Currently, if a controller fails or is taken down while it is the controller node for a job, the job will show up as failed and the job will not get any further output saved.  It would be beneficial if the execution node tried to connect to another available controller so that the output and job status were not lost.