`_get_extra_opts()` and `_get_branch_option()` were unnecessarily
quoting the value, causing it to be interpreted as a literal quote by
`subprocess.Popen()`.
Also, because there were separate helpers for repo options,
disableexcludes, branch options, and extra options, and specifically
because `_get_extra_opts()` parses *all* kwargs, any of the options from
the other helper funcs would end up being added to the command line
twice if `_get_extra_opts()` was used.
This commit consolidates all of the kwarg inspection and CLI opts
construction to a single helper function. It also adds unit tests to
make sure that we are formatting our commands properly.
Additionally, it makes a minor fix in `refresh_db()` which was not
accounted for when we changed the osmajorrelease grain to an integer in
2017.7.0.
This sorting was done mainly for the benefit of the test suite, but
Python 3 will raise an error when you try to sort a mixture of int and
tuple types, so sorting breaks down when there are UDP ports.
Instead, this just leaves them as an unsorted list when passed to the
API, and the test suite does the sorting before the assertEqual.
The logic which ensures that we expose ports which are being bound,
even when not explicitly configured, was done incorrectly. UDP ports
were being passed to the API as '1234/udp' instead of (1234, 'udp').
This results in the port not being exposed properly.
The logic has been corrected. Additionally both the "ports" input
translation function, as well as the post-processing code (where the
port numbers configured in port_bindings were being added) both
contained code to "fix" any ports which were configured using
'portnum/tcp', as these must be passed to the API simply as integers. To
reduce code duplication, this normalization is now only performed at the
very end of the post-processing function, after ports have been
translated, and any missing ports from the port_bindings have been
added.
The unit test for the port_bindings input translation code, which was
written based upon the same incorrect reading of the API docs that
resulted in the incorrect behavior, have been updated to confirm the
(now) correct behavior. The unit test for the ports input translation
code has been updated to reflect the new normalization behavior.
Finally, an integration test has been added to ensure that we properly
expose UDP ports which are added as part of the post-processing
function.
When running the unit tests with the locale set to POSIX, some Unicode
tests fail:
$ LC_ALL=POSIX python3 ./tests/runtests.py --unit
[...]
======================================================================
ERROR: test_list_products (unit.modules.test_zypper.ZypperTestCase)
[CPU:0.0%|MEM:73.2%]
----------------------------------------------------------------------
Traceback (most recent call last):
File "tests/unit/modules/test_zypper.py", line 236, in
test_list_products
'stdout': get_test_data(filename)
File "tests/unit/modules/test_zypper.py", line 53, in get_test_data
return rfh.read()
File "/usr/lib/python3.6/encodings/ascii.py", line 26, in decode
return codecs.ascii_decode(input, self.errors)[0]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position
828: ordinal not in range(128)
======================================================================
ERROR: test_non_ascii (unit.templates.test_jinja.TestGetTemplate)
[CPU:0.0%|MEM:73.2%]
----------------------------------------------------------------------
Traceback (most recent call last):
File "tests/unit/templates/test_jinja.py", line 341, in test_non_ascii
result = fp.read()
File "/usr/lib/python3.6/encodings/ascii.py", line 26, in decode
return codecs.ascii_decode(input, self.errors)[0]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 5:
ordinal not in range(128)
======================================================================
ERROR: test_non_ascii_encoding
(unit.templates.test_jinja.TestGetTemplate)
[CPU:0.0%|MEM:73.2%]
----------------------------------------------------------------------
Traceback (most recent call last):
File "tests/unit/templates/test_jinja.py", line 303, in
test_non_ascii_encoding
fp_.read(),
File "/usr/lib/python3.6/encodings/ascii.py", line 26, in decode
return codecs.ascii_decode(input, self.errors)[0]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 5:
ordinal not in range(128)
----------------------------------------------------------------------
Therefore open files in binary mode and explicitly decode them with
utf-8 instead of their default locale.
Fedora releases are integers, and Arch's osrelease is simply `rolling`,
so https://github.com/saltstack/salt/pull/45636 caused these tests to
regress. This fixes them by moving the osrelease check until after the
MacOS check.
It also fixes the windows check, which would _always_ evaluate to `True`
since it was not calling the function. Therefore, the `if` would just be
a simple boolean which would resolve to `True` since it was just checking
a function reference.
When boto and moto are not installed, the BotoVpcTestCaseBase() should
be skipped. _get_boto_version() and _get_moto_version() require that
boto and moto are installed, but are called anyway in the skipIf
condition.
