Comment 1 for bug 1154476

I think it is (partially) the same underlying problem as in bug #1135441. More specifically, even if innobackupex killed the child xtrabackup process when it encountered a fatal error, it did not wait for that process to actually terminate. Which means the process could still be running after innobackupex termination. So a file created by xtrabackup could interfere with 'rm' executed by the test.