Make ConfigChangeObserverTest poll for change

[pont-us]

Describe the bug
A clear and concise description of what the bug is.

ConfigChangeObserverTest.test_config_change_observer points a ConfigChangeObserver at a configuration, changes the configuration, then immediately checks that ConfigChangeObserver noticed the change. There isn't always time for this to happen. Sometimes the check is done before ConfigChangeObserver has had time to reach its next polling interval and load the new configuration from the file. This results in a spurious test failure.

To Reproduce
Steps to reproduce the behaviour:

Run the ConfigChangeObserverTest.test_config_change_observer repeatedly. Occasionally it fails. (Frequency of failures dependent on system configuration and current load.)

Expected behaviour
The test should pass every time.

Additional context
The simple but inefficient way to fix this is to introduce a fixed delay the the test between updating the configuration file and checking that the change has been registered. However, I have discovered that occasionally (rarely) this can take up to two seconds, and it would be inefficient to slow down every test run by two seconds for these rare cases. The test should poll for the change so it doesn't waste any time, with a fall-back time-out of two seconds or so, to catch genuine failures where ConfigChangeListener really doesn't register it.

[forman]

I don't like time.sleep solutions as they

• don't make the (test) code more deterministic and
• slow down test runs.

Consider mocking or testing at a different call level.