DTest* TestTopology.test_movement* is flaky. All of the testing so far (and thus all of the current known observed failures) have been when running against trunk. When the test fails, it always due to the assert_almost_equal assert.
The following CircleCI runs are 2 examples with dtests runs that failed due to this test failing it's assert:
p.s. assert_almost_equal has a comment "@params error Optional margin of error. Default 0.16". I don't see any obvious notes for why the default is this magical 16% number. It looks like it was committed as part of a big bulk commit by Sean McCarthy (who I can't find on JIRA). If anyone has any history on the magic 16% allowed delta please share!