Currently, the esmvaltool stops running when the output directory already exists. A common scenario where this happens is if you're running multiple recipes with the same name in parallel, e.g. on a SLURM cluster. This is inconvenient and it would be nice if the tool was able to come up with a good directory name by itself.