Currently there is a race condition at the end of shutdown_primary.js. In a parallel shell, the primary node is issued a shutdown command, which causes the shell to exit with a network error during the shutdown process. Sometimes the error is thrown early enough that the primary can still be connected to, causing the test to fail.

To account for this uncertainty, the test should retry connecting to the shutting down primary until it throws an error as expected (or times out, in which case the test would fail).