The reason is that we are using a sleep to wait for slave to reach the
slave_transaction_retries limit.
Fix: wait for the slave to stop instead. This is what we want to do, since
the slave stops when the limit is reached.