This still sets the root password but that is only to unlock the
account. The key is used instead to authenticate and access the board.

This should increase tests resilience as we can now spawn new shell
instance for every test and thus taint of shell from one test can't
influence others.

This of course needs changes in existing tests and that is not part of
this code. The update of existing code and removal of client_board
fixture in favor of board_access should be done in the future. For now
client_board is documented as obsolete and not suggested for the new
code.

The disadvantage of this change is that every fixture that wants to do a
simple call (such as uci or cat) has to create its own connection and
that can get spammy pretty fast. The solution for this is the new
fixture fixture_board_access. This provides one shell instance. In
reality it provides shell instance that can reconnect. This is due to
issues when client container would change its IP address (for example
when we change LAN addressing).

These setups are handy in general. They created files and diretories
that are automatically removed on revert.

The RootPassword is used in conftest.py and might be usable in other
places as well. Part of this are tests on OpenWrt container to make sure
that setup works as expected and also setups for installing packages in
both OpenWrt and Alpine.

This should allow us to configure and call updater in reasonably common
way. The setup currently covers features we already were using and also
one generic Updater centric setup that is not required at the moment but
is going to be for future testing of updater.

This replaces static strings with generated ones from lorem-text
library.

To make tests reproducible we have to se seed for random. This is pretty
common so this adds context function for it to toolbox.

Toolbox was labeled as common tasks run on router. Now this changes it
to just generic common toolbox and splits it to three submodules. There
is module for Alpine Linux as well as for OpenWrt. Then there is one
additional module for utilities needed in tests implementation.

The abstractproperty is deprecaterd since Python 3.3.

This tests commonly fails because syslog-ng takes its time to create the
/var/log/messages file even when it is already running. Moving it does
not ensure that it is run later on but at least that is what happens
right now.

The FDLogging uses poll to propagate data from one feed to the other.
This is not limited to only two file descriptors but technically any
number of them. Thus it is way more efficient to simply add new feeds to
existing thread over spawning the new one.

This improves cli handling in a wide variety of ways. Let's point one at
a time.

Change how we match sent command to not have issue with line break
caused limited number of columns. Now it simply expects new line and
return characters while it looks for characters from command. This makes
it possible to remove hack to increase number of columns in Shell.

The initial prompts in Shell were sanitized a little bit and commented.
The "weird" prompt it turns out is caused by TERM variable and remote
shell trying to control local terminal. One solution would be to set
different TERM but that complicates mterm usage while debugging. The
alternative and used option is to simply simplify regular expression to
not have problem with it. With exception of Busybox's default prompt
they are specific enough without having to match also new line.

For the same reason as in previous paragraph the nsfarm prompt was also
simplified. It now matches new lines only optionally. The reason is also
that not every program terminates with new line at the end and so it is
completely possible and valid that prompt is displayed at the end of
line not at the beginning. The "nsfprompt" is also specific enough word
that should not be anywhere else that we can rely only on it when we
look for our prompt.

This also fixes txt_write and bin_write as sendeof does work only on
pexpect instance but not on fdexpect. The more correct and used option
here is our own ctrl_d function that only signals EOF without actually
closing stdin.

The txt_read and bin_read methods gets option to not fail if file does
not exist. The txt_write method also got option to append instead of
overwrite.

The logging was improved by implementing our own code that splits lines.
The splitlines method is little bit too aggressive with splitting as it
expects pretty reasonable new lines consistency usage with standards.
The issue here is that for example '\r' that moves cursor to the left
can be spawned multiple times and has no effect and in some cases that
is what happens. The combination '\r\r\n' should be only one line not
two as detected by splitlines. The code we now use detects combination
of any number of '\r' and single '\n' as a one line split. It also
handles only '\r' as that is commonly used to overwrite existing line so
we consider it as new line as well. There is still a room for
improvements such as support for backspace and other control characters.

This makes this script support Mox as well as it reduces unnecessary
spam it creates in logs.

This sorts all includes in nsfarm project.

We want to make sure that we adhere to black formating and just also
sorted imports are reasonable request.

Jan Miksik authored 3 years ago and Karel Koci committed 3 years ago

Black is now the coding standard for Python projects in Turris. The only
exception over default is 120 columns limit.

nsfarm/lxd/utils: fix PermissionError on container clean

See merge request !37

The user might not have ability to send signal to that process but we
only check if process exists (is running) and not having ability means
that process is running.

Removed LXDConnection and refactored.
Connection replaced wit pylxd.Client()
and variables and logging moved into
lxd.__init__

Karel Koci authored 3 years ago and Jan Miksik committed 3 years ago

Instead of having hardcoded amd64 in image files we now try to detect
architecture instead.

Use simplestream to obtain images now instead old method.

The `lxc exec` expects `--` before the command. It seems to work even
without it but it behaves weirdly in cases when application relies on
its name (that is argv[0] and example of such application is Busybox).

This allows you to specify name for container. This name is used only in
logs and if not specified the image name is used. This should
primarily be used when you spawn multiple containers from same image
side by side. Then specifying name allows easy recognition of different
containers in the logs.

The '-' is invalid so this replaces it with 'x'. For consistency this
also changes it in such a way that it is always appended so the hidden
issue like this won't happen again. It also makes the naming scheme
consistent.

This tests txt_write, txt_read, bin_write and bin_read Shell methods.

There were also few issues with it that were discovered during the
testing. For example calling prompt twice causes match of new prompt.
The output of txt_read was incosistent with text prepresentation in
Python.

The new feature here is ability of txt_write to append instead of
overwrite.

This was causing issue when tests from multiple packages were run in one
session as serial console is not correctly closed on cleanup of board
fixture.

These should be session anyway as they do not cause any conflict with
anything else (compare to boot_serial that boots and reboots the
router). There is no reason why we would want the new instances of these
objects, rather the opposite.

replaced nonexisting debug mark
the test is now skipped due to some incompatibility
that will be investigated.