I'm using GNU Parallel version 20150810 when both --joblog and --dryrun together. This has the side effect of recording the dryrun as "successful" in the joblog. Therefore when removing dryrun and rerunning the job, the jobs do not actually run since the dryrun simulations are listed in the log as having been successfully ran! As a workaround one has to remove the appropriate lines from the joblog. I wonder it might make sense to avoid recording the --dryrun in the joblog altogether, or, if that information is important, another option may be to use some dryrun
simulation marker (as a stupid example, a special value in the exit code field) so that the real run ignores it.