I was relying on the newer Juju behavior (or local provider behavior?), where machines are destroyed by default when the service is destroyed. I'll put back the machine harvesting code I thought might not be needed.
test_basic_admin, test_roles_granted and test_upgrade_charm usually pass, and I suspect their failures was due to the provisioning issues. The more problematic tests are the failover ones, as they sometimes trigger a race condition I need to work out how to fix (if you failover too often too quickly, you might promote a unit to master that has not reached a recoverable state and things die).
bundletester is irrelevant here, as it is just the launcher. As far as bundletester is concerned there is just one huge test.
I've upgraded the Makefile rules to install the test dependencies the first time they are needed.
I was relying on the newer Juju behavior (or local provider behavior?), where machines are destroyed by default when the service is destroyed. I'll put back the machine harvesting code I thought might not be needed.
test_basic_admin, test_roles_granted and test_upgrade_charm usually pass, and I suspect their failures was due to the provisioning issues. The more problematic tests are the failover ones, as they sometimes trigger a race condition I need to work out how to fix (if you failover too often too quickly, you might promote a unit to master that has not reached a recoverable state and things die).
bundletester is irrelevant here, as it is just the launcher. As far as bundletester is concerned there is just one huge test.
I've upgraded the Makefile rules to install the test dependencies the first time they are needed.