The test_run unit tests end result when a command is not found. When
_run() is called without setting cwd, it will use the home directory as
working directory. When the home directory does not exist, the unit test
will fail:
Traceback (most recent call last):
File "tests/unit/modules/test_cmdmod.py", line 231, in test_run
ret = cmdmod._run('foo', use_vt=True).get('stderr')
File "salt/modules/cmdmod.py", line 536, in _run
.format(cwd)
salt.exceptions.CommandExecutionError: Specified cwd
'/sbuild-nonexistent' either not absolute or does not exist
Therefore set cwd to the current directory since the working directory
is not used in this test case.
This fixes one failing test of #45627.
Signed-off-by: Benjamin Drung <benjamin.drung@profitbricks.com>
This makes some further tweaks to make the test more stable under heavy
load. Firstly, the background job sleeps longer, and secondly, we make
up to 3 attempts run state.running instead of just the one, in case our
first attempt was too early. It also uses threading to make the job
sleep, since the method of adding a & to the command seemed to be
producing intermittent failures with returning clean JSON.
Without setting the man path, this would be checking that the installed salt
package used to bootstap has an spm man page, which is possible that it
doesn't, and also not what we want to check
the running tests run second because of alphabetical order, and stopping sshd
does not kill off the running connection, so we know that sshd has to be on
arch for us to connect, so stop it with dead and start it back with running.
It is also not possible to stop the systemd-journald process without shutting
down the full machine.
yum is no longer installed by default on fedora. apparently it has been
removed from the base installs, and only dnf is available. Stop depending on
the pkg manager, and just use bash.
yum is no longer installed by default on fedora. apparently it has been
removed from the base installs, and only dnf is available. Stop depending on
the pkg manager, and just use bash.
The behavior of the set union used to combine ext_nodes and top file
matches is not consistent on Python 3 (well, 3.5 at least), and this
produces flaky test results.
Since we do not change things like the state compiler in point
releases, the method used to combine the ext_nodes and top file matches
has been changed for Oxygen and is now fully deterministic. This test
will therefore be skipped for the remainder of the 2017.7 release cycle
on Python 3.
Fixes https://github.com/saltstack/salt-jenkins/issues/686.
test_exclude shares file paths with test_include, and while I can't
reproduce the failures, it is likely that improperly cleaned-up files
from test_include are causing the failures in test_exclude. This is
backed up by the fact that the state.sls return data from the
salt-runtests.log shows no trace of the "to-include-test" file
(suggesting it was excluded as expected), yet os.path.isfile() returns
True for this path, causing the test to fail.
This commit uses a distinct base dir for both tests, which should keep
this sort of failure from happening.
Carbon 1.1.1 for some reason added six to their setup.py, which breaks
this test since it's not installed into the virtualenv. This PR forces
this test to use a version of carbon which does not dep on six in its
setup.py.
Because the destructive test decorator was applied to the tearDown
function, the tests show a stacktrace instead of a nice skip message
when running without the `--run-destructive` flag. By applying the
decorator to the test class (since all tests are destructive anyway)
we can keep the output clean and still skip tests appropriately.
Cherry-picked and squashed the two commits from the develop branch
(details at the end of this message)
The unit test module didn't exist at all on 2017.7, hence a conflict during
cherry picking, resolved by introducing only the new tests (the other ones
don't pass as is on 2017.7)
Also, the mock in unit tests for mine.get had to be adapted,
because the return of salt.utils.minions.CkMinions._check_compound_minions()
takes a different form on 2017.7 (just the list of minions).
Original cherry-picked commits:
commit 115ebef6a089dedf8dbadd3b4cca768adbb4a710
Author: Georges Racinet <gracinet@anybox.fr>
Date: Wed Nov 8 18:02:29 2017 +0100
Unit test for backwards compatibility of mine.get
Could not find calls to RemoteFuncs in unit tests, therefore started a
new one, hope it's in the appropriate place.
The test replaces the caceh by its own (very limited) one. It's quite
possible it could be improved by reusing another one meant for unit tests.
commit dc884478399355f3fbea8626a8366b933580dee0
Author: Georges Racinet <gracinet@anybox.fr>
Date: Mon Oct 30 13:49:13 2017 +0100
Backwards compat for mine.get on pre-Nitrogen minions
With this change, the master will accept `expr_form` from
its minions doing a `mine.get`, which is what minions versions
before Nitrogen would do. This solves issue #42713.
In case both `tgt_type` and `expr_form` are present, the former
gets precedence.
Performance-wise, this adds a single dict lookup for minions
running Nitrogen onwards, and two for pre-Nitrogen minions.
This is, in my opinion, acceptable.
