See calibration/planning#34. Trying to carry on executing a notebook after an uncaught exception rarely works, because the later code often depends on things which didn't happen (variables not defined, files not written, etc.). This stops notebook execution on uncaught errors, as is the norm for Python (and, I think, most languages with exceptions). This should make for clearer output when things do go wrong (because there aren't a load of meaningless extra errors after the real problem).
Slurm still shouldn't treat the jobs as failing, because bash does essentially the same thing by default: unless you use set -e, it will ignore failing commands and carry on to the next one. So reports should still be generated and files moved as expected. I plan to address that in a later step.
Running xfel-calibrate at the command line. Output is in /gpfs/exfel/data/scratch/kluyvert/agipd-calib-900201-203-errstop.