Fix eni rendering for bridge parameters which require repeated key for values
There are a few bridge parameters which require repeating the key with each
value in the list when rendering eni. Extend the network unittests to cover
all of the known bridge parameters and check we render eni and netplan
correctly.
disk_setup: udev settle before attempting partitioning or fs creation.
This attempts to use udevadm settle to wait until devices have been
fully "realized". If a device exists, there may still be events in
the udev queue that would create its partition table entries.
We need to wait until those have been processed also.
The instance-level SSH key was named 'sshKeys' and now is 'ssh-keys'.
The project-level SSH key attribute has not changed so is intentionally
not changed here.
python2.6: fix unit tests usage of assertNone and format.
python2.6 unittest.TestCase does not have the assertIsNone or
assertIsNotNone. We just have to explicitly use the unittest2
version, which we get from helpers.
The desire to use assertIsNone comes from flake8 (through hacking,
I believe).
Also, fix "{}.format('foo')" which is not valid in python2.6.
If the user configured:
datasource_list: ["Ec2", "None"]
then ds-identify would write
datasource_list: ["Ec2", "None", "None"]
which would break the logic to avoid warning.
tox/build: do not package depend on style requirements.
When the style/checking dependencies were updated in
test-requirements.txt, the debian package build dependencies created
by ./packages/bddeb were also updated. Pycodestyle was added to the list
in order to pin its version. That broke the package build for 16.04. The
reason for this is simply that python3-pycodestyle is not available in
16.04.
The change here is to remove style dependencies from test-requirements,
and add them to the tox environments directly. We had previously changed
the package build process to not run pep8 or flake8 simply to avoid having
to code to N different versions of style checkers (3bcb72c593f).
The link between package build and test-requirements still exists, though.
So future breakage can occur if any package is added to
test-requirements.txt (or requirements.txt) if the target distro release
does not have a python3-<packagename> in its archive.
There is also a bit of a tox.ini cleanup here, in that we do not have to
explictly list '-rrequirements.txt' as the setup.py pulls those in. And
lastly, we drop the -rtest-requirements.txt from the base 'testenv', and
add these test requirements only to environments that need to run test.
Finally, a change to packages/debian/control.in to drop the build
dependencies that were listed for style checking and also a dependency
on iproute2 which was a bad unit test that has been previously fixed.
Any CiTestCase subclass can now set a class attribute with_logs = True and
tests can now make assertions on self.logs.getvalue(). This branch
restructures a bit of cc_ntp module to get better test coverage of the
module. It also restructures the handler_cc_ntp unit tests to avoid nested
mocks where possible. Deeply nested mocks cause a couple of issues:
- greater risk: mocks are permanent within the scope, so multiple
call-sites could be affected by package mocks
- less legible tests: each mock doesn't advertise the actual call-site
- tight coupling: the unit test logic to tightly bound to the actual
implementation in remote (unrelated) modules which makes it more
costly to maintain code
- false success: we should be testing the expected behavior not specific
remote method names as we want to know if that underlying behavior
changes and breaks us.