Klever

At the moment when decision of some sub-job fails decisions of other sub-jobs are terminated using standard means of Python multiprocessing. Although this likely works well it leads to a job status Corrupted since it doesn't deal carefully with reports uploading. Switching to Klever extensions to multiprocessing should help.

In addition I suggest to improve Klever extensions to Python multiprocessing so that they will be more robust, in particular there won't be any hang ups any more.