Currently, this tool is riddled with eclass strings (a specific instance of moodle in uofa), get rid of it and replace it with moodle. After all, this is meant to be agnostic of what moodle instance you are running.

This is helpful, specially in ANT and Jenkins where you can afford such flexibility without sacrificing simplicity of usage.

Python 2.7 is dumb this way.

This was unforeseen. dataroot and phpu_dataroot was setup later in ANT. This should be set appropriately.

Placing data in moodle-parallel-phpunit is stupid. The timing data get's erased, which slows down the tests again, causing massive embarrassment when demoing.

This is more appropriate to what the bash file is doing.

In cases where you have things like
docker exec ${container} service postgresql status
it would be good to use "${container}", in case the variable contains a space.

http://mywiki.wooledge.org/BashPitfalls#for_arg_in_.24.2A
http://unix.stackexchange.com/questions/131766/why-does-my-shell-script-choke-on-whitespace-or-other-special-characters

Currently, -u lmsadmin is appended to docker execution. Remove this.

Currently, code is intermingled with data related files/folder, e.g. .config phpu_moodledatas, etc. Move them all to data directory.

To implement the best performance wrt to scheduling: Sort phpunit by execution time and execute each phpunit in that order. That guarantees best time.

This will allow cases (deployment) to persist the config parameter instead of setting it every git pull.

Currently logging is python's or bash print or echo statement. Actually look for a logging library and do it that way.

Currently, pg_restore -j 16 is hard coded. Make 16 be $core or something.

Some files are hard coded, specially in phpunit_container.py and and all the bash files. Unless there is a concrete argument that they will never change, make them a parameter.

Currently, this program don't have any indication if there was any error in the test or not. This is useless for jenkins.

config.php is not constant between moodle instances.

The following is generic config.php for postgres instance.

that is that is

Some paths are hard coded. Make them all relative to the current directory.

Currently, only full phpunit tests are handled. This can be easily amended to handle subsets.

After some investigation, there is a thread lock in jenkins. For some reason, this lock is not being experience in my pc. But whatever the reason, the following is a semantic error.

def _pop_testsuite(self):
        self._testsuites_lock.acquire(True)

        testsuite = None
        if len(self.testsuites) > 0:
            testsuite = self.testsuites.pop()
            self._testsuites_lock.release()

        return testsuite

As you can see, lock is never released when len(self.testsuites) is 0 or empty, thus the lock always occur at the very end.

I've made a mistake of creating a default "execution time file" during setup. This destroys the past "execution time file". This must be amended by checking if an "execution time file" already exist, if so, don't touch it.

This is a different kind of starvation. Not halting the whole operation, but slowing it down. Only one core is doing the job, specifically in a two core Jenkins environment. I can't keep track if this is going on in an 8 core machine (I feel like I don't see some numbers showing up in the log, thus might be an evidence of starvation on some threads).

Currently, this number is printed after the test is done. Since this is test_assigned/test_total, last tests can have total_test/total_test, since at this point all tests are assinged. Either, way this is confusing. Two solutions:

Change this to test_completed/test_total or move this prior to starting test (or when test is assigned).

Now that docker containers are encapsulated in phpunit_container classes, setup.py is now much cleaner and straight forward. test.py is still rampant with direct bash execution and implementing test/tests method in the phpunit_container_abstract should clean test.py