the rc, output, and error from all commands need to be combined and evaluated so an accurate
and complete summary can be given at the end of module execution.
* Improve integration tests to cover this scenario.
* Break up user integration tests into smaller files
* Properly lock account when creating a new account and password is supplied
* Simplify rc collection in FreeBSD class
Since the _handle_lock() method was added, the rc would be set to None, which could make
task change reporting incorrect. My first attempt to solve this used a set and was a bit too
complicated. Simplify it my comparing the rc from _handle_lock() and the current value of rc.
* Improve the Linux password hash and locking behavior
If password lock and hash are provided, set the hash and lock the account by using a password
hash since -L cannot be used with -p.
* Ensure -U and -L are not combined with -p since they are mutually exclusive to usermod.
* Clarify password_lock behavior.
Tests should use the Python built-in ``venv`` module on Python 3 instead of the standalone ``virtualenv`` module.
On Python 2 the ``virtualenv`` module continues to be the only option.
The version installed is either the OS packaged version or the last release to support Python 2, which is version 16.7.10.
Change:
- Internally, use dnf.subject.Subject#get_best_query for state: absent
- Add a bunch of tests for removing packages, given a bunch of different
pkg specs (nv, nvr, nvra, wildcard, etc.)
Test Plan:
- New tests
- Local experiments with DNF API via PDB.
Tickets:
- Fixes#72809
Signed-off-by: Rick Elrod <rick@elrod.me>
Change:
- Changes to make F33 work with current tests
- Add F33 to docker.txt
- Add F33 to AZP config
Test Plan:
- ci_complete
Signed-off-by: Rick Elrod <rick@elrod.me>
Change:
- The FreeBSD release can contain -RC or -PRERELEASE in addition to
-RELEASE, -STABLE, or -CURRENT.
Test Plan:
- Added new fixed from an RC version of TrueNAS which uses a -PRERELEASE
version of FreeBSD.
Tickets:
- Fixes#72331
Signed-off-by: Rick Elrod <rick@elrod.me>
* arg_spec - move type checking lookup method to a function
* Change get_wanted_type name and behavior
Change the name to get_validator to bette describe what it is doing.
Change the interface to always return a value. This lines up with the behavior of get_*
functions always returning something or None and check_* functions raising an
Exception if something went wrong during the check.
* Add param to check_type_str()
Not meant to be a long term fix, but gets tests passing. More work is needed to figure
out how to solve this cleanly.
* Remove private attribute mapping types to validator
Since the function that needs it has moved to parameters.py, there is no need to have it as
a attribute of AnsibleModule.
Update tests that were referencing the private attribute.
* Use private method for 'str' type
To avoid having to put the string conversion warning behavior in the check_type_str() method,
use the private _check_type_str() method for 'str' type.
Import CHECK_ARGUMENT_TYPES_DISPATCHER for backwards compalitibility and store it as
a private attribute.
Revert changes to support plugins that are referencing serf._CHECK_ARGUMENT_TYPES_DISPATCHER.
* Add changelog
* Change function name to better reflect its... function
* Change dict name to better reflect its contents
CHECK_ARGUMENT_TYPES_DISPATCHER --> DEFAULT_TYPE_VALIDATORS
* Fix changelog
* systemd - do not overwrite unit name when searching
PR #72702 introduced a bug that changed the unit name when splitting it up for the purpose
of searching for the unit. This only happens on unit file templates on systems that have a 5.8
or newer kernel and a version of systemd that does not contain a bugfix that causes systmed
to fail to parse dbus.
* Use facts rather than a manual probe to determine if systmed is present
* Remov unnecessary block
* Use vars files instead of set_fact
* Add tests for using a templated unit file
* Update changelog fragment
* Use template to get correct path to sleep binary
* import_playbook - change additional params to deprecation
I incorrectly recommended this be set as a warning when it should have been a deprecation.
* Fix deprecation sanity test to not required a collection name when not inside a collection
Also pin virtualenv to 16.7.10 for older Mac OS X systems. This was the version being installed
anway with the previous constraint (<20).
On systems with Python 3, now prefer venv over virtualenv. Test to see if venv is functional since
some systems have a non-functional venv installation (such as Debian).
Brian Coca