Split type byrun doesn't check success of creating dataset.
When using split type byrun, poms creates a new derived dataset with an additional run_number constraint. POMS apparently doesn't check that the derived dataset was created successfully before submitting jobs. It needs to be taught to check the success of dataset creation and to not submit jobs if dataset creation was not successful.
#2 Updated by Marc Mengel about 2 years ago
Well, actually, we have workflows that rely on that behavior -- when we have a generating stage (i.e. a stage with no input dataset) that declares the dataset rather than POMS doing it... We have to figure out why the declaration fails and react accordingly.
Also, the launch scripts should fail if the dataset is emtpy anyhow...