This is a workaround for the functional suite running
slow, until there's a more proper solution.

Reduce the polling time from 1 second to 0.01 second,
so short lived tests end quicker.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Try as much as possible to turn timestamps into the
ISO format. In fact, iso requires hours and minutes
to be separated by a ':', but that makes URLs far
less convenient, requiring escape characters. So
let's replace it with a dot.

Also, put the short ID after the date, so date sort
can be properly done.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Make some changes to bring the concept of job ID as a
user visible information:

* Make the human output to display the job ID
* Make the json output to have the job ID
* Record the job id in a file inside the results dir

Adapt functional tests to that new schema.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Another change to clarify the purpose of each directory.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Instead of using 'debug log' we can use 'job log', as it
is a human readable log of the job execution.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

We used 'run' as a prefix for avocado job output directories
after virt-test (basically, inertia). Let's use something
better:

job-[first 8 characters of job id]-timestamp

This way the job id starts to be more prominent in the
human interface as well.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

We can aim for a more terse output than what we have
today. This is my proposal to simplify things a bit,
while still being clear.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Run avocado in a long test, interrupt it and verify if
the runner displays the appropriate behavior, in terms
of exit code and output.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

If we press Ctrl+C, the test that was being executed
still does not have a status, and that reflects in the
job log with a None printed in the final status.

So, if there's no status yet, we should safely (?)
assume that the test was interrupted.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Handle KeyboardInterrupt inside avocado, making
Ctrl+C to terminate the program, instead of generating
a python backtrace.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Add a special exception for when tests are interrupted
by Ctrl + C, and the required mapping in status.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Add send_signal(), that lets us to send arbitrary signals
to subprocesses, and rewrite kill() and terminate() in
terms of it.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

avocado: Bump to version 0.10.1

Bugfix release.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

avocado: Fix renamed attribute from avocado.core.output

The reference was incorrect, causing some of the plugins
to crash.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Version 0.10.0

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

The tests dir layout has changed, update setup accordingly.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Test feedback 2

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

This calls os.abort() as the test action, testing avocado
ability of handle this case correctly.
Signed-off-by: NRuda Moura <rmoura@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Previously, the throbber update method was being called
unconditionally, making throbber progress to appear even
when human output was not enabled. Fix this by making
a proxy method that calls it only if there is a human
plugin enabled, in TestResultProxy.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Put state of the initialized test in the multiprocess Queue,
so we have a lot of accurate information that does not need
to be mocked, we only have to mock the failure reason and
time elapsed. With this, aborted tests can be handled more
gracefully.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Introduce a TestAbortError, used for when an avocado test
aborts.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

If the test process is running, the trobber will give feedback about that
to the user.

This requires the process triggering and timeout handling to be completely
different than the previous implementation.
Signed-off-by: NCleber Rosa <crosa@redhat.com>

Signed-off-by: NCleber Rosa <crosa@redhat.com>

It's now a known fact that long test runs are can lead to doubts with
regards to the actual test run status (is it frozen or really busy?).

To give users a better experience, the OutputManager now implements
a throbber for progress feedback.
Signed-off-by: NCleber Rosa <crosa@redhat.com>

In some circurnstances, the journal may be accessed simultaneously by
multiple code paths. The database consistency should be enforced in any
case, and only initialized and populated when appropriate.
Signed-off-by: NCleber Rosa <crosa@redhat.com>

Since other more standard plugins have moved to test state in a previous commit
the journal plugin should also follow that.
Signed-off-by: NCleber Rosa <crosa@redhat.com>

Signed-off-by: NCleber Rosa <crosa@redhat.com>

This commit implements two very different things, but still very
much connected:

1) Separate test loading to the process that will be actually
running the test. This leads to a better separation of duties of
the process interacting with the user and the process that is
actually doing the (usually) heavy work.

2) Remove the pickling of the test instance and sending it back
and forth.

As an added bonus, it was noted during development, that both
processes load all necessary Python modules (including the test
module) when unmarshalling the pickled instance, resulting in
unecessary work. Actual data on performance improvements were not
collected though.

It has one change in behaviour (regression) though. Test timeout
now comes only from the test parameters. The reasoning is that
since there's now a better separation of roles and the fact that
the first process is not loading the test instance, it can not
look at the timeout defined there.

IMHO this is a sign of good design, since other types of test
(think of shell scripts) can not communicate their intended timeout.
Maybe a 'test.cfg' or something like that should be the solution to
that.
Signed-off-by: NCleber Rosa <crosa@redhat.com>

docs.source.WritingTests: Introduce environment variables.

Add new subsection about environment variables that
Avocado exports to the tests.

Plus fix some typos.
Signed-off-by: NRuda Moura <rmoura@redhat.com>