Jobs in the middle of the graph can become leaves (wr scheduler only)

[mgcam]

Example: run which got to qc complete even though there was a buried job 40759-archive_to_irods_ml_warehouse[2#888]. The job failed at 15:32:26 and the most recent status file is-rw-rw-r-- 1 gspipe gsanalysis 132 Sep 8 15:41 qc-complete.json

This was noted only for Heron runs. The pp data to iRODS are done for a subset of entities. So the PhiX job does not have a dependent job.

In general: passing through Nth element from previous node to next node when Nth doesn't exist in current node

[mgcam]

Immediate mitigation: move pp irods archival in parallel graph path to other irods jobs