The ImpalaShell class expects to start impala-shell and interact with it
by sending instructions over stdin and reading the results. This
assumption was incorrect when used for impala-shell batch sessions,
where the process exits on its own. If there's a delay in
ImpalaShell.__init__ - between starting the process and polling to see
that it's running - for a batch process, ImpalaShell will fail the
assertion that process_status is None. This can be easily reproduced by
adding a small (0.1s) sleep after starting the new process.
Most batch runs of impala-shell happen through `run_impala_shell_cmd`.
Updated that function to only wait for a successful connection when
stdin input is supplied. Otherwise the command is assumed to be a batch
function and any failures will be detected during `get_result`. Removed
explicit use of `wait_until_connected` as redundant.
Fixed cases in test_config_file that previously ignored WARNING before
the connection string because they did not specify
`wait_until_connected`.
Tested by running shell/test_shell_commandline.py with a 0.1s delay
before ImpalaShell polls.
Change-Id: I24e029b6192a17773760cb44fd7a4f87b71c0aae
Reviewed-on: http://gerrit.cloudera.org:8080/21598
Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
Reviewed-by: Jason Fehr <jfehr@cloudera.com>
Reviewed-by: Kurt Deschler <kdeschle@cloudera.com>
e1098a6a02