Make the integration tests non-flaky. We get some failures currently due to upstream Juju issues, but the test where we log a bunch in an action and then fail the action is unpredictable -- regularly, not all the lines are captured by Juju before the failure propagates to our test and the model is subsequently cleaned up. We could make the test a lot more lax (just assert we get some lines, not 1000), or add a wait in our test before the model is cleaned up (e.g. wrap the juju fixture). Or maybe log a bunch cleanly without a failure during the action, then fail the test afterwards.
Make the integration tests non-flaky. We get some failures currently due to upstream Juju issues, but the test where we log a bunch in an action and then fail the action is unpredictable -- regularly, not all the lines are captured by Juju before the failure propagates to our test and the model is subsequently cleaned up. We could make the test a lot more lax (just assert we get some lines, not 1000), or add a wait in our test before the model is cleaned up (e.g. wrap the
jujufixture). Or maybe log a bunch cleanly without a failure during the action, then fail the test afterwards.