* Change "unknown" user to "some_unknown_user_xyz" since "unknown"
*is* a valid user
* Add timeout to proxy test_exit_status_no_proxyid() since a known
failure case is to infinitely loop with an error.
* Always force the source code tree under test to be the first entry
in PYTHONPATH (excluding verbatim_env)
* Fix signal handling
We had a little mix-up with the args ordering for our signal handling.
This sends the proper signal to processes on cleanup.
I have also temporarily disabled the pytest engines because they were
causing the minions to try to to connect to a master IPC socket which could not be found.
This put the minions into a continual futex state which was not playing well with kill sigs.
* Lint
* Added support for instance start and stop actions in GCE
* added cloud events for rebooting, starting, and stopping of GCE instances
* Fixed lack of double newline between new functions in salt-cloud GCE provider.
Using a salt.runner.Runner() instance to run this test has exposed a
problem with run_run_plus, where the fileclient doesn't behave as
expected. Since the raw return data is not needed, using run_run here
lets the test pass while we continue to work on run_run_plus.
The return data from run_run_plus previously returned the runner return
in the ``fun`` key. This has since been corrected. This commit replaces
these references to the ``fun`` key with the corrected ``return`` key.
This allows one to pass a kwarg to run_run_plus called
``__reload_config`` which will be passed as the ``from_scratch``
argument to self.get_config(). This allows self.run_run_plus() to force
a reload of the master config for integration tests which require the
config to be reloaded, for instance, from master.d/*.conf.
The Windows distribution doesn't come with `six` as a pre-installed
python package. Since `salt.ext.six` already exists, it makes sense
to use that instead of adding `six` to the Windows distribution.
Signed-off-by: Sergey Kizunov <sergey.kizunov@ni.com>
The salt-master on Windows was broken by PR #35703 due to the change in
how the ReqServer object was started in a new process. The new way
failed to correctly pickle/unpickle the args/kwargs. To fix this, we
use `__setstate__` and `__getstate__` similar to how other objects in
the same file handle it (such as the `Maintenance` object).
Signed-off-by: Sergey Kizunov <sergey.kizunov@ni.com>
Reactor uses LocalClient to send message but do not read events that
produces memory leak in EventPublisher: sent messages are collected
there in the IOStream class.
