This reduces user's overhead to launch tests. It uses targets filtering.
This makes target specification optional.
At the moment user can specify that he wants to run tests on specific
board. In the future we should add filters for additional configured
board features, such as presence of Wi-Fi cards or USB storage device.

This not only moves target config parsing to nsfarm library but also it
changes it to be implementation defined not just access to configparser.

This should suppress mod of the day and other messages just to make logs
little bit less duplicate.

The current tests check if appropriate profiles exists but not if they
actually work. The nsfarm-root has to work when container at least
starts. The problem is with nsfarm-internet. It is required to test if
it is possible to access the Internet or not with that profile. That is
exactly what this test does.

This should have been is_symlink as it checks for symlinks.

There was for some reason condition that only directories are listed
when image hash is generated. In reality the code was prepared to handle
any path and should include all paths. This removes this invalid
condition and makes it so all files provided for image are included in
hash calculation.

This in effect makes it so new image build is triggered on any file
modification.

The Python PEP8 specifies that exceptions should be named *Error so this
renames the exceptions to match that.

This is just test that checks that we have all required devices on host
system. These devices are optional and additional modules are requires
to be installed and loaded to kernel.

At the moment only the /dev/ppp is tested there.

There is change that adds device exclusivity. Only for few devices we
need exclusive passtrough to container. In most cases it is enough to
pass only access to that device.

Another change is that there is possibility to specify parameter for
images. At the moment this allows only passtrough of Unix character
devices but that can be expanded in the future.

The arguments of Container class changed but they were  not updated
here.

The micro terminal is minimal connection between terminal and socket.
This way it can be hooked up to any socket in nsfarm and thus allowing
the direct access for CLI interfaces. The primary use is to access
board's serial console.

This replaces serial_miniterm with simpler and more generic implementation.

It turns out that nonblocking doesn't mean nonblocking in all cases. In
this case when there was no input the read was blocking.
The timeout set to zero solves this.

This now not only makes sense becuase we have live logging it also fixes
TODO pointed there, we just wait for file to appear (syslog-ng to
start).

Karel Koci authored 4 years ago and Karel Koci committed 4 years ago

Problem with logging trough pexpect is that we log output only when we
expect it. That is when we access it trough pexpect. This means that
time and synchronization when logs arrive from multiple sources can't be
archived and only one source logs at the time (the one that is used
currently by pexpect). This is not ideal as we are interested in
situation such as restarting network interfaces that produce some logs
not only on router but also in container at the same time.
Synchronization on time bases in such cases is important and we can't
wait for code to call pexpect to print it to logs later on.

This instead utilizes sockets and by reading source immediately and just
passing it to socket.

Upstream seem to have fixed invalid init script in iperf3-openrc package
so now this hack is no longer required.

Yeah it provides dict with fields where we are interested in name. The
rest of the function does that correctly but one line does not.

This replaces code that reads whole file and then updates content with
function that reads file in limited chunks.

This is just initial minimal set of tests. Nothing big.

This now separates image and container representation to two separate
classes. For users it is almost the same but it makes structure of lxd
module much better and improves overal readability of Container class.

It was pretty nasty how original LXD connection was designed. It is ok
to just wrap that to class and pass it around instead of calling
connection from various locations just to be sure that all is
initialized before we use it.

This is arguably better than raising NotImplementedError. It also
catches inconsistencies much sooner.

This is now only initial implementation. There should be also
specification of board and more but that is going to be added later on
once we are able to run tests on more than one board.

It is planned to have not only tests but also selftests and for that we
do not need configuration. Those arguments are solely for real tests.
Moving them to tests makes them required to only ten those tests.

First network test in set was in some probability failing. This just
adds dummy sleep to stabilize network before tests are executed.

This also incorporates feedback from ongoing review.

This reuses WAN tests for LAN as well to check that clients are able to
access internet trough router.

Watchdog is not running immeediatelly after boot as it seems and also it
is not essential for use at the moment to check for it.

Some of the non-essential fixtures should be scoped not for session but
for module only or even less because they can collide with other tests.

These are simple tests to check system that was just booted.

Static method can't be used this way. The only other clean alternative
is just changing it from method to function.

Upper case are constants and we do not consider these as constants so we
should not name them that way.