Currently impala-shell always reads a config file ( ~/.impalarc by default), even if it is run during an automatic test. This means that the settings in .impalarc can have side effects during tests - for example if "impalad" is set to remote cluster, the tests will run on that instead of the mini cluster.
I see two ways to fix this issue:
1. always set --config_file when running impala-shell during tests to point to somewhere else than ~/.impalarc
2. use an environment variable to affect the config file loading - for example IMPALARC_LOCATION could be set in every test
I think that 2 is somewhat better, because it could be done centrally (for example in impala_test_suite.py or impala-py.test), while 1 could not be enforced so easily.
A bonus feature for 2 could be to add the possibility of loading more than one config files, for example by setting IMPALARC_LOCATION to a comma separated list of file locations.