This test could use a little updating. We can avoid checking the number of indexes after a failed command, which may be affected by the current replication state, with a check on the error code in the failed command result.

This should enable the test to run more smoothly in passthrough test suites.