GoogleServiceTestSetUp test isn't robust against interrupted test run

[Francis J. Lacoste]

Symptoms
========
 GoogleServiceTestSetup (canonical.launchpad.testing.tests.googleserviceharness)

Failure in test GoogleServiceTestSetup (canonical.launchpad.testing.tests.googleserviceharness)
Failed doctest test for canonical.launchpad.testing.tests.googleserviceharness.GoogleServiceTestSetup
  File "/home/buildbot/slaves/launchpad/full/build/lib/canonical/launchpad/testing/tests/googleserviceharness.py", line 18, in GoogleServiceTestSetup

----------------------------------------------------------------------
File "/home/buildbot/slaves/launchpad/full/build/lib/canonical/launchpad/testing/tests/googleserviceharness.py", line 25, in canonical.launchpad.testing.tests.googleserviceharness.GoogleServiceTestSetup
Failed example:
    assert not service_is_available() # Sanity check.
Exception raised:
    Traceback (most recent call last):
      File "/home/buildbot/slaves/launchpad/full/build/utilities/../lib/zope/testing/doctest.py", line 1360, in __run
        compileflags, 1) in test.globs
      File "<doctest canonical.launchpad.testing.tests.googleserviceharness.GoogleServiceTestSetup[2]>", line 1, in ?
        assert not service_is_available() # Sanity check.
    AssertionError

Diagnosis
=========

GoogleServiceTestSetup doesn't recover gracefully from an interrupted test run. The old process remains running and we are using a predictable port for it.

Possible fixes
==============

* Use a dynamic port (a stale service would waste resources but not break test runs)

* Kill off the old service more reliably even when a test run is interrupted

[Robert Collins (lifeless)]

The behaviour enforced by the asserts is appropriate. However, an interrupted test run shouldn't leave the ephemeral service running, and we should be using ephemeral ports etc too. So I'm going to tweak the description.