SW-828: processed review comment: Made a strict seperation in goals/responsibilities. The PostgresTestDatabaseInstance gives you a running isolated postgres server, and the PostgresTestMixin uses such an instance automagically while unittesting.

SW-828: abstracted RADBCommonTestMixin into even more common PostgresTestMixin. adapted t_postgres.py and radb_common_testing.py to new PostgresTestMixin.

SW-826: only insert predecessor/successor links into the radb when needed, thus preventing unneeded error log lines about duplicate keys

SW-826: fixed the taskprescheduler by letting it respond to RA.TaskApproved events and by NOT inserting/updating the task into the RADB.

SW-826: the taskprescheduler should NOT insert the task into the RADB. Added that requirement to the test, which now fails. Next commit has the fix for this bug.

SW-818: fixed t_trigger_service.py and made sure that Popen is mocked away so the test does not make acual calls to the tbbctl.

SW-816: TaskPrescheduler now supplies the Specification class with an RADatabase instance instead of an RADBRPC

SW-816: reused radb connections. Testing showed that opening/closing db connection too quickly in succession can lead to timeouts while waiting for sockets in TCP_WAIT state.

SW-816: implemented truncate trigger which cascades a truncate on resource_claim to resource_usage and resource_usage_delta

SW-816: improved speed of the RADBCommonTestMixin by only creating a postgres instance upon setUpClass, and applying a fresh radb schema every test via setUp

SW-816: bonus feature: refactored RADBCommonTest into a RADBCommonTestMixin which takes care of setting up a testing postgres radb instance, and which can be used in conjunction with a unittest.TestCase to do unittests on a fresh RADB instance. By using a mixin instead of chained inheritance we have better seperation of responsibility, and tools like pycharm can handle these tests better as well.

SW-816: modified radb insertSpecificationAndTask into insertOrUpdateSpecificationAndTask. This has the advantage that we do not have to reconnect predecessor relations, and in the webscheduler a re-insert of a task will now be visible as an update on an existing task, instead of a delete (disappear) followed by an insert (reappear). Catch task_predecessor_unique PostgresDBQueryExecutionError in the Specification class of the resourceassigner, so scheduling can continue even on tasks which were already known and already had links to predecessors.

SW-816: added test test_do_assignment_does_not_raise_on_inserting_predecessors which reproduces the bug of SW-816. It fails with a 'duplicate key value violates unique constraint task_predecessor_unique' error from the RABD when scheduling a successor task for the second time.

SW-816: rewrote t_resourceassigner to use a testing-RADB, thereby removing many wrong assumptions made in the mocks. The resource_assgner and its schedulers can still be simplified significantly if we just use the RADB as designed; just insert the claims, and let the RADB check/set the conlict status.