first stop the slave, then run commands on the master that are
supposed to fail on the slave, then start the slave.

if you swap first two steps, the slave might get and execute those
commands before it's stopped, which will fail the test.

also, improve debugability