Some charms never implemented the workload status messages, these
charms have a wl status of unknown and wls message of ''. This fix
allows a wls of '' to be specified.
Currently interacting with functions in zaza.model requires the
model to be passed in. This has resulted in multiple calls to
get_juju_model(). It is cleaner to push these calls down into
the model wrapper and make the model_name an optional
argument. In addition, the current model name is now cached
without having to check the os.env each time.
Unfortunately this has resulted in the signature changing on a
great many function so this diff is bigger than would normally
be desirable.
Recently functions were added to zaza.model for adding and removing
models. These duplicated existing functions in zaza.controller.
Given these functions act against the controller, zaza.controller
feels like the correct location. In addition the implementations in
zaza.controller correctly handle disconnects which the new ones
did not.
This PR removes the duplicates and points any existing references
from zaza.model to zaza.controller. Finally, docstrings were added
and documentation links.
This pr adds a number of new fuctions to zaza.model. As discussed
zaza.model needs to be broken up as it has become too large. I would
like to continue the discussion around that and submit a subsequent
patch for that at a later date.
Without introducing mojoisms into zaza there are some changes that are
required to enable mojo's use of zaza.
The goal is to keep mojo specific utilities in openstack-mojo-specs and
to move any generalized tools here in zaza.
This change updates some generalized zaza tools to enable mojo to use
them.
Currently there is no way for the tests.yaml to specify if the
expected workload status message should be checked as a prefix
or as an exact match. I think the prefix approach actually covers
all our use cases so rather than complicate the tests.yaml
options lets just always use startswith for checks.
If the deployment state is assessed before any units have been
provisioned then the checks passes as there are no active units
or units with bad states etc. So, wait for at least one unit to
be provisioned before running checks
* Add the ability to wait for bespoke statuses
This change adds the ability to wait for bespoke work load
statuses and messages. These are defined in the charms tests.yaml
* Add unit tests
* Remove debug print
* Fix typos and add checks for errored units
* Restore juju_wait as openstack_utils imports it
The vault charm action to authorise the charm within vault needs to
be run on the leader. This mp adds run_action_on_leader to support
that and updates the tests.
Add tests for the authorize-charm action on the vault app. To
support this add get_action method to return an applications
actions. However, this is not implemented in libjuju yet so
fallback to subprocess
Bring over only those utilities required to run network_setup.py from
openstack-mojo-specs. These utilities have been generalized and python3
configured. Use libjuju whenever possible and pre-deprecate anything
else.
Add addition functions to the model.
The idea of this change is to put together helpers for including
functional tests for charms in a central location. The charm
declares the tests and bundles to be run to complete a
functional tests but the tests themselves live here in zaza.
To use this code the charm should have the following:
1) zaza in the test-requirements.txt
2) tox.ini should include a target like:
```
[testenv:func35]
basepython = python3
commands =
functest-bundle-deploy
```
3) Bundles which are to be used for the tests:
```
ls -1 tests/bundles/*
tests/bundles/xenial.yaml
```
4) A tests/tests.yaml file that describes the bundles to be run and
the tests
```
charm_name: vault
tests:
- zaza.charms_tests.vault.VaultTest
gate_bundles:
- xenial
dev_bundles:
- bionic
```
Tests can be run without running a deployment using functest-run-tests
with the list of test classes to be run:
```
functest-run-tests -t zaza.charms_tests.vault.VaultTest
```
Known Issues:
- The deploy_bundle and add_model methods should be using libjuju
- DEV_BUNDLES is currently ignored
- VaultUtils and VaultTest should probably be in separate files.
- When skipIfNotHA skips a test unittest does not pick up it has
been skipped
- A new model is created for each bundle even if an existing empty
model exists
- No model cleanup is performed.
Frode Nordahl
Liam Young
Chris MacNaughton
David Ames
Frode Nordahl
David Ames