v2.3.7

Bug Fixes

Jobs that could not be reset are now unstuck (FSC-4035)

A small number of jobs had got into a state where resetting them failed every time, and kept failing on every automatic retry afterwards. This happened when one of the job's passengers (a piece of collaborative demand) was removed by the demand engine while the job was mid-flight. The job's saved snapshot still pointed at that now-deleted passenger, so replaying the snapshot during a reset hit a dead reference and aborted the whole reset.

Resets now cope with this. Before replaying a job's saved state, we check which of the referenced records still exist and quietly skip any that have since been deleted, so the reset converges on the live, consistent state instead of failing outright.